Single cell sequencing

Abstract

The invention provides methods of performing single cell RNA sequencing, methods of designing or providing a composition for performing single cell RNA sequencing, methods of analysing single cell RNA sequencing data, and related systems and products The single cell RNA sequencing method comprises obtaining a sequencing library and comprises using a plurality of target-specific capture reagents associated with a respective plurality of target transcripts, to enrich the sequencing library for cDNA molecules corresponding to the plurality of target transcripts, wherein the concentration of a first target-specific capture reagent for a first target transcript is different from the concentration of a second target-specific capture reagent for a second target transcript.

Description

[0001]SINGLE CELL SEQUENCING This application claims priority from GB 2312683.2 filed 18 August 2023, the contents and elements of which are herein incorporated by reference for all purposes. FIELD OF DISCLOSURE The present invention relates to a method for performing single cell sequencing, to a method for providing or designing a composition for single cell sequencing and to related products. It is particularly, but not exclusively, concerned with single cell sequencing protocols and compositions where the concentrations of one or more targeting reagents for a first sequencing target are different from the concentrations of the one or more targeting reagents for another sequencing target. BACKGROUND Single-cell transcriptome analysis has become a very important tool in modern biology, due to its ability to provide a lens into cell-to-cell variability. A notable feature of scRNA-Seq data is the sparsity of the datasets, which is often caused by the relatively shallow sequencing of the transcriptome (Hagemann-Jensen et al., 2022). This can lead to uncertainty about whether a zero count represents the true absence of gene expression or simply low expression. Computational and experimental methods have been proposed to address this issue. For example, imputation approaches using machine learning such as neural networks, have been used to infer missing count data (Luecken and Theis, 2019). Such methods can introduce bias and spurious correlations that are not based on biology, as no causal reasoning is involved in the imputation process. Experimental methods proposed include SMART-Seq (Hagemann-Jensen et al., 2022; Replogle et al., 2020), a plate-based single-cell sequencing method that allows for deeper transcriptome sequencing but at the cost of lower throughput, and TAP-Seq (Schraivogel et al., 2020) a droplet- based method that uses nested PCR cycles to extract predetermined genes of interest by controlling the primers used in the PCR cycles. This allows for increased sequencing of targeted genes but at the cost of providing a very limited picture of the transcriptome (restricted to panels of a few hundred genes at most). Therefore, there is still a need for improved methods for performing single cell sequencing. SUMMARY OF THE DISCLOSURE The present inventors postulated that some of the problems associated with RNA sequencing, and single-cell sequencing data in particular, could be alleviated by modifying conventional sequencing protocols to enrich low expressed genes through target specific capture steps with capture reagent concentrations specifically set for this purpose (a process which they call “weighting”). They show that the weighted datasets contain more information than the current state of the art (non-biased single cell experiments), while preserving the statistical characteristics of observed counts and without degrading downstream analysis performance. Thus, according to a first aspect, there is provided a method of performing single cell RNA sequencing of a sample comprising: reverse-transcribing one or more RNA molecules present in the sample to obtain one or more corresponding cDNA molecules; and amplifying the cDNA molecules to obtain a sequencing library; wherein the method comprises one or more steps that use a plurality of target-specific capture reagents associated with a respective plurality of target transcripts, to enrich the sequencing library for cDNA molecules corresponding to the plurality of target transcripts, wherein the concentration of a first target-specific capture reagent for a first target transcript is different from the concentration of a second target-specific capture reagent for a second target transcript. The method may have any one or more of the following optional features. The plurality of target-specific capture reagents may comprise at least 2, at least 3, at least 4, at least 5, at least 6, at least 7, at least 8, at least 9 at least 10, at least 15, at least 20 or more different sets of target-specific capture reagents, wherein the target-specific capture reagents in a set have the same concentration as each other and a different concentration from that of the target-specific capture reagents in any other set. Thus, the plurality of target-specific capture reagents may comprises a plurality of target-specific capture reagents at different concentrations. The method may further comprise sequencing the sequencing library, thereby obtaining a plurality of reads each corresponding to a transcript of the plurality of target transcripts. The method may comprise capturing the one or more RNA molecules using target-specific reagents. The cDNA amplification may be performed using target-specific capture reagents. The method may comprise a capture step that uses target-specific capture reagents prior to cDNA amplification. The method may comprise a capture step that uses target-specific capture reagents prior to reverse transcription. In embodiments, the method comprises a capture step that uses target-specific capture reagents prior to or during reverse transcription. Performing target-specific enrichment prior to or during reverse transcription may be particularly beneficial as it may increase the chances of target transcripts being represented in the composition that is subject to amplification. Multiple capture steps may be performed in the same protocol. For example, a capture step may be performed during or prior to reverse transcription, and a further capture step may be performed during cDNA amplification. A capture step may comprise multiple substeps. For example, performing the cDNA amplification using target-specific capture reagents may comprise performing a plurality of PCR reactions each using respective target-specific capture reagents. For example, the cDNA amplification may use two semi-nested PCR steps, each using a respective target-specific capture reagent (e.g. a target specific outer primer and a target specific inner primer). A target-specific capture reagent may comprise: a primer, a pair of primers, a probe or a pair of probes. The concentration of the first and second target-specific capture reagents may be dependent on the expected expression level of the first and second target transcripts in the sample. The method may comprise selecting the concentration of the first and second target-specific capture reagents based on the expected expression level of the first and second target transcripts in the sample, or selecting a composition comprising said target-specific capture reagents at said concentrations. The concentration of a target-specific capture reagent for a subset or all of the plurality of target transcripts may be dependent on the expected expression level of respective transcripts in the sample. The method may comprise selecting the concentration of a subset or all of the target- specific capture reagents based on the expected expression level of each of the transcripts in the sample, or selecting a composition comprising said target-specific capture reagents at said concentrations. The expected expression level of a transcript in the sample may be based on the expression level of the transcript in a reference sample or set of samples. The expected expression level of a transcript in the sample may be the expression level of the transcript in a reference sample or set of samples. A reference sample may be a sample from the same organism as the organism from which the sample to be analysed originates. A reference sample may be a sample comprising cells from the same organ as an organ from which cells in the sample to be analysed originate, and / or a sample comprising cells from the same tissue as a tissue from which cells in the sample to be analysed originate. A reference sample may be a sample comprising cells from the same cell type as cells in the sample to be analysed. A reference sample may be a sample comprising cells from the same cell line as cells in the sample to be analysed. A reference sample may be a sample comprising cells from a cell line in the same category of cell lines as a cell line in the sample to be analysed, where categories of cell lines are defined by tissue of origin and / or one or more phenotypic characteristics. The expected expression level of the first and second or the plurality of transcripts in the sample may be based on a reference dataset comprising expression levels for the transcripts in one or more reference samples, wherein the expression levels in each reference samples are independently selected from: expression levels from bulk RNA sequencing and expression levels from single cell RNA sequencing. The plurality of target-specific capture reagents may comprise a target-specific capture reagent or a set of target-specific capture reagents for each of the plurality of target transcripts, wherein the concentration of each of the target-specific capture reagents in a set of target-specific capture reagents for a transcript is the same. A set of target-specific capture reagents may comprise a plurality of different target-specific capture reagents for the same target transcript. The relative concentrations of the target-specific capture reagents for a subset or all of the plurality of target transcripts may be dependent on the expected relative expression levels of respective transcripts in the sample. The method may comprise selecting the relative concentrations of a subset or all of the target-specific capture reagents based on the expected relative expression level of each of the transcripts in the sample, or selecting a composition comprising said target-specific capture reagents at said concentrations. The expected expression level of the first transcript in the sample may be higher than the expression level of the second transcript in the sample, and the concentration of the first target-specific capture reagent may be lower than the concentration of the second target-specific capture reagent. The concentration of each target-specific capture reagent may depend on the expected expression level of the target transcript of the target-specific capture reagent relative to the expected expression level of all other target transcripts. The concentration of a target-specific capture reagent that is specific for a target transcript that has a higher expected expression level relative to the expected expression level of another target transcript may be lower than the concentration of a target-specific capture reagent that is specific for the other target transcript that has a lower expected expression level. The concentrations of the first target-specific capture reagent and the second target-specific capture reagent may be such that the number of molecules corresponding to the first and second transcripts in the sequencing library are more similar to each other than the numbers of molecules that would be obtained in a sequencing library produced using equal concentrations of the first and second target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent such that the numbers of molecules corresponding to the first and second transcripts in the sequencing library are more similar to each other than the number of molecules that would be obtained in a sequencing library produced using equal concentrations of the first and second target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent such that the numbers of molecules corresponding to the first and second transcripts in the sequencing library are expected to be the same. The concentrations of all target-specific capture reagents may be such that the number of molecules corresponding to the respective target transcripts in the sequencing library are more similar to each other than the numbers of molecules that would be obtained in a sequencing library produced using equal concentrations of target-specific capture reagents. The method may comprise identifying concentrations of each target-specific capture reagent such that the numbers of molecules corresponding each of the target transcripts in the sequencing library are more similar to each other than the number of molecules that would be obtained in a sequencing library produced using equal concentrations of the target-specific capture reagents, and / or wherein the method comprises identifying concentrations of each of the target-specific capture reagents such that the numbers of molecules corresponding to each of the target transcripts in the sequencing library are expected to be the same. The plurality of target transcripts may comprise at least 100, at least 200, at least 500, at least 1000, at least 2000, at least 3000, at least 4000 or at least 5000 different transcripts, wherein the plurality of target transcripts comprise transcripts from at least 200, at least 500, at least 1000, at least 2000, at least 3000, at least 4000 or at least 5000 different genes. The plurality of target transcripts may comprise transcripts from at least 20%, at least 30%, at least 40%, at least 50%, at least 60%, at least 70%, or at least 80% of the genes expected to be expressed in the sample. The expected expression level of the first target transcript in the sample may differ from the expected expression level of the second target transcript in the sample by a factor of at least 2, at least 5 or at least 10. The plurality of transcripts may comprise transcripts that have expected expression levels in the sample that differ by a factor of at least 2, at least 5 or at least 10. The problem of sparsity that is present in scRNAseq experiments is believed to depend on both the number of genes that are being probed and the composition of gene expression of the genes sampled. For instance, consider a target panel comprised of one gene with low expression and four genes with high expression. In this scenario, sparsity would be observed in the low-expressed gene due to the larger number of available transcripts for sequencing highly expressed genes. Conversely, a target panel consisting of five low-expression genes would exhibit minimal sparsity since all genes have an equal chance of being sequenced. In other words, the likelihood that all genes will be sampled in a targeted panel assay is expected to vary depending on the alterations in the number of genes and the level of the expressed transcripts within those genes. The present methods improve this situation by biasing the likelihood of sampling a gene based on the expression level of the transcripts targeted. The magnitude of the improvement is related to the magnitude of the problem, and therefore the present methods may be particularly useful when the transcripts that are being targeted comprise transcripts that have very different levels of expression. The concentrations of the first target-specific capture reagent and the second target-specific capture reagent may be such that the sum of the expected percentages of the first and second transcripts in the sample that are not represented in the sequencing library is lower than the sum of the expected percentages of the first and second transcripts in the sample that are not represented in a sequencing library that would be obtained using equal concentrations of the first and second target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent such that the sum of the expected percentages of the first and second transcripts in the sample that are not represented in the sequencing library is lower than the sum of the expected percentages of the first and second transcripts in the sample that are not represented in a sequencing library that would be obtained using equal concentrations of the first and second target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent such that the sum of the expected percentages of the first and second transcripts in the sample that are not represented in the sequencing library is the lowest of the values expected for a plurality of candidate concentrations of the first and second target-specific capture reagents. The concentrations of each of the target- specific capture reagents may be such that the sum of the expected percentages of each of the target transcripts in the sample that are not represented in the sequencing library is lower than the sum of the expected percentages of the target transcripts in the sample that are not represented in a sequencing library that would be obtained using equal concentrations for all target-specific capture reagents. The method may comprise identifying concentrations of each of the target- specific capture reagents such that the sum of the expected percentages of each of the target transcripts in the sample that are not represented in the sequencing library is lower than the sum of the expected percentages of each of the target transcripts in the sample that are not represented in a sequencing library that would be obtained using equal concentrations for all target-specific capture reagents. The method may comprise identifying concentrations of each of the target- specific capture reagents such that the sum of the expected percentages of each of the target transcripts in the sample that are not represented in the sequencing library is the lowest of the values expected for a plurality of candidate sets of concentrations of the target-specific capture reagents. The one or more steps that use a plurality of target-specific capture reagents associated with a respective plurality of target transcripts may comprise: a capture step that precedes the step of reverse-transcribing the one or more RNA molecules present in the sample using a first plurality of target-specific capture reagents; and / or the step of amplifying the cDNA molecules, wherein the step of amplifying the cDNA molecules uses a second plurality of target-specific capture reagents. Thus, the method may comprise a step of that use a plurality of target-specific capture reagents associated with a respective plurality of target transcripts as a capture step that precedes the step of reverse-transcribing the one or more RNA molecules present in the sample. Instead or in addition to this, the method may comprise using a plurality of target-specific capture reagents associated with a respective plurality of target transcripts in the step of amplifying the cDNA molecules. An additional non-specific cDNA amplification step may be used. Thus, the one or more steps that use a plurality of target-specific capture reagents associated with a respective plurality of target transcripts may comprise: a capture step that precedes the step of reverse-transcribing the one or more RNA molecules present in the sample; and the step of amplifying the cDNA molecules, wherein the step of amplifying the cDNA molecules uses a plurality of target-specific capture reagents; and wherein the capture step that precedes the step of reverse-transcribing the one or more RNA molecules present in the sample uses a first plurality of target-specific capture reagents, the step of amplifying the cDNA molecules uses a second plurality of target-specific capture reagents, the concentration of a first target-specific capture reagent in the first plurality of target-specific capture reagents is different from the concentration of a second target-specific capture reagent in the first plurality of target-specific reagent, the concentration of a first target- specific capture reagent in the second plurality of target-specific capture reagents is different from the concentration of a second target-specific capture reagent in the second plurality of target- specific reagent. The targets of the first and second target-specific capture reagents in the first and second plurality of target-specific reagents may be the same or different from each other. In other words, the concentrations of the first and second target-specific capture reagents for any two respective transcripts may differ from each other in the first plurality of target-specific capture reagents, and the concentrations of the first and second target-specific capture reagents for any two respective transcripts may differ from each other in the second plurality of target-specific capture reagents. The one or more steps that use a plurality of target-specific capture reagents associated with a respective plurality of target transcripts may comprise: a capture step that precedes the step of reverse-transcribing the one or more RNA molecules present in the sample; and the step of amplifying the cDNA molecules, wherein the step of amplifying the cDNA molecules uses a plurality of target-specific capture reagents; and the capture step that precedes the step of reverse- transcribing the one or more RNA molecules present in the sample may use a first plurality of target-specific capture reagents, the step of amplifying the cDNA molecules may use a second plurality of target-specific capture reagents, wherein the first plurality of target-specific capture reagents comprises a plurality of target-specific capture reagents at different concentrations, and the second plurality of target-specific capture reagents comprises a plurality of target-specific capture reagents at different concentrations. The concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the first plurality of target-specific reagents may be such that the number of molecules corresponding to the first and second transcripts that proceed to reverse-transcription are more similar to each other than the numbers of molecules that would proceed to reverse- transcription using equal concentrations of the first and second target-specific capture reagents in the first plurality of target-specific capture reagents. The concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the first plurality of target-specific reagents may be such that the unique molecular counts represented in the sequencing library for the first and second transcripts are more similar to each other than the unique molecular counts that would be represented in a sequencing library obtained using equal concentrations of the first and second target-specific capture reagents in the first plurality of target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the first plurality of target-specific capture reagents such that the unique molecular counts for the first and second transcripts represented in the sequencing library are more similar to each other than the unique molecular counts that would be obtained in a sequencing library produced using equal concentrations of the first and second target-specific capture reagents in the first plurality of target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the first plurality of target-specific capture reagents such that the unique molecular counts corresponding to the first and second transcripts represented in the sequencing library are expected to be the same. The concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific reagents may be such that the number of amplified cDNA molecules corresponding to the first and second transcripts that proceed to sequencing are more similar to each other than the numbers of amplified cDNA molecules that would proceed to reverse-transcription using equal concentrations of the first and second target- specific capture reagents in the second plurality of target-specific capture reagents. The concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific reagents may be such that the number of cDNA molecules in the sequencing library corresponding to the first and second transcripts are more similar to each other than the number of cDNA molecules corresponding to the first and second transcripts in a sequencing library obtained using equal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of cDNA molecules corresponding to the first and second transcripts represented in the sequencing library are more similar to each other than the number of cDNA molecules corresponding to the first and second transcripts that would be obtained in a sequencing library produced using equal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of cDNA molecules corresponding to the first and second transcripts in the sequencing library are expected to be the same; and / or wherein the concentrations of the first target-specific capture reagent and the second target- specific capture reagent in the second plurality of target-specific reagents are such that the number of reads corresponding to the first and second transcripts obtained by sequencing the sequencing library are more similar to each other than the number of reads corresponding to the first and second transcripts that would be obtained by sequencing a sequencing library obtained using equal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of reads corresponding to the first and second transcripts obtained by sequencing the sequencing library are more similar to each other than the number of reads that would be obtained by sequencing a sequencing library produced using equal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents. The method may comprise identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of reads corresponding to the first and second transcripts obtained by sequencing the sequencing library are expected to be the same. Also described according to a second aspect is a single cell sequencing library obtained using the method of any embodiment of the first aspect. Also described according to a third aspect is a method of designing or providing a composition for performing RNA sequencing of a sample, the method comprising: Identifying a plurality of target- specific capture reagents associated with a respective plurality of target transcripts; Identifying an expected expression level of a first transcript of the plurality of transcript and a second transcript of the plurality of target transcripts in the sample, and Determining a concentration of a first target- specific capture reagent for the first target transcript and a concentration of a second target-specific capture reagent for the second target transcript using the expected expression level of the first and second transcripts in the sample. The method may comprise determining a concentration of a first target-specific capture reagent for the first target transcript and a concentration of a second target-specific capture reagent for the second target transcript using the expected expression level of the first and second transcripts in the sample comprises using the expected relative expression level of the first and second transcripts. The method may comprise providing one or more compositions comprising the target-specific capture reagents at the determined concentrations. The single cell RNA sequencing may comprise: Reverse-transcribing one or more RNA molecules present in the sample to obtain one or more corresponding cDNA molecules using a composition comprising target-specific capture reagents; and / or Amplifying the cDNA molecules to obtain a sequencing library using a composition comprising target-specific capture reagents; and / or Performing a capture step that uses a composition comprising target-specific capture reagents prior to or during cDNA amplification of reverse transcribed RNA molecules present in the sample and / or prior to reverse-transcription of RNA molecules present in the sample. The single cell RNA sequencing may comprises a plurality of steps using a composition comprising target-specific capture reagents, and wherein the method comprises determining a concentration of a first target- specific capture reagent for the first target transcript and a concentration of a second target-specific capture reagent for the second target transcript using the expected expression level of the first and second transcripts in the sample, separately for each composition comprising target-specific capture reagents. The method may have any of the features described in relation to the first aspect. For example, a target-specific capture reagent may be selected from: a primer or pair of primers and a probe or pair of probes. The method may comprise identifying an expected expression level of each of the plurality of target transcripts in the sample and determining a concentration of each target-specific capture reagent using the expected expression level of each of the plurality of target transcripts in the sample. Identifying an expected expression level of a first transcript of the plurality of transcript and a second transcript of the plurality of target transcripts, or each of the plurality of transcripts, in the sample may comprise obtaining a reference dataset comprising expression levels of the first and second transcripts or each of the plurality of target transcripts in one or more reference samples. A reference sample may be a sample from the same organism as the organism from which the sample to be analysed originates. A reference sample may be a sample comprising cells from the same organ as an organ from which cells in the sample to be analysed originate. A reference sample may be a sample comprising cells from the same tissue as a tissue from which cells in the sample to be analysed originate. A reference sample may be a sample comprising cells from the same cell type as cells in the sample to be analysed. A reference sample may be a sample comprising cells from the same cell line as cells in the sample to be analysed. A reference sample may be a sample comprising cells from a cell line in the same category of cell lines as a cell line in the sample to be analysed, where categories of cell lines are defined by tissue of origin and / or one or more phenotypic characteristics. The reference dataset may comprise expression levels for the transcripts in one or more reference samples, wherein the expression levels in each reference samples are independently selected from: expression levels from bulk RNA sequencing and expression levels from single cell RNA sequencing. The plurality of target-specific capture reagents may comprise a plurality of different target-specific capture reagents for at least one of the respective plurality of target transcripts (optionally in each of a plurality of compositions, when multiple compositions for use in respective capture steps are designed), wherein the concentration of each of the target-specific capture reagents for the same target transcripts is the same (optionally in the respective composition). Obtaining a reference dataset may comprise retrieving expression data for one or more samples from one or more database, receiving expression data for one or more samples from one or more computing devices or user interfaces, or subjecting one or more samples to an expression analysis experiment to determine expression levels for the plurality of transcripts. Determining a concentration of capture reagents for a set of target transcripts may comprise: determining an observed relative frequency of each target transcript in the set of target transcripts using the expected expression levels of the target transcripts in the sample; determining a target relative frequency of each target transcripts; and determining a weight for each target-specific capture reagent using the observed relative frequency and the target relative frequency for the target transcript of the target specific capture reagent. The weight may be the ratio of the target relative frequency and the observed relative frequency for the target, or a normalized version thereof. A concentration of a capture reagent may be the product of the determined weight for the capture reagent and a concentration associated with the expected expression level of the target transcript or a default concentration that is the same for all capture reagents for the set of transcripts. The observed relative frequency of a transcript may be the expected expression level of the transcript divided by the sum of the expected expression levels of all transcripts in the set. The target relative frequency may be the inverse of the number of transcripts in the set. Determining a weight for each target-specific capture reagent may comprise determining the ratio of the target relative frequency and the observed relative frequency for each target and normalizing the resulting weights by dividing each weight by the highest weight of the resulting weights. The weight may be used to reduce the concentration of capture reagents compared to the concentrations associated with the observed relative frequency of each target transcript in the set of target transcripts. The plurality of target transcripts may comprise at least 200, at least 500, at least 1000, at least 2000, at least 3000, at least 4000 or at least 5000 different transcripts. The plurality of target transcripts may comprise transcripts from at least 200, at least 500, at least 1000, at least 2000, at least 3000, at least 4000 or at least 5000 different genes. The plurality of target transcripts may comprise transcripts from at least 20%, at least 30%, at least 40%, at least 50%, at least 60%, at least 70%, or at least 80% of the genes expected to be expressed in the sample. Determining a concentration of capture reagents for a set of target transcripts may comprise: selecting a plurality of sets of candidate relative concentrations of the capture reagents; for each set of candidate concentrations, determining the sum of the expected percentages of each transcript in the set of transcripts that are not represented in a sequencing library obtained using a composition comprising the set of candidate relative concentrations; and selecting a set of candidate concentrations that has the lowest sum of the expected percentages of each transcript in the set of transcripts that are not represented in a sequencing library obtained using a composition comprising the set of candidate relative concentrations amongst the plurality of sets of candidate relative concentrations of the capture reagents. Determining the sum of the expected percentages of each transcript in the set of transcripts that are not represented in a sequencing library obtained using a composition comprising the set of candidate relative concentrations may comprise determining, for each target transcript and candidate relative concentration of the capture reagents for the transcript: the expected molecular count ( ^^^^) for the transcript after reverse transcription as a function of the candidate relative concentration of the capture reagents for the transcripts ( ^^^^), an expected total number of mRNA molecules in each of a plurality of cells ( ^^) and an expected number of mRNA molecules for the transcript in each of a plurality of cells ( ^^^^). The expected molecular count for the transcript may be determined as ^^^^= ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^( ^^, ^^, ^^^^, ^^^^) and the candidate relative concentration of the capture reagents for the transcripts ( ^^^^) may be a candidate concentration in a composition used for reverse-transcription or a capture step prior to reverse transcription. Determining the sum of the expected percentages of each transcript in the set of transcripts that are not represented in a sequencing library obtained using a composition comprising the set of candidate relative concentrations may comprise determining, for each target transcript and candidate relative concentration of the capture reagents for the transcript: the expected number of mRNA molecules for the transcript that are represented in the sequencing library after cDNA amplification ( ^^^^) as a function of the expected molecular count ( ^^^^) for the transcript after reverse transcription, and the candidate relative concentration of the capture reagents for the transcripts ( ^^^^).The number of mRNA molecules for the transcript that are represented in the sequencing library after cDNA amplification ( ^^^^) may be determined as and the candidate relative concentration of the capture reagents for the transcripts ( ^^^^) may be a candidate concentration in a composition used for cDNA amplification. The method may comprise determining a concentration of the plurality of target-specific capture reagents using the expected expression level of the plurality of target transcripts in the sample. The method may comprise identifying a plurality of groups of target transcripts, each group associated with a different range of expression levels and / or a plurality of groups of target-specific capture reagents, each group associated with a different range of determined concentrations, and selecting a common concentration for target-specific capture reagents in each group using the determined concentrations associated with the group. For example, target transcripts may be divided into multiple groups corresponding to quartiles, deciles, or any other subdivision of the expected expression level distribution across transcripts, and a common concentration for all capture reagents in a group of transcripts may be used (such as e.g. the concentration associated with a transcript with mean or median expression level in the group, or the mean or median concentration determined across transcripts in the group). The single cell RNA sequencing may comprise: Reverse-transcribing one or more RNA molecules present in the sample to obtain one or more corresponding cDNA molecules; and Amplifying the cDNA molecules to obtain a sequencing library; and the single cell RNA sequencing comprises a capture step that precedes the step of reverse-transcribing the one or more RNA molecules present in the sample using a first plurality of target-specific capture reagents; and / or the step of amplifying the cDNA molecules uses a second plurality of target-specific capture reagents, wherein the method comprises determining a concentration for each of the first and / or second plurality of target-specific reagents. The method may comprise determining a concentration for each of the first and / or second plurality of target-specific reagents, wherein determining a concentration of a first target-specific capture reagent for the first target transcript and a concentration of a second target-specific capture reagent for the second target transcript using the expected expression level of the first and second transcripts in the sample comprises: identifying a concentration of the first target-specific capture reagent and a concentration of the second target-specific capture reagent in the first plurality of target-specific capture reagents such that the unique molecular counts corresponding to the first and second transcripts represented in the sequencing library are expected to be the same; and / or identifying a concentration of the first target-specific capture reagent and a concentration of the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of cDNA molecules corresponding to the first and second transcripts in the sequencing library are expected to be the same; and / or identifying a concentration of the first target-specific capture reagent and a concentration of the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of reads corresponding to the first and second transcripts obtained by sequencing the sequencing library are expected to be the same. According to a fourth aspect, there is provided a composition for performing single cell RNA sequencing of a sample obtained using the method of any embodiment of the preceding aspect. According to a fifth aspect, there is provided a kit for performing single cell RNA sequencing of a sample comprising the composition of any embodiment of the fourth aspect. The method according to the first aspect may have any of the features described in relation to the third aspect. Thus, a method according to the first aspect may comprise performing the method of any embodiment of the third aspect. The method according to the third aspect may have any of the features described in relation to the first aspect. Thus, a method according to the first aspect may comprise performing the method of any embodiment of the third aspect. The method according to the third aspect may be computer implemented. Indeed, the process determining expected unique molecular counts and / or number of reads in a sequencing library over large panels of targets (e.g. hundreds or more genes) may be far beyond the capability of the human mind. According to a sixth aspect, there is provided a method of analysing single cell RNA sequencing data, the method comprising determining a count of RNA molecules associated with a target transcript in a sample from single cell RNA sequencing data obtained using a method comprising: Reverse-transcribing one or more RNA molecules present in the sample to obtain one or more corresponding cDNA molecules; and Amplifying the cDNA molecules to obtain a sequencing library; wherein the single cell RNA sequencing data method comprises one or more steps that use a plurality of target-specific capture reagents associated with a respective plurality of target transcripts, to enrich the sequencing library for cDNA molecules corresponding to the plurality of target transcripts, wherein the concentration of a first target-specific capture reagent for a first target transcript is different from the concentration of a second target-specific capture reagent for a second target transcript; and wherein determining the count of RNA molecules for a target transcript in a sample comprises estimating a true count or true count distribution based on an observed count and the relative concentrations of the plurality of target-specific capture reagents. The method may have any of the features described in relation to the first aspect. Estimating the true transcript counts may comprise estimating a distribution of transcript counts for one or more cell populations given observed counts for the respective one or more cell populations. The method may comprise estimating a distribution of transcript counts for a plurality of cell populations in the same or different samples and comparing the estimated distributions for two or more of the cell populations. Estimating a distribution of transcript counts for a cell population may comprise estimating an expected cell size for cells in the population, wherein the cell size is the sum of the number of transcripts in the cells for all of the transcripts measured (i.e. true total pool of the measured transcripts in a cell of the population). Estimating a cell size for cells in the population may comprise assuming that the sum of measured counts for all measured transcripts in the cells is sampled from the cell size using a beta-Poisson distribution, where λ is the expected cell size to be estimated and the α and β parameters represent the capture chemistry efficiency (efficiency of the RT-PCR) of the single-cell sequencing protocol. Estimating a distribution of transcript counts for a cell population may comprise obtaining an estimate of the true transcript counts for a plurality of cells in the population given the observed counts (or a sample mean of the observed counts for cells in the population) using an estimate of the expected cell size for the cells in the population and Cornfield’s approximation. This may comprise obtaining an estimate of the true transcript counts for a plurality of cells in the population assuming that the observed number of transcripts are sampled from the true transcript counts according to a multivariant Fisher Noncentral Hypergeometric distribution, wherein the weights of the distribution depend on the relative concentrations of the target-specific reagents. This may be used as an initial estimate that may be used as a starting point, to infer a true count distribution for each or one or more transcripts using e,g. a Markov Chain Monte Carlo (MCMC) or Variational Inference process. Methods of performing single cell RNA sequencing as described herein may comprise determining the true counts of one or more transcripts based on the concentrations of capture specific reagents used. As the skilled person understand, this is not a necessary step in a general embodiment as most single cell RNA sequencing data analysis pipelines explicitly or implicitly compare profiles between individual cells rather than comparing genes across cells sequenced. Therefore, the absolute counts for transcripts or even the relative counts between transcripts is not relevant to most conventional scRNAseq analysis pipelines. BRIEF DESCRIPTION OF THE FIGURES Figure 1 is a flow diagram showing, in schematic form, a method of performing single cell sequencing according to the disclosure. Figure 2 is a flow diagram showing, in schematic form, a method of providing or designing a composition for single cell sequencing according to the disclosure. Figure 3 shows an embodiment of a system for performing single cell sequencing and / or providing and / or designing a composition for performing single cell sequencing. Figure 4 shows the results of comparison of scRNAseq data simulators in terms of statistics of the real data (from Baron et al., 2016) and simulated data. Minerva=simulation method described herein. Top: violin plots showing a comparison of gene sparsity distribution (percentage of sparsity by gene) between simulators. Bottom: violin plots showing a comparison of gene dispersion distribution between simulators. Figure 5 shows the results of comparison of targeted scRNAseq data (TAP-seq data), whole transcriptome equivalents and simulated targeted scRNAseq data obtained using methods described herein, in terms of statistics of the data. Minerva=simulation method described herein. Figure 5A shows violin plots showing the gene dispersion distributions for real TAP -Seq data, whole transcriptome scRNA-Seq data, and simulated data from Minerva. Figure 5B shows violin plots showing the distribution of dispersion by gene for real TAP-Seq data, whole transcriptome scRNA-Seq data, and simulated data from Minerva. Figure 5C shows a comparison of the relationship between gene sparsity and mean expression between real TAP-Seq data, whole transcriptome scRNA-Seq data, and simulated data from Minerva.2-dimensional kernel density plots show the distribution of these variables for each condition. Figure 6 shows the results of an investigation of how different experimental protocols alter the characteristics of transcriptomics data observed. Figure 6A shows the sparsity distribution (percentage of zero counts (sparsity) by gene) by cell population. Figure 6B shows the gene dispersion estimates by cell population. Figure 6C shows the relationship between gene sparsity and mean expression. In each plot, data is shown for each of 4 different simulated experimental protocols: from left to right for each population: all steps biased (i.e. weighted sampling at both the chemistry capture step and the cDNA amplification step), biased chemistry (i.e. weighted sampling at the capture chemistry step), biased library (i.e. weighted sampling at the cDNA amplification step) and non-biased (state of the art, no weighted sampling at any step). Figure 6D shows, for each cell population, the number of observable genes in a dataset as a function of sequencing saturation. One data series is shown for each protocol. Figure 6D investigates the impact of different experimental protocols on the number of observable genes in a dataset across multiple cell populations as a function of sequencing saturation. This figure illustrates that using weighted experimental protocols enhances the number of observable genes, particularly as the sequencing saturation of the experiment increases. Significant differences are observed in the number of genes used when employing weighted protocols. The Weighted Library protocol exhibits the least increase, eventually reaching a plateau. Weighted Chemistry outperforms Weighted Library protocol with a substantial increase in the number of observed genes as the sequencing saturation increases and only begins to the plate near 100% sequence saturation. Finally, Both Steps Weighted protocol perform all of the Weighted transcriptomes methods with a continuous increase in the number of observed genes but only slightly more than Weighted Transcriptomes. Figure 7 shows the results of an investigation of the effect of different experimental protocols on the statistical characteristics of observed gene counts, simulated using a method as described herein. Figure 7A shows the Two-Dimensional KDE of Mean-Variance from Baron et al., 2016 by Type of Weighted Transcriptome. The Biological Variance is the mean and variance of the gamma distribution used by Minerva to simulate biological variance. While all of the others are different types of single-cell experiments. Non-Weighted is a normal single-cell experiment and the rest apply a different weighting protocol to enrich for low expressed genes. Figure 7B shows one- Dimensional KDE of Mean Expression by Dataset and Quartile. Only the Observed Mean expression distribution was plotted to compare weighted transcriptomes to the Non-Weighted transcriptome. Figure 7C shows one-Dimensional KDE of Variance by Dataset and Quartile. Only the Observed Mean expression distribution was plotted to compare weighted transcriptomes to the Non-Weighted transcriptome. Figure 7D shows one-Dimensional KDE of the Mean expression Log Fold Change by Dataset and Quartile. The Log Fold Change in mean expression was computed using the Non-Weighted gene mean expression as the baseline. The KDE estimates displayed in the figure illustrate the distribution representing the overall alteration in gene mean expression compared to the Non-Weighted Transcriptomes. Figure 7E shows one-Dimensional KDE of the Variance Log Fold Change by Dataset and Quartile. The Log Fold Change in variance was computed using the Non-Weighted gene variance as the baseline. The KDE estimates depicted in the figure illustrate the distribution representing the overall change in gene variance compared to the Non-Weighted Transcriptomes. B-E: Only the Observed Mean expression distribution was plotted to compare weighted transcriptomes to the Non-Biased transcriptome. Figure 7F shows one-Dimensional KDE of the gene observed count Mutual Information of the true count (eMI of the transcriptome) by Dataset. eMI of an individual gene was estimated with the simulated observed counts of a gene vs their simulated true counts. With simulated true counts, the inventors can quantify how much information simulated observed counts contain about the original true. In this particular figure, the inventors estimated KDE by Dataset. Figure 7G shows one-Dimensional KDE of the gene observed count Mutual Information of the true count by Dataset and Quartile. eMI of an individual gene was estimated with the simulated observed counts of a gene vs their simulated true counts. With simulated true counts, the inventors can quantify how much information simulated observed counts contain about the original true. In this particular figure, the inventors estimated KDE by Dataset and Quantile. Figure 8 shows the results of an investigation of the effect of methods of performing single cell sequencing according to the disclosure, compared to prior art single cell sequencing methods, using simulated data. Figure 8A shows the expected Observed Gene Count per Cell as a function of average UMI per cell. Figure 8B shows violin plots showing the log-transformed mean expression distribution across cell populations and experimental protocol. Figure 8C shows Kernel Density estimate of the empirical distribution of the log-transformed mean expression distribution. Figure 9 shows the results of a validation of the concept of weighted transcriptomes using qPCR. All plots show ratios of Ct for pairs of query and housekeeping genes. The housekeeping gene used is GAPDH in all cases. 20X taqman = normal Taqman probe concentration for both the housekeeping gene and the query gene(s), 10X Taqman = half normal Taqman probe concentration for both the housekeeping gene and the query gene(s), 40X Taqman = double normal Taqman probe concentration for both the housekeeping gene and the query gene(s). Variable Taq = variable concentrations of TaqMan probes between Housekeeping and target gene. Figure 9A shows data for MCF7 cells, query gene: FOXA1. Figure 9B shows data for MCF7 cells, query gene: GATA3. Figure 9C shows data for MCF7 cells with estradiol treatment, query genes from left to right: FOXA1, GATA3, NCOA1, RARA. Figure 9D shows data for MCF7 cells without estradiol treatment, query genes from left to right: FOXA1, GATA3, NCOA1, RARA. Figure 9E shows data for T47D cells with estradiol treatment, query genes from left to right: FOXA1, GATA3, NCOA1, RARA. Figure 9F shows data for MDA-MB-231 cells with estradiol treatment, query genes from left to right: FOXA1, GATA3, NCOA1, RARA. Figure 10 shows a step-by-step cell workflow, starting from cell lysis and capturing mRNA transcripts, leading to the creation of the final end sequencing pool. Throughout the process, there is a progressive reduction in the amount of information as the number of transcripts decreases at each step. Notably, RT-PCR, a common technique, typically results in approximately 50% loss of captured transcripts. Following that, the successfully converted cDNA molecules are labeled with UMIs (Unique Molecular Identifiers and amplified using PCR, generating the sequencing pool from which the reads are ultimately sequenced. Figure 11 shows a comparison of the relationship between gene sparsity and mean expression between simulators (Minerva, SPARSim, and Splatter) and real scRNA-Seq data from Baron et al. across multiple and cell populations. Two-dimensional kernel density plots show the distribution of these variables for each dataset and cell population. Figure 12 shows a comparison of gene dispersion and sparsity between simulators (Minerva, SPARSim, and Splatter) and real data from Baron et al. Figure 12a shows a first set of violin plots that show the gene sparsity distribution for each simulator and real data. Figure 12b shows Violin plots showing the gene dispersion for each simulator and real data. Figure 13 compares the observed mean expression of target panel genes between the Whole and Simulated Target Transcriptome is a proxy for evaluating PCR efficiency. Genes associated with PCR primers were categorized as more or less efficient by assessing the observed TAP-Seq mean expression, comparing it to the observed Whole Transcriptome mean expression from Schraivogel et al., 2020. A) illustrates a comparison of observed mean expression between the Target and Whole Transcriptome. B) presents a comparison of observed mean expression between the Target and Simulated Target Transcriptomes. C) showcases the observed mean expression comparison between the Simulated and Whole Transcriptome, highlighting that Minerva's observed mean expression is comparable to that of the Whole Transcriptome. Figure 14 explores how different Experimental Protocols alter the sparsity in Single Cell Data across multiple simulated cell populations from Baron at al. The current state of scRNA-Seq experimental protocols involves the Non-Weighted Experiment Type. The other experimental protocols are proposed as theoretically possible experimental protocols, wherein gene transcripts are weighed to capture chemistry, cDNA amplification, or both. Figure 14a shows a first set of violin plots that shows the gene sparsity distribution for each type of weighted transcriptome. Figure 14b shows a second set of violin plots that show the gene dispersion for each type of weighted transcriptome. Figure 15 shows the results of an investigation into the impact of various experimental protocols on Gene Dispersion Estimates in Single Cell Data across multiple simulated cell populations from Baron et al. This figure demonstrates that the gene dispersion distributions remain consistent across the different experimental protocols, suggesting that applying a weighing protocol in the experiment does not significantly affect the overall gene dispersion distribution across multiple simulated cell types. Figure 16a shows a One-dimensional KDE of the log-transformed mean for each type of Weighted Transcriptome. The KDEs were fitted separately and plotted based on mean expression quantiles. Figures 16b shows a one-dimensional KDE of the log-transformed variance for each type of Weighted Transcriptome. The KDEs were fitted separately, and the log-transformed variance was plotted according to mean expression quantiles. Across both subfigures, the inventors observe a discernible relationship between mean and variance across the quantiles. Though it's important to note that the various weighted transcriptomes exhibit different distributions, with all weighted transcriptomes showing an increase in mean and variance compared to non-weighted transcriptomes in the first, second, and third quantiles. In the fourth quantile, weighted transcriptomes are more constrained, with both Steps-Weighted transcriptomes being the most constrained in both the mean and variance, although not entirely eliminated. Figure 17a shows one-dimensional KDE of observed mean expressions LFC for each type of Weighted Transcriptome. Figure 17b shows one-dimensional KDE of observed variance LFC for each type of Weighted Transcriptome. Both subfigures: LFC was first calculated between Non- Weighted Mean and Variance and Weighted Transcriptomes for each gene, then a KDE was fitted to get a sense of the overall changes in distributions. Figure 18 shows results of analysis of one-dimensional KDE of the eMI of the transcriptome. eMI of an individual gene was estimated with the simulated observed counts of a gene vs their simulated true expression. With a simulated true mean expression, the inventors can quantify how much information simulated observed counts contain about the original true. Figure 18a shows the KDE estimate of the overall MI contained in a weighted transcriptome across the protocols the inventors observe and increase MI with Both Steps Weighted showing the greatest increase. Figure 18b shows the KDE estimate of the MI by mean expression quantile as previously observed MI increases for all weighted transcriptomes with Both Steps Weighted showing the most significant increase regardless of quantile. Figure 19 shows boxplots depicting the ARI for Single Cell Normalization Methods categorized by the Type of Weighted Transcriptomes. The ARI serves as a measure of the consistency with which Single Cell methods cluster and identify cell populations, using simulated cell populations as a ground truth for evaluation. Figure 20 compares gene dispersion and sparsity between simulators (Minerva, SPARSim, and Splatter) and real data from Baron et al., 2016 (Comparing statistical characteristics of real and simulated data). Figure 20a shows a first set of violin plots that show the gene sparsity distribution for each simulator and real data. Figure 20b shows a second set of violin plots that show the gene dispersion for each simulator and real data. Figure 21 compares the relationship between gene sparsity and mean expression between simulators (Minerva, SPARSim, and Splatter) and real data from Baron, Zhao, and Macosko. Two- dimensional kernel density plots show the distribution of these variables for each dataset and cell population. Figure 22 shows identification of optimal weights for simulating target transcriptomes via Mean Square Error comparison between Minerva's Observed Mean Counts and Schraivogel l1000 Target Panels Observed Mean Counts. Figure 22a displays all of the weights tested from 1 to 50 and Figure 22b focuses along the y-axis to show the elbow stops between the weights 25 and 30. Figure 23 shows a scatter Plot comparing the Log2 Transformed observed mean count (OMC) of Targeted (TAP-seq) and Normal scRNA-Seq in a pilot experiment. This plot allows to visualize the experimental noise injected into the pilot experiment by TAP-Seq. A gene is determined to be efficient if the OMC from the Target panel is greater than or equal to Normal scRNA-Seq OMC. Figure 24A shows a scatter Plot comparing the Log2 Transformed OMC of Weighted and Targeted Transcriptomes. Using this plot, we can visualize the experimental efficiency of Weighted Transcriptomes and get a sense of their higher efficiency. A gene is determined to be efficient if the OMC from Weighted is greater than or equal to the Targeted OMC. Figure 24B shows the distributions of the Log2 Fold Change (L2FC) between the OMC of Weighted and Targeted Transcriptomes. Left is the overall distribution of the L2FC and shows a general increase in the OMC of weighted transcriptomes compared to the Targeted Panel. Right shows the same comparison but separated by quantile. The greatest increases in OMC of Weighted Transcriptomes occur in the first and second quantiles, followed by the third. The fourth quantile distribution is bimodal, most likely due to the suppression of the highest expressed genes that occurs when using Weighted Transcriptomes. Figure 25 shows boxplots comparing the Log2 Transformed Sparsity of Weighted and Targeted Transcriptomes on a quantile basis. This show how sparsity changes on an individual quantile basis, with the largest changes in the sparsity occurring in the firs and second quantiles. Figure 26A shows a scatter Plot comparing the Log2 Transformed OMC of Weighted Efficient Genes and Normal scRNA-Seq Transcriptomes. A gene is determined to be efficient if the OMC of Weighted is greater than or equal to the Targeted OMC. Figure 26B shows the distributions of the Log2 Fold Change (L2FC) between the OMC of Weighted Efficient Genes and Normal scRNA-Seq Transcriptomes. Left is the overall distribution of the L2FC and shows a general increase in the OMC of weighted transcriptomes compared to the Targeted Panel. Right shows how the L2FC distributions vary by quantile. The greatest increases in OMC of Weighted Transcriptomes occur in the first and second quantiles, followed by the third. The fourth quantile distribution lacks a bimodal distribution, likely due to removing the genes that Weighted Transcriptomes suppressed. Figure 27A shows a scatter Plot comparing the Log2 Transformed Sparsity of Efficient Weighted and Normal scRNA-Seq Transcriptomes. A gene is determined to be efficient if the OMC of from Weighted is greater than or equal to the Normal scRNA-Seq Transcriptome OMC. Figure 27B shows boxplots comparing the Log2 Transformed Sparsity of Efficient Weighted and Normal scRNA-Seq Transcriptomes on a quantile basis. This shows how sparsity changes on an individual quantile basis, with the largest changes in the sparsity occurring in the first and second quantiles. Figure 28A shows a scatter Plot comparing the Log2 Transformed OMC of L2FC adjusted Weighted and Whole Transcriptomes. A gene is determined to be efficient if the OMC of from Weighted is greater than or equal to the Whole OMC. Figure 28B shows distributions of the L2FC between the OMC of adjusted Weighted and Whole Transcriptomes. Left is the overall distribution of the L2FC and shows a general increase in the OMC of weighted transcriptomes compared to the Targeted Panel. Right shows how the L2FC distributions vary by quantile (gene expression level quartiles). The greatest increases in OMC of Weighted Transcriptomes occur in the first and second quantiles, followed by the third. The fourth quantile distribution is slightly bimodal; likely due to an over-correction of the suppression of the highest expressed genes that occurs when using Weighted Transcriptomes. Figure 29A shows a scatter plot comparing the Log2 Transformed Sparsity of L2FC Corrected Weighted and Normal scRNA-Seq Transcriptomes. A gene is determined to be efficient if the OMC of from Weighted is greater than or equal to the Normal scRNA-Seq Transcriptome OMC. Figure 29B shows boxplots comparing the Log2 Transformed Sparsity of L2FC Corrected Weighted and Normal scRNA-Seq Transcriptomes on a quantile basis. This shows how sparsity changes on an individual quantile basis, with the largest changes in the sparsity occurring in the first and second quantiles. Figure 30A shows a scatter plot comparing the Log2 Transformed OMC of Regression Corrected Weighted and Whole Transcriptomes. A gene is determined to be efficient if the OMC of from Weighted is greater than or equal to the Whole OMC. Figure 30B shows distributions of the Log2 Fold Change (L2FC) between the OMC of Regression Corrected Weighted and Whole Transcriptomes. Left is the overall distribution of the L2FC and shows a general increase in the OMC of weighted transcriptomes compared to the Targeted Panel. Right shows how the L2FC distributions vary by quantile. The greatest increases in OMC of Weighted Transcriptomes occur in the first and second quantiles, followed by the third. The fourth quantile distribution is slightly bimodal; likely due to an over-correction of the suppression of the highest expressed genes that occurs when using Weighted Transcriptomes. Figure 31A shows a scatter plot comparing the Log2 Transformed Sparsity of Regression Corrected Weighted and Normal scRNA-Seq Transcriptomes. A gene is determined to be efficient if the OMC of from Weighted is greater than or equal to the Normal scRNA-Seq Transcriptome OMC. Figure 31B shows boxplots comparing the Log2 Transformed Sparsity of Regression Corrected Weighted and Normal scRNA-Seq Transcriptomes on a quantile basis. This shows how sparsity changes on an individual quantile basis, with the largest changes in the sparsity occurring in the first and second quantiles. Figure 32 shows two dimensional kernel density estimates of the mean and variance of expressed genes across all three experimental protocols: Whole (normal), Targeted, and Weighted Transcriptomes. The heteroskedasticity is preserved in Weighted Transcriptomes and exhibits shape distributions similar to those of the other methods. DETAILED DESCRIPTION Throughout the specification and claims, the following terms take the meanings explicitly associated herein, unless the context clearly dictates otherwise. “and / or” where used herein is to be taken as specific disclosure of each of the two specified features or components with or without the other. For example “A and / or B” is to be taken as specific disclosure of each of (i) A, (ii) B and (iii) A and B, just as if each is set out individually herein. The phrase “in one embodiment” as used herein does not necessarily refer to the same embodiment, though it may. Furthermore, the phrase “in another embodiment” as used herein does not necessarily refer to a different embodiment, although it may. Thus, as described below, various embodiments of the invention may be readily combined, without departing from the scope or spirit of the invention. The features disclosed in the foregoing description, or in the following claims, or in the accompanying drawings, expressed in their specific forms or in terms of a means for performing the disclosed function, or a method or process for obtaining the disclosed results, as appropriate, may, separately, or in any combination of such features, be utilised for realising the invention in diverse forms thereof. It must be noted that, as used in the specification and the appended claims, the singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from “about” one particular value, and / or to “about” another particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values are expressed as approximations, by the use of the antecedent “about,” it will be understood that the particular value forms another embodiment. The term “about” in relation to a numerical value is optional and means for example + / - 10%. Throughout this specification, including the claims which follow, unless the context requires otherwise, the word “comprise” and “include”, and variations such as “comprises”, “comprising”, and “including” will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps. Other aspects and embodiments of the invention provide the aspects and embodiments described above with the term “comprising” replaced by the term “consisting of” or ”consisting essentially of”, unless the context dictates otherwise. In describing the present invention, the following terms will be employed, and are intended to be defined as indicated below. Single cell sequencing refers to technologies that identify sequence information from single cells in a sample. The sequence information is typically obtained by next generation sequencing. In the context of the present disclosure, the term “single cell sequencing” applies to any single cell sequencing approach that comprise one or more target sequence selection steps using target specific reagents. In particular, the term may be interpreted to refer to single cell transcriptome sequencing (also referred to as scRNA-seq or scRNAseq) unless context indicates otherwise. scRNA-seq protocols typically perform the same steps as bulk-RNA sequencing, i.e. reverse transcription (RT), amplification, library generation and sequencing, but by contrast to bulk RNA sequencing, at least the reverse transcription step is performed in a reaction volume that contains a single cell on average (e.g. single cells in respective wells or encapsulated in droplets in a microfluidic device), and at least one of the steps that is performed in a single cell reaction volume (typically the RT) incorporates a cell-specific barcode in the reaction product. After a barcoded reaction product has been obtained (e.g. barcoded cDNA), subsequent steps can be performed in a combined reaction volume. Amplification is typically performed using either polymerase chain reaction (PCR) or in vitro transcription (IVT). Multiple scRNA-seq protocols have been described and the invention of the present disclosure applies equally to any of them. These include, e.g. the protocols described in Tang et al. (2009), STRT (single-cell tagged reverse transcription, described in Islam et al., 2011), SMART-seq (described in Ramsköld et al., 2012), SORT-seq (described in Muraro et al., 2016), CEL-seq (described in Hashimshony et al., 2012), RAGE-seq (described in Singh et al, 2019), Quartz-seq (described in Sasagawa et al., 2013), and C1-CAGE (described in Kouno et al., 2019). Some protocols include unique molecular identifiers (UMIs), allowing true molecule counting. Protocols that do not include UMIs typically only allow relative quantification of transcripts. An example of a droplet based scRNA-seq platform is the 10x Genomics where single cells are encapsulated into droplets of oil comprising gel beads which release oligonucleotides comprising a PCR primer, a barcoded oligonucleotide, a unique molecular identifier (UMI, for counting), and a polydT for polyA capture. Cells are lysed and undergo RT in the droplet, the products are then pooled, the oil removed, and a sequencing library is prepared then sequenced using the Illumina dye sequencing method. By default, single cell RNA sequencing is not target specific, beyond selection of RNA molecules comprising a polyA tail. Reference to “target-specific” or “targeted” protocols refer to single cell sequencing protocols where at least one step of the protocol selects molecules that have predetermined sequences. The selection can be performed through any means, including using gene-specific primers in one or more PCR steps, or using probes that hybridise with a target sequence and are associated or can be associated with a substrate or pull-down component (e.g. biotinylated probes that are recognised by antibodies), or selectively amplified only when they have recognised their target. For example, selection can be performed using pairs of target specific probes that hybridise to adjacent location on a target transcript. Only when both probes in a pair have successfully hybridised to their target transcript can they be ligated. The probes can each contain a region that hybridises to a target sequence and a region that comprises primers (e.g. universal primers, sample specific primers, etc.) for amplification. Thus, only successfully hybridised and ligated probes may be amplifiable. The probes may also contain one or more additional elements such as e.g. barcodes, unique molecule identifiers (UMIs), etc. Target specific protocols are typically used to select and sequence transcripts from a predetermined set of genes. The methods of the present disclosure typically use targeted single cell RNA sequencing. Targeted single cell RNA sequencing have been described. For example, targeted perturb-seq (TAP-seq) is a protocol described by Schraivogel et al. (2020) which amplifies predetermined genes of interest using two semi-nested multiplex PCRs with gene specific primers after the RT step, rather than amplifying the whole transcriptome for single cells. The protocol was developed as an alternative to untargeted single cell RNA sequencing in the context of perturb-seq screens (which combine pooled CRISPR screens and single-cell transcriptomics), to limit the number of gene expression change hypothesis tests to be performed and improve measurements of lowly expressed genes and small effects. However, the protocol does not apply any specific adjustment of the amplification conditions depending on the target genes or their expected expression levels. Instead, all target genes are treated equally. A pair of primers (an outer primer and an inner primer) is provided for each gene, and all single oligos are pooled in equimolar ratio (see Schraivogel et al., 2020, Methods, sections “TAP-seq primer panel design” and “PCR1 and PCR2 with gene- specific outer and inner primers”). The present inventors recognised that this effectively relies on the number of target genes being small enough to allow lowly expressed genes to be detected even in the presence of more highly expressed genes. The present inventors further recognised that as the number of target genes increases, it becomes increasingly likely that lowly expressed genes will not be detected. In other words, the sparsity problem that is familiar to conventional scRNA-seq protocols still exists, it is just smaller for small gene panels and increasingly larger as the size of the gene panel increases. As another example, the 10x Single Cell Flex Kit was designed to enable single cell whole transcriptome analysis of fixed samples. The protocol fixes the cells with paraformaldehyde (unless the cells are already fixed, e.g. cells from a FFPE tissue sample) then hybridises samples to probe sets, the probe sets including gene-specific pairs of probes with a sample barcode. Multiple samples are then pooled. Single cells are then encapsulated in droplets comprising a fixed cell and a functionalised gel bead, where probes undergo ligation and extension to generate 10x barcoded products that contain a cell specific barcode and UMIs. The resulting barcoded library (comprising sample barcodes, cell barcodes, and UMIs) can then be used in any conventional single cell whole transcriptome sequencing protocol. The probe panel is designed for whole transcriptome analysis, although user-designed probe pairs can be added to the panel. Two probe panels are available: a human probe set and a mouse probe set. Each probe panel is transcriptome-wide, only excluding specific genes such as highly variable genes (HLA, TCR joining and variable regions, etc.). Genes are targeted with either “3-fold coverage” or “1-fold coverage”. The “coverage” refers to the number of different probe pairs that target a gene. In other words, most genes are targeted using multiple pairs of probes with different sequences, and some highly expressed genes are targeted with a single pair of probes. The present inventors recognised that this approach is fundamentally limited as it simply distinguishes between genes that are “easy to detect” and all other genes, but does not have the flexibility to take into account differing expression levels between genes (as increasing the number of different probe pairs is an inflexible way to increase the likelihood of capturing targets, and also does not scale in the same way for all targets and probe sets because of probe specific effects). As the vast majority of genes across the transcriptome have the same number of probe sets, the approach fundamentally fails to address the sparsity problem with scRNA-seq protocols, and many lowly expressed genes will fail to be detected. A “sample” as used herein may be any sample suitable for single cell sequencing analysis. A sample may be a cell or tissue sample (e.g. a biopsy), or a sample comprising genetic material derived from single cells where genetic material from individual cells is associated with a cell specific barcode (e.g. a sample that has already been partially processed using a single cell sequencing protocol). The sample may be one which has been freshly obtained from a subject or may be one which has been previously obtained from a subject and processed and / or stored (e.g. frozen, fixed or subjected to one or more purification, enrichment or extractions steps provided that these are compatible with single cell sequencing). Alternatively, the sample may be a cell or tissue culture sample. As such, a sample as described herein may refer to any type of sample comprising cells or genomic material derived therefrom, whether from a biological sample obtained from a subject, or from a sample obtained from e.g. a cell line. The sample may be from a mammalian (such as e.g. a mammalian cell sample or a sample from a mammalian subject, including in particular a model animal such as mouse, rat, etc.), or a human (such as e.g. a human cell sample or a sample from a human subject). Further, the sample may be transported and / or stored, and collection may take place at a location remote from the processing (e.g. any step from collection to library preparation) and / or sequencing location. Further, any computer-implemented method steps described herein may take place at a location remote from the sample collection location and / or remote from the processing location and / or remote from the sequencing location (e.g. the computer-implemented method steps may be performed by means of a networked computer, such as by means of a “cloud” provider). The term “sequence data” refers to information that is indicative of the presence and / or amount of genomic material in a sample that has a particular sequence. Such information may be obtained using sequencing technologies, such as e.g. next generation sequencing (NGS). The sequence data typically comprises a set of sequencing reads and / or a count of the number of sequencing reads that have a particular sequence. Sequence data may be mapped to a reference sequence, for example a reference genome or transcriptome, using methods known in the art (such as e.g. Bowtie, available at bowtie-bio.sourceforge.net / index.shtml). Thus, counts of sequencing reads may be associated with a particular genomic location. The term “genomic location” refers to genomic coordinates in a reference genome or transcriptome. The present disclosure relates in part to method of performing single cell sequencing of a sample using target-specific capture reagents, where the concentration of a first capture reagent for a first target differs from the concentration of a second capture reagent for a second target. In other words, the method comprises at least one step that uses a plurality of target-specific capture reagents, and the plurality of target-specific capture reagents comprise 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20 or more (e.g.50, 100, or any number up to the number of different targets captured) different sets of target-specific capture reagents, wherein the target-specific capture reagents in a set have the same concentration as each other and a different concentration from that of the target-specific capture reagents in any other set. A single cell RNA sequencing protocol may comprise a plurality of steps: a first step in which RNA molecules to be subjected to sequencing library are captured (this may be referred to as “Single Cell Chemistry”, SCC), a second step in which a sequencing library is obtained from captured RNA molecules (this may also be referred to herein as “Sequencing Protocol”, SP), and a third step in which the sequencing library is sequenced. Sequencing may use any sequencing technology known in the art, typically next generation sequencing such as Illumina sequencing. Capture of mRNA molecule may be sequence specific (targeted) or non-specific (untargeted), such as e.g. capturing all molecules comprising a polyA tail. The step of obtaining a sequencing library typically comprises a reverse transcription step that produces cDNA from the captured mRNA molecules, and a cDNA amplification step that amplifies individual cDNA molecules to increase the amount of material subject to sequencing. cDNA amplification can be target specific (e.g. using target specific primers) or non-specific (e.g. using universal primers). The term “target-specific” refers to a capture reagent that is specific for a particular genomic target (e.g. a particular gene or transcript). A non-specific capture reagent is a capture reagent that is not specific for a particular genomic target, and instead captures a plurality of targets that share a common feature, such as e.g. a polyA tail. In the context of a protocol, the term “target-specific” refers to a protocol that makes use of a plurality of target-specific capture reagents. By contrast, a non-specific protocol is a protocol that does not use target-specific capture reagents. Such a protocol may for example use capture reagents that capture all RNA molecules, or all RNA molecules comprising a polyA tail. A capture reagent may comprise an oligonucleotide probe or primer. Thus, a target-specific capture reagent may comprise an oligonucleotide probe or primer comprising a sequence that specifically hybridises with a particular genomic target. Reference to a genomic target typically refers to DNA or RNA molecules that are associated with a particular genomic locus and therefore have a known sequence for which target-specific capture reagents can be designed. A primer may be a PCR primer. A probe may be any molecule that can be used to separate or selectively enrich a molecule having a predetermined sequence from molecules that do not have the predetermined sequence. Probes may be provided in solution or may be associated with a surface. Separation of molecules captured by a probe from other molecules may be performed by physically separating the captured molecule, such as e.g. using a pulldown mechanism whereby the probe is directly or indirectly associated with a physical surface (e.g. a surface of a column, a surface of a magnetic bead, etc.) or pull-down component (e.g. antibody). For example, probes may be associated with a biotin moiety and physical separation may be effected using streptavidin coated beads or biotin-recognising antibodies. As another example, probes may be directly associated with beads. Separation of molecules captured by a probe from other molecules may be performed by selective degradation of the molecules that have not been captured by the probe. Separation of molecules captured by a probe from other molecules may be performed by selectively amplifying captured molecules. Probes may be used to selectively enrich a target molecule, for example using ligation dependent probe amplification. For example, a pair of probes comprising a target specific sequence and a PCR priming site may be hybridised to a target sequence (through the target specific sequences) such that the probes are adjacent to each other and can be ligated. PCR amplification via the PCR priming sites can then be used to enrich the sample for the target sequence. The concentration of the first and second capture reagents may be determined based on the expected expression level of the first and second targets in the sample to be analysed. The concentration of the first and second capture reagents may be determined based on a reference dataset. Thus, the expected expression level of the first and second targets in the sample to be analysed may be determined based on a reference dataset. A reference dataset may comprise expression levels for the first and second targets in one or more reference samples. The expression levels may have been obtained using any expression analysis method known in the art, including e.g. sequencing (bulk RNA sequencing or single cell RNA sequencing), expression arrays, quantitative RT-PCR (qRT-PCR) or any molecular counting assay known in the art. In embodiments, the expression levels have been obtained using sequencing, such as bulk RNA sequencing or single cell RNA sequencing. The reference dataset may have been obtained or may be obtained from a database. The reference dataset may have been obtained or may be obtained by analysing a reference sample (e.g. pilot experiment) using any expression analysis method known in the art. A reference sample may be a sample from the same organism as the organism from which the sample to be analysed originates. For example, when performing single cell sequencing of a sample comprising human cells, a reference dataset comprising expression levels determined in one or more human samples may be used. Similarly, when performing single cell sequencing of a sample comprising mouse cells, a reference dataset may be a dataset comprising expression levels determined in mouse samples. A reference sample may be a sample comprising cells from the same organ as the organ from which the cells in the sample to be analysed originate. For example, when performing single cell sequencing of a sample comprising liver cells, a reference dataset comprising expression levels determined in one or more liver samples may be used. The cells in a sample to be analysed and / or in a reference sample may be from a primary tissue or from a cell line. A reference sample may be a sample comprising cells from the same tissue as the tissue from which the cells in the sample to be analysed originate. For example, when performing single cell sequencing of a sample comprising cardiac muscle tissue (myocardium), a reference dataset comprising expression levels determined in one or more myocardium samples may be used. A reference sample may be a sample comprising cells from the same (or a similar) cell type(s) or cell line(s) as the cells in the sample to be analysed. For example, when performing single cell sequencing of a sample comprising pancreatic beta cells, a reference dataset comprising expression levels determined in one or more pancreatic beta cell samples may be used. When performing single cell sequencing of a sample comprising a breast cancer cell line, a reference dataset comprising expression levels determined in one or more samples of a breast cancer cell line or the same breast cancer cell line may be used. When performing single cell sequencing of a sample comprising a plurality of cell lines or cell types, a reference dataset comprising expression levels determined in one or more samples comprising the plurality of cell lines / cell types in similar proportions (as may be the case e.g. in samples that are obtained from the same tissues), in known proportions or in isolation (i.e. each sample comprising one cell line or cell type) may be used. A similar cell line may be a cell line from the same tissue. A similar cell type may be a cell type from the same group of cell types, such as e.g. subtypes of T lymphocytes may be considered to be similar cell types. Reference to cell types may refer to cell types defined at various levels of granularity. For example, lymphocytes may be considered as a cell type, even though further more specific cell types (e.g. T lymphocytes and B lymphocytes, CD4+ T lymphocytes and CD8+ T lymphocytes, etc. can be distinguished). Thus, for example, when performing single cell sequencing of a sample comprising lymphocytes (which can comprise mixtures of multiple subtypes of lymphocytes, or single subtypes of lymphocytes), a reference dataset comprising expression levels determined in one or more lymphocyte samples (which can comprise mixtures of multiple subtypes of lymphocytes, or single subtypes of lymphocytes) may be used. A reference sample may be a sample comprising cells from the same cell line as the cells in the sample to be analysed. For example, when performing single cell sequencing of a sample comprising MCF7 cells, a reference dataset comprising expression levels determined in one or more MCF7 cell samples may be used. A reference sample may be a sample comprising cells from a cell line in the same category of cell lines as the cell line in the sample to be analysed. For example, when performing single cell sequencing of a sample comprising MCF7 cells, a reference dataset comprising expression levels determined in one or more breast cancer cell line samples may be used. Categories of cell lines may be defined based on the tissue or organ of origin of the cell line. Categories of cells lines may further be defined based on the healthy or diseased status of the cell line (e.g. tumour vs healthy). Instead or in addition to this, categories of cell lines may be defined based on one or more phenotypic characteristics, such as their sensitivity to one or more drugs and / or dependency on one or more proliferation factors (where a proliferation factor may be e.g. a hormone, oncogene, etc). For example, oestrogen dependent cell lines may be considered to represent a category of cell lines. In embodiments, the sample to be analysed is a sample comprising cells from one or more cell lines and the concentrations of the capture reagents are determined using a reference dataset comprising expression levels for the targets to be captured in one or more samples comprising cells of the one or more cell lines. In embodiments, the sample to be analysed is a tissue sample and the concentrations of the capture reagents are determined using a reference dataset comprising expression levels for the targets to be captured in one or more samples of the same tissue, preferably the same tissue from the same organism. The methods described in the present disclosure may be implemented in a computer system. As used herein, the term “computer system” includes the hardware, software and data storage devices for embodying a system or carrying out a method according to the above-described embodiments. For example, a computer system may comprise a processing unit such as a central processing unit (CPU) and / or graphical processing unit (GPU), input means, output means and data storage, which may be embodied as one or more connected computing devices. Preferably the computer system has a display or comprises a computing device that has a display to provide a visual output display. The data storage may comprise RAM, disk drives or other computer readable media. The computer system may include a plurality of computing devices connected by a network and able to communicate with each other over that network. It is explicitly envisaged that computer system may consist of or comprise a cloud computer. The present disclosure provides methods that may be provided as computer programs or as computer program products or computer readable media carrying a computer program which is arranged, when run on a computer, to perform the method(s) described herein. As used herein, the term “computer readable media” includes, without limitation, any non-transitory medium or media which can be read and accessed directly by a computer or computer system. The media can include, but are not limited to, magnetic storage media such as floppy discs, hard disc storage media and magnetic tape; optical storage media such as optical discs or CD-ROMs; electrical storage media such as memory, including RAM, ROM and flash memory; and hybrids and combinations of the above such as magnetic / optical storage media. The methods described herein may be computer implemented unless indicated otherwise, such as e.g. where wet steps are performed. In particular, the present disclosure provides methods for designing compositions for single cell sequencing, and methods for analysing single cell sequencing data. These methods may be computer implemented unless wet steps are explicitly included. Indeed, all these methods may include sequence data analysis steps, the complexity of which is far beyond the capability of the human mind (e.g. thousands to hundreds of thousands of reads are analysed per sample). Single cell sequencing In embodiments of the present invention, a sample is sequenced using single cell RNA sequencing, where one or more steps of the sequencing protocol use target specific capture reagents and the concentrations of the target specific reagents for at least two targets are different from each other. Figure 1 is a flow diagram showing, in schematic form, a method of performing single cell sequencing according to an embodiment of the disclosure. At optional step 10, a sample comprising cells is obtained. At optional step 12, one or more compositions comprising capture reagents are provided in which the concentrations of capture reagents for different target genes differ. This may be performed as explained below by reference to Figure 2. This may comprise a step 12a of identifying a plurality of gene specific weights for a respective plurality of target genes, and a step 12b of providing one or more compositions based on the identified weights. The composition(s) comprise target-specific capture reagents, including a set of target-specific capture reagents (comprising one or more different target-specific reagents) for each gene to be targeted in the sequencing assay (i.e. each gene / transcript to be sequenced). The composition(s) are used in one or more steps of the sequencing assay that use a plurality of target-specific capture reagents associated with a respective plurality of target transcripts, to enrich the sequencing library for cDNA molecules corresponding to the plurality of target transcripts. In each of the one or more compositions, the concentration of a first target-specific capture reagent for a first target transcript is different from the concentration of a second target-specific capture reagent for a second target transcript. In other words, the plurality of target-specific capture reagents comprises a plurality of target-specific capture reagents (or sets of reagents, one set for each target) at different concentrations. When multiple steps that use a plurality of target-specific capture reagents are used, each step may use a respective composition in which the concentrations of a first and second one (or all) of the plurality of target-specific capture reagents has been determined by identifying gene specific weights. At step 14, the compositions are used to obtain a sequencing library instead of default compositions comprising equal concentrations of capture reagents for all genes. This may comprise an optional capture step 14A using a first composition, a reverse transcription (or reverse transcription and PCR amplification) step 14B, and a cDNA amplification step 14C using a second composition. Any one or more of these steps may be targeted or untargeted. At least one of these steps is targeted, i.e. uses target-specific capture reagents. In particular, the capture step may be untargeted in that it is designed to capture any mRNA, and not mRNA targets that have a specific sequence. For example, the capture step may use reagents that capture RNA molecules that have a polyA tail. The capture step may be targeted in that it is designed to capture RNA molecules that have particular sequences, for example using target-specific probes (e.g.10x flex probes). The reverse transcription step 14B may be targeted or untargeted. The reverse transcription step 14B is typically untargeted. Targeted RT may include an amplification step that uses primers that are target specific (e.g. hybridising to specific target sequences). Untargeted RT may include an amplification step that uses primers that are not specific to targeted sequences (e.g. universal primers). The cDNA amplification step may be targeted or untargeted. Targeted cDNA amplification may use primers that are target specific (e.g. hybridising to specific target sequences). Untargeted cDNA amplification may use primers that are not specific to targeted sequences (e.g. universal primers). The method may further comprise sequencing the library obtained at step 16, thereby obtaining sequencing data comprising a plurality of reads, from which observed transcript counts can be determined. The method may further comprise analysing the sequencing data. This may comprise step 18 of analysing the observed transcript counts, such as e.g. by estimating a true transcript count for one or more transcripts, based on the observed transcript count obtained by sequencing the library. Instead or In addition to estimating the true transcript counts, the transcript counts or the sequencing data (e.g. raw or processed reads) may be analysed using any method known in the art for single cell RNA sequencing data. For example, the transcript counts may be clustered to identify cell populations. Note that many such analyses do not require determination of true transcript counts as they rely on differential patterns of expression between different populations of cells, which are fully preserved in the sequencing data obtained using the methods described herein, and are in many cases enhanced by inclusion of genes in these profiles that would not have been detected using prior art sequencing approaches. Estimating a true count based on an observed count and the relative concentrations of the plurality of target-specific capture reagents may comprise determining an observed number of transcripts for the transcript for each cell, and determining a corresponding true transcript count assuming that the observed number of transcripts are sampled from the true transcript counts according to a multivariant Fisher Noncentral Hypergeometric distribution, wherein the weights of the distribution depend on the relative concentrations of the target-specific reagents. Determining a corresponding true transcript count assuming that the observed number of transcripts are sampled from the true transcript counts according to a multivariant Fisher Noncentral Hypergeometric distribution, wherein the weights of the distribution depend on the relative concentrations of the target-specific reagents may be performed using Cornfield’s approximation method. Determining a true transcript count for a cell may comprise: (i) obtaining an initial estimate of a true transcript count assuming that the observed number of transcripts are sampled from the true transcript counts according to a multivariant Fisher Noncentral Hypergeometric distribution, wherein the weights of the distribution depend on the relative concentrations of the target-specific reagents; and (ii) obtaining a refined estimate of the true transcript count using the initial estimate to parameterise a distribution from which transcript counts are sampled to provide refined estimates of the true transcript count using an iterative algorithm, optionally wherein the iterative algorithm is a Markov Chain Monte Carlo algorithm. At each iteration of the iterative algorithm, the distribution may have a parameter that is the estimated transcript count at the preceding iteration. At the first iteration the parameter may be set to the initial estimate of the true transcript count. The iterative algorithm may comprise setting the estimate of the true transcript count to the estimate of the true transcript count obtained at the current iteration when an acceptance criterion is met, and setting the estimate of the true transcript count to the estimate of the true transcript count obtained at the preceding iteration when an acceptance criterion is not met, wherein the acceptance criterion compares the likelihood of the estimates of the true transcript count at the current and preceding iteration assuming that true transcript counts are sampled from a multivariant Fisher Noncentral Hypergeometric (MFNH) distribution, wherein the weights of the distribution depend on the relative concentrations of the target-specific reagents. The MFNH distribution may be a distribution ^^^^~ ^^ ^^ ^^ ^^( ^^, ^^, ^^) where Mj is the transcript count for gene j in cell i, n is the estimated mRNA content for cell i, o is a weight for gene j which depends on the relative concentrations of the target- specific reagents. Determining a true transcript count for a cell may further comprise: obtaining an initial estimate of the mRNA content for each cell based on the cell library size and an estimated sampling efficiency for the cell, and obtaining a refined estimate of the mRNA content using the initial estimate to parameterise a multi-variant Poisson distribution from which mRNA content values are sampled to provide refined estimates of the mRNA content using the iterative algorithm. At each iteration of the iterative algorithm, the distribution may have a parameter that is the estimated mRNA content of the cell at the preceding iteration. At the first iteration the parameter may be set to the initial estimate of the mRNA content of the cell. The distribution may be a multivariate Poisson distribution. For example, estimating a true transcript count for one or more transcripts may comprise: (i) obtaining an initial estimate of a true transcript count assuming that the observed number of transcripts are sampled from the true transcript counts according to a multivariant Fisher Noncentral Hypergeometric distribution, wherein the weights of the distribution depend on the relative concentrations of the target-specific reagents, and obtaining an initial estimate of the mRNA content for each cell based on the cell library size and an estimated sampling efficiency for the cell; and (ii) obtaining a refined estimate of the true transcript count using the initial estimate to parameterise a multi-variant Poisson distribution from which transcript counts are sampled to provide refined estimates of the true transcript count, and obtaining a refined estimate of the mRNA content using the initial estimate to parameterise a multi-variant Poisson distribution from which mRNA content values are sampled to provide refined estimates of the mRNA content, using an iterative algorithm, such as a Markov Chain Monte Carlo algorithm. The iterative algorithm may comprise setting the estimate of the true transcript count to the estimate of the true transcript count obtained at the current iteration when an acceptance criterion is met, and setting the estimate of the true transcript count to the estimate of the true transcript count obtained at the preceding iteration when an acceptance criterion is not met. The acceptance criterion can for example compare the likelihood of the estimates of the true transcript count at the current and preceding iteration assuming that true transcript counts are sampled from a multivariant Fisher Noncentral Hypergeometric (MFNH) distribution, wherein the weights of the distribution depend on the relative concentrations of the target-specific reagents and the estimate of the mRNA content at the current or preceding iteration. Estimating the true transcript counts may comprise estimating a distribution of transcript counts for one or more cell populations given observed counts for the respective one or more cell populations. Analysing the observed transcript counts at step 18 may comprise estimating a distribution of transcript counts for a plurality of cell populations in the same or different samples and comparing the estimated distributions for two or more of the cell populations. Estimating a distribution of transcript counts for a cell population may comprise estimating an expected cell size for cells in the population, wherein the cell size is the sum of the number of transcripts in the cells for all of the transcripts measured (i.e. true total pool of the measured transcripts in a cell of the population). Estimating a cell size for cells in the population may comprise assuming that the sum of measured counts for all measured transcripts in the cells is sampled from the cell size using a beta-Poisson distribution, where λ is the expected cell size to be estimated and the α and β parameters represent the capture chemistry efficiency (efficiency of the RT-PCR) of the single-cell sequencing protocol. Estimating a distribution of transcript counts for a cell population may comprise obtaining an estimate of the true transcript counts for a plurality of cells in the population given the observed counts (or a sample mean of the observed counts for cells in the population) using an estimate of the expected cell size for the cells in the population and Cornfield’s approximation. This may comprise obtaining an estimate of the true transcript counts for a plurality of cells in the population assuming that the observed number of transcripts are sampled from the true transcript counts according to a multivariant Fisher Noncentral Hypergeometric distribution, wherein the weights of the distribution depend on the relative concentrations of the target-specific reagents. This may be used as an initial estimate that may be used as a starting point, to infer a true count distribution for each or one or more transcripts using e,g. a Markov Chain Monte Carlo (MCMC) or Variational Inference process. The method may further comprise a step 20 of providing the results of any one or more of the preceding steps (such as e.g. sequencing data, observed transcript counts for one or more transcripts, estimated true transcript counts for one or more transcripts) to a user, for example through a user interface. Providing reagents for single cell sequencing The present disclosure also describes methods of providing compositions for single cell sequencing, and methods of designing compositions (e.g. reagent mixtures) for single cell sequencing. Figure 2 is a flow diagram showing, in schematic form, a method of providing or designing a composition for single cell sequencing according to the disclosure. At optional step 20, reference expression levels (e.g. transcript counts or any other metric of expression level) for a plurality of target genes are provided, for example from a reference dataset or plurality of datasets. At optional step 22, capture reagent concentrations associated with the expression levels obtained at step 22 are obtained. These may be for example the concentrations or relative concentrations of capture reagents that have been used to obtain the reference expression levels. The reference expression levels may have been obtained using one or more assays that do not use target-specific capture reagents. In such cases, default concentrations may be used. By default these concentrations may be equal, or assumed to be equal. At step 24, gene specific weights are obtained for at least one, preferably all of the target genes, as explained further below. At step 26, concentrations of capture reagents are determined using the results of steps 22 and 24. When the expression levels obtained at step 22 were obtained using target-specific capture reagents, the same target-specific capture reagents may be used but their concentrations may be adjusted from the concentration used to obtain the reference expression levels using the weights obtained at step 24. Other target-specific capture reagents may be used instead or in addition to those that were used to obtain the reference expression levels, or when not such reagents were used to obtain the reference expression levels. In such cases, the concentrations of step 22 may be set to default concentrations for all of these reagents. The gene specific weights may be values by which the capture reagent concentrations obtained at step 22 are to be multiplied in order to perform a single cell RNA sequencing method according to embodiments of the disclosure. At optional step 28, one or more compositions (such as e.g. a composition for biased capture and / or a composition for biased cDNA amplification) are prepared using the weights identified at step 26. In particular, capture reagents may be mixed in relative concentrations equal to the weights identified, or to the weights identified multiplied by the concentrations obtained in step 22. One or more sets of gene specific weights may be identified, each set corresponding to a composition comprising respective target-specific capture reagents at concentrations determined based on a respective set of weights. Each composition may be used in a respective capture step. For example, the sequencing may comprise: performing a capture step that uses a composition comprising target-specific capture reagents prior to reverse-transcription of RNA molecules present in the sample, reverse-transcribing one or more RNA molecules present in the sample to obtain one or more corresponding cDNA molecules; and amplifying the cDNA molecules to obtain a sequencing library using a composition comprising target-specific capture reagents. As another example, the capture step prior to reverse-transcription may be omitted or may use capture reagents that are not target-specific, and the step of amplifying the cDNA molecules to obtain a sequencing library may use a composition comprising target-specific capture reagents. As another example, the capture step prior to reverse-transcription may be present and may use capture reagents that are target-specific, and the step of amplifying the cDNA molecules to obtain a sequencing library may use a composition comprising non-target-specific capture reagents (e.g. universal primers). The concentrations of the target-specific capture reagents are determined using the expected expression level of the first and second transcripts in the sample. In particular, the concentrations of the target-specific capture reagents are determined by identifying concentrations of target-specific capture reagents that are such that the number of molecules corresponding to the target transcripts in the sequencing library are more similar to each other than the numbers of molecules that would be obtained in a sequencing library produced using the same concentration of the first and second target-specific capture reagents. The concentrations of the target-specific capture reagents may be selected such that the numbers of molecules corresponding to the target transcripts in the sequencing library are expected to be the same. There are multiple ways to identify such concentrations. A first way will be described by reference to optional steps 24A-24F, and a second way will be described by reference to optional step 24G. At step 24A, an observed relative frequency may be determined for each transcript to be captured (together referred to as a “set” of transcripts). An observed relative frequency for a transcript may be obtained as the expected expression level of the transcript divided by the sum of the expected expression levels of all transcripts in the set. At step 24B, a target relative frequency may be determined for each transcript. This may be any frequency that is such that the target frequency for the transcripts in the set are closer to each other than the corresponding observed relative frequencies. For example, the target relative frequency for all transcripts in the set may be the inverse of the number of transcripts in the set. At step 24C, a gene specific weight is obtained for each transcript in the set using the observed and target relative frequencies. This may be obtained as the ratio of the target relative frequency and the observed relative frequency for the target. The method of steps 24A-24C may be applied in any embodiment in which a single target-specific capture step is used. In embodiments where a capture step is provided prior to reverse transcription, and a the cDNA amplification step also uses target-specific capture reagents, the method of steps 24A-24C may be used to determine gene specific weights for the a capture step is provided prior to reverse transcription, and the method may proceed to steps 24D-24F through which gene specific weights can be determined for the cDNA amplification step. At step 24D, an expected number of unique molecules after capture and / or reverse transcription using target- specific capture reagents is obtained. This may be estimated based on the expected reverse transcription efficiency and the expected number of molecules captured for each transcript during the target-specific capture step. At step 24E, the observed relative frequency of unique molecules (number of times that each unique molecule corresponding to a transcript is represented in the reference sequencing dataset) is determined. A target relative frequency is also determined, typically such that the relative frequency for each unique molecule is equal. For example, the target relative frequency of a unique molecule may be the inverse oft eh number of unique molecules corresponding to the set of target transcripts, in the reference sequencing data. At step 24F, a gene specific weight is obtained for each transcript in the set using the observed and target relative frequencies. This may be obtained as the ratio of the target relative frequency and the observed relative frequency for the target. At step 24G, gene specific weights for one or more target-specific capture steps may be determined simultaneously by identifying one or more sets of weights (e.g. one set for each target- specific capture step) that result in the lowest sum of the expected percentages of each transcript in the set of transcripts that are not represented in a sequencing library obtained using one or more compositions comprising the relative concentrations of the capture reagents obtained using the sets of weights, amongst a plurality of candidate sets of weights. This may comprise selecting a candidate set of weights, which may comprise a set for each respective target-specific capture step (e.g. a set for a target-specific capture step prior to reverse transcription and / or a set for target- specific cDNA amplification step), determining an expected unique molecular count for each gene based on a set of weights used at a target-specific capture step prior to or during reverse transcription (if such a step is used), determining an expected count for each unique molecule based on the expected unique molecular count and a set of weights used at a target-specific cDNA amplification step (if such a step is used), determining the percentage of unique molecules not sequenced for each gene based on the expected count of unique molecules and determining the sum of these percentages corresponding to the set(s) of weights. The entire process may then be repeated for a further candidate set of weights and the sum of percentages corresponding to the candidate sets of weights may be compared to identify an optimal set of weight. Any optimisation routine known in the art may be used to suggest candidate sets of weights for testing at each iteration of the method, such as e.g. a greedy algorithm. Systems Figure 3 shows an embodiment of a system for performing single cell sequencing and / or for providing or designing compositions for single cell sequencing, according to the present disclosure. The system comprises a computing device 1, which comprises a processor 101 and computer readable memory 102. In the embodiment shown, the computing device 1 also comprises a user interface 103, which is illustrated as a screen but may include any other means of conveying information to a user such as e.g. through audible or visual signals. The computing device 1 is communicably connected, such as e.g. through a network, to sequence data acquisition devices 3, such as a sequencing machine, and / or to one or more databases 2 storing single cell sequencing data, and / or to one or more sample processing devices 4 (e.g. one or more robots that can perform one or more respective sample processing steps to obtain a sequencing library from single cells, and / or one or more robots that can execute instructions to prepare a composition for single cell sequencing designed using a method as described herein). The one or more databases 2 may further store one or more of: target related information (e.g. expected and / or measured expression levels for a plurality of targets), primer related information (e.g. sequences, physico-chemical properties, etc.), composition related information (e.g. batch, concentrations of reagents, etc.) and / or sample related information, etc. The computing device may be a smartphone, tablet, personal computer or other computing device. The computing device is configured to implement a method for performing single cell sequencing and / or for designing compositions for single cell sequencing, as described herein. In alternative embodiments, the computing device 1 is configured to communicate with a remote computing device (not shown), which is itself configured to implement a method for performing single cell sequencing and / or for designing compositions for single cell sequencing, as described herein. In such cases, the remote computing device may also be configured to send the result of the methods to the computing device. Communication between the computing device 1 and the remote computing device may be through a wired or wireless connection, and may occur over a local or public network 6 such as e.g. over the public internet. The sequence data acquisition device 3 and / or the sample processing device 4 may be in wired connection with the computing device 1, or may be able to communicate through a wireless connection, such as e.g. through WiFi and / or over the public internet, as illustrated. The connection between the computing device 1 and the sequence data acquisition device 3 and / or sample processing device 4 may be direct or indirect (such as e.g. through a remote computer). The sequence data acquisition device(s) 3 are configured to acquire sequence data from nucleic acid samples, for example sequencing libraries extracted from single cells. The sequence data acquisition means is preferably a next generation sequencer. The following is presented by way of example and is not to be construed as a limitation to the scope of the claims. EXAMPLES Example 1 – Introduction Single-Cell RNA Sequencing (scRNA-Seq) technologies suffer from sparsity - the presence of large amount of “zeros” (genes / transcripts that are not detected). Some of these will be truly zero (non-expressed genes) and others likely are genes that are expressed but not detected. In particular, highly expressed genes are overrepresented in the data and many lowly expressed genes are being missed. The present inventors recognised that sparsity emerges due to (i) the sampling process that is occurring and (ii) the inherently shallow sequencing of individual cells (i.e. the transcriptome of each cell is shallow sequenced even though the library as a whole may be deep sequenced – because there is a limited amount of sequencing capability). The sampling process of a single cell sequencing experiment operates according to a ‘Sample Without Replacement’ scheme. This means that when a molecule is sampled it’s gone and is not being ‘placed back’ in the cell. In other words, each sampling event (conversion of an initial transcript molecule all the way through the protocol and its sequencing) is destructive. As a result, sequencing a transcript of Gene A reduces the probability of sequencing another transcript of Gene A. In shallow sequencing, where you have only very few chances to ‘catch’ a transcript, lowly expressed genes are thus often missed. This sparsity means that the full transcriptome is never measured in single cells, which fundamentally limits the current applicability of scRNA-Seq to tasks like cell populations identifications, which can rely upon the variance of highly expressed genes (marker genes). More granular understanding of the expression patterns within cells are typically very limited. The present inventors devised a solution to this problem. The solution described in the present examples employs Single Cell Chemistry (SCC) and / or Sequencing Protocols (SP) that use estimates of gene-specific weights that down-weight highly expressed genes and up-weight lowly expressed genes, thus overall making the sampling process more uniform across the transcriptome. SCC refers to the steps involved in capturing the target molecules (e.g. all mRNAs or selected mRNAs) to be subjected to sequencing library preparation. SP refers to the steps involved in sequencing library preparation from captured mRNA molecules. They demonstrate the power of this approach using simulation and experimental proofs of concept. In particular, they develop and demonstrate a new approach to simulating scRNA-seq data that is able to replicate the properties of real data while also enabling to test the effect of changing the gene-specific weights employed to do either or both of the SCC and SP steps (Example 2). Specifically, they model scRNA-Seq data using a Gamma-MFNH-Binomial-MFNH Hierarchical model, where: (i) the biological heterogeneity of cell populations is modelled using a Gamma distribution, (ii) technical variance of experimental steps that can be manipulated are modelled with two independent steps (one for each of SCC and SP) modelled using respective Multivariate Fisher Noncentral Hypergeometric (MFNH) distributions, and (iii) RT-PCR is modelled as a Binomial downsampling process. They use this approach to demonstrate that the weighted single cell data (which changes the weights of one or both of the MFNH distributions for at least some of the genes, in order to reduce sparsity in the data) preserves important statistical properties of the unweighted data (Example 3), such that raw expression levels can be inferred back from data from weighted protocols. They then demonstrate an approach to estimating the weights to be applied to obtain improved scRNA-seq data that addresses the sparsity problem above (Example 4). The method uses a pilot experiment or a reference dataset that provides a rough estimate of the transcript relative abundances (i.e. transcript frequencies). Based on these transcript frequencies, they show that it is possible to estimate the sampling weights to be used to obtain any desired target frequency for each transcript, using the ratio of target frequency-to-observed frequency as the sampling weight. The sampling weight can then be used to determine how much to increase or decrease the concentration of the reagents used in the enrichment protocol at hand (e.g. PCR primers, Antibodies, and / or Probes). They finally validate the approach experimentally using qPCR and TAP-seq (Schraivogel et al. 2020) (Example 5). In particular, they validate that when manipulating enrichment methods reagent concentrations it is possible to alter the qPCR Curves for both the reference housekeeping and target gene. In particular, they show that it is possible to move these curves closer together, i.e. making it closer to an equal likelihood that genes will be sampled even if they are expressed at very different levels. Further, they conduct a scRNA-seq experiment on peripheral blood mononuclear cells (PBMCs) using a modified version of the TAP-seq protocol where custom sampling weights are calculated to adjust the concentration of nested PCR primers. They use these results to show that the approach can shift the mean expression distribution of genes observed. Single-cell RNA Sequencing (scRNA-seq) has evolved into a standard tool for assessing cellular heterogeneity. A diverse array of specialized single-cell sequencing protocols is being continually developed to tackle specific challenges and cater to various applications. One such method is targeted transcriptomes, which aim to enhance overall resolution and facilitate the detection of specific genes of interest. As the adoption of this technology increases, there is a growing need for continual refinement of workflows and protocols. While targeted panels offer one avenue for experimental manipulation in Single Cell Experiments, as explained above and further below, the present inventors recognized that they are not the sole option; enrichment protocols are also feasible. To efficiently explore the design space of Single Cell Experiments, the inventors developed a new scRNA-seq simulator (named “Minerva”) designed to accommodate the intricacies of experimental protocols and enhance scRNA-seq methodologies. Using the Multivariant Fisher Noncentral Hypergeometric Distribution, Minerva conceptualizes scRNA-Seq as a weighted sampling process. It models the progressive reduction in the number of transcripts at four critical experimental steps: capture chemistry, RT-PCR, cDNA amplification, and sequencing. The inventors’ findings (see Examples 2 and 3) indicate that Minerva reliably replicates observed scRNA-Seq statistical characteristics with performance equivalent to or surpassing the current state of the art. Furthermore, Minerva demonstrates its capability to simulate targeted panels and their corresponding statistical characteristics. Discrepancies between Minerva's simulation and previous TAP-seq experiments revealed inefficiencies in the PCR primers used to amplify target transcripts. Moreover, using Minerva, the inventors uncovered a novel set of theoretically possible experimental protocols collectively termed 'Weighted Transcriptomes.' These protocols leverage enrichment techniques to manipulate the weighted sampling steps of single-cell experiments, aiming to enhance the resolution and detection of genes across transcriptomes. These results emphasize Minerva's utility in modeling various scRNA-Seq protocols and assessing their performance, providing researchers with a benchmarking method for wet lab experiments. Example 2 – scRNA-seq simulation for improved single-cell experiments In this example, the inventors propose a new method to model an entire single-cell experiment and account for both biological and technical noise in scRNA-seq data. They compare the performance of this simulator to two other state of the art simulators, and show that the proposed method is able to faithfully recapitulate the relationship between gene sparsity, mean expression, and gene dispersion. The results showed that the proposed method could nearly perfectly simulate the sparsity of targeted transcriptome data. The method is the only available method that can flexibly simulate experimental protocols that involve biased enrichment steps as proposed herein. These results suggest that the introduction of enrichment steps in the single cell experiment can significantly improve the observable gene counts and characterization of statistical properties of low-expressed genes. Overall, the method represents a reliable way to explore novel experimental protocols and assess the effects of these on statistical properties of single-cell data. Introduction Single cell RNA sequencing (scRNA-Seq) provides a high-resolution perspective on cellular heterogeneity within tissues, organs, responses to perturbations, and other biological contexts (Tang et al., 2009; Hwang et al., 2018). However, numerous technical challenges remain. The most significant obstacle is the sparsity of scRNA-seq data, primarily caused by technical dropouts. Technical dropouts arise when a gene is expressed within a cell, yet the scRNA-Seq experiment fails to detect any gene transcripts, resulting in an erroneous count of zero transcripts. Technical dropouts are intrinsically linked to the experimental protocol of scRNA-Seq and are a consequence of the sequential downsampling that occurs during the workflow (Fig 10). A scRNA-Seq experimental protocol involves four critical downsampling steps: capture chemistry, RT-PCR, cDNA amplification, and sequencing (Tang et al., 2009; Zheng et al., 2017). At each stage, the number of transcripts is progressively reduced, yielding a relatively limited portion of the cell transcriptome available for sequencing by the end of the experiment. A single-cell experiment is a complex process that can be seen as a succession of the following key stages: - Biological Condition: this refers to the subject of the experiment, such as a cell population, condition, or perturbation (i.e. true biological variation rather than technical noise); - Chemistry Capture: dissociation of cells and marking of material from individual cells for later sequencing by capture of mRNA molecules; - RT-PCR: conversion of captured mRNA molecules into copy or complementary DNA (cDNA) through RT-PCR (which is inherently inefficient and can destroy up to 50% of the starting mRNA molecules); - cDNA Amplification and Sequencing: amplification of cDNA to a level suitable for sequencing through PCR (which can amplify differences in cDNA amounts due to either biological or technical noise, resulting in an imbalance between high and low-expression genes, and making it difficult to observe low counts of certain genes), and this adds noise to the data but this noise is well characterized and easier to correct than the previous stages. The present example proposes a cutting-edge Single Cell Simulator that addresses the limitations of current Single Cell Simulators by modeling every step of the Single Cell Experiment. A key advantage of the present method is the use of a Multivariant Fisher Noncentral Hypergeometric Distribution (MFNH, Fog, 2008), which allows for the consideration of potential experimental manipulations (i.e. enrichment protocols, etc.). In particular, the method uses a Gamma-MFNH- Binomial-MFNH Hierarchical Model where the Gamma distribution models the biological variance, the Binomial distribution models RT-PCR conversion of mRNA to cDNA, and MFNH models the technical noise introduced by Chemistry Capture Efficiency, cDNA Amplification, and Sequencing. The use of the MFNH distribution makes it possible to capture the “sampling-without-replacement” process that takes place in these protocols, and provides the ability to model the effect of manipulation of experimental protocols, such as selective enrichment. The method makes it possible to understand the impact of different enrichment protocols on the experiment outcomes and to design more effective and efficient experiments. It also serves as a tool to validate the reliability and accuracy of new computational methods’ ability to analyze the data these new experimental protocols generate, ensuring that they produce trustworthy results. The following work: (i) validates the simulator by comparing its output to real-world scRNA-Seq data, ensuring that it can simulate a typical single-cell experiment accurately and faithfully; (ii) demonstrates its ability to simulate targeted panels, and (iii) demonstrates the potential of the method as a tool for theoretical experimentation, by allowing users to manipulate different parameters and studying the effects of changes in experimental protocols. The process of transcript downsampling in this study adheres to a specific sampling approach referred to as `sampling without replacement.' While technically, all bulk RNA-Seq and Microarray experiments also conform to `sampling without replacement,' the quantity of molecules involved in the experiment introduces variations in the underlying behavior and dynamics of the sampling process. In particular, 'sampling without replacement' assumes paramount significance when the number of molecules engaged in reactions is relatively limited, especially compared to the comprehensive pool of molecules being measured, such as an individual cell's transcriptome. In contrast, conventional bulk methods operate on population averages, encompassing millions to billions of molecules from thousands to hundreds of thousands of cells. This substantial scale results in the sampling procedure of bulk and microarray-based methods behaving akin to “sampling with replacement”. To understand the impact this sampling scheme has, think of the cell as a pool of mRNAs, which appear at different frequencies. Sampling without replacement removes one mRNA copy from the pool at every step and thus changes the probability of capturing this gene again. This change will be subtle for highly expressed genes that appear in many copies, but it will be substantial for lowly expressed genes that only had a few copies to start with. This observation explains why scRNA- seq data are biased towards highly expressed genes and miss lowly expressed genes. This bias does not affect applications that only depend on a small number of highly expressed genes, like identifying cell populations by marker genes, but prevents a comprehensive view of the transcriptome and limits our ability to comprehensively characterise cell populations or perturbations responses. Researchers have devised various computational and experimental strategies to tackle the issue of technical dropouts, each with its unique advantages and drawbacks. One common computational approach involves imputation, wherein machine learning models, such as neural networks, are employed to estimate missing count data (Luecken and Theis, 2019). However, this method can introduce bias and spurious correlations that lack a biological basis, as it doesn't involve causal inference. Experimental methods to combat sparsity include SMART-Seq and targeted panel sequencing (TAP-Seq) (Hagemann-Jensen et al., 2022; Replogle et al., 2020; Schraivogel et al., 2020). SMART-Seq is a plate-based single-cell sequencing technique that enables deeper transcriptome sequencing, albeit with lower throughput. Where TAP-Seq is a droplet-based method that utilizes nested PCR cycles to extract specific genes of interest by controlling the primers used in the PCR cycles, this approach enhances the likelihood of observing transcripts from the target gene panel across a high-throughput number of cells, although it comes at the expense of resolution. None of these methods fundamentally resolve the inherent winner- take-all effects present in scRNA-Seq experiments. In essence, computational methods alone cannot eliminate this issue. What is required is the development of a new and optimized experimental protocol for scRNA-Seq that specifically aims to address the sampling without replacement effects. Here the inventors present a new Single Cell Simulator that comprehensively models every significant downsampling step in a scRNA-Seq experiment while accounting for potential experimental manipulations. The simulator takes into account each stage of the scRNA-Seq process, from modelling the cell's original 'true expression' using a gamma distribution to simulating the binomial downsampling process of RT-PCR. However, it's important to note that Capture Chemistry and cDNA Amplification steps differ from true expression and RT-PCR, as they introduce the potential for experimental manipulations, such as enrichment protocols. These manipulations can be treated as a form of weighted sampling. However, it's not feasible to employ standard distributions used for modelling weighted sampling, like the Dirichlet-Multinomial, because they are tailored for sampling with replacement schemes. Instead, an alternative distribution is required to simulate weighted sampling within a sampling without replacement scheme. The MFNH provides a suitable solution for this purpose (Fog, 2008). Consequently, Minerva employs a Hierarchical model that combines a Gamma distribution, two MFNH distributions, and a Binomial distribution to simulate the scRNA-Seq experimental protocol. This model ensures that Capture Chemistry and cDNA Amplification are accurately represented, and it merges cDNA Amplification and Sequencing into a joint step using MFNH (Baruzzo et al., 2020). Methods Properties of Noncentral Hypergeometric distributions. The simulator utilizes Noncentral Hypergeometric distributions to model the sampling process of single-cell experiments, which involves weighted sampling without replacement. Specifically, simulation employs the Fisher Noncentral Hypergeometric (FNH) distribution to model the sampling processes for capture chemistry and cDNA amplification. To provide an intuition of the behaviour of this distribution and highlight the differences from its sibling distribution, the Wallienus Noncentral Hypergeometric (WNH), we explain below what Noncentral Hypergeometric distributions are, how they extend the Hypergeometric distribution, and present the mean, variance, and probability mass function (PMF) of the FNH. Differences between the Wallienus and Fisher Noncentral Hypergeometric distributions. The Hypergeometric distribution models a specific sampling process known as sampling without replacement. When discussing the Hypergeometric distribution and its related noncentral distributions, it is common to use a visual analogy of drawing balls from an urn, where the urn contains all the available balls for sampling. In this analogy, the number of balls in the urn is represented by N, and there are K populations whose individual populations, denoted as m, add up to N. Unlike the binomial distribution and most statistical distributions, when a ball is drawn, it is removed from the urn and not replaced. Noncentral Hypergeometric distributions are an extension of the Hypergeometric distribution that can accommodate weighted sampling scenarios. In these scenarios, each population has a weight ω that alters the probability of a class being sampled, with an increasing weight leading to a higher probability of selection. There are two primary types of Noncentral Hypergeometric distributions: the WNH distribution and the FNH distribution (Fog 2008). The main distinction between the WNH and FNH distributions lies in the sampling process of the model. In the case of the WNH distribution, sampling is done sequentially, meaning that the ordering of sample draws affects the probability of sampling the next ball drawn based on the ball drawn and its assigned weight. In addition, because the WNH sampling process is sequential, it is possible to determine a desired sample size or the number of balls to be drawn in advance (Fog 2023). The FNH distribution assumes no dependence between draws. Therefore, when sampling from an FNH distribution, all balls are drawn simultaneously from the urn without any prior knowledge of how many balls will be sampled (Fog 2023). An insightful metaphor to understand these differences was proposed by Fog in his work on Biased Urn Theory. He likened the sampling process of the WNH distribution to fishing with a fishing rod, where only one fish can be caught at a time. The probability of catching a particular fish species increases with its weight. In this scenario, the desired number of fish to catch can be decided in advance. By contrast, the FNH distribution models a process where a fishing net is used. The fishing net is cast into a lake, left for a period of time, and then pulled back in. The probability of catching a specific fish species improves with its weight, but it is impossible to know how many fish will be caught in advance. Based on the differences between these two distributions, the simulations in the present examples utilize the Multivariant version of the FNH (MFNH) due to no dependence between samples, as the fishing net metaphor fits the single-cell sequencing scenario to a greater extent than the WNH. Parameters and Functions of the Fisher Noncentral Hypergeometric Distribution. The FNH distribution is parameterized as follows: for a simple univariate FNH distribution, it is characterized by two populations, denoted as ^^1and ^^2, both of which are non-negative integers ( ^^1, ^^2^^ ^^ ∈ ℕ). The total number of elements in the populations is represented by N, which is the sum of ^^1and ^^2, i.e., N = ^^1+ ^^2. Each population has a weight, denoted as ω, which is a positive real number (ω ∈ ℝ), where an increase in the weight increases the probability of drawing a specific population. When the weights are close to one, the FNH distribution collapses into the hypergeometric distribution. Finally, there are n samples to be drawn, where n is an integer within the range [0, N), and x is the number of observations that are from population m1(n-x being the number of observations that are from population m2). The terms ^^0, ^^1, and ^^2are used in the calculation of the mean and variance of the FNH distribution. They are defined as follows: The mean of the FNH distribution, denoted as^^1^^0, is calculated as ^^1divided by ^^0. The variance of the FNH distribution, denoted as^^2^^0− (^^1^^0)2, is calculated as ^^2divided by ^^0minus the square of^^1^^0. Please note that ^^^^ ^^ ^^and ^^^^ ^^ ^^represent the minimum and maximum values of x in the summation. The Probability Mass Function (PMF) for the MFNH is defined as follows: Where P(x1,…,xk) is the joint probability of observing xk balls from population k, and is given by: The mean of the MFNH distribution, denoted by μk, is calculated as follows: where r is the unique positive solution to the equation: These equations define the MFNH distribution and allow for the analysis of biased sampling processes involving multiple populations. Modelling Cell-Specific Parameters: To model the various sources of variance within a single-cell experiment requires a ground truth to start from, which in the present methods starts with estimating cell specific parameters. Within the modelling framework, there are five key parameters for cell j: the expected mRNA content of the cell population p that cell j belongs to, capture chemistry efficiency ^^^^, RT-PCR efficiency ^^^^, its sequencing saturation ^^^^(a measure of the proportion of the library complexity – i.e. number of different transcripts in the library for a cell - that are actually sequenced, which depends on both the sequencing depth (typically common to all cells) and the library complexity for the cell (typically cell specific)), ^^^^ ^^and which are the mean expression and dispersion for gene i in a given cell population p. Estimating Expected mRNA Size of Cell Populations: The simulator treats cells as samples from a cell population, where the mRNA content of cells (the total number of transcripts within a cell) from a given cell population randomly fluctuates around the expected mRNA content of the population. This biological characteristic is of significant importance as it can have a profound impact on the observed counts and subsequent processes in a single-cell experiment. To identify cell populations from a given dataset, the simulator employs the state-of-the-art single-cell quality control (QC), normalization, and cell population identification pipeline described by Lueken et al. 2019. Briefly, QC is performed by filtering out cells with a mitochondrial percentage of UMIs (fraction of counts from mitochondrial genes per barcode, where high fractions are indicative of cytoplasmic mRNA having leaked out of the cell through a broken membrane) greater than 10% and cells with library size larger than three median absolute deviations. The filtered cells' library sizes are then normalized (e.g. by normalising count data using a size factor proportional to the count depth per cell, or using the median count depth per cell in the dataset). In the present method, normalisation is performed using the pooled size factor normalization method developed by Lun et al.2016. Next, the genes exhibiting the top 10% of variance are selected as representative of the cell populations and undergo PCA to reduce the dimensionality to 50 dimensions. Once compressed, a K-nearest neighbours (KNN) graph is constructed based on the Euclidean distance between cells in the reduced space. Finally, the Louvain community detection algorithm is applied to identify the cell populations (Luecken et al.2019). Once the cell populations have been identified, a slightly modified version of the heuristic developed by Ye et al. 2019 is used to estimate the expected mRNA content of each cell population. Instead of estimating cell-specific sampling efficiencies, the expected mRNA content of the cell population itself is estimated. To estimate the expected cell library size of a population ( ^^^^), which serves as a proxy for the mRNA content of that population relative to others, the following assumptions are made. Suppose a cell library is smaller or larger than other cell populations. In that case, it indicates that the mRNA content of the cell population is expected to be smaller or larger compared to the others. Based on this assumption, the simulated mRNA content of a cell population is estimated using the user- defined minimum mRNA content ( ^^^^ ^^ ^^) and maximum mRNA content ( ^^^^ ^^ ^^) in conjunction with the library size. First, a logarithmic transformation (base 10) is performed on all expected library sizes and the minimum observed expected library size ( ^^^^ ^^ ^^) and the maximum observed expected library size ( ^^^^ ^^ ^^) are determined. Then, a library size weight is calculated using the equation shown in equation 1 to obtain the library weight ( ^^^^) for each cell population. Finally, the expected mRNA content ( ^^[ ^^^^]) of a cell population is calculated using equation 2: Sampling Cell Specific Capture Chemistry efficiency: The capture chemistry efficiency of a cell is considered to be a random variable that is not specific to a particular cell population. Instead, it is a technical variable that exhibits random fluctuations around the expected capture chemistry efficiency of a specific single-cell sequencing protocol. In order to simulate this process, the simulator utilizes sampling from a beta distribution. The parameters ^^^^and ^^^^are user-defined and determine the shape of the beta distribution (the default used are 4 and 18). Other parameter values are possible depending on the capture chemistry kit used. The parameters may be set such that the reported capture chemistry efficiency is equal to the mean of the distribution. In other words, the expectation of the beta distribution should be the reported or desired value of the expected capture chemistry efficiency of the experiment. ^^^^~ ^^ ^^ ^^ ^^(∝^^, ^^^^) Sampling Cell Specific RT-PCR. The RT-PCR efficiency of a cell is considered to be a random variable that is not specific to a particular cell population. Instead, it is a technical variable that exhibits random fluctuations around the expected RT-PCR efficiency of the RT-Polymerase used in the experiment. To simulate this process, the simulator employs sampling from a beta distribution, which is parameterized by user-defined parameters ^^^^and ^^^^. By default, these parameters are set to 18, 18, resulting in a beta distribution with an expectation of 0.5. These values enable the simulator to capture the stochastic nature of the RT-PCR, which typically samples about 50% of the population. While users have the flexibility to modify these parameters, it is worth noting that manipulating RT-PCR efficiency is not recommended unless you are simulating an improved RT-PCR. ^^^^~ ^^ ^^ ^^ ^^(∝^^, ^^^^) Sampling Cell Specific Sequence Saturation: For a given cell, cell-specific sequencing saturation occurs, where sequencing saturation represents the percentage of unique transcripts observed in the counts obtained from the sequencing. For a given cell, there is a cell-specific sequencing saturation that is assumed to be a randomly fluctuating variable. However, unlike capture chemistry efficiency and RT-PCR, this is simulated based on the sampling efficiency of the cell. The sampling efficiency ( ^^^^) of cell j is calculated by dividing the cell's library size ( ^^^^) by its population's expected mRNA content ( ^^^^), as shown in equation 3. Using the previously sampled values of ^^^^and ^^^^, a cell's sequencing saturation ( ^^^^) is calculated as the division of ^^^^by the product of ^^^^and ^^^^, as demonstrated in equation 4. Estimating Gene-Specific Parameters for a Given Cell Population: When the simulator is provided with a reference dataset (e.g. a scRNA-Seq reference dataset of a cell line or biological setting (i.e. tissue)), the data is utilized to generate a half-normal sampling distribution for each gene class and fit a mean-dispersion curve. To accomplish this, the simulator follows a series of steps. First, it takes a count matrix X from a single-cell dataset and normalizes it using the 'poscount' library size normalization method from the DESeq package (Love et al.2014). Specifically, ‘poscount’ normalisation works by calculating genes geometric mean across all samples (in this case cells) and dividing gene count by this mean. The median of these ratios in a given sample is taken as the size factor for the sample. Next, it empirically estimates the mean and variance in expression directly from the normalized counts and uses them to estimate the gene expression dispersion. Then, it fits a mean-dispersion curve using Maximum Likelihood Estimation (MLE) to estimate the asymmetric dispersion (aDisp) and extra Poisson (ePois) properties from the observed data. This enables to model dispersion as a parameter dependent on a gene’s mean expression. The estimated aDisps and ePois enable the estimation of a gene dispersion given its mean, allowing the Simulator to sample mean expression and calculate the corresponding dispersion based on the properties of the mean-dispersion curve. Once the mean expression and dispersion values are estimated, the simulator proceeds to scale the mean expression so that the sum total of all genes' mean expression is equal to the expected mRNA content. This scaling enables the biological variance of samples to fluctuate randomly around the expected mRNA content cell population, thereby providing a more realistic simulation of transcriptional behaviour in a single-cell experiment. Theoretical Model of Single Cell Experiments. Single-cell experiments are an invaluable approach for analysing gene expression of cell populations in heterogeneous biological settings and processes. However, the outcomes of single-cell experiments can be affected by a diverse set of sources of variability. This requires the use of a hierarchical model that effectively captures the different levels of biological and technical variance. In addition, to variability, it is important to model the specific type of sampling without replacement that occurs in these experiments, where all transcripts are sampled simultaneously without any dependence between draws. To develop a precise theoretical model for single-cell experiments, to the inventors thoroughly considered and addressed each source of variability and sampling bias, providing a new model that accounts for each of these. The first crucial aspect to consider is biological variability, which encompasses the inherent stochastic differences between cells and represents the true expression distribution we aim to estimate in an experiment. To model this variability, the model uses a Gamma distribution. The Gamma distribution has been extensively used in modelling biological variance in both differential analysis and simulator methods, as demonstrated in previous studies (e.g. Zappia et al. 2017, Baruzzo et al.2020). The Gamma distribution offers significant flexibility, allowing it to effectively capture variations in both highly and lowly expressed genes, regardless of whether they have large or small variances. Once the Gamma distribution parameters for all genes are estimated from single-cell data, the simulator scales them to a given cell population's expected mRNA content, thereby enabling these Gamma distributions to simulate true transcript counts. The second crucial factor to consider is the technical variability originating from the capture chemistry step in single-cell experiments. Capture chemistry is a step of a single-cell experiment that has been improved upon in multiple iterations of single cell sequencing technologies. For example, previous versions like 10xv1 chemistry captured only 10% of the mRNA molecules, while the current 10xv3 chemistry captures approximately 30%. Capture chemistry involves a sampling without replacement process within the experimental protocol, where only a fraction of the mRNA molecules within a given cell is captured. Furthermore, it serves as the initial step in the experiment where enrichment or amplification techniques can be applied, potentially introducing a weighted sampling process. To accurately model the sampling process in the capture chemistry step of single-cell experiments, the chosen sampling distribution must accommodate sampling without replacement while allowing for the potential application of weights. While the Hypergeometric distribution has been used previously (see e.g. Baruzzo et al.2020), it does not precisely reflect the sampling process that occurs during capture chemistry and lacks the flexibility of weighted samplings. However, the MFNH distribution fits the simulators use case as it captures the sampling without replacement process in which the draws are independent, and it can account for variations in the probability of sampling different classes within the specified distribution by utilizing a 'weight' parameter. Increasing the weight increases the probability of sampling the corresponding transcript. By utilizing the MFNH distribution, the simulator can effectively model both the sampling without replacement process in the experiment and the manipulation of a gene's probability of being sequenced. The next significant source of variation to consider is the RT-PCR step, where captured mRNA molecules are converted into cDNA for downstream amplification in single cell experiments. RT- PCR is a critical step in the experiment, and its inefficiency is the primary cause of sparsity in the resulting data. This inefficiency leads to the loss of approximately 50% of the captured molecules, resulting in a significant reduction in the counts, which affects the detection of low-expressed genes. To model this step, the simulation employs a binomial downsampling approach, where the probability that a molecule with n copies has k copies (UMIs) after downsampling (P(X=k), where X~B(n,p)) can be calculated using the probability density function for the bionmial distribution (f(k,n, p)), where n of represents the count of captured molecules for a given gene, and p represents the RT-PCR efficiency. By sampling from a binomial distribution with these parameters to simulate the sequencing pool count, where the sampled value indicates the number of unique molecular identifiers (UMIs) that remain for cDNA amplification and subsequent sequencing. Finally, cDNA amplification bias is another source of variability that arises from biases introduced during the PCR amplification step, leading to differences in cDNA production and gene expression measurements. To simulate the impact of cDNA amplification and sequencing, it is important to consider the nature of the core reaction driving both of these steps is PCR. PCR exhibits stochastic exponential behaviour, which introduces variability in its efficiency on a per-gene basis, influenced by factors such as GC content and temperatures. In the context of single-cell sequencing experiments, the presence of unique molecular identifiers (UMIs) attached to mRNA molecules ensures the deduplication of reads and the identification of unique transcripts. However, issues arise when trying to detect poorly amplified transcripts due to the Polya process PCR follows, which can result in a decrease in the number of UMIs observed for the affected transcripts, making them more challenging to detect within a given cell. Another important aspect to consider during cDNA amplification is the second opportunity in a single cell experiment to introduce customized amplification and enrichment protocols, similar to the capture chemistry step discussed earlier. To capture these effects and potential experimental alterations, the simulator utilises the MFNH distribution. Using the MFNH distribution, it is possible to incorporate weights to represent experimental alterations applied during cDNA amplification. Higher primer concentrations, for example, can increase the odds of a transcript being sequenced, providing a more accurate representation of the single-cell experimental process. The resulting hierarchical model utilized by the simulator is a Gamma-MFNH-Binomial-MFNH model. Each individual distribution used captures both biological and technical processes occurring at various steps of a single-cell experiment: This model accounts for the estimation of biological variance that researchers aim to quantify, as well as the potential experimental manipulations that can take place during different stages of the experiment. Modelling Biological Variability. Let N be the number of genes and M be the number of cells to simulate, describing the entire count matrix or single experimental condition. Let ^^^^ ^^be a random variable representing the expression level of gene i in cell j(^^ = 1, … , ^^; ^^ = 1, … , ^^). Let ^^^^ ^^be a random variable representing the count value (read or UMI count) of gene i in cell j. In a real scenario, only ^^^^ ^^(the observed value of ^^^^ ^^) is known, while ^^^^ ^^(the realization of ^^^^ ^^) is unknown and to find it is often the primary objective in a scRNA-seq experiment. In the present method, for a given gene i, the expression levels ^^^^ ^^are modelled using a gamma distribution, where ^^^^is the "average" expression level of gene i and Φ^^is a parameter describing the biological variability in the expression level of gene i: Modelling Capture Chemistry. Next, the previously sampled biological variance for cell j is sampled to create its capture molecule pool ^^^^using the MFNH parameterized by: the ^^^^gene expression vector, the size of cell j capture molecule pool ^^^^which is calculated in the following manner ^^^^=∑^^^^∗ ^^^^, and ^^^^, which is a vector containing weights that represent any potential experimental enrichment of a particular set of transcripts being applied at the capture chemistry step of the simulation. By default, all ^^^^are set to 1, which simulates no biasing of the sampling process. Sampling the MFNH creates the vector ^^^^which is the Captured Pool of ^^^^. This can be expressed formally as: ^^^^~ ^^ ^^ ^^ ^^( ^^ = ^^^^, ^^ = ^^^^, ^^ = ^^^^) Modelling Reverse Transcription of mRNA. The captured transcripts in ^^^^are now transformed into a cell j sequencing pool, denoted as ^^^^. This pool represents the number of transcripts successfully converted into cDNA molecules. To simulate this conversion process, we sample the cell j capture pool vector ^^^^using a Binomial distribution. Here, the number of trials denoted as n, corresponds to the individual gene's molecule count ^^^^ ^^, while the success probability, denoted as p, is parameterized by the cell j RT-PCR efficiency ^^^^. We take one sample per gene, and the resulting sampled value represents the number of mRNA molecules for a specific gene creating the sequence pool vector ^^^^which represents mRNA converted to cDNA, as shown in the following equation: ^^^^~ ^^ ^^ ^^ ^^ ^^( ^^ = ^^^^, ^^ = ^^^^) Modelling Copy DNA Amplification. Finally, the cDNA molecules in the ^^^^are sampled to create cell j observed counts ^^^^. To do this sampling, the model uses an MFNH distribution that is parameterised so that: ^^^^are the populations, ^^^^is the number of samples to draw which is calculated by ^^^^= ∑ ^^^^∗ ^^^^, and ^^^^is a weight vector representing any potential experimental enrichment of a particular set of transcripts being applied during cDNA amplification. By default, ^^^^are all set to 1, which simulates no biasing of the sampling process. This is expressed formally as: ^^^^~ ^^ ^^ ^^ ^^( ^^ = ^^^^, ^^ = ^^^^, ^^ = ^^^^) Simulating Targeted Transcriptomes. In these examples, targeted transcriptomes are simulated by modifying the ^^^^during the cDNA amplification step. When a gene is part of a panel, its weight is set to a default value of 30. If a gene is not part of the targeted panel, its weight is set to 1. This approach allows for flexibility and enables the model to simulate slight off-target effects, although these effects may differ and vary for each gene from the weights currently used. The default weights of 30 for target panel genes were selected utilising Mean Square Error to assess the distance between the log-transformed observed means of the simulator’s simulated transcriptomes and the observed mean expression of targeted panel data from Schraivogel et al 2020 (see Fig. 22). A similar approach can be used to modify the capture chemistry weights for targeted transcriptomes that are obtained with a target specific capture step. Note that targeted transcriptomes are not weighted transcriptomes as described herein as no target specific weights are applied. All of the protocols simulated use the same normal 10x kit until the cDNA amplification step, in which target specific (non-weighted) primers are used. Simulating Weighted Transcriptomes. In the simulator, there are two steps in the single-cell sequencing experiment where enrichment protocols can be applied: capture chemistry and cDNA amplification. How and at which step the gene-specific odds are applied depends on the type of experiment that is being simulated. To compute the weights for a weighted transcriptome, a reference dataset or a pilot experiment are used, from which an observed relative frequency can be obtained. Currently, the weighted transcriptomes are calculated using a target frequency, denoted as t. This target frequency is determined as the reciprocal of the number of non-zero 1 expressed genes ( ^^ = ^^ ^^ ^^ ^^ ^^ ^^) in the pilot experiment or reference dataset. To simulate a weighted transcriptome where only one of the steps is weighted (i.e. to conduct either a weighted chemistry or library experiment), the following process can be applied. The weight for gene ^^^^is calculated by dividing the desired target relative frequency by the observed relative frequency of gene ^^^^. The observed relative frequency of a gene is calculated by dividing a given gene's mean expression ^^^^by the sum of all genes' mean. To normalize the weights within the range of 0 to 1, all weights are divided by the maximum calculated weight (with a default maximum weight value of 100, so any weights exceeding this value are automatically set to 100). This can be calculated rather straightforwardly using the following equations: To correct for bias in All Steps (i.e. both capture and amplification) of a single-cell experiment, we first calculate the weights ^^^^(collectively the weighted mean, ^^^^) as described above. Next, in order to determine the weights during cDNA amplification ( ^^^^), an estimate of the expected number of captured molecules ( ^^[ ^^^^]) is required. To calculate this, the expected number of captured molecules in ^^[ ^^^^] (calculated as the expectation of the MNFH distribution for captured molecules given a number of transcripts) is multiplied by the expected RT-PCR efficiency (set to 0.5 in the present examples). Subsequently, we use the equations 5 and 6 above to calculate ^^^^. Evaluation of model performance. Performance was evaluated by comparing synthetic count tables produced to real count matrices in terms of: - distributions of count intensity, - distributions of count variability, - level of sparsity, - count intensity of a gene calculated as the log-transformed mean of the normalized counts across all samples, - count variability of a gene calculated as the variance of the normalized counts across all samples, - global sparsity of the count matrix calculated as the percentage of zero values in the matrix, and - sparsity by gene and cell was calculated as the percentage of zero count values. These metrics are described in more detail in Baruzzo et al., 2020. In particular, to evaluate the performance of the simulator and compare it to count matrices generated by other simulators and real datasets, we employed comparison methods similar to those used in Baruzzo et al., 2020. The performance was assessed by comparing the statistical properties of simulated counts with real count matrices in terms of gene sparsity, dispersion, and log-transformed mean expression on a per dataset and cell population basis. Gene sparsity was measured as the fraction of zeros observed in cells belonging to a specific cell population. Gene dispersion was calculated as the variance of the normalized counts for each gene within a cell population. Finally, genes' mean expression values were calculated and subsequently log-transformed. To analyze gene sparsity and dispersion, we estimated the distributions using KDE and visualised them using violin plots. These plots effectively illustrate the variations in these statistical characteristics across different simulators and real datasets. Additionally, we generated a two-dimensional KDE estimate to explore the relationship between sparsity and log mean expression. The present simulator was compared to two state of the art simulators: SPARSim (Baruzzo et al., 2020) and Splatter (Zappia et al., 2017), used with their default parameters. Both of the simulators were used with their default parameters. SPARSim is a simulator that uses a Gamma-Multivariant Hypergeometric Distribution to model single-cell data, with a gamma distribution for biological variance and a multivariant hypergeometric distribution for technical variability. Splatter simulates gene means expression levels from a Gamma distribution, and the corresponding count values are obtained from a Poisson distribution. In addition, Splatter simulates high-expression outlier genes and enforces a mean-variance trend previously utilized in bulk RNA-Seq simulations. Finally, a logistic function describes the relationship between gene expression level and sparsity per gene. This function is used to calculate the probability of a dropout event. A notable limitation of Splatter is its inability to account for the dynamics of a sampling process, which the simulator described herein can naturally accommodate. Data. The datasets used in this study were obtained from Wu et al. (2019), Zhao et al. (2020), and Baron et al. (2016). These datasets were preprocessed using standard single-cell quality control and normalization procedures with scater and scran (Mc-Carthy et al., 2017; Lun et al., 2016). The first four cell populations were used for each dataset. This covered a diverse range of single-cell sequencing modalities and biological variance. Results are only shown for the Baron et al. data but similar results were obtained for the other datasets. Results Statistical comparison to other simulators – Validation of Statistical Characteristics of Simulations scRNA-Seq simulations are a valuable tool for researchers developing methods of analyzing single-cell gene expression. However, accurately capturing key features of real scRNA-Seq data is essential for simulations to be useful. In this section, the inventors performed a statistical comparison of the new simulator to two other simulators: SPARSim and Splatter. The primary objective of this analysis was to assess the performance of the present and comparative simulators in capturing three key relationships found in real scRNA-Seq data: the distribution of sparsity, the distribution of the biological variance of genes (commonly referred to as gene dispersion), and the relationship between a gene’s sparsity (sparsity is used as a proxy for technical dropouts) and mean expression (which shows a decrease in sparsity as mean expression increases). Violin plots were used to show these distributions, enabling comparison between the real distributions and the corresponding ones obtained by each of the simulators. Looking at the distribution of sparsity for each dataset and each cell population (Figure 4, top), the method described in this example and SPARSim performed extremely well, with near-identical violin plots and high overlap with the real datasets. However, Splatter displayed a greater instability in the sparsity it generated in lower-depth sequencing datasets generated using 10x sequencing and showed inconsistent performance on these datasets. Looking at gene dispersion (Figure 4, middle), the performance of the method described in this example and SPARSim was significantly better than Splatter. Splatter consistently overestimated gene dispersion (distribution consistently exceeding that of the real data and the other simulators). The method described in this example and SPARSim had comparable performance in many cases, but there were instances where the present method performance appears to be slightly better than SPARSim’s. To compare the performance of the simulators in how they capture the relationship between gene sparsity and mean expression, kernel density estimates were fitted to each of the datasets and the probability of getting a value was plotted as a series of topologically shaded colours (2-dimensional kernel density plots showing the distribution of the percentage of sparsity by gene as a function of log mean expression of the gene, each point being coloured by the probability of observing a value for the gene). The results show (Figure 11) that Splatter did not fit the real data well, with its kernel density being tight and shifted to higher sparsity and lower mean expression per gene compared to real datasets. The new simulator and SPARSim followed the real data’s kernel distribution nearly perfectly. However, SPARSim’s kernel contained a series of unexplained outliers in the at least one of the datasets, while the new method did not have such outliers. This suggests that the new method is slightly better than SPARSim at explaining the relationship between gene sparsity and mean expression. The new simulator (Minerva) consistently captures the distribution of sparsity by gene per cell population across various datasets, as evidenced by the near-identical violin plots and substantial overlap with real datasets. This ability to accurately model gene sparsity is also observed in SPARSim and Splatter (see Fig 12a and Fig 20a). Nevertheless, the fidelity of simulated data sparsity is influenced by factors such as sequencing depth, capture chemistry, and biological context. Notably, when utilizing the Zhao et al., 2020 data as input, none of the simulators performs exceptionally well when compared to the real data. SPARsim and Minerva tend to underestimate sparsity, whereas Splatter tends to overestimate it (see Fig 21a). Conversely, in the Baron et al., 2016 dataset, all simulators accurately reflect gene sparsity (see Fig 12a). Collectively, these findings demonstrate that Minerva is capable of simulating gene sparsity across diverse biological settings; however, the quality of the dataset significantly impacts its simulation capability. The second comparison focused on gene dispersion, which quantifies the biological variability of genes. In this comparison, both Minerva and SPARSim outperformed Splatter significantly. Splatter consistently displayed an overestimation of gene dispersion or exhibited a distinct bimodal distribution with a cluster of high dispersion and another cluster with no gene dispersion (see Fig 12b and Fig 20b). This is evident in its distribution, which consistently deviated from the real data and the other simulators across all datasets and cell populations. While Minerva and SPARSim exhibited comparable performance overall, there were instances where Minerva's performance appeared slightly superior to SPARSim's. Specifically, across most cell populations of Macosko et al., 2015, SPARSim consistently underestimated gene dispersion (see Fig 12b). Conversely, Minerva did not exhibit any pronounced discrepancies in modeling gene dispersion compared to other simulators, although minor differences were observed between the simulated gene dispersion of Minerva and the real data in Zhao et al., 2020. These findings suggest that Minerva accurately captures gene dispersion compared to real datasets and other simulators. The final comparison delves into the relationship between gene sparsity and mean expression, which is the key relationship in scRNA-Seq data, as gene sparsity and mean expression are inversely related with sparsity decreasing with mean expression and vice versa. To evaluate the performance of the three simulators, the inventors utilized Kernel Density Estimates (KDEs) for each dataset and cell population, representing the probability of a given value through a series of topologically shaded colors (see Fig.11 and Fig 21). This analysis reveals that, once again, while Splatter roughly captures the overall characteristics, it does not reproduce them perfectly or as effectively as Minerva. Splatter's KDE is concentrated and shifted towards higher sparsity and lower mean expression per gene compared to real datasets. In contrast, Minerva and SPARSim closely resemble the KDE of real data. However, SPARSim's KDE exhibits a set of unexplained outliers in the Baron et al., 2016 and Macosko et al., 2015 datasets, while Minerva does not have such outliers (see Fig.11 and Fig 21). Moreover, all simulators overestimate the majority of the mean expression distribution in comparison to real data for Zhao et al., 2020. However, each simulator demonstrates different behavior: Splatter significantly underestimates mean expression, while both Minerva and SPARSim overestimate it, although Minerva's overestimation is the closest to the real data among the three (see Figure 11). These findings affirm that Minerva adeptly models the relationship between gene sparsity and mean expression across diverse biological settings and experimental contexts. These results demonstrate that Minerva accurately captures key features such as sparsity distribution, gene dispersion, and the relationship between gene sparsity and mean expression. Minerva performed on par with its primary rival, SPARSim, and surpassed the state-of-the-art simulator Splatter across these major metrics. These results demonstrate that Minerva is a reliable and accurate tool for researchers who wish to simulate gene expression at the single-cell level. Simulation of targeted transcriptomics data Single-cell targeted panel sequencing techniques, like TAP-seq, represent a recent advancement protocoled to improve the detection of genes expressed at low levels. These methods can be conceptualized as a binary manipulation of the likelihood of sequencing gene transcripts, depending on whether a gene is part of a specific targeted gene panel or not. By concentrating cDNA amplification and sequencing on a specific gene panel, this technique enables comprehensive characterization of their expression distribution. Targeted panel sequencing (TAP- Seq, Schraivogel et al., 2020) is a scRNA-Seq technique designed to increase the representation of selected genes including lowly expressed genes. To the best of the inventors’ knowledge, the new method described herein is the only simulator capable of simulating this data type. Here the inventors examine the ability of Minerva to simulate gene sparsity, dispersion, and the relationship between sparsity and mean expression using whole transcriptome data obtained from Schraivogel et al. This simulator was used to explore the relationship between gene targeting and mean expression levels to understand better the capabilities and limitations of TAP-Seq. Targeted and whole transcriptome data from Schraivogel et al., (2020) were compared to the data obtained using the methods described. The use of targeted transcriptome data was paired with whole transcriptome data, enabled the inventors to compare Minerva's simulated targeted transcriptomes to a ground truth. The primary objective is to validate Minerva's capability to accurately reproduce the statistical characteristics specific to targeted transcriptome data derived from whole transcriptome datasets The distribution of percentage of sparsity by gene was compared using violin plots as described above. The results on Figure 5A show that the present method could nearly perfectly simulate the sparsity of targeted transcriptome data. Additionally, both the targeted data simulated using the present method and targeted transcriptome data showed a substantial decrease in sparsity, which is the main goal of targeting panel sequencing. These findings demonstrate the effectiveness of the presently described method in simulating this type of data and the potential benefits of using targeting panel sequencing for scRNA-Seq studies. When comparing gene dispersion, the simulated targeted data again matched the real targeted data almost perfectly (Figure 5B). Both the simulated and targeted panels showed an overall increase in the dispersion of genes which comes from a more accurate estimation of a gene’s mean expression, typically from the better representation of the gene (increase in the number of detected UMIs).When comparing the relationship between gene dispersion and mean expression (again using 2-D kernel density plots), the data show a substantial overlap between the simulated and real targeted transcriptome data (Figure 5C). These results show that the present method is able to simulate targeted scRAN seq data, and that targeting specific genes can reduce sparsity and improve accuracy of single cell gene expression measurements. Initially, the inventors compared the gene dispersion distribution of real targeted transcriptome data and Minerva's simulated targeted transcriptome using the paired whole transcriptome as a reference (see Fig 5a). The results depicted in Figure 5a indicate that Minerva's simulated targeted transcriptome successfully generates realistic gene sparsity compared to the real targeted transcriptome data. Both the Minerva simulated and targeted transcriptome data display a significant reduction in sparsity, which aligns with the primary objective of targeting panel sequencing. However, there are discernible differences. Specifically, Minerva's simulated targeted transcriptome exhibits an elongated violin plot with a lower density of genes exhibiting near 100% sparsity. This observation suggests that Minerva's approach to simulating targeted transcriptomes, particularly in terms of probe efficiency and off-target behavior, may differ from real targeted transcriptomes. Next, the inventors compared the gene sparsity in real targeted transcriptome data, whole transcriptome data, and the simulated targeted transcriptome generated by Minerva. The results of this comparison revealed excellent performance from Minerva, as the simulated data nearly perfectly matched the targeted transcriptome data. The violin plots in Figure 5b show that Minerva's simulated data exhibit a slightly higher dispersion estimate than the real targeted data, which is likely due to miscalibration in the off-target effects and probe efficiency during the simulation process with Minerva. Despite these slight differences, both the simulated and targeted panels exhibited an overall increase in gene dispersion. This increase can be attributed to a more accurate estimation of a gene's mean expression, usually resulting from a better representation of the gene due to an increase in the number of detected UMIs. Consequently, more of the expression distribution can be observed. These results demonstrate the effectiveness of Minerva in simulating targeted panel sequencing data and highlight the potential benefits of this technique for improving the accuracy of gene expression measurements. Finally, the inventors examined the relationship between gene sparsity and mean expression in the targeted transcriptome, both real and simulated, in comparison to the whole transcriptome. To accomplish this, the inventors generated two-dimensional KDE density plots. The results of this analysis, illustrated in Figure 5c, reveal a substantial overlap in density between the Minerva simulated data and the targeted transcriptome data. Furthermore, significant differences are observed between the real and simulated targeted transcriptomes compared to the whole transcriptome. However, it is important to note that Minerva exhibits a higher estimated mean expression and reduced sparsity, evident in the slight leftward and downward shift compared to the actual targeted transcriptome data. This discrepancy is attributed to probe inefficiencies, off- target effects, and offers and interesting opportunities that suggest Minerva's simulated target panels can serve as a theoretical benchmark. Benchmarking Target Panels via Simulation Single-cell targeted panels have emerged as a promising experimental approach to address inherent limitations in single-cell experiments, particularly the sparsity of gene expression and the necessity to enhance the observed mean expression of genes within target panels. This is achieved through the use of capture probes or custom PCR primers, selectively capturing transcripts corresponding to genes within a specified target panel. The primary objective of employing targeted panels is to augment the observed mean counts of targeted genes, thus bolstering statistical power in downstream analyses relative to the cost and required sequencing. However, current evidence does not support the notion that targeted panels effectively enhance the observed mean counts of the specified gene panel. To evaluate whether the observed count of a gene is more efficient than a whole transcriptome experiment, the inventors conducted a straightforward comparison by examining if the observed counts of its target panel were higher than those of the whole transcriptome (refer to Fig 13a). The figure clearly indicates that the majority of observed mean counts for genes in the target panel are lower than those in the whole transcriptome data. Notably, the whole transcriptome data is deeply sequenced, and these results suggest that while target panels offer similar performance at a reduced sequencing cost, they often fall short in elevating the observed mean counts of genes. Exceptions do exist, as a minority of genes exhibit a notable increase in observed mean counts, implying that observed outcomes may be linked to inefficiencies in the PCR primers employed for targeted amplification. Despite the identified inefficiencies in Single Cell Targeted Panel experiments, optimizing the experiment with a deeply sequenced whole transcriptome reference set for comparison is cost- prohibitive due to the extreme depth of sequencing required. Minerva provides a theoretical benchmarking approach for targeted panels, effectively simulating the statistical characteristics of deeply sequenced whole transcriptomes when appropriately parameterized. As a proof of concept, the inventors compared the target panel to Minerva's simulated targeted panel and observed a remarkably similar correspondence to the whole transcriptome (see Fig 13b). To further validate this, the inventors plotted the simulated target panel against the whole transcriptome's observed mean counts and observed a strong correlation between the two (see Fig 13c). This suggests that Minerva can generate simulated targeted panels serving as a bare minimum benchmark for target panel performance using reference whole transcriptome datasets from public data. While single-cell target panels were developed to overcome the limitations of single-cell sequencing experiments, questions persist, particularly regarding probe efficiency when utilizing custom PCR primers. In the previous section, the inventors validated the statistical characteristics of Minerva's simulated target panel compared to TAP-Seq data generated by Schraivogel et al. In this section, the inventors delve deeper into this analysis, demonstrating that disparities between simulated and observed TAP-Seq data are attributable to PCR primer inefficiencies. The inventors propose that Minerva's simulated target panels can serve as a theoretical benchmark for evaluating the performance of target panels, shedding light on the impact of PCR primer efficiencies in interpreting experimental results. Weighted sampling of gene transcripts – Removing sparsity from single cell experiments The above shows that the simulator described is able to faithfully recapitulate the statistical properties of current single cell experiments and binary manipulations of sequencing probabilities in targeted transcriptomes. In particular, the inventors wished to explore theoretically possible single-cell experiments involving continuous manipulations of sequencing probabilities. A continuous manipulation would finely adjust the probability of sequencing specific gene transcripts across all expressed genes. Continuous manipulation of sequencing probabilities offers a potential solution to one of the significant challenges faced in current single-cell experiments: the technically induced gene sparsity, which disproportionately affects genes expressed at low to medium levels. Unfortunately, many biologically interesting gene types, such as transcription factors, kinases, receptors, and others, fall into this category. As a result of their low expression, these genes often remain concealed, making it arduous for single-cell experiments to characterize them accurately. Minerva's hierarchical model for simulating each technical step of the experiment reveals two significant stages: Capture Chemistry and cDNA Amplification, during which the sequencing probabilities can be altered. These stages give rise to a range of theoretically possible experimental protocols, namely Non-Weighted, Weighted Library, Weighted Chemistry, and Both Steps Weighted. Therefore, the model was used to explore various theoretically possible single-cell experiments, using real counts data from various datasets (as explained in Methods above) to parameterize the simulations. Many genes of very high biological interest (such as transcription factors, kinases, receptors) are expressed at low levels and therefore often not detected in scRNAseq experiments. A series of potential enrichment protocols were explored using the present methods: - Non-Biased Protocol (also referred to herein as “Non-Weighted experiment”): this is equivalent to current non targeted scRNAseq protocols – note that that current targeted scRNAseq protocols do apply some enrichment but although only some genes may be selected, these are all selected to the same extent (i.e. there is no tailored enrichment depending on the expected expression level of the genes); - Biased cDNA Amplification Protocol (also referred to herein as “Weighted Library”): primer concentrations are calculated for the PCR reaction (or any other alternative method such as using biotinylated hybridisation probes which can be enriched via an antibody pulldown) to obtain approximately one-to-one ratios of sequencing-able reads after cDNA Amplification; in other words, Weighted Library manipulates the ratio of PCR amplified reads, achieved through antibody pull-down or custom PCR probes. This method focuses on adjusting the read-to-UMI ratio to increase or decrease the probability of sequencing a specific UMI- labeled transcript; - Biased Capture Chemistry Protocol (also referred to herein as “Weighted Chemistry”): mRNA probe concentration is calculated to enrich low-expressed genes and reduce high- expressed genes in terms of the number of transcripts to be reverse-transcribed and sequenced; in other words in a Weighted Chemistry experiment, the manipulation relies on probe-based modifications of mRNA molecule capture probabilities during the Capture Chemistry step of scRNA-Seq experiment; and - Biased All Steps Protocol (also referred to herein as “Both Steps Weighted”): Primer / probe concentrations are calculated for the Capture Chemistry and PCR steps; in other words, Both Steps Weighted of experiments manipulate the probability of capturing a molecule and the read-to-UMI ratio. To assess the potential of these experimental protocols, the inventors implemented a straightforward heuristic for computing transcript weights. This heuristic is grounded in predefined 1 target frequencies, set to ^^ ^^ ^^ ^^ ^^ ^^ (where nGenes is the number of genes measured), as outlined in the Methods above for comprehensive details. Adjusting the target frequency in this manner allows to naturally assign higher weights to genes with moderate and low expression levels while assigning lower weights to highly expressed genes. In a uniform cell population, this weighting strategy would compel all observed mean counts per gene to be equal. However, in a diverse environment like tissue, the inventors anticipate variations as the relative frequency of each gene in a cell population naturally fluctuates. Exploring the efficacy of Weighted transcriptomes in mitigating gene sparsity Here, the inventors compared the impact of continuously weighted experimental protocols on the distribution of gene sparsity observed in simulated counts in comparison to Non-Weighted single- cell experiments. Looking at the distribution of percentage of sparsity by gene (Figure 6A), the results show that substantial reduction in sparsity can be achieved through any of the enriched protocols described, compared to a normal non-biased single cell experiment. In other words, all experimental protocols that continuously manipulated sequencing probabilities exhibited significant reductions in gene sparsity compared to the standard Non-Weighted protocols (see also Fig 14a). This is particularly apparent for the biased chemistry and biased all steps protocols (i.e. Weighted Chemistry and Both Steps Weighted simulated counts displayed the most substantial decreases in observed sparsity). This is believed to be because these methods increase the number of gene transcripts fed into RT-PCR. Since RT-PCR is the most wasteful part of a Single Cell Experiment, (on average the step is only 50% effective) any increase in the number of transcripts per gene going into the RT-PCR improves the probability of being sequenced, whereas after RT-PCR there is only a limited ability to bias cDNA amplification to reduce sparsity (because many genes may have already been lost irretrievably by that point). In other words, the low efficiency of RT-PCR limits the capacity to bias cDNA amplification steps during or after PCR to minimize sparsity relative to the number of transcripts that passed through RT-PCR. Therefore, any increase in the number of transcripts per gene that successfully undergoes RT-PCR has the largest impact on improving the probability of being sequenced and observed in the experiment. Consequently, it is evident that the most effective approach to achieve the greatest reduction in gene sparsity is to focus primarily on enhancing or manipulating the Capture Chemistry and RT- PCR steps within single-cell experiments. Next, the inventors explored how the weighted experimental protocols may warp the gene dispersion compared to current Non-Weighted protocols. Looking at dispersion (Figure 6B), the results show surprisingly little general warping of the gene dispersion in weighted data compared to normal data. See also Fig.14b, i.e. contrary to initial expectations, there appears to be little to no substantial general warping of the overall gene dispersion distribution in continuously manipulated protocols. This is believed to be due to the increased representation of low and medium-expressed genes and the nature of sampling without a replacement process. However, it is still possible that highly expressed genes may still suffer from an artificially reduced gene dispersion. Such potential warping of gene dispersion will also vary based on when the weighting is applied in the experiment. Despite these concerns, the results are extremely promising and suggest that weighted methodologies can enrich low and medium-expressed genes while preserving their dispersion. Finally, the inventors explored the effect Weighted Transcriptomes had on the gene sparsity and observed the mean count relationship. Looking at the relationship between sparsity and log-mean expression (Figure 6C), the results show that all weighted methods increase the log-mean expression and reduce sparsity simultaneously (see also Fig.15, which shows that all weighted methods exhibited simultaneous increases in mean expression and reductions in sparsity). Although the Biased Library protocol did increase log-expressed genes and reduce sparsity, the over- all pattern of the log-mean sparsity distribution observed in the Non-Biased Experimental protocol remained unchanged. On the other hand, compared to the Non-Biased distribution, the Biased Chemistry and All Steps Biased protocols exhibited the most significant changes, with substantial sparsity reduction and a tighter log mean expression distribution due to weighting to the Capture Chemistry. This, combined with the preserved gene dispersion, suggests that Biasing of Capture Chemistry and Biasing All Steps experimental protocols can enrich low-expressed genes while preserving the dispersion. As shown on Fig.15, Particularly in the lower range of mean expression for all weighted protocols, there was a noticeable shift towards the left and downward direction. However, the degree of improvement varied among the weighting protocols. While the Weighted Library protocol did elevate the mean expression of genes and decrease sparsity, the overall pattern of the mean sparsity distribution observed in the Non-Weighted Experimental protocol remained unchanged, especially in terms of the distribution tail of highly expressed genes (see Fig 15). Meanwhile, Weighted Chemistry protocols exhibited more significant changes than the Non-Weighted and Weighted Library distribution, with substantial reductions in sparsity and a narrower distribution of mean expression, effectively removing the tail of highly expressed genes observed in Weighted Library and Non-Weighted data. This observed reduction can be attributed to the weighting applied to the Capture Chemistry stage (see Fig 15). The Both Steps Weighted protocol simultaneously manipulates both Capture Chemistry and cDNA Amplification steps of the scRNA-Seq experiment with the goal of increasing the probability of observing a given gene transcript. The inventors observed a significant increase in the mean counts of transcripts and a substantial reduction in gene sparsity to a greater degree than any other weighted protocol, including Weighted Chemistry. This is illustrated by a substantial decrease in the lower observed mean counts tail in Fig 15. However, this enhancement comes at the highest cost compared to other weighted protocols, significantly constraining highly expressed genes, which enables a more pronounced reduction in observed mean expression in the most highly expressed genes compared to other protocols due to the applied weighting across multiple steps. Consequently, Both Steps Weighted Single Cell Experiments effectively minimize gene sparsity but may restrict the observed mean counts of highly expressed genes. Weighted transcriptomes increase the overall observable number of genes in single cell transcriptomics experiments A significant limitation of current single-cell experiments is that the number of observable genes is primarily limited to genes with a high mean expression. Therefore, proper characterization of gene distributions of medium to low expressed genes due to the inherent "winner takes all" effects of sampling without replacement is impossible. The inventors hypothesized that continuous manipulation of the probability of sequencing a transcript should result in an increase in both observed mean expression and the number of observed genes compared to Non-Weighted experiments. To test this, they examined how the number of genes with a non-zero count increased as a function of sequencing saturation (Figure 6D). The inventors found that, across all weighted protocols, there was a general augmentation in the number of genes with non-zero counts, regardless of sequencing saturation, compared to Non- Weighted experiments (refer to Fig 6D). Biasing cDNA Amplification demonstrated an initial significant increase in the number of observable genes, particularly in lower sequencing saturations. However, as sequencing saturation increased, the observed count of Weighted libraries eventually plateaued and approached that of the Non-Weighted experiment. This suggests that there is an effective limit to the ability of Weighted Library protocols to enhance the observability of low and medium-expressed genes effectively (see Fig 6D). While Weighted Chemistry and Both Steps Weighted protocols exhibited a substantial increase in the number of detectable genes compared to both Weighted Library and Non-Weighted protocols from the outset, this increase persisted as sequencing saturation increased, eventually plateauing closer to a sequencing saturation of 0.9 and 1 (see Fig 6D). A significant difference can be observed between Both Steps, Weighted protocols and Weighted Chemistry protocols, regarding the number of detectable genes, particularly in the second and fourth cell populations. However, in first and third cell populations, the difference between Weighted Chemistry and Both Steps Weighted drops. It remains unclear why this is the case and whether the differences in the number of observed genes are driven by inherit differences in the cell populations or another reason. Despite this, the overall results indicate that single-cell experiments employing Both Steps Weighted protocols can achieve the most significant reduction in sparsity, increased observed mean expression, and the largest increase in the number of non-zero expressing genes. However, it's worth noting that fine-tuning the optimal weights may be useful to harness these benefits fully. Discussion Existing scRNAseq data simulators aim to provide a ground truth for benchmarking that accurately recapitulates the statistical characteristics observed in Single Cell Data. A new simulator was developed with a different purpose than other simulators in the single-cell biology ecosystem. Its main goal is to be a wet lab-aware simulator that accurately reflects the statistical characteristics of single-cell data while considering the potential experimental manipulations that could be performed. This feature allows users to explore the theoretical space of single-cell experiments and quickly observe the statistical characteristics of the resulting data. The simulator can simulate various single-cell experiments, including targeted or weighted transcriptomes, using normal scRNA-Seq as input. This capability enables the user to efficiently search for experimental protocols that optimize single-cell data to reduce sparsity and mean intensity. The data provided in this example shows that the simulator is able to recapitulate normal scRNA- Seq data faithfully (equally well or better than existing state-of-the-art simulators). The data further shows that the simulator can simulate targeted panel scRNAseq data without any priors beyond the target gene list. Finally, the simulator was used to explore the space of theoretically possible single-cell experiments to search for experimental protocols that could reduce sparsity and increase detectable mean expression in genes with low expression. This showed that three enrichment approaches tested (Biased library -also referred to as biased cDNA amplification and Weighted Library, Biased chemistry also referred to as Weighted Chemistry, and all steps biased also referred to as Both Steps Weighted) can effectively reduce sparsity and increase the detectable mean expression. Biased Chemistry (Weighted Chemistry) and All Steps Biased (Both Steps Weighted) protocols showed additional advantages over the Biased Library (Weighted Library) protocol, likely due to the enrichment occurring after RT-PCR in the former. RT-PCR is only about 50% efficient in converting mRNA transcripts into cDNA, making it functionally similar to a filter or a shedder, which removes nearly half of the captured transcripts from being sequenced in a single-cell experiment. This has a greater impact on the lowest expressed genes. Additionally, the Biased Library step enriches reads generated from PCR, shifting the nature of the enrichment from sampling without replacement to sampling with addition, also known as the Polya urn problem. This subtle shift is because we are biasing a PCR reaction with uniquely labelled transcripts, which means we are altering the sequencing library’s composition while we are enriching reads of those transcripts. This results in an increase in the probability of observing low-expressed gene transcripts that have made it through RT-PCR, but it doesn’t enrich the transcripts themselves. Despite these limitations, Biasing the library step of a Single Cell Experiment is easier than Biasing the Capture Chemistry and still provides some improvements in terms of sparsity and detectable mean expression. Single-cell simulators are traditionally developed and employed to create benchmarking datasets with predefined ground truths, proving invaluable for method development. Many simulators have been developed, each with varying underlying models, yet they all share the common goal of generating synthetic datasets mirroring the statistical characteristics of real-world scRNA-Seq data. Notable examples of such simulators include Splatter and SPARsim, both dedicated to simulating scRNA-Seq datasets. Despite their effectiveness in generating synthetic datasets that closely resemble reality, these simulators have inherent limitations due to the choices made in their theoretical model for simulating scRNA-Seq. The proposed novel simulator not only accounts for the statistical characteristics of scRNA-Seq datasets but also models each step of the scRNA-Seq experiment while accommodating potential experimental manipulations. Such a simulator not only enables the generation of realistic synthetic datasets but also facilitates theoretical analysis of scRNA-Seq experiments and the potential to optimise experimental protocol against theoretical performance benchmarks. By utilizing the simulator, the inventors were able to explore the impact of different enrichment protocols on the occurrence of technical dropouts within specific experimental protocols, facilitating the search for optimal experimental configurations. To this end, the inventors conducted a series of simulation studies that initially focused on validating the statistical characteristics of datasets generated by Minerva for both typical and targeted scRNA- Seq experiments. Once this validation was achieved, the inventors investigated hypothetical experimental protocols, which are referred to as 'Weighted Transcriptomes.' (see next Example) These protocols centre on using enrichment methods at various stages of the scRNA-Seq experiment in different combinations. Ultimately, the inventors devised three distinct Weighted Transcriptomes: Weighted Library, Weighted Chemistry, and Both Steps Weighted. In the Weighted Library protocol, the inventors simulated the application of enrichment methods exclusively during the cDNA Amplification step. Weighted Chemistry involved applying enrichment to the Capture Chemistry step, while the Both Steps Weighted approach simulated the use of enrichment protocols at both the Capture Chemistry and cDNA Amplification stages. This study illustrates the development of statistical models that accurately mirror experimental protocols and their statistical properties, optimising the experimental protocol for scRNA-Seq experiments. These proposed methods hold the theoretical potential to mitigate technical dropouts, thus allowing scRNA-Seq experiments to comprehensively characterize individual cells' transcriptomes. Example 3 – Statistical properties of weighted single cell data In this example, the inventors used the simulation method described in Example 2 to investigate the properties of scRNAseq data obtained using protocols designed to enrich low expressed genes to address the problem of sparsity of scRNAseq data. They show that the weighted datasets contain more information than the current state of the art (non-biased single cell experiments), while preserving the statistical characteristics of observed counts and without degrading downstream analysis performance. The results in Example 2 have shown that weighted transcriptomes dramatically reduce sparsity and increase mean expression and the observed gene counts for a given single-cell experiment by manipulating the probability of sequencing a gene transcript depending on its relative frequency in the transcriptome. This results in highly expressed genes being unenriched while low and medium-expressed genes are enriched, which improves the overall visibility of the transcriptome in single-cell experiments. Moreover, this approach increases the number of observable genes and the observed mean expression of the enriched low to medium-expressed genes while potentially constraining the mean expression and variance of highly-expressed genes. However, the extent of these effects is unclear. Do they affect only a few or a lot of genes? Are there varying degrees to which genes are affected? Introduction Single-cell experiments provide researchers with a high-throughput and high-resolution view of biological systems, enabling them to identify rare cell populations, characterize tissue properties, discover unique cell states, and study their response to perturbation. However, despite the great promise of single-cell experiments, there are fundamental limitations in their ability to characterize the transcriptome of individual cells. This is due to the inefficiencies in the single-cell experimental protocols, combined with shallow sequencing, which results in sampling without a replacement process that fundamentally limits our ability to detect low-expressed genes, let alone accurately characterize the statistical properties of their gene expression distribution. Here, the inventors proposed the concept of weighted single-cell experiments: a set of theoretical (in this example – experimental implementations are described further below) wet lab protocols designed to enrich low-expressed genes, enabling better characterization and observance of these genes. In Example 2, the inventors showed that these types of experimental protocols effectively remove sparsity from single-cell datasets and increase the observed mean expression. In this example, a more in-depth investigation into the statistical properties of the weighted single-cell datasets is provided, to test whether enrichment protocols alter the statistical properties and information content of these datasets. They confirm that weighted single-cell experiments’ statistical characteristics are like normal single-cell datasets such that current downstream normalisation and analysis methods can be used (with the added benefit of the additional information obtained through the new protocol). To study these properties, the simulation method described in Example 2 was used. As in Example 2, three major types of weighted transcriptomes were investigated: library-biased, chemistry biased, and all-steps-biased. The library-biased protocol applies enrichment methods to the cDNA amplification step of a single-cell experiment. The chemistry-biased protocol is when enrichment is applied to the capture chemistry step of the experiment. Finally, all steps biased protocol is when enrichment is applied to both capture chemistry and cDNA amplification. The experimental protocols were simulated across eight different datasets that are publicly available, spanning different biological contexts. By simulating across such a diverse range of biological settings, the inventors were able to obtain a very good sense of how weighted single-cell experiments affect the statistical characteristics of the observed counts. They show that in weighted single-cell experiments, most genes across the expression distribution exhibit no alteration or effects of manipulating the mean-variance relation. In addition, weighted datasets contain more information than non-biased (normal) single-cell experiments, allowing to better characterize the true expression distribution using observed counts from scRNAseq. Finally, they validate that current single-cell normalization methods are able to remove technical variance and identify cell populations in weighted datasets. No degradation in performance was observed. These results suggest that weighted single-cell experiments do not alter the statistical characteristics of genes observed counts and provide more information on the transcriptome than non-biased (normal) single-cell experimental protocols. In other words, they surprisingly show that weighted single-cell datasets do not violate or alter the statistical assumptions and characteristics assumed during single-cell analysis, and therefore enable user to perform substantially all types of downstream analysis that are currently possible with scRNAseq, with the additional benefit of dramatically enriched information about the transcriptome, particularly in relation to low expressed genes. Methods Simulating Data. Four experimental protocols including were simulated across eight Single Cell Datasets in various biological contexts, using the simulation method described in Example 2. As demonstrated in Example 2, this method is able to accurately simulate targeted transcriptomes (where only selected genes are captured and sequenced, but no relative weighting of the target genes is performed) or weighted transcriptomes (where relative weighting of genes is implemented, whether the protocol aims to capture a full transcriptome or a subset of it – although the benefits of weighted transcriptomes are particularly stringent in cases where large sets of genes are captured, i.e. large panels (multiple hundreds or thousands of genes) or full transcriptome experiments). The simulated experimental protocols were: - Non-Biased: no manipulations - this is a standard experimental approach for scRNAseq; - Library Biased: cDNA Amplification step is weighted either during or after PCR by manipulating primer concentrations during PCR or hybridization probes after PCR; this manipulation can be used to obtain more even composition of reads across genes that are represented in the cDNA (i.e. genes that are successfully converted to cDNA by RT-PCR – which as discussed above is unlikely to include all genes present in the original transcriptome); - Chemistry Biased: this involves weighting of hybridization probes during the Capture Chemistry Step; and - All Steps Biased: this involves weighting both the Capture Chemistry and cDNA Amplification steps. Datasets. The following datasets were used: Baron et al. (2016): Studied the cell populations of pancreatic islets in both mice and humans. Protocol: inDrop, cell count:8569. Macosko et al. (2015): Introduced the Drop-Seq method and utilized it to identify and characterize retina cell populations. Protocol: Drop-seq, cel count: 49300. Jessa et al. (2019): Characterized pediatric cancer brain tumours and studied how different cellular lineages alter tumour behaviour. Protocol: 10x, cell count: 61595. Kotliarov et al. (2020): Utilized CITE-Seq to study the immune system response between vaccinated and non-vaccinated individuals to identify an immune baseline. Protocol: 10x v2, cell count: 58654. Wu et al. (2019): Identified and characterized cell populations in the adult kidney. Protocol: snDrop-Seq, cell count: 17542. Zhao et al. (2020): Explored the cellular heterogeneity of resident immune cells in the liver. Protocol: 10x, cell count: 68100. Zeisel et al. (2018): Studied the mouse nervous system, identified and characterized cell populations, and examined their findings’ implications for the general mammalian nervous system. Protocol: 10x v1, cell count: 160796. Zilionis et al. (2019): Studied tumour-infiltrating myeloid cells in lung cancer in both humans and mice. Protocol: iDrop, cell count: 173954. These datasets span various biological contexts, from paediatric brain cancer to the pancreas, and use different droplet-based single-cell isolation techniques (i.e., Drop-seq, 10x, and inDrop). The datasets were used to identify cell populations, estimate the mean expression and dispersion for each gene, as described below. Describing the Observed Count Distributions - Moment-Based Statistics. To assess the effects of weighted transcripts and identify any potential distortions they may apply to observed count distributions, the first (mean) and second (variance) moments of each gene’s observed count distribution were calculated for any given cell population within a dataset. Single-cell count matrices follow a gene-by-cell format, where the count in a given row represents the expression count of a gene in a particular cell, and the count in a given column represents the expression count of all genes in a particular cell. To calculate the mean and variance, a subset of the count matrix (cell population count matrix) is extracted comprising counts for a given cell population (i.e. counts for all genes in all cells labelled as forming part of a particular population). A cell population is a set of ^^ cells that belong to a given population. Using this cell population count matrix, the empirical mean µ and variance ^^2is estimated for a given gene ^^ by iterating over the rows of the matrix X: where ^^^^, ^^is the expression count of gene ^^ in cell ^^. Once all the gene’s mean expressions were calculated, each gene was characterised by the quantile (quartile) that it belongs to. This was performed as follows. First, all genes with a mean expression less than 0.001 were removed. This removed a small subset of extreme outliers that was on the low end of the estimated mean expression and prevents these from skewing the quartiles to lower values. Then, the genes were ranked by mean expression from lowest to highest and the empirical Cumulative Frequency Distribution (eCFD) was calculated (Dekking, 2005). Finally, the 25th, 50th, and 75th percentiles were identified and genes were assigned to a given quartile. Determining the quartile a gene belongs to allows to explore the effects of weighted transcriptomes on the gene mean and variance as a greater level of resolution (i.e. separately within each quartile), as their effects may not be the same across all of the quartiles of the mean expression distribution. Describing the Observed Count Distributions - Calculating the Log Fold Change in Weighted Transcriptomes. To compare how the different weighting protocols effect mean gene expression (typically increased) and variance (typically decreased), the log-fold change of both the mean and variance between a given weighted transcriptome and the Non-Biased transcriptome were calculated using the following formulae: log(FC(mean)) = log(meanweighted) − log(meannormal) log(FC(variance)) = log(varianceweighted) − log(variancenormal) This log-fold change was calculated on a per-gene, cell population, and dataset basis. Once calculated, a Kernel Density Estimate (KDE) was fitted to these to visualize the changes that occurred and determine the overall directionality of the effects of weighted transcriptomes on gene expression and variance. Describing the Observed Count Distributions - Measuring the Strength of the Relationship Between Observed and True Count. While comparing moments can be informative and provide observational evidence of how the observed count distribution has changed, it doesn’t indicate the amount of information that the observed count distribution contains about the true count distribution. Therefore, the empirical mutual information between the observed counts of each experimental method and the true gene expression distribution was also calculated per cell populating in a given dataset. The formula for calculating mutual information between two random variables ^^ and ^^ is: where ^^ is the observed counts of a particular experimental method for a given cell population in a given dataset, and ^^ is the true gene expression distribution for the given cell population and dataset (simulated “biological variance” representing the true counts for a gene expression, which are downsampled to obtain simulated observed counts). ^^( ^^, ^^) is the joint probability distribution of ^^ and ^^, and ^^( ^^) and ^^( ^^) are the marginal probability distributions of ^^ and ^^ (empirically calculated by counting the number of counts that fall within each of a plurality of bins and estimating the percentage of counts for each bin as the probability for that bin). Calculating mutual information allowed the inventors to explore the mutual dependency between the observed and true data and better understand how different experimental protocols influence the observed counts of single-cell experiments, Mutual Information is a nonlinear measurement of the relationship between the observed and biological variance count distributions; this allows the inventors to capture a wider range of the relations between the two. Although Mutual Information cannot indicate whether the relationship is positive or negative, this is of little consequence for this analysis since the primary question is whether weighted transcriptomes have more information about true gene expression than non-weighted transcriptomes from current single-cell protocols. Assessing Performance of Normalization Methods - Normalization Methods. Normalization is a critical step in the downstream analysis of Single Cell Data Analysis. This normalises the data to account for sequencing depth. There are various approaches for normalizing single-cell data, including delta, residual, and count-based normalization. The delta approach efficiently transforms data by adding a small pseudo count (i.e., 0.001) to the count matrix and applying a log transformation. The optimal statistical environment for applying the delta transformation is when the variance primarily depends on the mean. Within this context, it provides a quick and easy way of removing heteroskedasticity. Mathematically, the delta transformation can be written as: ^^′^^, ^^= ^^ ^^ ^^2( ^^^^, ^^∗ ^^^^+ ^^) where ^^^^ ^^is the count for gene ^^ in cell ^^, ^^^^is the cell specific scaling factor to adjust ^^^^ ^^by, α is the pseudo count, and x’i,j is the transformed value used for downstream analysis. A cell specific factor can be set as the ratio of the cell library size divided by the mean cell library size. Here, variations of the delta approach were tested, including the Sum Pooled Factor and 10k scaling log transformation. These are two different methods of calculating cell specific scaling factors. Sum Pooled Factor works estimates the contribution a given cell to a pseudo cell through summing together gene counts. Linear regression is used to model this and then used to predict the cell specific scaling factor. This accounts for technical differences in cell library size. 10k scaling calculates scaling factors in relation to 10,000 then scales all cell counts to 10,000 (As if all cells have 10k UMIs overall), then log-transforms the obtained values. Residual-based normalization methods stabilize variance by fitting a null model on a gene-specific level (Hafemeister and Satija, 2019). The motivation for developing these methods comes from the inability of the delta approach to deal with genes with an extremely low mean expression (typically with a mean less than 0.01) (Hafemeister and Satija, 2019; Ahlmann-Eltze and Huber, 2023). Once the null model is fitted, the observed count data is transformed into a residual, typically a Pearson residual, which is then used for downstream analysis. The Pearson residual is calculated as the difference between the observed count and the expected count based on the null model, normalized by the square root of the variance of the expected count. Mathematically, the Pearson residual can be calculated as follows: where ^^^^ ^^is the observed count for gene ^^ in cell ^^, ^^^^is the expected count based on the null model, = µ^^+ ^^ ⋅ µ2^^is the variance of the expected count (variance of a negative binomial). Residual-based normalisation was performed using the SCTransform package (github.com / satijalab / sctransform, Hafemeister & Satija, 2019). Count-based normalization takes the observed counts and immediately applies GLM-PCA to identify latent structures for subsequent downstream analysis (Townes et al., 2019). Through a comparison of these orthogonal methods for normalizing weighted single- cell datasets and an assessment of their effectiveness in accurately identifying differentiating features between cell populations, the inventors aimed to determine the capability of current single-cell bioinformatic methods in removing technical variance from weighted transcriptomes. Assessing the performance of normalisation methods - Adjusted Rand Index. scRNA-Seq is primarily utilized to discover novel cell populations and subpopulations, heavily reliant on the ability of current single-cell bioinformatics methods to identify genes with the highest variance. These high-variance genes are then used as features for dimensionality reduction and population visualization. However, the introduction of weighted transcriptomes has the potential to modify the statistical properties, possibly leading to a decrease in the performance of existing single-cell methods. To investigate whether weighted transcriptomes alter or diminish the performance of these methods, the inventors employed the ARI. The Adjusted Rand Index (ARI) was used to evaluate the ability of each normalization method to accurately identify cell populations from the weighted data. The ARI is a statistical measure of the similarity between two clustering results. It considers all pairs of data points and calculates the proportion of pairs assigned to the same cluster in both results, normalized by the maximum possible agreement based on chance. In the present case, the two clusters being compared are the identified cluster by a normalization method and the ground truth cell population that was simulated. The ARI can be calculated using a contingency table, where ^^^^ ^^is the number of data points simultaneously assigned to cluster ^^^^in the first clustering result and to cluster ^^^^in the second clustering result. Let ^^^^be the total number of data points assigned to cluster ^^^^in the first clustering result, and be the total number of data points assigned to cluster ^^^^in the second clustering result. Then, the ARI is given by the following formula: where n is the total number of data points. The ARI ranges between -1 and 1: a value of 1 indicates perfect agreement between the two clustering results, 0 indicates agreement no better than chance, and a negative value indicates disagreement worse than chance (Rand, 1971). ARI is a proven method for evaluating the performance of clustering algorithms and their results. In previous studies, ARI has been used to assess the performance of single-cell normalization methods in unbiased single-cell experiments (Townes et al., 2019). ARI has been used previously to evaluate the performance of normalization methods because single-cell normalisation aims to remove the technical variance introduced during the experiment, enabling users to identify clusters driven by biological variance. In this example, ARI is used to assess the effect a weighted transcriptome has on the ability of current standard Single Cell Normalization methods to remove technical variance and identify biological clusters. Results Assessing the Effects of Weighted Transcriptomes on Gene Expression Mean and Variance. Weighted Single Cell Experiments evaluated here are experimental protocols where the probability of sequencing a gene transcript is manipulated depending on its relative frequency in the transcriptome. Specifically, highly expressed genes are unenriched while lower expressed genes are enriched. This approach increases the number of observable genes and the observed mean expression of lower-expressed genes while constraining the mean expression and variance of highly-expressed genes. To investigate the extent of these effects (i.e. do they affect only a few or a lot of genes, and are there varying degrees to which genes are affected), the simulation method described in Example 2 was used, across a variety of datasets (see Methods). Simulating across a large variety of biological and technical contexts allows to determine the true effects of weighted transcriptomes on observed gene expression by searching for general changes across these settings. The most important relationship that needs to be checked is the mean-variance relationship. In biological settings, it is commonly known that variance is often larger than the mean and increases as the mean increases. If observed count statistics in weighted transcriptomes violate this relationship, it would invalidate most statistical assumptions of current techniques for analysing biological count data (Anders and Huber, 2010). The mean-variance relationship is of utmost importance and requires thorough examination, as it characterizes the gene expression distribution key aspect for downstream bioinformatic methods in both single-cell and bulk RNA-Seq data analysis. These methods heavily rely on assumptions related to this relationship. Notably, it is assumed that as the mean expression increases, so does the variance, with the variance often surpassing the mean (known as heteroskedasticity). Hence, when identifying cell populations, genes exhibiting high variance are utilized for clustering and the identification of cell populations. This choice is based on the hypothesis that genes with high variance represent multiple expression distributions, each corresponding to a distinct cell population. In biological settings, it is widely acknowledged that variance is frequently larger than the mean and tends to increase as the mean expression rises. If the observed count statistics in weighted transcriptomes deviate from this relationship, it would undermine the validity of most statistical assumptions used in current techniques for analyzing weighted transcriptome data (Anders et al., 2010). Therefore, the mean and variance of the simulated Non-Biased and weighted transcriptomes were estimated on a per-cell population and dataset basis. A two-dimensional KDE was then used to create a density estimation of the joint distribution of gene expression’s mean and variance, in order to visualise the mean-variance relationship. Figure 7A (showing the data for the Baron et al. dataset, the work was performed for all datasets, showing substantially similar results) shows that the Non-Biased Single Cell data does not follow the mean-variance relationship observed in the true mean-variance (simulated biological variance mean-variance). This suggests that Non-Biased Single Cell Data (current state of the art) does not fully reflect the true statistical behaviours of genes' biological variance but rather follows a more constrained shadow of the true transcriptome. This is most likely attributed to the technical noise introduced during single-cell experiments, where transcripts undergo multiple rounds of downsampling. As a result, the statistical properties of low to medium-expressed genes are altered, leading to reductions in both the mean and variance of gene expression. However, the overall heteroskedasticity of the mean-variance relation holds for Non-Biased transcriptomes. Figure 7A further shows that when comparing the various weighted transcriptomes (Library Biased, Chemistry Biased, and All Steps Biased), their mean-variance relationship all followed the same general trend as the Non-Biased transcriptome. However, this relationship was more constrained particularly for the upper right outliers (high expressed genes following a regression to the mean effect), implying that a small amount of information may be lost for these very highly expressed genes. This observation aligns with the expectation that weighted transcriptomes would limit the count variance of genes with high mean expression. This constraining of the outliers was not the same across the differently weighted transcriptomes. In particular, the impact of different experimental protocols for weighting the transcriptome on the count variance of highly expressed genes varied, primarily depending on whether the weighting step occurred before or after the RT-PCR stage in the single-cell experiment. Library-biased transcriptomes had the largest outliers among the weighted transcriptomes and were the closest to the Non-Biased transcriptomes compared to Chemistry Biased and All steps Biased. The Chemistry and All Steps Biased outliers were far more constrained. Despite this, the area with the greatest density appears to follow the same trend as the Non-Biased transcriptome. This suggests that genes in the upper percentile are constrained, but the bulk of the distribution is unaffected (i.e. the bulk of genes with a high mean expression appear to be unaffected). To better understand how the effects of weighted transcriptomes may constrain the mean and variance of individual genes, the inventors explored the distribution at a higher resolution. To do this, the inventors estimated the empirical Cumulative Frequency Distribution (eCDF) of the gene expression distribution for each cell population in each dataset and experimental protocol. Using the eCDF genes were grouped into quartiles by converting their mean expression distribution into an empirical percentile. Next, one-dimensional KDEs were estimated for both the mean and variance per quartile and dataset (Figures 16, 7B, 7C). Figures 7B, C show that both the mean and the variance of all experimental protocols, including both weighted and Non-Weighted transcriptomes, follow a similar pattern: the mean and variance tend to be slightly higher in the first quartile for weighted transcriptomes, while in the second and third quartiles all of the weighted transcriptomes have a similar, if not the same, expression distribution as the non-biased transcriptome for both mean and variance. Finally, both moments were more constrained in the fourth quartile, with the distribution’s right tail thinner than the Non-Biased transcriptome. Overall, these results show that the mean-variance relationship is preserved for most genes. Thus, the data on Figures 7B and 7C suggest that weighting the transcriptome doesn’t alter the moments of the gene expression distributions. The only notable distinction observed in the weighted transcriptomes was a higher mean and variance in the first and second quantiles compared to Non-Weighted transcriptomes. In the third quantile, the expression distribution of all weighted transcriptomes closely resembled, if not identical to, that of the Non-Weighted transcriptome in terms of both mean and variance. Additionally, in the fourth quantile, the mean and variance of weighted transcriptomes exhibited greater constraints, resulting in a thinner right tail compared to the Non-Weighted transcriptome. These findings suggest that the mean-variance relationship is largely preserved across most genes in the weighted transcriptomes. It is important to highlight that the KDE distributions depicted in Fig.16 indicate that weighting the transcriptome does not significantly alter the moments of the distribution. However, these methods are meant to enrich the data with genes that were previously unobserved. To validate that weighted transcriptomes effectively enrich genes, the Log-Fold Change (LFC) of both moments of the weighted transcriptomes against the Non-Biased transcriptome was calculated. The distribution of LFC for both moments was then visualised by fitting a one- dimensional KDE. An expected outcome of weighting transcriptomes is a notable increase in the mean expression of genes within the 1st and 2nd quantiles. Concurrently, one can anticipate an increase in the variance of gene expression due to the inherent heteroscedasticity property of gene expression distributions. Both the mean and variance exhibit positive skewness, signifying that weighted transcriptomes enhance the observed gene counts (refer to Fig 17 and Fig 7D, 7E). What was unexpected was the observation that the increase in both the mean and variance of gene expression occurred across all quantiles. A left tail (representing a decrease in the LFC of a gene's mean and / or variance) of LFC KDE is most pronounced in the third and fourth quartiles, showing that some highly expressed genes are being constrained when weighting is applied. These characteristics differ based on the experimental protocol used to weigh the transcriptome. The Weighted Library exhibits the smallest left tail compared to Weighted Chemistry and Both Step Weighted. Notably, Weighted Chemistry and Both Steps Weighted protocols showed only a superficial difference in their left tails. It's worth emphasizing that the left tail of Weighted Transcriptomes remains relatively small. In other words, applying the enrichment protocols during capture chemistry appears to result in higher (though still small) numbers of constrained genes compared to enriching after the RT-PCR (i.e. in the cDNA amplification step). As explained in Example 2, this is expected since by manipulating the capture chemistry it is possible to have greater effects on Single Cell Count Data, since the enrichment occurs before RT-PCR. Thus, tis outcome aligns with prior research, which demonstrated that manipulating capture chemistry has the most substantial impact on single-cell data, particularly in enriching before RT-PCR. Nevertheless, the overall effect appears relatively minor and primarily concentrated on a select group of highly expressed genes. Notably, low to medium expressed genes display a significant boost in observed mean expression and maintain their mean-variance relationship. This suggests that the overall information content of the weighted transcriptome exceeds that of standard scRNA- Seq experiments. Determining the Information Content of Weighted Transcriptomes. The primary purpose of a single-cell experiment is to obtain a high-resolution and high-throughput view of biological systems. In such experiments the observed counts of the experiment are assumed to reflect the true expression distribution, thereby offering valuable insights into the biological processes at play. The greater the extent to which the observed counts distribution reveals the true expression distribution of genes, the more valuable the obtained data becomes. Quantifying the amount of information that the observed counts contain about the true expression distribution in a biological setting is currently impossible. However, in a simulation, this is relatively easy to calculate (i.e. assessing the extent to which observed counts reflect biological variance is feasible when conducting simulations, as both the observed and true counts are known). This enables the quantification of the information contained within observed counts concerning biological variance in simulation. To measure this information, a commonly used measure is Mutual Information (MI). Mutual Information (MI) is an information theory-based statistic that quantifies the amount of information knowing variable X tells you about variable Y. When variable X provides us with no information about Y, MI is 0. If X provides perfect information about Y, then MI is 1. In an ideal experiment, we want the maximum information we can get about the true expression distribution from the observed count. Using simulated data, it is possible to calculate how much information differently weighted transcriptomes contain by calculating the empirical MI (eMI) between the simulated observed counts and corresponding simulated biological variance ('ground truth’ counts, simulated true gene expression counts). To assess the overall amount of information that a given weighted transcriptome contains, the eMI between the observed and biological variance (simulated observed and ground truth counts) per gene for a given cell population and dataset was calculated. In other words, the inventors first calculated the eMI between the observed and simulated true counts per gene for a given cell population and dataset to assess the overall amount of information a given weighted transcriptome contains. A KDE was then fitted across the calculated eMI to visualize the overall distribution of eMI for a given type of transcriptome for each dataset. As shown in Figure 7F and Figure 18, the eMI distribution of weighted transcriptomes consistently contains more information than non- biased (Non-Weighted, normal) single-cell transcriptomes. This is observed in both a positive shift in the overall density of the distribution and the increase in tail thickness of the right tail of the distribution. There are differences in the amount of skew and increase in tail thickness depending on how the transcriptome was weighted. Specifically, Library Biased experiments show less of an increase in the overall informational content than Chemistry Biased and All-Step Biased experiments. Both Steps Weighted transcriptomes eMI distribution varied from the other weighted protocols in terms of the shape of its distribution consistently forming a parteo-like distribution with a thick right tail across multiple datasets (see Fig 18 and Fig 7F). However, it's important to point out that Boths Steps Weighted density at the lower end of the eMI distribution is significantly lower than all of the other weighted protocols. Indicating that Both Steps Weighted substantially increase the overall informational content compared to other weighted transcriptomes. Despite this, all weighted transcriptomes significantly increase the amount of information contained about the real gene levels in the transcriptome, compared to non-biased (normal) single-cell experiments. In other words, it's important to note that both Weighted Library and Chemistry also significantly increased the information they contain compared to Non-Weighted single-cell experiments as well. Finally, the amount of information in weighted transcriptomes was assessed on a quartile basis (using the same process as above, but applied separately for each gene quartile). The results on Figure 7G show that weighted transcriptomes contain more information than Non-Biased transcriptomes (overall shift in the density of the eMI distribution and thicker tails) in all quartiles. However, the data shows that the largest gains in terms of information for weighted transcriptomes come from the first and second quartile. These quartiles have a larger shift in the distribution and thicker right tails. In contrast, in the third and fourth quartiles, the information gain primarily comes from an increase in the thickness of the right tail rather than a large shift in the density of the distribution. This indicates that most information gains come from the low to medium expressed genes – as expected. Overall, these results clearly demonstrate that weighted transcriptomes contain more information about the true expression distribution than Non-Biased (normal) single- cell experiments. This information gain primarily comes from the first and second quartiles of the expression distribution, but there are still gains in the third and fourth quartiles. Performance of Single Cell Normalization Methods in Weighted Transcriptomes. Normalization is a critical step in the analysis of Single-Cell experiments. Its primary goal is to eliminate technical variance while preserving biological variance. Sources of technical variance can include differences in sequencing depth, batch effects, and so on. Cell populations can be identified and characterised by removing variance from technical sources. Previous studies have assessed the performance of normalization techniques by quantifying their ability to eliminate technical variance and applying clustering algorithms to the normalized data for cell population identification (Townes et al., 2019; Ahlmann-Eltze et al., 2023). scRNA-Seq is primarily utilized to discover novel cell populations and subpopulations, heavily reliant on the ability of current single- cell bioinformatics methods to identify genes with the highest variance. These high-variance genes are then used as features for dimensionality reduction and population visualization. Here the inventors evaluated the impact of different normalization techniques on the identification of pre- characterized cell populations using Adjusted Rand Index (ARI) as a performance metric. In this study, four different normalization methods (SCTransform, GLM-PCA, Delta, and Sum Pooled Factors) were tested across datasets and transcriptome types (Non-Biased and Weighted Transcriptomes). The results on Figure 19 show no substantial degradation in the scRNA-Seq method's ability to identify cell populations accurately, and no substantial differences between the performance of the various methods across the different Weighted Transcriptomes. This is not surprising given the previous observation that there were no substantial alterations to the statistical characteristics of the weighted transcriptomes, except for the most highly expressed genes. Therefore, the data show that these alterations of highly expressed genes should not affect the identification of cell populations, and indicate that normal single-cell normalization and downstream analysis methods can be applied to weighted transcriptomes. In other words, the weighted transcriptomes contain additional information about the transcriptome particularly in relation to genes that are at the low end of the expression spectrum, while still preserving all relevant information that is exploited by current methods for e.g. cell population identification. These results are encouraging as they suggest that standard single-cell normalization and downstream analysis techniques can be applied to weighted transcriptomes without modification. Figure 8 shows further results obtained using the data described above. Figure 8A shows the excepted observed gene count per cell (number of detectable genes) as a function of the average UMI per cell (which represents the sequencing capacity in the experiment). The data show that all weighted sampling-based methods quickly increase the number of observed genes compared to the “non-biased” (prior art) approach, and provide nearly double the amount of observed genes at all values of average UMI. This is indicative of the superior performance of these methods over the current single-cell protocols. Figure 8B shows the log-transformed mean expression distribution across cell populations and experimental protocols. The primary goal of this plot is to show that the weighted sampling constrains highly expressed genes and increases the number of low-expressed genes that can be detected. This results in a violin plot that is less dispersed. In addition, the data show an increase in the mean expression of genes when weighted sampling is applied. Figure 8C shows a Kernel Density estimate of the distribution of log transformed mean expression for each gene observed. A kernel density estimate allows us to plot the empirical distribution of the log-transformed mean expression distribution. The data shows that the double- weighting protocol (“Biased All Steps”) results in a greater enrichment of genes and constrain their expression to a median of 3.86. In comparison, the Non-Biased median mean expression is 0.9.. The data show that the proposed methods with enrichment can “push” many lowly expressed genes outside of the ‘0’ (failed to be observed) peak, thereby increasing the median expression across genes. Discussion In this example, the effects of weighted Single Cell Data on the statistical characteristics of observed gene expression counts were investigated. The inventors demonstrated that the statistical relationship between the mean and variance was not disturbed for most genes in all weighted protocols. This suggests that the weighted protocols enhance the representation of gene expression distribution. A few very highly expressed genes that were outliers in the non-biased protocols were more constrained in the weighted transcriptomes. The inventors further showed that Weighted Single Cell Datasets contained a clear increase in the information the observed count data contains about the true expression distribution used for simulations. This indicates that Weighted datasets provide an overall gain in information, suggesting that the trade-off of potentially losing some limited information on the most highly expressed genes is worth it, as it greatly improves the overall amount of information available. Thus, the inventors showed by looking at statistical characteristics of gene count distributions that most genes are unaffected, suggesting that current normalization methods can work with weighted datasets. The inventors validated this hypothesis by verifying the ability to identify cell populations using different normalization methods and showing that all tested methods performed well. These results imply that Weighted Single Cell Experiments are a great way to increase the overall representation of the transcriptome with little to no cost (as investigations that are possible with non-biased protocols can still be performed equally accurately, while enabling new investigations that could not be performed in the absence of information on expression of lowly expressed genes). Example 4 – Determining gene specific weights and recovering true counts In this example, the inventors demonstrate how gene specific weights can be obtained for performing weighted experiments as described in Examples 2 and 3. They further show how the effect of weighting can be taken into account to recover expected true counts from weighted transcriptome data. Optimizing Gene Weights Weighted transcriptomes are designed to improve the detection and representation of low to medium-expressed genes in scRNA-Seq experiments. This can be performed using enrichment methods, either through PCR (modifying PCR primer concentrations) or any kind of capture step (e.g. antibody pulldowns, probes, etc). These enrichment protocols modify the PCR primer or probe concentrations based on (i) an excepted relative frequency between transcripts captured, and (ii) a target relative frequency between transcripts captured. To achieve a target relative frequency, we calculate ’weights’ that tell us how to modify the concentrations of primers / probes to achieve the desired target relative frequency. The present example demonstrates two methods for calculating weight. The first method (uniform relative frequency) is a simple heuristic method used in Examples 2, 3 and 5. The second method is a more formally optimised method. Uniform Relative Frequency. In the simulation method described in Example 2, there are two steps in the single-cell sequencing experiment where enrichment protocols can be applied: Capture Chemistry and cDNA Amplification. By default, the gene odds are all set to one, which represents a non-biased (or normal) single-cell sequencing experiment. To weigh only one of the steps to conduct either a chemistry or library-biased experiment, we calculate the Odds for gene ^^ by dividing the desired target relative frequency by the observed relative frequency of gene ^^, as explained in Example 2. The observed relative frequencies are obtained from a reference dataset (which can be e.g. from a database of scRNAseq data, or from a pilot experiment). As explained above, the relative frequency of a gene is calculated as the µgfor the gene divided by the sum of µgacross all genes observed in the reference dataset the target frequency is calculated as one over the number of non-zero expressed genes in the pilot experiment / reference dataset. For genes that are not observed in the pilot experiment / reference dataset, the weight may be set to the maximum weight used. This approach sets a uniform target frequency, i.e. it aims to bring all genes to a level where they represent approximately equal proportions of the material that is sequenced in a sample that comprises a single population of cells. A gene specific weight can then be calculated as the ratio of the target frequency relative to the observed frequency. In a sample that comprises a plurality of populations of cells, by setting the weights using the uniform frequency approach this allows natural differences between the cell populations to emerge as the target genes may not have the same proportions in cells from different populations. Thus, although the weighting may result in overall more similar proportions of sequenced materials between target genes (e.g. “rescuing” low expressed genes), the actual proportions for each transcript may vary between cells as a result of natural variation within and between cell populations. To bias both steps of a single-cell sequencing experiment, the gene-specific odds for the cDNA amplification step are adjusted according to the expected molecules in the sequencing pool after applying weights to the capture chemistry and loss of molecules due to RT-PCR inefficiencies, as explained in Example 2 above. Budget Optimization. The method above is a simple heuristic that provides a reasonable starting point, and as demonstrated in Example 3 already provides significant benefit when used to obtain weighted transcriptome data. However, it is possible to improve on this approach for calculating gene weights ^^^^. In the present method, the problem of calculating gene weights is reframed as a budget optimization problem with the aim to minimize ^^^^, the percentage of molecules not sequenced on a per-gene basis. By focusing on a per-gene basis instead of minimising the total percentage of molecules not sequenced, this allows low and medium-expressed genes to have equal importance as high-expressed genes. By summing all of the ^^^^genes together, we get ^^, providing us with the basis for an optimization function to minimize: based on the number of reads ^^ that will be sequenced. In order to minimise ^^, two sets of weights are optimized: ^^^^and ^^^^. ^^^^represents the weights at the capture chemistry step of a single-cell sequencing experiment. Weight ^^^^represent the weights applied at the cDNA amplification step. A cell’s expected transcripts ^^^^is estimated using a Dirichlet distribution multiplied by the tool number of mRNAs content of the cell (which is the total number of mRNA molecules in the cell). The capture chemistry step is modelled using a Multivariant Fisher Non-Central Hypergeometric (MFNCH) distribution. The expected molecular count ^^^^as a function of ^^^^can be calculated for a sample size of ^^ mRNA in the cell comprising ^^^^mRNA for gene ^^ as: ^^^^= ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^^( ^^, ^^, ^^^^, ^^^^) To speed up optimisation, this can be precalculated for a range of values of ^^^^for a sample size of ^^ mRNA in the cell comprising ^^^^mRNA for gene ^^. The relative frequency of the molecules does not change after RT-PCR as manipulating the cDNA step of the single-cell sequencing experiment only alters the reads to umi ratio for a given gene (i.e. it does not alter how many unique molecules are represented but how many times each unique molecule is represented). Therefore, ^^^^is fixed once estimated for the RT-PCR step. To account for the cDNA amplification step, the expected UMI count ^^ is calculated as a function of the number of reads that belong to a given gene ^^, which can be manipulated via the weight ^^^^: where B is the total number of reads, or a factor set to 1. Based on ^^ (expected observed UMI count) and the known number of mRNA molecules for gene ^^ ( ^^^^) (where Xg is the expected transcript count per gene, parameterised using a Dirichlet distribution as explained above), we can calculate ^^^^as: A set of weights ^^^^and ^^^^comprising a pair of weights for each gene g can therefore be identified using any optimisation process known in the art, which minimises objective function ^^. Removing all technical biases introduced by weighted transcriptomes While the above analysis of weighted transcriptomes indicates that the technical biased introduced by weighted transcriptome is minimum, a new method to account for any potential bias that may exist explicitly can be useful. This new method aims to accurately estimate the distribution of transcript counts ^^^^given the observed counts ^^^^and then to be able to compare the distribution between the cells of one or populations or conditions and identify any differences between the distribution. To achieve this, the following assumptions of a cell expression distribution are made: 1. A given gene’s expression distribution is at a steady state within a given cell population or condition. 2. The number of mRNA transcripts of a given gene randomly fluctuates according to a given gene’s expression distribution. 3. The number of mRNA molecules within a cell is the sum of all the randomly fluctuating transcripts. This is referred to below as the “cell size”. 4. A given cells library size is the product of: (i) the sum of all mRNA molecules in the cell and (ii) the capture chemistry efficiency. 5. The observed counts of a given gene per cell are the result of both cell and gene-specific capture chemistry. Based on these assumptions, the differential expression method accounts for the technical bias introduced on both the cell and gene levels and remove these effects. Cell-specific bias is currently viewed in terms of sequencing depths (i.e. the difference in library size); to account for this during normalization, a size factor is estimated and all cells are then adjusted to the same sequencing depth. Gene-specific biases are commonly viewed in terms of the heteroskedasticity of high expressed genes; to account for this, a log transformation is applied to the size factor adjusted counts (Luecken et al 2018). However, this view of normalization only partly addresses the problem and fails to see the immediately observed UMI counts as a source of technical bias themselves. For weighted transcriptomes, properly accounting for technical biases further requires accounts for the technique bias introduced by gene-specific weights upon the observed counts for a group of cells from a given condition. The easiest way to account for and remove this bias is to infer a given gene’s true transcript count distribution, which requires an estimate of the expected cell size for a group of cells. During single-cell sequencing experiments, a number of mRNA from the true transcript counts are sampled randomly based on the experiment’s capture chemistry efficiency. The sum of sampled counts is referred to as the cell library size, which can be thought of as a sample from a conjugate distribution where the percent of molecules captured reflect capture efficiency from a cell’s randomly fluctuating cell size. Previously, this has been described using a Beta-Binomial where the Beta distribution represents capture chemistry efficiency, and the binomial distribution represents the downsampling based on a population p sampled from a cell size n (Ye et al.2019). However, the Beta-Binomial distribution implicitly limits the cell size of a cell population. In keeping with the above assumption, the cell size is a random variable; a Beta-Poisson distribution would better fit biological differences in cell size and technical differences in capture chemistry efficiency. In this model, the lambda of the Beta-Poisson would be the excepted cell size for a given cell population, and the alpha and beta would describe the capture chemistry efficiency of the single-cell sequencing protocol. In addition, a Beta-Poisson’s parameters are easier to infer than a Beta-Binomial. Multiple options are also available to infer the parameters, including method of moment, maximum likelihood estimate, and approximate Bayesian computations as described in Tang et al., 2023. Due to the difficulties of removing gene-specific bias on the observed counts, it would be easier to infer the latent transcript count distribution of a given cell population conditions on the observed counts and gene-specific weights used in the experiment. To do this, an initial estimate of the transcript counts given the observed counts is needed; this can be achieved using Cornfield’s Approximation as described below. To approximate transcript counts, we use an estimate of the expected cell size lambda parameter from the estimated Beta-Poisson and the sample mean of the observed counts. This provides a reasonable starting point to then use Markov Chain Monte Carlo (MCMC) or Variational Inference to infer the latent transcript count distribution of a given gene. The primary difference between these two methods is that Variational Inference requires a conjugate prior distribution assumed for the transcript count distribution, which would be gamma distribution. For example, to provide accurate estimates of transcript counts with low uncertainty under the above assumptions, Cornfield’s approximation can be used to provide an initial estimate of transcripts and then a Markov Chain Monte Carlo (MCMC) is used to refine the estimate and calculate uncertainty. Cornsfield’s approximation is a method of inferring the number of balls in an Biased Urn (which is what the MFNH models). However, this introduces an implicit bias as the above approximation assumes that the transcript counts follow a normal distribution. Therefore, the above estimates can be used as initial guesses of transcript count in cell i. Using an MCMC algorithm, the initial transcript count estimates can be used to parameterize a distribution (e.g. a Beta Poisson distribution). This can be sampled from, to refine transcript count estimates further. Example 5 – Experimental validation In this example, the inventors used the methods described in Example 4 (Optimising gene weights by uniform relative frequency) to calculate gene weights used in proof-of-concept experiments to show that by manipulating primer concentrations in qPCR experiments, and probe concentrations in TAPseq experiments, it is possible to bring observed expression levels of low and high expressed genes in the same sample to similar levels. This demonstrates that the principles theoretically explored in Examples 2 and 3 can be applied in real experimental protocols. qPCR proof of concept In this experiment, qPCR was used to investigate whether it is possible, by modifying the concentrations of qPCR primers for query genes relative to a housekeeping gene, to bring the readout (number of amplification cycles) for the query gene closer to that of the housekeeping gene. In other words, this experiment investigates whether by attempting to obtain a uniform representation of genes in a sequencing library as described in Example 4, it is in fact possible to do so. This demonstrates the concept of using weighted transcriptomes to make low expressed genes discoverable by scRNAseq where they would normally have been very likely not be sequenced. Methods Cell culture and RNA extraction. MCF7 and TAMR cell lines (tamoxifen resistant, in particular: T47D and MDA-MB-231 cell lines) were cultured according to guidelines, RNA was extracted using QAIGEN RNA plus easy mini kit (Cat.No. / ID: 74134) and quantified via qubit assay (ThermoFisher, Cat.No. / ID: Q10210). cDNA synthesis. This was performed using the SuperScript VILO cDNA synthesis Kit (Cat.No: 11754-050). cDNA was synthesised as according to manufacturers guidelines. The following protocol has been optimized for generating first-strand cDNA for use in two-step qRT–PCR. The reaction volume may be scaled as needed up to 100 μL. 1. For a single reaction, combine the following components in a tube on ice: a. 5X VILO Reaction Mix: 4 µL. b. 10X SuperScript Enzyme Mix: 2 µL. c. RNA (up to 2.5 ug): x µL. d. DEPC-treated water: to 20 µL. For multiple reactions, prepare a master mix without RNA. 2. Gently mix tube contents and incubate at 25°C for 10 minutes. 3. Incubate tube at 42°C for 60 minutes. 4. Terminate the reaction at 85°C at 5 minutes. 5. Use diluted or undiluted cDNA in qPCR, or store at –20°C until use. TaqMan qPCR. A. Prepare the PCR reaction mix. Use the same amount of cDNA for all samples (1 to 100ng per 10ul reaction). - For each sample (to be run in triplicate), pipet the following into a nuclease-free 1.5mL microcentrifuge tube (Volume per 10ul reaction (ul) for Single reaction; Four replicates): 20X TaqMan Gene Expression Assay: 0.5 + 0.5; 2 + 2 2X TaqMan Master Mix: 5; 20 cDNA template : 0.5; 2 RNase-free water: 3.5; 14 - Cap the tube and invert it several times to mix the reaction components. - Centrifuge the tube briefly. B. Load the Plate: (i) transfer 10ul of PCR reaction mix into each well of a 384-well reaction plate. (ii) Seal the plate with the appropriate cover. (iii) Centrifuge the plate briefly. (iv) Load the plate into the instrument (Quantstudio 6). C. Run the plate: (i) Create experiment / plate document for the run using desired parameters. (ii) Run the plate. QuantStudio 6 Run Method is provided in Table 1. Stage Step Ramp rate Temperature Time Hold Stage Step 1 1.9oC / s 95oC 20 seconds PCR Stage Step 1 1.9oC / s 95oC 1 second Number of Cycles: 40 Step 2 1.6oC / s 60oC 20 seconds Enable AutoDelta: Unchecked (default) Starting cycle: Disabled when Enable AutoDelta is unchecked Table 1. QuantStudio 6 Run Method parameters. D. Analyse the result. Refer to user guide for real-time PCR instrument. PCR Mastermixes. The composition of the mastermixes used in these experiments is provided in Tables 2-7. Single Four PCR Reaction Mix Component Reaction Replicates 20x tawman gene expression assay 0.5 + 0.5 2 + 2 2x tawman master mix 5 20 cDNA template 0.5 2 Rnase-free water 3.5 14 Table 2. Normal concentrations (also referred to as “20x Taqman”). Single Four PCR Reaction Mix Component Reaction Replicates 20x tawman gene expression assay 0.3 + 1.34 1.22 + 5.36 2x tawman master mix 5 20 cDNA template 0.5 2 Rnase-free water 2.86 11.44 Table 3. Variable NCOA3 (gapdh vs ncoa3) Single Four PCR Reaction Mix Component Reaction Replicates 20x tawman gene expression assay 0.25 + 3.8 1 + 15.2 2x tawman master mix 5 20 cDNA template 0.5 2 Rnase-free water 0.45 1.8 Table 4. Variable RARA (gapdh vs rara) Four PCR Reaction Mix Component Single Reaction Replicates 20x tawman gene expression assay 0.28 + 2.23 1.12 + 8.92 2x tawman master mix 5 20 cDNA template 0.5 2 Rnase-free water 1.99 7.96 Table 5. Variable FOXA1 (gapdh vs foxa1) Four PCR Reaction Mix Component Single Reaction Replicates 20x tawman gene expression assay 0.32 + 1.16 1.28 + 4.64 2x tawman master mix 5 20 cDNA template 0.5 2 Rnase-free water 3.02 12.08 Table 6. Variable GATA3 (gapdh vs gata3) Four PCR Reaction Mix Component Single Reaction Replicates 20x tawman gene expression assay 0.5 + 0.5 2 +2 2x tawman master mix 5 20 cDNA template 0 0 Rnase-free water 4 16 Table 7. No template control. Calculation of variable concentrations: To compute the weights for this experiment, the inventors utilised a Bulk RNA-Seq dataset for each cell line from Ensembl as reference dataset to estimate a given genes relative frequency. The target frequency of this experiment is 0.5. To calculate the weights for housekeeping genes and target gene I calculated the weight for gene ^^^^is calculated by dividing the desired target relative frequency by the observed relative frequency of gene ^^^^. The relative frequency of a gene is calculated by dividing a given gene's mean expression ^^^^by the sum of all genes' mean. Results The results of these experiments are shown on Figure 9, which show ratio of Ct observed for two different query genes relative to a housekeeping gene, with different concentrations of TaqMan probes. The TaqMan probes are either used at the same concentration for the housekeeping gene and the query gene (mimicking a non-biased protocol), or at different concentrations (mimicking a weighted protocol). Figures 9A and 9B show for two different genes that changing the concentration of the TaqMAn probes without changing the ratio of the concentration for the query (FOXA1 and GATA3, respectively) and housekeeping genes (i.e. non-biased protocols) does not significantly change the ratio of Ct observed for the genes. However, when the concentrations for the query and housekeeping genes are weighted to compensate for the fact that the housekeeping gene is expressed at much higher levels than the query gene, Ct ratios much closer to 1 can be observed. This shows that it is possible to approach uniform representation of transcripts in a library by biasing probe concentrations. Figures 9C and 9D compare Ct ratios for query and housekeeping genes obtained with non- biased (20x taqman) and biased (Variable taq) protocols. Figures 9C and 9D show data in two different biological conditions: with the cells cultured in the presence of estradiol (Fig.9C) or without estradiol (Fig. 9D). The MCF7 cell line is sensitive to estradiol which stimulates proliferation. Figures 9E and 9F compare Ct ratios for query and housekeeping genes obtained with non-biased (20x taqman) and biased (Variable taq) protocols. Figures 9E and 9F show data for two different cell lines that are both insensitive to estradiol. In all conditions the biased protocols were able to bring the readout for the non-housekeeping genes much closer to those of the housekeeping genes, i.e. approaching more equal representation. TAPseq Proof of concept In this experiment, the present inventors used the approach described in Example 4 (Uniform Relative Frequency) to identify gene specific weights for targeted single-cell RNA sequencing using the TAP-seq protocol. TAP-seq (Shraivogel et al.2020) is a targeted single cell RNA sequencing protocol where cDNA product from single cells is amplified in a nested multiplex PCR reaction targeting selected genes of interest. This is followed by Illumina adapter addition and sequencing. In particular, cellular and molecular barcodes as well as a universal PCR handle are added to all mRNAs during reverse transcription (i.e. using RT primers comprising a dT section followed by a unique molecular identifier (UMI), a cell barcode and a universal PCR handle). The cDNA is then purified and gene specific primers are combined with the universal PCR handle to amplify transcripts of interest, using two semi-nested multiplex PCTs (i.e. a first PCR uses a universal primer and a gene-specific outer primer, and a second PCR uses the universal primer and a gene specific inner primer that also contains a common sequence for addition of the Illumina adapters in the next PCR). A Third PCR is used to add Illumina sequencing adapters. All primers are used at the same concentration for all genes. In a TAP-seq protocol, all targeted genes are typically detected, because all target genes are enriched and the target list is restricted to a number such that all enriched genes can be sequenced. However, no gene specific enrichment is performed and primer concentrations are the same for all genes. Therefore, as the number of target genes increase, the likelihood that low expressed genes will be detected decreases. The weighted transcriptome methods described herein remedy this, enabling larger amounts of genes to be probed simultaneously. This demonstrates the concept of using weighted transcriptomes to make low expressed genes discoverable by scRNAseq where they would normally have been very likely not be sequenced. Methods TAP-seq was performed as described in Schraivogel et al. 2020b available at protocolexchange.researchsquare.com / article / pex-864 / v1 with one difference which is that the concentrations of the gene specific outer and inner primers were adjusted using a method as described herein. This was applied to the L1000 panel of genes as demonstrated in Schraivogel et al., 2020. Thus, by reference to the above, a library biased experiment was implemented. Determining gene specific weights. The method in Examples 2 and 4 (Uniform Relative Frequency) was used to determine gene specific weights for each of the 973 genes listed below. Inner and outer primer sequences for each gene were as described in Schraivogel et al. 2020 (See Supplementary Table S1 which provides all primer sequences for the L1000 gene panel). All genes were assigned a target relative transcript count of 0.001027749 (1 / 973, i.e. target equal representation of all genes). The list below provides for each gene: gene ID: normalised target count: reference primer concentration: primer concentration. Entries for each gene are separated by “ / / ”. AARS:1.68E-04:6.1:6.1 / / ABCB6:2.05E-05:50.12:30 / / ABCC5:1.57E-04:6.56:6.56 / / ABCF1:0.001045438:0.98:0.98 / / ABCF3:2.24E-04:4.59:4.59 / / ABHD4:5.58E-05:18.43:18.43 / / ABHD6:1.02E-04:10.12:10.12 / / ABL1:1.49E-04:6.92:6.92 / / ACAA1:0.001155635:0.89:0.89 / / ACAT2:3.51E-04:2.92:2.92 / / ACBD3:8.48E-04:1.21:1.21 / / ACD:1.84E-04:5.59:5.59 / / ACLY:5.65E- 04:1.82:1.82 / / ACOT9:6.24E-04:1.65:1.65 / / ADAM10:0.001663309:0.62:0.62 / / ADAT1:2.37E- 04:4.33:4.33 / / ADGRE5:0.003044705:0.34:0.34 / / ADGRG1:2.67E-04:3.84:3.84 / / ADH5:0.001112706:0.92:0.92 / / ADI1:0.001058278:0.97:0.97 / / ADO:2.07E-04:4.96:4.96 / / ADRB2:4.58E-04:2.24:2.24 / / AGL:2.97E-04:3.46:3.46 / / AKAP8:3.54E-04:2.9:2.9 / / AKAP8L:5.47E- 04:1.88:1.88 / / AKR7A2:7.68E-04:1.34:1.34 / / AKT1:6.25E-04:1.64:1.64 / / ALAS1:2.85E-04:3.6:3.6 / / ALDH7A1:9.39E-05:10.94:10.94 / / ALDOA:3.36E-04:3.06:3.06 / / ALDOC:8.83E-05:11.63:11.63 / / AMDHD2:2.38E-04:4.31:4.31 / / ANKRD10:6.60E-04:1.56:1.56 / / ANO10:1.28E-04:8.04:8.04 / / ANXA7:0.001421833:0.72:0.72 / / APBB2:4.98E-06:206.26:30 / / APOE:1.34E-06:766.1:30 / / APP:8.55E-04:1.2:1.2 / / APPBP2:4.47E-04:2.3:2.3 / / ARFIP2:2.36E-04:4.36:4.36 / / ARHGAP1:6.75E- 04:1.52:1.52 / / ARHGEF12:9.03E-05:11.39:11.39 / / ARHGEF2:9.46E-04:1.09:1.09 / / ARID4B:0.002675209:0.38:0.38 / / ARID5B:8.45E-04:1.22:1.22 / / ARL4C:0.002927992:0.35:0.35 / / ARNT2:9.77E-06:105.15:30 / / ARPP19:8.37E-04:1.23:1.23 / / ASAH1:0.003999493:0.26:0.26 / / ASCC3:7.21E-04:1.43:1.43 / / ATF1:4.79E-04:2.15:2.15 / / ATF5:6.75E-04:1.52:1.52 / / ATF6:8.93E- 04:1.15:1.15 / / ATG3:0.002491035:0.41:0.41 / / ATMIN:4.68E-04:2.19:2.19 / / ATP11B:5.75E- 04:1.79:1.79 / / ATP1B1:1.16E-04:8.89:8.89 / / ATP2C1:2.71E-04:3.79:3.79 / / ATP6V0B:0.003574227:0.29:0.29 / / ATP6V1D:7.42E-04:1.39:1.39 / / AURKA:3.33E-05:30.82:30 / / AURKB:1.19E-05:86.5:30 / / AXIN1:2.27E-04:4.54:4.54 / / B4GAT1:4.16E-05:24.71:24.71 / / BACE2:8.09E-05:12.71:12.71 / / BAD:3.11E-04:3.31:3.31 / / BAG3:1.17E-04:8.81:8.81 / / BAMBI:1.53E-06:670.34:30 / / BAX:0.001873547:0.55:0.55 / / BCL2:0.001328501:0.77:0.77 / / BCL7B:5.29E-04:1.94:1.94 / / BDH1:1.26E-04:8.13:8.13 / / BECN1:9.46E-04:1.09:1.09 / / BHLHE40:2.10E-04:4.88:4.88 / / BID:0.001772357:0.58:0.58 / / BIRC2:8.73E-04:1.18:1.18 / / BIRC5:2.34E-05:43.96:30 / / BLCAP:6.39E-04:1.61:1.61 / / BLMH:2.85E-04:3.61:3.61 / / BLVRA:0.001564419:0.66:0.66 / / BMP4:2.11E-06:487.52:30 / / BNIP3:3.11E-04:3.31:3.31 / / BNIP3L:0.002026673:0.51:0.51 / / BPHL:8.43E-05:12.19:12.19 / / BRCA1:9.66E-05:10.64:10.64 / / BTK:9.69E-04:1.06:1.06 / / BUB1B:7.09E-06:144.94:30 / / BZW2:5.94E-04:1.73:1.73 / / C2CD2:9.95E- 05:10.33:10.33 / / C2CD2L:2.92E-04:3.52:3.52 / / C2CD5:3.92E-04:2.62:2.62 / / C5:4.58E-05:22.44:22.44 / / CAB39:0.001026656:1:1 / / CALM3:0.002667351:0.39:0.39 / / CALU:5.63E-04:1.83:1.83 / / CAMSAP2:7.03E-05:14.61:14.61 / / CANT1:3.48E-04:2.95:2.95 / / CAPN1:8.36E- 04:1.23:1.23 / / CARMIL1:1.06E-04:9.72:9.72 / / CASC3:4.83E-04:2.13:2.13 / / CASK:2.48E- 04:4.15:4.15 / / CASP10:6.60E-04:1.56:1.56 / / CASP2:4.46E-04:2.31:2.31 / / CASP3:5.17E- 04:1.99:1.99 / / CASP7:3.19E-04:3.22:3.22 / / CAST:0.006030958:0.17:0.17 / / CAT:0.00132026:0.78:0.78 / / CBLB:4.06E-04:2.53:2.53 / / CBR1:8.70E-04:1.18:1.18 / / CBR3:8.74E- 05:11.76:11.76 / / CCDC85B:0.002060786:0.5:0.5 / / CCDC86:8.60E-05:11.94:11.94 / / CCDC92:2.11E-04:4.88:4.88 / / CCL2:8.28E-05:12.41:12.41 / / CCNA1:3.83E-07:2681.35:30 / / CCNA2:2.53E-05:40.63:30 / / CCNB1:3.16E-05:32.5:30 / / CCNB2:1.36E-05:75.53:30 / / CCND1:2.22E-05:46.23:30 / / CCND3:0.002714113:0.38:0.38 / / CCNE2:1.94E-05:53.1:30 / / CCNF:1.88E-05:54.72:30 / / CCNH:7.02E-04:1.46:1.46 / / CCP110:2.53E-04:4.06:4.06 / / CD320:2.93E-04:3.51:3.51 / / CD40:5.16E-04:1.99:1.99 / / CD44:0.006168561:0.17:0.17 / / CD58:4.29E-04:2.4:2.4 / / CDC20:7.09E-06:144.94:30 / / CDC25A:1.92E-06:536.27:30 / / CDC25B:4.83E-04:2.13:2.13 / / CDC42:0.009834395:0.1:0.1 / / CDC45:8.05E-06:127.68:30 / / CDCA4:1.39E-04:7.41:7.41 / / CDH3:5.75E-07:1787.57:30 / / CDK1:1.25E-05:82.5:30 / / CDK19:4.38E- 04:2.34:2.34 / / CDK2:4.12E-05:24.94:24.94 / / CDK4:4.02E-04:2.55:2.55 / / CDK5R1:4.50E- 05:22.82:22.82 / / CDK6:5.82E-04:1.76:1.76 / / CDK7:1.48E-04:6.93:6.93 / / CDKN1A:2.40E- 04:4.28:4.28 / / CDKN1B:0.001163109:0.88:0.88 / / CDKN2A:3.66E-05:28.08:28.08 / / CEBPA:8.65E- 04:1.19:1.19 / / CEBPD:0.005224313:0.2:0.2 / / CEBPZ:0.001012091:1.02:1.02 / / CENPE:4.77E- 05:21.54:21.54 / / CEP57:9.16E-04:1.12:1.12 / / CERK:3.81E-04:2.7:2.7 / / CETN3:2.41E-04:4.27:4.27 / / CFLAR:0.003464797:0.3:0.3 / / CGRRF1:2.34E-04:4.4:4.4 / / CHAC1:3.83E-07:2681.35:30 / / CHEK1:2.97E-05:34.6:30 / / CHEK2:7.03E-05:14.61:14.61 / / CHERP:2.92E-04:3.52:3.52 / / CHIC2:5.32E-04:1.93:1.93 / / CHMP4A:0.002023415:0.51:0.51 / / CHMP6:3.56E-04:2.88:2.88 / / CHN1:1.40E-05:73.46:30 / / CHP1:0.001089325:0.94:0.94 / / CIAO3:1.69E-04:6.08:6.08 / / CIAPIN1:3.79E-04:2.71:2.71 / / CIRBP:0.004369181:0.24:0.24 / / CISD1:4.21E-04:2.44:2.44 / / CLIC4:2.71E-04:3.79:3.79 / / CLPX:5.72E-04:1.8:1.8 / / CLSTN1:4.44E-04:2.32:2.32 / / CLTB:0.001048504:0.98:0.98 / / CLTC:0.001771015:0.58:0.58 / / CNDP2:0.001561161:0.66:0.66 / / CNOT4:8.98E-04:1.14:1.14 / / CNPY3:0.003000626:0.34:0.34 / / COASY:3.79E-04:2.71:2.71 / / COG2:4.84E-04:2.12:2.12 / / COG4:3.14E-04:3.27:3.27 / / COG7:1.34E-04:7.68:7.68 / / COL1A1:9.58E-07:1072.54:30 / / COL4A1:5.75E-07:1787.57:30 / / COPB2:0.00115161:0.89:0.89 / / COPS7A:4.65E-04:2.21:2.21 / / COQ8A:2.93E-04:3.51:3.51 / / CORO1A:0.010915863:0.09:0.09 / / CPNE3:0.001341725:0.77:0.77 / / CPSF4:2.01E-04:5.1:5.1 / / CREB1:0.001081084:0.95:0.95 / / CREG1:9.64E-04:1.07:1.07 / / CRELD2:4.87E-04:2.11:2.11 / / CRK:6.58E-04:1.56:1.56 / / CRKL:3.12E- 04:3.29:3.29 / / CRTAP:0.001564802:0.66:0.66 / / CRYZ:1.39E-04:7.42:7.42 / / CSK:0.002721971:0.38:0.38 / / CSNK1A1:0.003273149:0.31:0.31 / / CSNK1E:2.73E-04:3.77:3.77 / / CSNK2A2:6.05E-04:1.7:1.7 / / CSRP1:3.82E-04:2.69:2.69 / / CTNNAL1:1.42E-05:72.47:30 / / CTNND1:4.69E-04:2.19:2.19 / / CTSD:0.004577311:0.22:0.22 / / CTSL:3.58E-04:2.87:2.87 / / CTTN:1.78E-05:57.66:30 / / CXCL2:5.56E-06:184.92:30 / / CXCR4:0.001974162:0.52:0.52 / / CYB561:1.28E-04:8.04:8.04 / / CYCS:0.001851891:0.55:0.55 / / CYTH1:0.001773315:0.58:0.58 / / DAG1:1.01E-04:10.21:10.21 / / DAXX:6.51E-04:1.58:1.58 / / DCK:9.79E-04:1.05:1.05 / / DCTD:4.04E- 04:2.54:2.54 / / DCUN1D4:1.57E-04:6.55:6.55 / / DDB2:3.38E-04:3.04:3.04 / / DDIT4:5.53E- 04:1.86:1.86 / / DDR1:1.72E-05:59.59:30 / / DDX10:3.73E-04:2.76:2.76 / / DDX42:0.001043904:0.98:0.98 / / DECR1:0.001670017:0.62:0.62 / / DENND2D:9.54E-04:1.08:1.08 / / DERA:3.96E-04:2.59:2.59 / / DFFA:5.03E-04:2.04:2.04 / / DFFB:1.07E-04:9.58:9.58 / / DHDDS:2.46E- 04:4.19:4.19 / / DHRS7:0.001495426:0.69:0.69 / / DHX29:7.37E-04:1.39:1.39 / / DLD:5.70E-04:1.8:1.8 / / DMAC2L:5.90E-04:1.74:1.74 / / DMTF1:6.90E-04:1.49:1.49 / / DNAJA3:3.06E-04:3.36:3.36 / / DNAJB1:7.24E-04:1.42:1.42 / / DNAJB2:2.83E-04:3.63:3.63 / / DNAJB6:0.001514016:0.68:0.68 / / DNAJC15:0.001762583:0.58:0.58 / / DNM1:1.51E-05:67.88:30 / / DNM1L:7.81E-04:1.32:1.32 / / DNMT1:0.001178633:0.87:0.87 / / DNMT3A:3.64E-04:2.83:2.83 / / DNTTIP2:9.59E-04:1.07:1.07 / / DPH2:1.26E-04:8.15:8.15 / / DRAP1:0.003493735:0.29:0.29 / / DSG2:6.52E-06:157.73:30 / / DUSP11:4.17E-04:2.46:2.46 / / DUSP14:5.04E-05:20.39:20.39 / / DUSP22:7.03E-04:1.46:1.46 / / DUSP3:5.90E-04:1.74:1.74 / / DUSP4:9.20E-06:111.72:30 / / DUSP6:0.0067757:0.15:0.15 / / DYNLT3:5.08E-04:2.02:2.02 / / DYRK3:8.62E-06:119.17:30 / / E2F2:3.66E-05:28.08:28.08 / / EAPP:0.001311253:0.78:0.78 / / EBNA1BP2:4.93E-04:2.09:2.09 / / EBP:4.48E-04:2.29:2.29 / / ECD:3.18E-04:3.23:3.23 / / ECH1:0.001299754:0.79:0.79 / / EDEM1:5.01E-04:2.05:2.05 / / EDN1:1.23E-05:83.79:30 / / EED:3.36E-04:3.06:3.06 / / EFCAB14:0.001208913:0.85:0.85 / / EGF:2.49E-06:412.52:30 / / EGFR:0:Inf:30 / / EGR1:1.71E-04:6:6 / / EIF4EBP1:8.30E-04:1.24:1.24 / / EIF4G1:0.00103413:0.99:0.99 / / EIF5:0.002729061:0.38:0.38 / / ELAC2:2.53E-04:4.06:4.06 / / ELAVL1:7.62E-04:1.35:1.35 / / ELOVL6:4.10E-05:25.06:25.06 / / ELP1:1.45E-04:7.08:7.08 / / EML3:2.80E-04:3.67:3.67 / / ENOPH1:3.54E-04:2.9:2.9 / / ENOSF1:2.90E-04:3.54:3.54 / / EPB41L2:2.34E-04:4.4:4.4 / / EPHA3:0:Inf:30 / / EPHB2:3.87E-05:26.55:26.55 / / EPN2:5.48E- 05:18.75:18.75 / / EPRS:0.001332909:0.77:0.77 / / ERBB2:3.49E-05:29.47:29.47 / / ERBB3:7.86E- 06:130.8:30 / / ERO1A:9.65E-04:1.06:1.06 / / ETFB:0.001391361:0.74:0.74 / / ETS1:0.002825652:0.36:0.36 / / ETV1:7.67E-07:1340.68:30 / / EVL:0.003785423:0.27:0.27 / / EXOSC4:2.27E-04:4.52:4.52 / / EXT1:2.22E-04:4.63:4.63 / / EZH2:1.72E-04:5.99:5.99 / / FAH:8.74E- 05:11.76:11.76 / / FAIM:1.18E-04:8.68:8.68 / / FAM20B:2.76E-04:3.72:3.72 / / FAS:3.48E-04:2.95:2.95 / / FASTKD5:1.60E-04:6.43:6.43 / / FAT1:0:Inf:30 / / FBXL12:2.10E-04:4.9:4.9 / / FBXO11:7.84E- 04:1.31:1.31 / / FBXO21:4.13E-04:2.49:2.49 / / FBXO7:9.79E-04:1.05:1.05 / / FCHO1:2.17E- 04:4.73:4.73 / / FDFT1:8.19E-04:1.26:1.26 / / FEZ2:6.73E-04:1.53:1.53 / / FGFR2:3.07E-06:335.17:30 / / FGFR4:1.15E-06:893.78:30 / / FHL2:1.17E-05:87.91:30 / / FIS1:0.002052929:0.5:0.5 / / FKBP14:1.03E-04:10.02:10.02 / / FKBP4:3.94E-04:2.61:2.61 / / FOS:0.024032798:0.04:0.04 / / FOSL1:3.83E-06:268.14:30 / / FOXJ3:4.67E-04:2.2:2.2 / / FOXO3:6.60E-04:1.56:1.56 / / FOXO4:6.94E-05:14.81:14.81 / / FPGS:2.07E-04:4.96:4.96 / / FRS2:2.00E-04:5.13:5.13 / / FSD1:1.19E-05:86.5:30 / / FUT1:5.75E-07:1787.57:30 / / FYN:0.002217937:0.46:0.46 / / FZD1:1.73E- 04:5.93:5.93 / / FZD7:1.53E-06:670.34:30 / / G3BP1:0.001530689:0.67:0.67 / / GAA:6.17E- 04:1.67:1.67 / / GABPB1:2.32E-04:4.44:4.44 / / GADD45A:1.25E-04:8.24:8.24 / / GADD45B:0.001186873:0.87:0.87 / / GALE:9.29E-05:11.06:11.06 / / GAPDH:0.017745607:0.06:0.06 / / GATA2:1.15E-05:89.38:30 / / GATA3:3.38E-04:3.04:3.04 / / GDPD5:9.58E-05:10.73:10.73 / / GET1:1.92E-04:5.36:5.36 / / GFOD1:1.14E-04:9.03:9.03 / / GFPT1:2.33E-04:4.41:4.41 / / GHR:0:Inf:30 / / GLI2:0:Inf:30 / / GLOD4:8.51E-04:1.21:1.21 / / GLRX:0.004695941:0.22:0.22 / / GMNN:1.02E- 04:10.04:10.04 / / GNA11:6.55E-05:15.68:15.68 / / GNA15:2.46E-04:4.18:4.18 / / GNAI1:2.78E- 05:36.98:30 / / GNAI2:0.009451292:0.11:0.11 / / GNAS:0.009874833:0.1:0.1 / / GNB5:3.06E- 04:3.36:3.36 / / GNPDA1:3.68E-04:2.8:2.8 / / GOLT1B:3.62E-04:2.84:2.84 / / GPATCH8:0.001082234:0.95:0.95 / / GPC1:2.30E-06:446.89:30 / / GPER1:1.23E-05:83.79:30 / / GRB10:1.72E-05:59.59:30 / / GRB7:3.83E-07:2681.35:30 / / GRN:0.005008709:0.21:0.21 / / GRWD1:3.00E-04:3.42:3.42 / / GSTM2:7.38E-05:13.93:13.93 / / GSTZ1:1.98E-04:5.2:5.2 / / GTF2A2:0.001558095:0.66:0.66 / / GTF2E2:4.37E-04:2.35:2.35 / / GTPBP8:3.25E-04:3.16:3.16 / / H2AFV:0.001969562:0.52:0.52 / / HACD3:3.12E-04:3.29:3.29 / / HADH:2.43E-04:4.23:4.23 / / HAT1:5.98E-04:1.72:1.72 / / HDAC2:8.39E-04:1.23:1.23 / / HDAC6:1.78E-04:5.77:5.77 / / HDGFL3:5.27E-05:19.5:19.5 / / HEATR1:3.16E-04:3.25:3.25 / / HEBP1:7.20E-04:1.43:1.43 / / HERC6:2.52E-04:4.08:4.08 / / HERPUD1:0.002326027:0.44:0.44 / / HES1:2.41E-05:42.56:30 / / HIF1A:0.00121677:0.84:0.84 / / HIST1H2BK:1.80E-05:57.05:30 / / HIST2H2BE:1.67E-04:6.17:6.17 / / HK1:0.001066135:0.96:0.96 / / HLA-DMA:0.003288289:0.31:0.31 / / HLA-DRA:0.034909373:0.03:0.03 / / HMG20B:4.45E-04:2.31:2.31 / / HMGA2:2.30E-06:446.89:30 / / HMGCR:3.03E-04:3.39:3.39 / / HMGCS1:2.81E-04:3.66:3.66 / / HMOX1:8.60E-04:1.2:1.2 / / HOMER2:4.01E-05:25.66:25.66 / / HOOK2:1.94E-04:5.3:5.3 / / HOXA10:2.17E-05:47.46:30 / / HOXA5:2.11E-06:487.52:30 / / HPRT1:5.64E-04:1.82:1.82 / / HS2ST1:1.98E-04:5.19:5.19 / / HSD17B10:9.93E-04:1.03:1.03 / / HSD17B11:0.002019774:0.51:0.51 / / HSPA1A:0.001468595:0.7:0.7 / / HSPA4:9.22E-04:1.11:1.11 / / HSPA8:0.008523718:0.12:0.12 / / HSPB1:0.001053487:0.98:0.98 / / HSPD1:0.00229498:0.45:0.45 / / HTATSF1:7.55E-04:1.36:1.36 / / HTRA1:6.71E-06:153.22:30 / / HYOU1:2.80E-04:3.67:3.67 / / IARS2:7.05E-04:1.46:1.46 / / ICAM1:3.21E-04:3.2:3.2 / / ICAM3:0.002825077:0.36:0.36 / / ICMT:1.67E- 04:6.16:6.16 / / ID2:0.002185549:0.47:0.47 / / IDE:2.65E-04:3.88:3.88 / / IER3:3.18E-04:3.23:3.23 / / IFNAR1:0.001448472:0.71:0.71 / / IFRD2:2.92E-04:3.52:3.52 / / IGF1R:2.55E-04:4.02:4.02 / / IGF2BP2:2.11E-04:4.88:4.88 / / IGF2R:0.001004617:1.02:1.02 / / IGFBP3:2.24E-05:45.84:30 / / IGHMBP2:8.89E-05:11.56:11.56 / / IKBKB:7.69E-04:1.34:1.34 / / IKBKE:3.26E-04:3.15:3.15 / / IKZF1:0.004537065:0.23:0.23 / / IL13RA1:0.001050612:0.98:0.98 / / IL1B:3.79E-04:2.71:2.71 / / IL4R:7.18E-04:1.43:1.43 / / ILK:9.69E-04:1.06:1.06 / / INPP1:1.59E-04:6.48:6.48 / / INPP4B:6.02E- 04:1.71:1.71 / / INSIG1:4.35E-04:2.36:2.36 / / INTS3:3.83E-04:2.68:2.68 / / IPO13:8.07E- 05:12.74:12.74 / / IQGAP1:0.005290431:0.19:0.19 / / ISOC1:1.98E-04:5.19:5.19 / / ITFG1:5.41E- 04:1.9:1.9 / / ITGAE:6.50E-04:1.58:1.58 / / ITGB1BP1:8.81E-04:1.17:1.17 / / ITGB5:1.07E-05:95.76:30 / / JADE2:4.55E-04:2.26:2.26 / / JMJD6:3.36E-04:3.06:3.06 / / JPT2:1.46E-04:7.05:7.05 / / JUN:0.005193649:0.2:0.2 / / KAT6A:0.001378713:0.75:0.75 / / KAT6B:9.86E-04:1.04:1.04 / / KCNK1:3.26E-06:315.45:30 / / KCTD5:2.58E-04:3.98:3.98 / / KDELR2:0.001165792:0.88:0.88 / / KDM3A:3.20E-04:3.22:3.22 / / KDM5A:0.002105249:0.49:0.49 / / KDM5B:6.87E-04:1.5:1.5 / / KEAP1:2.18E-04:4.71:4.71 / / KHDC4:6.08E-04:1.69:1.69 / / KIAA0100:6.85E-04:1.5:1.5 / / KIAA0355:3.80E-04:2.7:2.7 / / KIAA0753:7.40E-05:13.89:13.89 / / KIF14:1.05E-05:97.5:30 / / KIF1BP:1.34E-04:7.69:7.69 / / KIF20A:1.34E-06:766.1:30 / / KIF2C:1.07E-05:95.76:30 / / KIF5C:5.23E- 05:19.64:19.64 / / KIT:6.13E-06:167.58:30 / / KLHDC2:5.43E-04:1.89:1.89 / / KLHL21:1.07E- 04:9.65:9.65 / / KLHL9:3.84E-04:2.68:2.68 / / KTN1:0.002849991:0.36:0.36 / / LAGE3:5.62E- 04:1.83:1.83 / / LAMA3:1.92E-07:5362.7:30 / / LAP3:0.00533911:0.19:0.19 / / LBR:0.00160524:0.64:0.64 / / LGALS8:8.92E-04:1.15:1.15 / / LGMN:9.22E-05:11.15:11.15 / / LIG1:1.67E-04:6.14:6.14 / / LIPA:0.001340383:0.77:0.77 / / LOXL1:9.39E-06:109.44:30 / / LPAR2:2.13E-04:4.82:4.82 / / LPGAT1:0.001435632:0.72:0.72 / / LRP10:0.0010073:1.02:1.02 / / LRRC41:3.11E-04:3.31:3.31 / / LSM5:0.001475111:0.7:0.7 / / LSM6:0.001804745:0.57:0.57 / / LSR:1.01E-04:10.18:10.18 / / LYN:0.004584402:0.22:0.22 / / LYPLA1:0.001569977:0.65:0.65 / / LYRM1:3.79E-04:2.71:2.71 / / MACF1:0.002054462:0.5:0.5 / / MALT1:9.67E-04:1.06:1.06 / / MAMLD1:1.15E-06:893.78:30 / / MAN2B1:0.001420875:0.72:0.72 / / MAP2K5:1.50E-04:6.84:6.84 / / MAP3K4:3.66E-04:2.81:2.81 / / MAP4K4:0.00138427:0.74:0.74 / / MAP7:3.99E-05:25.78:25.78 / / MAPK13:2.17E-04:4.75:4.75 / / MAPK1IP1L:9.48E-04:1.08:1.08 / / MAPK9:0.000329251:3.12:3.12 / / MAPKAPK2:6.10E-04:1.68:1.68 / / MAPKAPK3:0.001021673:1.01:1.01 / / MAPKAPK5:4.79E- 04:2.14:2.14 / / MAST2:3.81E-05:26.95:26.95 / / MAT2A:0.001018799:1.01:1.01 / / MBNL1:0.006419619:0.16:0.16 / / MBNL2:3.79E-04:2.71:2.71 / / MBOAT7:9.23E-04:1.11:1.11 / / MBTPS1:7.60E-04:1.35:1.35 / / MCM3:3.70E-04:2.78:2.78 / / MCOLN1:4.14E-04:2.49:2.49 / / MCUR1:3.50E-04:2.94:2.94 / / ME2:0.001496576:0.69:0.69 / / MEF2C:0.002558687:0.4:0.4 / / MELK:5.37E-06:191.53:30 / / MEST:4.79E-05:21.45:21.45 / / METRN:1.96E-04:5.25:5.25 / / MFSD10:7.53E-04:1.36:1.36 / / MICALL1:6.65E-05:15.45:15.45 / / MIF:0.006789307:0.15:0.15 / / MINDY1:1.17E-04:8.75:8.75 / / MKNK1:4.81E-04:2.14:2.14 / / MLEC:0.001034514:0.99:0.99 / / MLLT11:1.61E-04:6.38:6.38 / / MMP1:0:Inf:30 / / MMP2:0:Inf:30 / / MNAT1:3.28E-04:3.13:3.13 / / MOK:2.18E-05:47.04:30 / / MPC2:8.84E-04:1.16:1.16 / / MPZL1:2.48E-04:4.14:4.14 / / MRPL12:4.92E-04:2.09:2.09 / / MRPL19:5.33E-04:1.93:1.93 / / MRPS16:8.82E-04:1.16:1.16 / / MRPS2:3.27E-04:3.14:3.14 / / MSH6:3.51E-04:2.92:2.92 / / MSRA:2.49E-04:4.12:4.12 / / MTA1:2.82E-04:3.64:3.64 / / MTERF3:1.44E-04:7.12:7.12 / / MTF2:9.43E-04:1.09:1.09 / / MTFR1:9.54E-05:10.77:10.77 / / MTHFD2:7.13E-04:1.44:1.44 / / MUC1:1.32E-05:77.72:30 / / MVP:0.001317385:0.78:0.78 / / MYBL2:7.00E-05:14.69:14.69 / / MYC:9.55E-04:1.08:1.08 / / MYCBP:6.48E-04:1.59:1.59 / / MYCBP2:0.003329685:0.31:0.31 / / MYL9:1.94E-05:53.1:30 / / MYLK:3.47E-05:29.63:29.63 / / MYO10:2.09E-05:49.2:30 / / NCAPD2:2.47E-04:4.16:4.16 / / NCK1:7.94E-04:1.29:1.29 / / NCK2:7.75E- 04:1.33:1.33 / / NCOA3:0.00153548:0.67:0.67 / / NENF:7.66E-04:1.34:1.34 / / NET1:5.86E- 05:17.53:17.53 / / NFATC3:5.37E-04:1.91:1.91 / / NFATC4:9.58E-07:1072.54:30 / / NFE2L2:0.001510375:0.68:0.68 / / NFIL3:2.36E-04:4.36:4.36 / / NFKB2:4.84E-04:2.13:2.13 / / NFKBIA:0.001746293:0.59:0.59 / / NFKBIB:2.90E-04:3.54:3.54 / / NFKBIE:2.49E-04:4.13:4.13 / / NGRN:5.54E-04:1.85:1.85 / / NIPSNAP1:2.32E-04:4.43:4.43 / / NIT1:3.49E-04:2.94:2.94 / / NMT1:9.84E-04:1.04:1.04 / / NNT:6.63E-04:1.55:1.55 / / NOL3:4.73E-05:21.71:21.71 / / NOLC1:4.91E- 04:2.09:2.09 / / NOS3:1.40E-05:73.46:30 / / NOSIP:0.003205114:0.32:0.32 / / NOTCH1:4.16E- 04:2.47:2.47 / / NPC1:2.53E-04:4.06:4.06 / / NPDC1:1.69E-04:6.07:6.07 / / NPEPL1:3.25E- 04:3.17:3.17 / / NPRL2:2.86E-04:3.6:3.6 / / NR1H2:7.09E-04:1.45:1.45 / / NR2F6:8.99E-05:11.43:11.43 / / NR3C1:0.001773315:0.58:0.58 / / NRAS:4.42E-04:2.33:2.33 / / NRIP1:8.90E-04:1.15:1.15 / / NSDHL:1.16E-04:8.85:8.85 / / NT5DC2:3.93E-05:26.16:26.16 / / NUCB2:0.001473387:0.7:0.7 / / NUDCD3:4.09E-04:2.51:2.51 / / NUDT9:2.49E-04:4.13:4.13 / / NUP133:2.18E-04:4.72:4.72 / / NUP62:9.30E-04:1.11:1.11 / / NUP85:2.32E-04:4.44:4.44 / / NUP88:3.77E-04:2.73:2.73 / / NUP93:3.65E-04:2.82:2.82 / / NUSAP1:5.58E-05:18.43:18.43 / / NVL:2.17E-04:4.74:4.74 / / ORC1:7.09E-06:144.94:30 / / OXA1L:0.001245518:0.83:0.83 / / OXCT1:2.38E-04:4.32:4.32 / / OXSR1:5.06E-04:2.03:2.03 / / P4HA2:5.94E-06:172.99:30 / / P4HTM:1.89E-04:5.43:5.43 / / PACSIN3:7.67E-07:1340.68:30 / / PAF1:3.08E-04:3.33:3.33 / / PAFAH1B1:0.001303204:0.79:0.79 / / PAFAH1B3:3.10E-04:3.31:3.31 / / PAICS:3.88E-04:2.65:2.65 / / PAK1:0.001364147:0.75:0.75 / / PAK4:7.15E-05:14.38:14.38 / / PAK6:4.02E-06:255.37:30 / / PAN2:1.39E-04:7.38:7.38 / / PARP1:0.001335209:0.77:0.77 / / PARP2:8.85E-05:11.61:11.61 / / PAX8:2.99E-05:34.38:30 / / PCBD1:3.34E-04:3.08:3.08 / / PCCB:1.64E-04:6.28:6.28 / / PCK2:2.32E-04:4.43:4.43 / / PCM1:0.001885237:0.55:0.55 / / PCMT1:0.001500984:0.68:0.68 / / PCNA:4.28E-04:2.4:2.4 / / PDGFA:1.23E-05:83.79:30 / / PDHX:2.40E-04:4.29:4.29 / / PDIA5:5.94E-05:17.3:17.3 / / PDLIM1:3.62E-04:2.84:2.84 / / PDS5A:9.67E-04:1.06:1.06 / / PECR:1.15E-04:8.95:8.95 / / PEX11A:2.28E-05:45.06:30 / / PFKL:8.25E-04:1.25:1.25 / / PGAM1:0.001767949:0.58:0.58 / / PGM1:2.67E-04:3.85:3.85 / / PGRMC1:3.65E-04:2.82:2.82 / / PHGDH:5.33E-05:19.29:19.29 / / PHKA1:2.30E-06:446.89:30 / / PHKB:0.001243409:0.83:0.83 / / PHKG2:4.09E-04:2.51:2.51 / / PIGB:1.34E-04:7.67:7.67 / / PIH1D1:7.27E-04:1.41:1.41 / / PIK3C2B:9.45E-05:10.88:10.88 / / PIK3C3:5.40E-04:1.9:1.9 / / PIK3CA:5.47E-04:1.88:1.88 / / PIK3R3:2.57E-05:40.02:30 / / PIK3R4:1.72E-04:5.99:5.99 / / PIN1:9.55E-04:1.08:1.08 / / PIP4K2B:2.32E-04:4.42:4.42 / / PKIG:1.69E- 04:6.07:6.07 / / PLA2G15:1.27E-04:8.06:8.06 / / PLA2G4A:1.12E-04:9.2:9.2 / / PLCB3:8.60E- 05:11.94:11.94 / / PLEKHJ1:8.93E-04:1.15:1.15 / / PLEKHM1:3.62E-04:2.84:2.84 / / PLK1:1.19E- 05:86.5:30 / / PLOD3:2.26E-04:4.54:4.54 / / PLP2:0.002580918:0.4:0.4 / / PLS1:1.30E-05:78.86:30 / / PLSCR1:0.002077651:0.49:0.49 / / PLSCR3:5.49E-04:1.87:1.87 / / PMAIP1:5.35E-04:1.92:1.92 / / PMM2:4.10E-04:2.51:2.51 / / PNKP:6.06E-04:1.7:1.7 / / PNP:6.82E-04:1.51:1.51 / / POLB:3.84E- 04:2.68:2.68 / / POLD4:0.00222062:0.46:0.46 / / POLE2:1.11E-05:92.46:30 / / POLG2:2.55E- 04:4.04:4.04 / / POLR1C:1.72E-04:5.99:5.99 / / POLR2I:0.001108298:0.93:0.93 / / POLR2K:0.001078401:0.95:0.95 / / POP4:5.37E-04:1.92:1.92 / / PPARD:1.27E-04:8.06:8.06 / / PPARG:1.34E-05:76.61:30 / / PPIC:5.75E-07:1787.57:30 / / PPIE:5.90E-04:1.74:1.74 / / PPOX:1.04E- 04:9.89:9.89 / / PPP1R13B:5.04E-05:20.39:20.39 / / PPP2R3C:7.42E-04:1.38:1.38 / / PPP2R5A:8.94E- 04:1.15:1.15 / / PPP2R5E:0.001131104:0.91:0.91 / / PRAF2:1.48E-04:6.96:6.96 / / PRCP:8.13E- 04:1.26:1.26 / / PRKACA:8.81E-04:1.17:1.17 / / PRKAG2:0.001384654:0.74:0.74 / / PRKCD:9.49E- 04:1.08:1.08 / / PRKCH:6.95E-04:1.48:1.48 / / PRKCQ:2.28E-04:4.51:4.51 / / PRKX:0.001049845:0.98:0.98 / / PROS1:7.86E-06:130.8:30 / / PRPF4:3.97E-04:2.59:2.59 / / PRR15L:0:Inf:30 / / PRR7:1.33E-04:7.74:7.74 / / PRSS23:1.19E-04:8.61:8.61 / / PRUNE1:1.16E-04:8.86:8.86 / / PSIP1:0.001787305:0.58:0.58 / / PSMB10:0.004440474:0.23:0.23 / / PSMB8:0.004336218:0.24:0.24 / / PSMD10:3.08E-04:3.34:3.34 / / PSMD2:9.45E-04:1.09:1.09 / / PSMD4:0.001601982:0.64:0.64 / / PSMD9:0.001058853:0.97:0.97 / / PSME1:0.00943481:0.11:0.11 / / PSME2:0.007643289:0.13:0.13 / / PSMF1:9.66E-04:1.06:1.06 / / PSMG1:2.99E-04:3.44:3.44 / / PSRC1:3.01E-05:34.16:30 / / PTGS2:1.58E-04:6.48:6.48 / / PTK2:1.45E-04:7.09:7.09 / / PTK2B:0.001290555:0.8:0.8 / / PTPN1:0.001088367:0.94:0.94 / / PTPN12:9.81E-04:1.05:1.05 / / PTPN6:0.004073661:0.25:0.25 / / PTPRC:0.022096966:0.05:0.05 / / PTPRF:1.32E-05:77.72:30 / / PTPRK:4.08E-05:25.18:25.18 / / PUF60:0.001127654:0.91:0.91 / / PWP1:5.76E-04:1.78:1.78 / / PXMP2:1.08E-04:9.54:9.54 / / PXN:7.42E-04:1.38:1.38 / / PYCR1:9.01E-06:114.1:30 / / PYGL:9.01E- 04:1.14:1.14 / / RAB11FIP2:3.99E-04:2.57:2.57 / / RAB21:0.001132254:0.91:0.91 / / RAB27A:0.001464379:0.7:0.7 / / RAB31:0.001929508:0.53:0.53 / / RAB4A:7.75E-04:1.33:1.33 / / RAC2:0.005875531:0.17:0.17 / / RAD51C:2.55E-04:4.03:4.03 / / RAD9A:2.75E-04:3.73:3.73 / / RAE1:4.41E-04:2.33:2.33 / / RAI14:7.67E-07:1340.68:30 / / RALA:6.94E-04:1.48:1.48 / / RALB:9.30E- 04:1.11:1.11 / / RALGDS:2.53E-04:4.06:4.06 / / RAP1GAP:1.92E-07:5362.7:30 / / RASA1:3.08E- 04:3.34:3.34 / / RB1:0.001012474:1.02:1.02 / / RBKS:3.72E-05:27.64:27.64 / / RBM15B:3.63E- 04:2.83:2.83 / / RBM34:1.35E-04:7.63:7.63 / / RBM6:0.001650086:0.62:0.62 / / REEP5:0.002175775:0.47:0.47 / / RELB:3.64E-04:2.82:2.82 / / RFC2:2.77E-04:3.71:3.71 / / RFC5:9.22E-05:11.15:11.15 / / RFNG:3.12E-04:3.3:3.3 / / RFX5:2.92E-04:3.52:3.52 / / RGS2:0.003237503:0.32:0.32 / / RHEB:0.001018032:1.01:1.01 / / RHOA:0.00783417:0.13:0.13 / / RNF167:7.74E-04:1.33:1.33 / / RNH1:0.002867239:0.36:0.36 / / RNMT:0.001228269:0.84:0.84 / / RNPS1:0.001644336:0.63:0.63 / / RPA1:4.55E-04:2.26:2.26 / / RPA2:0.001013432:1.01:1.01 / / RPA3:7.57E-04:1.36:1.36 / / RPIA:4.78E-04:2.15:2.15 / / RPL39L:1.05E-04:9.77:9.77 / / RPN1:0.001142028:0.9:0.9 / / RPP38:4.90E-04:2.1:2.1 / / RPS5:0.035157173:0.03:0.03 / / RPS6:0.072718182:0.01:0.01 / / RPS6KA1:0.001092583:0.94:0.94 / / RRAGA:6.30E-04:1.63:1.63 / / RRP12:3.80E-04:2.7:2.7 / / RRP1B:6.50E-04:1.58:1.58 / / RRP8:3.32E-04:3.09:3.09 / / RRS1:3.52E- 04:2.92:2.92 / / RSU1:7.04E-04:1.46:1.46 / / RTN2:2.26E-05:45.45:30 / / RUVBL1:4.73E-04:2.17:2.17 / / RXYLT1:2.18E-04:4.7:4.7 / / S100A13:4.91E-05:20.95:20.95 / / S100A4:0.043490201:0.02:0.02 / / SACM1L:6.48E-04:1.59:1.59 / / SATB1:0.001188598:0.86:0.86 / / SCAND1:0.001949631:0.53:0.53 / / SCARB1:9.91E-05:10.37:10.37 / / SCCPDH:2.88E-04:3.57:3.57 / / SCP2:0.00179248:0.57:0.57 / / SCRN1:2.43E-04:4.24:4.24 / / SCYL3:2.44E-04:4.22:4.22 / / SDHB:0.001153527:0.89:0.89 / / SENP6:0.001471853:0.7:0.7 / / SERPINE1:1.53E-06:670.34:30 / / SESN1:2.68E-04:3.84:3.84 / / SFN:1.72E-06:595.86:30 / / SGCB:6.52E-05:15.77:15.77 / / SH3BP5:0.001460546:0.7:0.7 / / SHB:6.13E-06:167.58:30 / / SHC1:3.49E-04:2.94:2.94 / / SIRT3:8.91E-05:11.53:11.53 / / SKIV2L:2.01E-04:5.11:5.11 / / SKP1:0.005183109:0.2:0.2 / / SLC11A2:3.11E-04:3.31:3.31 / / SLC1A4:9.14E-05:11.24:11.24 / / SLC25A13:2.52E-04:4.08:4.08 / / SLC25A14:5.63E-05:18.24:18.24 / / SLC25A4:1.14E-04:9.01:9.01 / / SLC25A46:4.16E-04:2.47:2.47 / / SLC27A3:5.67E-04:1.81:1.81 / / SLC2A6:6.57E-04:1.56:1.56 / / SLC35A1:4.90E-04:2.1:2.1 / / SLC35A3:6.39E-04:1.61:1.61 / / SLC35B1:4.15E-04:2.47:2.47 / / SLC35F2:3.64E-05:28.22:28.22 / / SLC37A4:4.52E-05:22.72:22.72 / / SLC5A6:8.22E-05:12.5:12.5 / / SMAD3:4.48E-04:2.3:2.3 / / SMARCA4:6.53E-04:1.57:1.57 / / SMARCC1:0.001050612:0.98:0.98 / / SMARCD2:7.30E-04:1.41:1.41 / / SMC1A:0.001315086:0.78:0.78 / / SMC3:0.00137488:0.75:0.75 / / SMC4:5.34E-04:1.93:1.93 / / SMNDC1:8.56E-04:1.2:1.2 / / SNAP25:5.75E-07:1787.57:30 / / SNCA:1.63E-04:6.3:6.3 / / SNX11:3.22E-04:3.19:3.19 / / SNX13:5.82E-04:1.77:1.77 / / SNX6:0.00282546:0.36:0.36 / / SNX7:1.72E-06:595.86:30 / / SOCS2:1.46E-04:7.05:7.05 / / SORBS3:1.55E-04:6.61:6.61 / / SOX2:3.83E-07:2681.35:30 / / SOX4:7.09E-04:1.45:1.45 / / SPAG4:4.98E-06:206.26:30 / / SPAG7:9.31E-04:1.1:1.1 / / SPDEF:0:Inf:30 / / SPEN:0.001201822:0.86:0.86 / / SPP1:0:Inf:30 / / SPR:1.71E-05:60.26:30 / / SPRED2:2.86E-05:35.99:30 / / SPTAN1:8.08E-04:1.27:1.27 / / SPTLC2:0.001308953:0.79:0.79 / / SQOR:0.001549471:0.66:0.66 / / SQSTM1:0.002176733:0.47:0.47 / / SRC:2.24E-04:4.59:4.59 / / SSBP2:4.26E-04:2.41:2.41 / / ST3GAL5:0.001043521:0.98:0.98 / / ST6GALNAC2:2.11E-06:487.52:30 / / ST7:5.60E-05:18.37:18.37 / / STAMBP:3.96E-04:2.59:2.59 / / STAP2:8.43E-06:121.88:30 / / STAT1:0.007257694:0.14:0.14 / / STAT3:0.001946564:0.53:0.53 / / STAT5B:6.63E-04:1.55:1.55 / / STIMATE:2.41E-04:4.27:4.27 / / STK10:0.001817202:0.57:0.57 / / STK25:3.96E-04:2.6:2.6 / / STMN1:7.17E-04:1.43:1.43 / / STUB1:0.001358015:0.76:0.76 / / STX1A:8.82E-06:116.58:30 / / STX4:6.09E-04:1.69:1.69 / / STXBP1:2.11E-05:48.75:30 / / STXBP2:0.002392912:0.43:0.43 / / SUPV3L1:2.55E-04:4.04:4.04 / / SUV39H1:3.32E-05:31:30 / / SUZ12:0.001175758:0.87:0.87 / / SYK:0.001798421:0.57:0.57 / / SYNE2:0.00153778:0.67:0.67 / / SYNGR3:4.02E-06:255.37:30 / / SYPL1:0.001019565:1.01:1.01 / / TARBP1:2.89E-04:3.55:3.55 / / TATDN2:2.90E-04:3.55:3.55 / / TBC1D31:1.49E-04:6.9:6.9 / / TBC1D9B:3.72E-04:2.76:2.76 / / TBP:2.21E-04:4.66:4.66 / / TBPL1:3.95E-04:2.6:2.6 / / TBX2:0:Inf:30 / / TBXA2R:3.85E-05:26.68:26.68 / / TCEA2:2.81E-04:3.66:3.66 / / TCEAL4:4.08E-04:2.52:2.52 / / TCERG1:8.19E-04:1.26:1.26 / / TCFL5:8.26E-05:12.44:12.44 / / TCTA:2.32E-04:4.43:4.43 / / TCTN1:8.30E-05:12.38:12.38 / / TENT4A:2.45E-04:4.2:4.2 / / TERF2IP:0.001836175:0.56:0.56 / / TERT:7.67E-07:1340.68:30 / / TES:0.001741885:0.59:0.59 / / TESK1:1.78E-04:5.78:5.78 / / TEX10:1.58E-04:6.52:6.52 / / TFAP2A:3.83E-07:2681.35:30 / / TFDP1:4.03E-04:2.55:2.55 / / TGFB3:1.49E-05:68.75:30 / / TGFBR2:0.001544105:0.67:0.67 / / THAP11:5.68E-04:1.81:1.81 / / TIAM1:5.21E-04:1.97:1.97 / / TICAM1:1.75E-04:5.89:5.89 / / TIMELESS:5.54E-05:18.56:18.56 / / TIMM17B:6.84E-04:1.5:1.5 / / TIMM22:2.28E-04:4.51:4.51 / / TIMM9:5.87E-04:1.75:1.75 / / TIMP2:0.00164012:0.63:0.63 / / TIPARP:2.95E-04:3.48:3.48 / / TJP1:0:Inf:30 / / TLCD3A:3.83E-06:268.14:30 / / TLE1:1.53E- 04:6.74:6.74 / / TLK2:5.39E-04:1.91:1.91 / / TLR4:0.001284039:0.8:0.8 / / TM9SF2:0.001493318:0.69:0.69 / / TM9SF3:0.001311828:0.78:0.78 / / TMCO1:0.001743993:0.59:0.59 / / TMED10:0.00174016:0.59:0.59 / / TMEM109:6.69E-04:1.54:1.54 / / TMEM50A:0.00246133:0.42:0.42 / / TMEM97:7.24E-05:14.19:14.19 / / TNFRSF21:3.56E- 05:28.83:28.83 / / TNIP1:8.11E-04:1.27:1.27 / / TOMM34:1.08E-04:9.53:9.53 / / TOMM70:4.18E- 04:2.46:2.46 / / TOP2A:3.20E-05:32.11:30 / / TOPBP1:4.82E-04:2.13:2.13 / / TOR1A:4.98E- 04:2.06:2.06 / / TP53:8.27E-04:1.24:1.24 / / TP53BP1:1.81E-04:5.68:5.68 / / TP53BP2:3.28E- 04:3.13:3.13 / / TPD52L2:5.40E-04:1.9:1.9 / / TPM1:2.26E-04:4.55:4.55 / / TRAK2:3.97E-04:2.59:2.59 / / TRAM2:1.22E-04:8.46:8.46 / / TRAP1:2.58E-04:3.99:3.99 / / TRAPPC3:8.03E-04:1.28:1.28 / / TRAPPC6A:0.000931024:1.1:1.1 / / TRIB1:2.26E-04:4.55:4.55 / / TRIB3:9.81E-05:10.47:10.47 / / TRIM13:5.24E-04:1.96:1.96 / / TRIM2:1.65E-05:62.36:30 / / TSC22D3:0.004847151:0.21:0.21 / / TSEN2:1.19E-04:8.62:8.62 / / TSKU:0:Inf:30 / / TSPAN3:7.14E-04:1.44:1.44 / / TSPAN4:3.08E- 04:3.33:3.33 / / TSPAN6:5.75E-06:178.76:30 / / TSTA3:5.03E-04:2.04:2.04 / / TUBB6:6.55E- 05:15.68:15.68 / / TWF2:0.001334442:0.77:0.77 / / TXLNA:4.10E-04:2.51:2.51 / / TXNDC9:6.19E- 04:1.66:1.66 / / TXNL4B:1.73E-04:5.93:5.93 / / TXNRD1:6.35E-04:1.62:1.62 / / UBE2A:9.41E- 04:1.09:1.09 / / UBE2C:1.05E-05:97.5:30 / / UBE2J1:0.001779831:0.58:0.58 / / UBE2L6:0.003388329:0.3:0.3 / / UBE3B:2.56E-04:4.02:4.02 / / UBE3C:4.65E-04:2.21:2.21 / / UBQLN2:7.33E-04:1.4:1.4 / / UBR7:1.69E-04:6.07:6.07 / / UFM1:0.001429499:0.72:0.72 / / UGDH:1.18E-04:8.71:8.71 / / USP1:8.90E-04:1.15:1.15 / / USP14:8.34E-04:1.23:1.23 / / USP22:7.88E- 04:1.3:1.3 / / USP6NL:2.77E-04:3.71:3.71 / / USP7:0.001187832:0.87:0.87 / / UTP14A:3.39E- 04:3.03:3.03 / / VAPB:3.40E-04:3.02:3.02 / / VAT1:1.08E-04:9.47:9.47 / / VAV3:4.98E-...

Claims

CLAIMS 1. A method of performing single cell RNA sequencing of a sample comprising: Reverse-transcribing one or more RNA molecules present in the sample to obtain one or more corresponding cDNA molecules; and Amplifying the cDNA molecules to obtain a sequencing library; wherein the method comprises one or more steps that use a plurality of target-specific capture reagents associated with a respective plurality of target transcripts, to enrich the sequencing library for cDNA molecules corresponding to the plurality of target transcripts, wherein the concentration of a first target-specific capture reagent for a first target transcript is different from the concentration of a second target-specific capture reagent for a second target transcript.

2. The method of claim 1, wherein the method comprises capturing the one or more RNA molecules using target-specific reagents, and / or the cDNA amplification is performed using target-specific capture reagents, and / or the method comprises a capture step that uses target-specific capture reagents prior to cDNA amplification and / or the method comprises a capture step that uses target- specific capture reagents prior to reverse transcription.

3. The method of any preceding claim, wherein a target-specific capture reagent comprises: a primer, a pair of primers, a probe or a pair of probes.

4. The method of any preceding claim, wherein the concentration of the first and second target- specific capture reagents are dependent on the expected expression level of the first and second target transcripts in the sample.

5. The method of any preceding claim, wherein the method comprises selecting the concentration of the first and second target-specific capture reagents based on the expected expression level of the first and second target transcripts in the sample, or selecting a composition comprising said target-specific capture reagents at said concentrations.

6. The method of any preceding claim, wherein the concentration of a target-specific capture reagent for a subset or all of the plurality of target transcripts is dependent on the expected expression level of respective transcripts in the sample, and / or wherein the method comprises selecting the concentration of a subset or all of the target-specific capture reagents based on the expected expression level of each of the transcripts in the sample, or selecting a composition comprising said target-specific capture reagents at said concentrations.

7. The method of any of claims 4 to 6, wherein the expected expression level of a transcript in the sample is based on the expression level of the transcript in a reference sample or set of samples, or wherein the expected expression level of a transcript in the sample is the expression level of the transcript in a reference sample or set of samples.

8. The method of claim 7, wherein a reference sample is a sample from the same organism as the organism from which the sample to be analysed originates, and / or a sample comprising cells from the same organ as an organ from which cells in the sample to be analysed originate, and / or a sample comprising cells from the same tissue as a tissue from which cells in the sample to be analysed originate, and / or a sample comprising cells from the same cell type as cells in the sample to be analysed, and / or a sample comprising cells from the same cell line as cells in the sample to be analysed, and / or a sample comprising cells from a cell line in the same category of cell lines as a cell line in the sample to be analysed, where categories of cell lines are defined by tissue of origin and / or one or more phenotypic characteristics.

9. The method of any of claims 4 to 8, wherein the expected expression level of the first and second or the plurality of transcripts in the sample is based on a reference dataset comprising expression levels for the transcripts in one or more reference samples, wherein the expression levels in each reference samples are independently selected from: expression levels from bulk RNA sequencing and expression levels from single cell RNA sequencing.

10. The method of any preceding claim, wherein the plurality of target-specific capture reagents comprise a target-specific capture reagent or a set of target-specific capture reagents for each of the plurality of target transcripts, wherein the concentration of each of the target-specific capture reagents in a set of target-specific capture reagents for a transcript is the same, optionally wherein a set of target-specific capture reagents comprises a plurality of different target-specific capture reagents for the same target transcript.

11. The method of any preceding claim, wherein the relative concentrations of the target-specific capture reagents for a subset or all of the plurality of target transcripts are dependent on the expected relative expression levels of respective transcripts in the sample, and / or wherein the method comprises selecting the relative concentrations of a subset or all of the target-specific capture reagents based on the expected relative expression level of each of the transcripts in the sample, or selecting a composition comprising said target-specific capture reagents at said concentrations.

12. The method of any preceding claim, wherein the expected expression level of the first transcript in the sample is higher than the expression level of the second transcript in the sample, and wherein the concentration of the first target-specific capture reagent is lower than the concentration of the second target-specific capture reagent.

13. The method of any preceding claim, wherein the concentration of each target-specific capture reagent depends on the expected expression level of the target transcript of the target-specific capture reagent relative to the expected expression level of all other target transcripts, and / or wherein the concentration of a target-specific capture reagent that is specific for a target transcriptthat has a higher expected expression level relative to the expected expression level of another target transcript is lower than the concentration of a target-specific capture reagent that is specific for the other target transcript that has a lower expected expression level.

14. The method of any preceding claim, wherein the concentrations of the first target-specific capture reagent and the second target-specific capture reagent are such that the number of molecules corresponding to the first and second transcripts in the sequencing library are more similar to each other than the numbers of molecules that would be obtained in a sequencing library produced using equal concentrations of the first and second target-specific capture reagents.

15. The method of any preceding claim, wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent such that the numbers of molecules corresponding to the first and second transcripts in the sequencing library are more similar to each other than the number of molecules that would be obtained in a sequencing library produced using equal concentrations of the first and second target-specific capture reagents, and / or wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent such that the numbers of molecules corresponding to the first and second transcripts in the sequencing library are expected to be the same.

16. The method of any preceding claim, wherein the plurality of target transcripts comprise at least 100, at least 200, at least 500, at least 1000, at least 2000, at least 3000, at least 4000 or at least 5000 different transcripts, wherein the plurality of target transcripts comprise transcripts from at least 200, at least 500, at least 1000, at least 2000, at least 3000, at least 4000 or at least 5000 different genes, or wherein the plurality of target transcripts comprises transcripts from at least 20%, at least 30%, at least 40%, at least 50%, at least 60%, at least 70%, or at least 80% of the genes expected to be expressed in the sample, and / or wherein the expected expression level of the first target transcript in the sample differs from the expected expression level of the second target transcript in the sample by a factor of at least 2, at least 5 or at least 10, or wherein the plurality of transcripts comprises transcripts that have expected expression levels in the sample that differ by a factor of at least 2, at least 5 or at least 10.

17. The method of any preceding claim, wherein the concentrations of the first target-specific capture reagent and the second target-specific capture reagent are such that the sum of the expected percentages of the first and second transcripts in the sample that are not represented in the sequencing library is lower than the sum of the expected percentages of the first and second transcripts in the sample that are not represented in a sequencing library that would be obtained using equal concentrations of the first and second target-specific capture reagents.

18. The method of any preceding claim, wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent such that the sum of the expected percentages of the first and second transcripts in the sample that are not represented in the sequencing library is lower than the sum of the expected percentages of the first and second transcripts in the sample that are not represented in a sequencing library that would be obtained using equal concentrations of the first and second target-specific capture reagents.

19. The method of any preceding claim, wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent such that the sum of the expected percentages of the first and second transcripts in the sample that are not represented in the sequencing library is the lowest of the values expected for a plurality of candidate concentrations of the first and second target-specific capture reagents.

20. The method of any preceding claim, wherein the one or more steps that use a plurality of target- specific capture reagents associated with a respective plurality of target transcripts comprises: a capture step that precedes the step of reverse-transcribing the one or more RNA molecules present in the sample using a first plurality of target-specific capture reagents; and / or the step of amplifying the cDNA molecules, wherein the step of amplifying the cDNA molecules uses a second plurality of target-specific capture reagents.

21. The method of claim 20, wherein the concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the first plurality of target-specific reagents are such that the number of molecules corresponding to the first and second transcripts that proceed to reverse-transcription are more similar to each other than the numbers of molecules that would proceed to reverse-transcription using equal concentrations of the first and second target-specific capture reagents in the first plurality of target-specific capture reagents.

22. The method of claim 20 or claim 21, wherein the concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the first plurality of target-specific reagents are such that the unique molecular counts represented in the sequencing library for the first and second transcripts are more similar to each other than the unique molecular counts that would be represented in a sequencing library obtained using equal concentrations of the first and second target-specific capture reagents in the first plurality of target-specific capture.

23. The method of any of claims 20 to 22, wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the first plurality of target-specific capture reagents such that the unique molecular counts for the first and second transcripts represented in the sequencing library are more similar to each other than the unique molecular counts that would be obtained in a sequencing library produced using equalconcentrations of the first and second target-specific capture reagents in the first plurality of target- specific capture reagents.

24. The method of any of claims 20 to 23, wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the first plurality of target-specific capture reagents such that the unique molecular counts corresponding to the first and second transcripts represented in the sequencing library are expected to be the same.

25. The method of any of claims 20 to 24, wherein the concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target- specific reagents are such that the number of amplified cDNA molecules corresponding to the first and second transcripts that proceed to sequencing are more similar to each other than the numbers of amplified cDNA molecules that would proceed to reverse-transcription using equal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents, and / or wherein the concentrations of the first target-specific capture reagent and the second target- specific capture reagent in the second plurality of target-specific reagents are such that the number of cDNA molecules in the sequencing library corresponding to the first and second transcripts are more similar to each other than the number of cDNA molecules corresponding to the first and second transcripts in a sequencing library obtained using equal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents; and / or wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of cDNA molecules corresponding to the first and second transcripts represented in the sequencing library are more similar to each other than the number of cDNA molecules corresponding to the first and second transcripts that would be obtained in a sequencing library produced using equal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents, and / or wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of cDNA molecules corresponding to the first and second transcripts in the sequencing library are expected to be the same; and / or wherein the concentrations of the first target-specific capture reagent and the second target- specific capture reagent in the second plurality of target-specific reagents are such that the number of reads corresponding to the first and second transcripts obtained by sequencing the sequencing library are more similar to each other than the number of reads corresponding to the first and second transcripts that would be obtained by sequencing a sequencing library obtained usingequal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents; and / or wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of reads corresponding to the first and second transcripts obtained by sequencing the sequencing library are more similar to each other than the number of reads that would be obtained by sequencing a sequencing library produced using equal concentrations of the first and second target-specific capture reagents in the second plurality of target-specific capture reagents, and / or wherein the method comprises identifying concentrations of the first target-specific capture reagent and the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of reads corresponding to the first and second transcripts obtained by sequencing the sequencing library are expected to be the same.

26. A method of designing or providing a composition for performing RNA sequencing of a sample, the method comprising: Identifying a plurality of target-specific capture reagents associated with a respective plurality of target transcripts; Identifying an expected expression level of a first transcript of the plurality of transcript and a second transcript of the plurality of target transcripts in the sample, and Determining a concentration of a first target-specific capture reagent for the first target transcript and a concentration of a second target-specific capture reagent for the second target transcript using the expected expression level of the first and second transcripts in the sample.

27. The method of claim 26, wherein determining a concentration of a first target-specific capture reagent for the first target transcript and a concentration of a second target-specific capture reagent for the second target transcript using the expected expression level of the first and second transcripts in the sample comprises using the expected relative expression level of the first and second transcripts.

28. The method of any of claims 26 to 27, wherein the single cell RNA sequencing comprises Reverse-transcribing one or more RNA molecules present in the sample to obtain one or more corresponding cDNA molecules using a composition comprising target-specific capture reagents; and / or Amplifying the cDNA molecules to obtain a sequencing library using a composition comprising target-specific capture reagents; and / or Performing a capture step that uses a composition comprising target-specific capture reagents prior to or during cDNA amplification of reverse transcribed RNA molecules present in the sample and / or prior to reverse-transcription of RNA molecules present in the sample,29. The method of claim 28, wherein the single cell RNA sequencing comprises a plurality of steps using a composition comprising target-specific capture reagents, and wherein the method comprises determining a concentration of a first target-specific capture reagent for the first target transcript and a concentration of a second target-specific capture reagent for the second target transcript using the expected expression level of the first and second transcripts in the sample, separately for each composition comprising target-specific capture reagents.

30. The method of any of claims 26 to 29, wherein identifying an expected expression level of a first transcript of the plurality of transcript and a second transcript of the plurality of target transcripts, or each of the plurality of transcripts, in the sample comprises obtaining a reference dataset comprising expression levels of the first and second transcripts or each of the plurality of target transcripts in one or more reference samples.

31. The method of any of claims 26 to 30, wherein determining a concentration of capture reagents for a set of target transcripts comprises: determining an observed relative frequency of each target transcript in the set of target transcripts using the expected expression levels of the target transcripts in the sample; determining a target relative frequency of each target transcripts; and determining a weight for each target-specific capture reagent using the observed relative frequency and the target relative frequency for the target transcript of the target specific capture reagent.

32. The method of claim 31, wherein the weight is the ratio of the target relative frequency and the observed relative frequency for the target or a normalized version thereof, and / or wherein a concentration of a capture reagent is the product of the determined weight for the capture reagent and a concentration associated with the expected expression level of the target transcript or a default concentration that is the same for all capture reagents for the set of transcripts, and / or wherein the observed relative frequency of a transcript is the expected expression level of the transcript divided by the sum of the expected expression levels of all transcripts in the set, and / or wherein the target relative frequency is the inverse of the number of transcripts in the set.

33. The method of any of claims 31 or 32, wherein determining a weight for each target-specific capture reagent comprises determining the ratio of the target relative frequency and the observed relative frequency for each target and normalizing the resulting weights by dividing each weight by the highest weight of the resulting weights.

34. The method of any of claims 26 to 33, comprising determining a concentration of the plurality of target-specific capture reagents using the expected expression level of the plurality of target transcripts in the sample, and further comprising:identifying a plurality of groups of target transcripts, each group associated with a different range of expression levels and / or a plurality of groups of target-specific capture reagents, each group associated with a different range of determined concentrations, and selecting a common concentration for target-specific capture reagents in each group using the determined concentrations associated with the group.

35. The method of any of claims 26 to 34, wherein the single cell RNA sequencing comprises: Reverse-transcribing one or more RNA molecules present in the sample to obtain one or more corresponding cDNA molecules; and Amplifying the cDNA molecules to obtain a sequencing library; and the single cell RNA sequencing comprises a capture step that precedes the step of reverse- transcribing the one or more RNA molecules present in the sample using a first plurality of target- specific capture reagents; and / or the step of amplifying the cDNA molecules uses a second plurality of target-specific capture reagents, wherein the method comprises determining a concentration for each of the first and / or second plurality of target-specific reagents, wherein determining a concentration of a first target-specific capture reagent for the first target transcript and a concentration of a second target-specific capture reagent for the second target transcript using the expected expression level of the first and second transcripts in the sample comprises: (i) identifying a concentration of the first target-specific capture reagent and a concentration of the second target-specific capture reagent in the first plurality of target-specific capture reagents such that the unique molecular counts corresponding to the first and second transcripts represented in the sequencing library are expected to be the same; and / or (ii) identifying a concentration of the first target-specific capture reagent and a concentration of the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of cDNA molecules corresponding to the first and second transcripts in the sequencing library are expected to be the same; and / or (iii) identifying a concentration of the first target-specific capture reagent and a concentration of the second target-specific capture reagent in the second plurality of target-specific capture reagents such that the number of reads corresponding to the first and second transcripts obtained by sequencing the sequencing library are expected to be the same.

36. The method of any of claims 1 to 25, further comprising performing the method of any of claims 26 to 35.

Images