Modules for single-cell data conversion to transformed bulk-cell data and methods of using the same

Abstract

Introduced here is a module for converting single-cell data sets to transformed bulk-cell datasets for use in augmenting single-cell data, identifying cell type subsets, and in assessing factors associated with diseases. Such cell type subsets include immune cell subsets.

Description

MODULES FOR SINGLE-CELL DATA CONVERSION TO TRANSFORMEDBULK-CELL DATA AND METHODS OF USING THE SAMECROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 584,827, filed September 22, 2023, the entire contents of which are incorporated herein by reference in its entirety.TECHNICAL FIELD

[0002] Various embodiments concern approaches to single-cell dataset transformation for use in identifying disease, drug targets, risk factors, biomarkers, and cell type subsets, and non-transitory mediums for implementing the same.BACKGROUND

[0003] Despite rapid development of single-cell technologies, analysis of singlecell data remains highly challenging. In particular, interpretation of single-cell data in the context of biology is highly under addressed and lacks field consensus. Different computational methods exist for cell type annotation, but the field has not yet aligned on best approach, even for blood samples where cell type heterogeneity is lower than in tissue. Current single-cell analytics almost exclusively focus on Ribonucleic acid (RNA) sequencing (RNA-seq) data, but even the most current commercial single-cell methods are limited to, at most, two genomic profiling methods run simultaneously: RNA-seq with Assay for Transposase-Accessible Chromatin with high-throughput sequencing (ATAC-seq), or RNA-seq with Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq). Additionally, the 10x Genomics (10xG) platform has been widely used for single-cell profiling. Each of these workflows seek to provide meaningful readouts about a complex cellular admixture, with the information from RNA-seq at its core. However, most single-cell analytics are accompanied by high rates of dropout events, false discovery, reduced cell type subset identification, and biological noise, which are typically reduced with bulk-cell sequencing. To date, there are no sequencing technologies having studies which simultaneously possess the advantages of both single-cell sequencing and bulk-cell sequencing.BRIEF DESCRIPTION OF THE DRAWINGS

[0004] FIG. 1 illustrates a sequencing study of the present technology, combining Xseq, subset single-cell data, and bulk single-cell data.

[0005] FIGs. 2A-C illustrate gene detection using the sequencing study shown in FIG. 1. FIG. 2A includes a Venn diagram comparing genes detected with Xseq (left), bulk-single-cell sequencing (right), or both (center). FIG. 2B includes a stacked bar chart comparing detected gene biotypes using Xseq (xseq_unique), subset single-cell sequencing (sc_unique), or both methods (shared). FIG. 2C includes a histogram illustrating unique protein coding genes detected using Xseq (xseq_unique) or subset single-cell sequencing (sc_unique).

[0006] FIGs. 3A-D illustrate gene expression and epigenome analyses using the sequencing study shown in FIG. 1 . FIG. 3A illustrates an aggregated-level view of gene expression across subject subsets using different methods of the sequencing study. FIG. 3B illustrates a density graph for genes expressed at various mean expression values across subsets in single-cell data (top) and a histogram illustrating the number of genes found in different cell number categories (bottom). FIG. 3C includes a bar chart showing the mean ± standard deviation (SD) for transcription start site (TSS) scores for bulk single-cell sequencing compared to Xseq. FIG. 3D includes a bar chart illustrating an internal quality control metric for bulk single-cell sequencing compared to Xseq. Statistics shown are calculated using Mann-Whitney U, ***** represents p < 0.0001 .

[0007] FIG. 4 illustrates the predicted total gene drug targets per cell type subset identified using subset single-cell sequencing or Xseq.

[0008] FIGs. 5A-C illustrate cell type subset composition enriched by different methods of the sequencing study shown in FIG. 1 using peripheral blood mononuclear cells (PBMCs). FIG. 5A illustrates bulk PBMC classification by subset cell types classified using Xseq compared to bulk sequencing classification. FIG. 5B includes a Uniform Manifold Approximation and Projection (UMAP) illustrating aggregated data of cell types found using bulk-enriched sequencing, subset-enriched sequencing, or both sequencing methods (overlay). FIG. 5C illustrates predicted cell types by frequency of total PBMCs that may be captured by standard single-cell workflows on bulk PBMCs.

[0009] FIGs. 6A-C illustrate single-cell shifts in subpopulation composition informed by Xseq. FIG. 6A includes a bar chart illustrating expression of differentially regulated genes in effector memory CD8+T cells (T8em) in control subjects relative to rheumatoid arthritis (RA) subjects. FIG. 6B includes a UMAP illustrating distribution of T8em signatures in control subjects (hv) relative to RA subjects (ra) using subset singlecell sequencing data.

[0010] FIGs. 7A-B illustrate gene expression changes at the subpopulation level using subset single-cell sequencing data. FIG. 7A includes a bar chart illustrating differentially expressed genes in Gammadelta T (gadT) cells having effector-like signatures in control subjects (hv) relative to RA subjects (ra). FIG 7B illustrates the location of gadT cells having an effector-like signature in the context of all immune cell type subset populations samples in the blood (left) by UMAP and the density of expression of ADGR1 and ZNK683 differentially expressed genes in RA subjects (right).

[0011] FIGs. 8A-B illustrate differentially expressed genes (DEGs) based on cell type subset labeling. FIG. 8A illustrates the fraction of DEGs identified based on predicted cell type subset labels that are actual DEGs based on actual cell type subset labels (left) and the fraction of DEGs identified based on actual cell type subset labels that are correctly identified based on predicted cell type subset labels (right) before regrouping (left graph, precision). FIG. 8B illustrates the fraction of DEGs identified based on predicted cell type subset labels that are actual DEGs based on actual cell type subset labels (left) and the fraction of DEGs identified based on actual cell type subset labels that are correctly identified based on predicted cell type subset labels (right) after regrouping (right graph, recall).

[0012] FIGs. 9A-R illustrate performance (precision and recall) comparing predicted and actual cell subset labels. FIG. 9A includes boxplots illustrating fraction of predicted subset labels that are correct based on actual subset labels (precision) before regrouping. FIG. 9B includes boxplots illustrating fraction of actual subset labels corrected predicted by the model (recall) before regrouping. FIG. 9C includes boxplots illustrating F1 scores the subset label predictions of FIGs. 9A-B. FIG. 9D-R repeat FIGs. 9A-C with different regrouping strategies.

[0013] FIG. 10 illustrates standardized variance (y-axis) and average expression (x-axis) for highly variable genes relative to non-highly variable genes for total genes (left) and labeling with most variable genes (right).

[0014] FIG 1 1. Illustrates Iog10 expression (x-axis) and dropout rate (y-axis) for genes having high rate of dropout events.

[0015] FIG. 12 provides a heatmap matrix illustrating predicted cell type subsets within PBMC samples based on single-cell RNA-seq data alone.

[0016] FIGs. 13A-F illustrate cell type subset predictions among 1 .27 million cells from PBMC samples of 982 control donors. FIG. 13A illustrates cell type subset marker genes for cells predicted to be T cell subsets. FIG. 13B illustrates cell type subset marker genes for cells predicted to be B cell subsets. FIG. 13C illustrates cell type subset marker genes for cells predicted to be subsets of myeloid cells. FIG. 13D illustrates cell type subset marker genes for cells predicted to be NK subsets. FIG. 13E provides a Principal Component Analysis (PCA) plot of transformed bulk-cell dataset samples. Individual dots represent an aggregated subset transformed bulk-cell dataset from one subject. FIG. 13F provides a PCA plot of transformed bulk-cell dataset samples for NK and T cell subsets.

[0017] FIGs. 14A-D illustrate identification of single nucleotide polymorphisms (SNPs) associated with rheumatoid arthritis (RA). FIG. 14A illustrate the identification of SNPs associated with RA using ATAC-seq, HIChIP, and RNA sequencing analysis. FIG. 14B illustrates expression of the SNPs of FIG. 14A across different immune cell type subsets. FIG. 14C illustrates immune cell type subset-level resolution of expression quantitative trait loci (eQTLs) for identifying SNP-expression associations. FIG. 14D illustrates eQTLs linked to autoimmune disease.

[0018] FIG. 15 illustrates Gene K is a differentially expressed gene in T cell subsets of RA subjects relative to control subjects.

[0019] FIG. 16 illustrates that Gene K alone can stimulate inflammatory cytokine production. Triangles represent cytokine production in response to increasing recombinant Gene K exposure, and circles represent cytokine production in response to increase lipopolysaccharide (LPS) plus interferon gamma (INF-g). Each dot represents the mean of 2 experimental replicates.

[0020] FIGs. 17A-D illustrate the effect of Gene K deletion in CD4+T cells on cytokine production. FIG. 17A provides a bar chart illustrating the percent of edited Gene K reads (y-axis) for each subject sample (x-axis). FIG. 17B illustrates Gene K levels in T cell lysates for both wild type CD4+T cell controls and CD4+T cell Gene K knockouts. FIG. 17C illustrates TNFa levels measured in the supernatant of wild type CD4+T cell controls and CD4+T cell Gene K knockouts. FIG. 17D illustrates IL-6 levels measured in the supernatant of wild type CD4+T cell controls and CD4+T cell.

[0021] FIG. 18 illustrates Gene K SNP expression across different immune cell type subsets.

[0022] FIG. 19 provides a model for Gene K signaling and modulation.

[0023] FIGs. 20A-B illustrate Gene X expression across different immune cell type subsets. FIG. 20A provides RNA sequencing expression of Gene X across different immune cell type subsets. FIG. 20B provides results of ATAC-seq of the Gene X transcription start site across different immune cell type subsets.

[0024] FIGs. 21 A-D illustrate Gene X deletion in Treg cells following CRISPR deletion. FIG. 21 A provides a bar chart illustrating the percent of edited Gene X reads (y-axis) for each subject sample (x-axis). FIG. 21 B illustrates the distribution of insertiondeletion (indel) sizes in the Gene X loci. FIG. 21 C provide the RNA sequencing data of Gene X for wildtype Treg cells (left) relative to Gene X knockout Treg cells (right). FIG. 21 D provides Western Blot images of the Gene X protein for wildtype Treg cells (left) relative to Gene X knockout Treg cells (right).

[0025] FIGs. 22A-C illustrate that Gene X deletion in Tregs increases suppression function. FIG. 22A illustrates a study design schematic where wildtype Treg cells (WT Treg), Gene X knockout Treg cells (X-KO Treg), and Teff cells (Teff control) were treated with leukopaks and assessed for Teff proliferation. FIG. 22B illustrates cell trace for Teff stimulation for cells treated according to FIG. 22A. FIG. 22C illustrated percent suppression (y-axis) of Teff cells for cells treated according to FIG. 22A. Each dot represents an individual sample.

[0026] FIG. 23 provides boxplots illustrating expression of differentially expressed genes for wildtype Treg cells (WT) compared to Gene X knockout Treg cells (KO).

[0027] FIG. 24 provides a model for Gene X signaling and modulation.

[0028] FIG. 25 includes a flow diagram of a pipeline for converting single-cell datasets to bulk-cell datasets that may then be stored on a memory medium.

[0029] FIG. 26 includes a block diagram illustrating a transformation module of the present technology.

[0030] Various features of the technology will become more apparent to those skilled in the art from a study of the Detailed Description in conjunction with the drawings. In the drawings, embodiments are illustrated by way of example and not limitation for the purpose of illustration. Those skilled in the art will recognize that alternative embodiments may be employed without departing from the principles of the present disclosure. Accordingly, while specific embodiments are shown in the drawings, the technology is amenable to various modifications.DETAILED DESCRIPTION

[0031] Traditionally, there have been two approaches to sequencing ribonucleic acid (“RNA”), namely, single-cell sequencing and bulk sequencing. A main difference between single-cell sequencing and bulk sequencing is that in the former the sequencing library represents a single-cell while the latter represents a population of cells.

[0032] Single-cell sequencing has become increasingly popular as a tool for examining single-cell gene expression, largely because it offers high efficiency at lower cost than bulk sequencing. Moreover, because single-cell sequencing cover cell-to-cell variation rather than population-level metrics of cells as in bulk sequencing, rich datasets can be produced that are helpful in various contexts (e.g., pharmaceutical development).

[0033] However, there are downsides to single-cell sequencing. For example, single-cell sequencing may be complicated by increased dropout events, increased false discovery rates, reduced cell-type subset characterization, and reduced total gene detection relative to bulk sequencing.

[0034] Introduced here is an approach to transforming single-cell datasets into bulk-cell datasets that address the limitations of single-cell sequencing and deepen the ability to gain insights into biological processes through analysis of the underlying information. For example, the transformed bulk-cell datasets produced in accordancewith the approach may allow more accurate and efficient (e.g., in terms of time and consumption of computational resources) identification of cell type subsets and factors associated with disease. The transformed bulk-cell datasets may also be combined with additional datasets, such as subset single-cell datasets and bulk-cell datasets to provide a more comprehensive view of biological systems (and therefore, allow for a more comprehensive understanding of biological systems).

[0035] As further discussed below, the approach could be implemented by a computer program called a “multi-omics analysis platform” or simply “analysis platform” that is either connected to one or more sequencing modules or connected to one or more datastores in which single-cell datasets generated by one or more sequencing modules are stored. In operation, the analysis platform may implement the approach through execution of its modules, each of which may be responsible for performing one or more steps to progress toward generation of a transformed bulk-cell dataset.

[0036] Embodiments may be described with reference to particular types of clinical conditions, dataset formats, platform architectures, and the like. However, those skilled in the art will recognize that these features are similarly applicable to other types of clinical conditions, dataset formats, and platform architectures. For example, embodiments may be described in the context of a given disease or disease more generally for the purpose of illustration. However, these embodiments are more generally applicable to clinical conditions unless otherwise specified. The term “clinical condition” may be used to refer to a disease or other anomaly of the body or mind with recognizable, measurable, or describable clinical signs or symptoms. As another example, embodiments may be described in the context of an analysis platform that is responsible predicting the onset or progression of a disease based on an analysis of health data maintained by a healthcare facility (e.g., in the form of EHRs), insurance facility (e.g., in the form of claims), or another entity. However, the relevant features may be similarly applicable regardless of the source or contents of the health data.

[0037] Moreover, while embodiments may be described in the context of computer-executable instructions, aspects of the technology could be implemented via hardware or firmware instead of, or in addition to, software. For example, the analysis platform may reside on a computer server in the form of a desktop application that is responsible for obtaining, processing, and examining health data to predict the onset orprogress of disease in corresponding patients. However, aspects of the analysis platform could alternatively be implemented in hardware or firmware. For example, aspects of the approach described herein may be executed by an application-specific integrated circuit (“ASIC”) that is customized to do so. The ASIC could be implemented in a specialized computing device that is provided to the patient, for example, as part of a clinical trial or a treatment regimen.Terminology

[0038] References in the present disclosure to “an embodiment” or “some embodiments” means that the feature, function, structure, or characteristic being described is included in at least one embodiment. Occurrences of such phrases do not necessarily refer to the same embodiment, nor do they necessarily refer to alternative embodiments that are mutually exclusive of one another.

[0039] The terms “comprise” and “comprising” are to be construed in an inclusive sense rather than an exclusive or exhaustive sense (i.e. , in the sense of “including but not limited to”).

[0040] The term “based on” is also to be construed in an inclusive sense rather than an exclusive or exhaustive sense. Thus, unless otherwise noted, the term “based on” is intended to mean “based at least in part on.”

[0041] The terms “connected,” “coupled,” and variants thereof are intended to include any connection or coupling between two or more elements, either direct or indirect. The connection or coupling can be physical, logical, or a combination thereof. For example, elements may be electrically or communicatively connected to one another despite not sharing a physical connection.

[0042] The term “module” may refer broadly to software, firmware, hardware, or combinations thereof. Modules are typically functional components that generate one or more outputs based on one or more inputs. A computer program may include or utilize one or more modules. For example, a computer program may utilize multiple modules that are responsible for completing different tasks, or a computer program may utilize a single module that is responsible for completing multiple tasks.

[0043] The term “Xseq” refers to the sequencing methods and associated systems, compositions, and techniques, disclosed in U.S. Patent No. 11 ,352,714, the entirety of which is hereby incorporated by reference.

[0044] When used in reference to a list of items, the word “or” is intended to cover all of the following interpretations: any of the items in the list, all of the items in the list, and any combination of items in the list.

[0045] The disclosure of all publications, patents, and published patent applications listed herein are hereby incorporated by reference. To the extent any materials incorporated by reference herein conflict with the present disclosure, the present disclosure controls.Overview of Single-Cell Transformation Module

[0046] At a high level, transforming the individual datasets may bring consistency about the conclusions, or augment available datasets, such as single-cell datasets. For example, a single-cell dataset may be transformed into a bulk-cell dataset. Transformation of the single-cell dataset to the bulk-cell dataset may be carried out via execution of a routine by the analysis platform. This routine - which is representative of a sequence of executable code - may be called the “transformation module.” In some embodiments, the transformed bulk-cell dataset is combined with a second bulk-cell dataset to generate a combined bulk-cell dataset.

[0047] Figure 25 illustrates a series of software-implemented operations - commonly called a “pipeline” 100 - that includes obtaining a single-cell dataset 101 that is uploaded or downloaded to a datastore 102. The single-cell dataset 101 is derived from single-cell sequencing which includes isolation or separation of individual cells prior to sequencing. Nonlimiting examples of single-cell sequencing techniques include single-cell RNA sequencing, single-cell DNA sequencing, single-cell spatial transcriptomics, single-cell clonal lineage tracing, and single-cell transcriptomic sequencing. A transformation module 103 processes the single-cell dataset from datastore 102 to convert the single-cell dataset 101 to a transformed bulk-cell dataset 104. The transformed bulk-cell dataset 104 may be stored on memory medium 106. The transformed bulk-cell dataset 104 reflects bulk datasets_derived from bulk-cell sequencing which includes pooling of individual cells into a single sample prior tosequencing. Nonlimiting examples of bulk-cell sequencing techniques include pooled RNA sequencing, pooled DNA sequencing, DNA methylation sequencing, pooled ATAC-sequencing, bisulfite sequencing, transcriptome sequencing, exome sequencing, and microbiome sequencing.

[0048] As shown in Figure 26, the transformation module 103 can comprise a processing module 1 10, an augmentation module 120, and an imputation module 130. Alternatively, these modules could be external but accessible to the transformation module 103. For example, the processing module 110 may be separately executable by an analysis platform of which the transformation module 103 is a part. In such embodiments, outputs produced by the processing module 1 10 may be supplied to the transformation module 103 as inputs. The processing module, augmentation module 120, and imputation module 130 may be responsible for completing different tasks that enable the transformation module 103 to transform a single-cell dataset 101 into a transformed bulk-cell dataset 104. As an example, the processing module 110 may comprise the steps of applying a machine learning model to the single-cell dataset 101 . The machine learning model may be configured to predict a cell type subset label for each cell. The cell type subset labels may be collected and further processed by augmentation model 120, which categorizes the cell type subset labels to generate an augmented single-cell dataset. Augmentation model 120 may further comprise a normalization step and / or a step of performing a negative binomial distribution test to identify one or more differential features from the single-cell dataset relative to a control bulk-cell dataset. As another example, imputation module 130 may comprise a step of imputing additional values for the augmented single-cell dataset by employing an algorithm that utilizes matrix regularization. Imputation module 130 may further comprise a step of compiling the augmented single-cell dataset and any additional values into a data structure that is representative of a bulk-cell dataset. The step of compiling the augmented single-cell dataset and the additional values into the data structure may be repeated one or more times, thereby generating one or more data structures that are representative of one or more bulk-cell datasets.

[0049] In some embodiments, at least 1 , at least 2, at least 3, at least 4, at least 5, at least 6, at least 8, at least 10, at least 12, at least 14, at least 16, at least 18, at least 20, at least 22, at least 24, at least 25, at least 30, at least 35, at least 40, or at least 50 data structures that are representative of one or more bulk-cell datasets are generated.

[0050] In some embodiments, about 1 , about 2, about 3, about 4, about 5, about 6, about 8, about 10, about 12, about 14, about 16, about 18, about 20, about 22, about 24, about 25, about 30, about 35, about 40, or about 50 data structures that are representative of one or more bulk-cell datasets are generated.

[0051] In some embodiments, at least about 1 , at least about 2, at least about 3, at least about 4, at least about 5, at least about 6, at least about 8, at least about 10, at least about 12, at least about 14, at least about 16, at least about 18, at least about 20, at least about 22, at least about 24, at least about 25, at least about 30, at least about 35, at least about 40, or at least about 50 data structures that are representative of one or more bulk-cell datasets are generated.

[0052] These steps generate the transformed bulk-cell dataset 104, which may be stored on a memory medium or examined via an interface generated by a graphical user interface (GUI) module 150.A. Identifying Limitations of Traditional Approach

[0053] A common approach to characterizing or identifying changes in cell type subsets is to utilize single-cell data or bulk-cell data alone. FIG. 1 illustrates a schematic for a sequencing study that compares datasets generated through different sequencing types (e.g., Xseq, subset single-cell data, and bulk single-cell data). Peripheral blood mononuclear cells (PBMCs) were used to profile the immune system of control subjects (n = 3) and RA subjects (n = 3). PBMCs from each subject were sorted into 3 groups, with equal numbers of cells targets towards Xseq processing for 24 subsets / individual, single-cell processing (subset single-cell) for 24 subsets / individual, and single-cell processing (bulk-single-cell) on 10,000 live PBMCs per subject. In each group (control subjects vs. RA subjects), each cell type subset was captured in at least two of the three subjects. Alignment of sequenced reads was performed using the same reference genome.

[0054] Subject PBMC samples had a mean cell viability of 91 .2%. 24 immune cell type subsets were captured (Table I) for the Xseq workflow, sorting 800 cells per subset per subject (“Xseq”, FIG. 1 ). The same 24 subsets were captured for single-cell RNA- seq (single-cell RNA sequencing) using the 10xG 3' Chromium Next GEM single-cell 3' kit v3.1 on a Chromium controller system (“subset single-cell”, FIG. 1 ) with an 800-celloutput target per subset per subject. A target of 10,000 bulk FACS-enriched, live PBMCs per subject, stained with a panel of 134 different antibodies were captured (“bulk single-cell”, FIG. 1 ).

[0055] Four immune subsets were captured from two separate subjects to compare chromatin accessibility (ATAC-seq) between single-cell and Xseq workflows. 200 cells of each cell type were processed through the Xseq workflow, with technical replicates formed from two sorts of 200 cells per individual. In parallel, 10,000 cells of each cell type were processed through the 10xG Chromium Next GEM single-cell ATAC v2 kit.Table I: Xseq sorted subset nomenclaturei. Sequencing Sensitivity is Influenced by Sequencing Method

[0056] At a high level, many single-cell technologies are tethered by readouts from RNA-seq (Xseq is not strand-specific and does not contain UMIs, while the 10xG workflow is strand-specific and does contain UMIs). To compare Xseq RNA-seq to bulk single-cell (FIG. 2A), many shared genes were detected by at least one read by both Xseq and bulk single-cell technologies. However, Xseq detected 4-fold the total number of unique genes that bulk single-cell did (13,173 genes were uniquely detected by Xseq while 3,255 were unique to bulk single-cell). The majority of genes shared between Xseq and bulk single-cell were protein-coding (FIG. 2B). Bulk single-cell detected few protein-coding genes that were not also detected by Xseq. Xseq, however, detected many protein-coding genes, as well as pseudogenes, lincRNA, and antisense RNA biotypes that single-cell methods did not detect. When the data from the six subjects were aggregated together, Xseq consistently detected 2.7 to 7.6-fold more proteincoding genes than subset single-cell (FIG. 2C), suggesting that Xseq is uniquely sensitive when profiling cell populations such as immune cell populations.

[0057] Aggregated distribution of all protein-coding gene expression generated by the sequencing study of FIG. 1 was next examined. Density of expression values of genes unique to subset single-cell or Xseq or shared between subset single-cell and Xseq were evaluated (FIG. 3A). Studies have directly compared UMI counts to TPM values when comparing gene expression. Shared protein-coding genes between subset single-cell and Xseq data showed a bi-modal distribution, with roughly half of the shared genes having higher expression values. Unique protein-coding genes in each technology have low expression values, though Xseq overall detects higher numbers of these genes.

[0058] For drug discovery, identification of “druggable” targets is dependent on both the gene being expressed and the number of cells that express that gene. Though Xseq uniquely identifies a larger number of genes when compared to subset single-cell data, it’s unknown whether these unique genes are expressed in one cell or in many cells. Using subset single-cell data as a proxy, expression of protein-coding genes uniquely found in the single-cell data by the number of cells in which the genes are expressed was evaluated (FIG. 3B). Gene expression values mainly segregated into three ranges: genes expressed at low levels, usually only detected in one cell; genesdetected at moderate levels (2-8 UMI counts), detected in under 10 cells; and genes detected in more than 10 cells, which also tend to be expressed at higher levels.

[0059] These data suggest that the level of gene expression in bulk RNA-seq data closely follows the number of cells in which the gene is expressed. Genes expressed at moderate-to-high levels in bulk RNA-seq data are likely not expressed in a single cell at an abnormally elevated level, but rather are expressed in many cells at a uniform level. Increasing the sensitivity of detecting protein-coding genes overall, whether in bulk or single-cell RNA sequencing workflows, will increase the likelihood of capturing drug targets expressed in a larger proportion of cells.

[0060] Next, ATAC-seq metrics between Xseq and single-cell were evaluated. The transcription start site (TSS) score, which is generally associated with high quality processing owing to greater capture of areas of known biological importance, was significantly higher in Xseq than in single-cell (FIG. 3C). In addition, the “region mean count normalized” metric, a QC metric associated with the quantification of accessibility around conserved “house-keeping” loci, was significantly higher in the Xseq dataset than in the single-cell data (FIG. 3D).II. Sequencing Method Impacts Drug Target Discovery

[0061] It’s hypothesized that increasing the sensitivity of gene detection will, in theory, expand the number of detected disease-associated genes, thereby increasing the detected number of genes that are, or are associated with, potential drug targets. To assess this theory, a drug target set list stratified by position in the drug development pipeline (5). Over 22% of genes detected by Xseq fell into a “druggable” category across all subsets. Overall, Xseq was able to consistently capture 30-50% more drug target genes across subsets when compared to single-cell (FIG. 4).

[0062] When considering the unique protein-coding genes per category, Xseq outperformed single-cell RNA sequencing across all tiers of drug targets, including targets of approved small molecule or biotherapeutic drugs and clinical-phase drug candidates (Tier 1 ), known bioactive drug-like small molecule binding partners (Tier 2), or drug targets encoding secreted or extracellular proteins that were in close proximity to genome-wide association single-nucleotide polymorphisms (Tier 3A) or not (Tier 3B). This highlights that Xseq, as a drug discovery tool, is capable of capturing more genetargets than conventional single-cell methods used in the field. Nonetheless, single-cell RNA sequencing may be useful in niche cases where targeting specific subpopulations of cells is necessary. ill. Sequencing Method Impacts Cell Type Subset Enrichment

[0063] Several single-cell RNA sequencing approaches have been used to survey the immune system, but most single-cell studies sample tissue (such as peripheral blood) in bulk. While single-cell RNA sequencing on bulk PBMCs allows users to capture the distribution of immune cell types in the blood, this is often assessed by flow cytometry at a fraction of the cost (assuming the cell populations have known biomarkers). Furthermore, running single-cell RNA sequencing on bulk PBMCs may cause users to have datasets dominated by the most frequently found immune populations in the blood, which may or may not have meaningful biological relevance to a health condition.

[0064] Xseq captures 24 immune cell type subset populations, which broadly cover and enrich a number of relevant cell populations. To assess the overall coverage of known immune cell types using Xseq and traditional single-cell workflows, PBMCs in bulk (bulk single-cell) and PBMCs sorted into Xseq subsets followed by single-cell RNA sequencing processing (subset single-cell) were evaluated. Bulk single-cell and subset single-cell data for the six subjects were combined, normalized, and integrated using standard single-cell RNA sequencing workflows. Subset single-cell data was used to categorize bulk single-cell data into Xseq subsets using label transfer and a prediction score cutoff method standard to single-cell integration workflows. Bulk PBMC workflows were primarily characterized by high frequency cell types present in the blood, such as CD56-low NK cells (NKIo), classical monocytes (MoCI), and central memory CD4 and CD8 T cell types (T4cm, T8cm) (FIG. 5A). Cell type subset enrichment, as done in the Xseq workflow, provided better coverage across a number of biologically-relevant cell types, such as plasmablasts, Tregs, dendritic cells (DCs), and effector memory CD4+and CD8+T cell types (FIG 5B). Bulk capture identified some immune cell types (up to 10-15% of total) which are not obtained by Xseq isolation.

[0065] To further classify what other immune cell type subset populations may be better captured by bulk PBMC workflows, cells specific to bulk single-cell datasets to published cell type labels were mapped and frequency was assessed (FIG. 5C). Mostcell types specific to bulk PBMC workflows were CD4+and CD8+memory-like T cell types, which typically express higher levels of CCL5 and IL7R relative to other immune cell type subsets. Cell types that are reported in bulk single-cell workflows, such as hematopoietic stem progenitor cells (HSPCs), platelets, and proliferating cell types were also found to be present, albeit at very low frequencies. iv. Sequencing Method Impacts Subpopulation Resolution in the Blood

[0066] Xseq correctly captures well-defined immune populations by FACS and obtains multi-omic profiling of gene expression, chromatin accessibility, and cell surface protein expression within a single sample. However, Xseq may not be able to estimate changes in cell type frequency within the population that it samples. Single-cell workflows can detect variations in the composition of a population of cells, but the identification of individual cell types can be challenging due to closely related immune cell types exhibiting similar gene expression profiles despite having slight differences in cell surface markers. This is partially overcome by using cell surface protein measurements (antibody-based sequencing, such as CITE-seq) to categorize cells into literature-defined cell types. There may be difficulty in bulk single-cell applications without cell surface protein expression data and / or well-defined methods to use and incorporate this data into the analyses.

[0067] Immune populations involved in RA using the Xseq and single-cell workflows, were assessed. 16 differentially expressed genes (DEGs) in effector memory CD8+T cells (T8em) were identified using the Xseq workflow with an adjusted p-value below 0.05 between control subjects and RA subjects. Several of these genes were elevated in T8em cells of RA subjects (FIG. 6A) and all were involved in T cell activation (FOS, FOSB, and JUN aW directly play a role in T cell activation, while EGR1 and TNF may either lead to or be the result of T cell activation). This suggests that RA subjects may exhibit increased frequencies of activated (or cytotoxic) CD8+T cells in the blood.

[0068] To further explore the possibility of an activation signature in the T8em compartment, the single-cell data was examined to determine whether this activation signature was present across all T8em cells, or if the signature was restricted to a smaller subpopulation within T8em. Bulk single-cell data was first assessed, using a combination of antibody labels and gene expression to identify T8em from bulk PBMCs(CD3+CD8+CD45RA- CD45RO+by cell surface protein, and CCR7~ SELL: by singlecell RNA sequencing). T8em derived from bulk single-cell data fell into three gene expression signatures based on previously published cell type signatures. In bulk single-cell data, RA subjects presented slight decreases in the frequency of T8em with an effector-like signature (characterized by the expression of GZMK, CCL5, DLISP2, LYAR), and large decreases in T8em with a terminal- or MAIT-like signature (characterized by the expression of GZMK, KLRG1, KLRB1, and SLC4A10) relative to control signatures (FIG. 6B).

[0069] Next, the subset single-cell data that was enriched for immunology-relevant cell types (including T8em) by the Xseq workflow was assessed to determine whether increased sampling in this compartment would reveal additional signs of an activation signature. All T- and NK-lineage subset data was combined and integrated, again annotating clusters / signatures based on previously published cell type signatures. Subset single-cell data revealed that RA subjects exhibited an increase in T8em cells with an effector-like signature relative to control subjects. This was accompanied by increases in T8em with an activation-like signature (characterized by GATA3 and LTB expression, which may either lead to or be a result of T cell activation) in RA subjects to control subjects. Like bulk single-cell data, RA subjects had decreased frequencies of T8em with a terminal- or MAIT-like signature (FIG. 6C). The shifts in frequencies of cells with these signatures were consistent across subjects. Together, these data highlight the sensitivity of Xseq in detecting signatures (such as T cell activation) that can be further resolved using single-cell approaches. v. Different Sequencing Methods Identify Different Subpopulations in the Blood

[0070] Recognizing that single-cell studies can suffer from performance issues and high false discovery rates when employing many commonly used differential expression analyses and the sample size limitations, the subset single-cell data for gammadelta T cells (gadT) from RA subjects, which have been implicated in RA pathology, were next evaluated.

[0071] gadT in control subjects and RA subjects exhibited an effector-like signature, characterized by the expression of TRDC, TRGC1, KLRG1, CCL5, NKG7, and KLRB1. To false discovery bias in single-cell data, gadT cells were converted with effector-like signature into pseudo-bulk for DEG analysis. These comparisons revealedthat RA subjects exhibited differential expression of effector-related genes FCGR3A (CD16), ADGRG1 (GPR56), CX3CR1, KLRF1, ZNF683 (HOBIT), and IFNG (interferon gamma) relative to control subjects (adjusted p-value < 0.05, FIG. 7A).

[0072] Two genes, ADGRG1 (GPR56) and ZNF683 (HOBIT), have been implicated in RA pathology previously GPR56 is categorized as a Tier 3B targetable protein and both GPR56 and HOBIT were found to be specifically expressed in gadT cells with an effector signature (FIG. 7B).B. Motivation

[0073] Consider the general problem of utilizing single-cell sequencing to assess a sample. The resulting data will provide reduced total gene detection, cell type subset recognition, and an increased number of dropout events relative to bulk-cell data. This problem is referred to as “the single-cell problem.”

[0074] Now, consider this problem under the lens of healthcare problems, and in particular health data relating to diagnoses and / or identifying drug targets. In such a scenario, an attempt may be made to predict or identify a disease, yet only a small fraction of cell type subsets or genes may be observed. This is because traditional single-cell sequencing methods, while still valuable, are limited in their data output.C. Steps of the Module

[0075] For the purposes of illustration, consider the sequencing study provided in FIG. 1 , which employs multiple sequencing methods, each of which provide a unique conclusion about the sample source. These sequencing methods may be modified by translating a single-cell dataset to an augmented, transformed bulk-cell dataset using the transformation module of the present technology.

[0076] The transformation module is executed using a non-transitory medium with instructions stored thereon that, when executed by a processor of a computing device, causes the computing device to perform steps for transforming a single-cell dataset into a transformed bulk-cell dataset, comprising the steps of:(i) applying, to the single-cell dataset, a machine learning model that is configured to predict a cell type subset label for each cell, so as to create an augmented single-cell dataset;(ii) normalizing the augmented single-cell dataset from (i) and performing a negative binomial distribution test to identify one or more differential features from the single-cell dataset relative to a control bulk-cell dataset;(Hi) imputing additional values for the augmented single-cell dataset by employing an algorithm that utilizes matrix regularization;(iv) compiling the augmented single-cell dataset from (i) and the additional values from (iii) into a data structure that is representative of the bulk-cell dataset; and(v) storing the data structure in a memory.

[0077] The non-transitory medium may further comprise a step of performing a batch correction operation based on the one or more differential features from (ii) and / or step (vi), training a machine learning model using a training dataset that includes the bulk-cell dataset. Nonlimiting examples of the batch correction operation include identifying pairs of mutual nearest neighbors (MNN), canonical correlation analysis (CCA), negative binomial regression, surrogate variable analysis, distance-based correlation, and principal component analysis (PCA) correction.

[0078] The machine learning model may be developed using one or more algorithms. Nonlimiting examples of the one or more algorithms which may be used to develop the machine learning model include a Ridge logistic regression algorithm, a random forest algorithm, an extreme Gradient Boosting (XGBoost) algorithm, and a Single-cell Prediction of Population-level Effects from Single-cell Data (scPred) algorithm.D. Dataset Sources

[0079] The transformed bulk-cell dataset may be combined with a second bulkcell dataset to generate a combined bulk-cell dataset. The second bulk-cell dataset may comprise data from one or more of an Xseq dataset, a genome wide association study (GWAS), a database of transcription factor binding profiles, a quantitative trait loci (QTL) mapping, and an expression QTL (eQTL) mapping.

[0080] In some embodiments, the second bulk-cell data set is generated by a method comprising:a) contacting a cell-containing sample with a plurality of antigen-binding molecules under conditions that promote specific binding of the antigen-binding molecules to target antigens of the cells in the sample; wherein each antigenbinding molecule in the plurality binds to a unique target antigen in one or more cells of the sample, wherein each antigen-binding molecule is conjugated to an oligonucleotide, and wherein the oligonucleotide comprises (i) a unique molecular identifier sequence, (ii) an antigen-binding molecule identifier sequence corresponding to the unique target antigen, and (iii) an assay molecular identifier sequence; b) tagmenting genomic DNA in the cells to produce a plurality of double-stranded genomic DNA fragments comprising an oligonucleotide adaptor sequence at the 5' end of each genomic DNA fragment; c) lysing the cells in the sample; d) isolating mRNA transcripts from the cell lysate to produce a first fraction comprising the mRNA transcripts and a second fraction comprising tagmented genomic DNA fragments and the antigen-binding molecule-bound antigens from the cell lysate; e) generating a cDNA library by reverse transcribing the mRNA transcripts in the first fraction; f) amplifying (i) the unique molecular identifier sequences; (ii) the antigen-binding molecule identifier sequences; (iii) the assay molecular identifier sequences, and the (iv) tagmented genomic DNA fragments in the second fraction; g) analyzing the amplified unique molecular identifier sequences, the assay molecular identifier sequences, and the antigen-binding molecule identifier sequences to identify at least one target antigen in the second fraction to identify the expression profile of the sample; h) analyzing the tagmented genomic DNA fragments in the second fraction to identify the epigenetic profile of the sample; and i) analyzing the cDNA library generated from the first fraction to identify the transcriptional profile of the sample.

[0081] One or more of the single-cell dataset, the transformed bulk-cell dataset, the control bulk-cell dataset, or the second bulk-cell dataset may be derived from a genomic dataset, a transcriptomic dataset, an epigenomic dataset, a spatial genomics dataset, a chromatin accessibility dataset, a proteomics dataset, or a metabolomics dataset. In some embodiments, any one of the datasets of the present technology is derived from a publicly available database, including, but not limited to GenBank, Ensembl, RefSeq, UCSC Genome Browser, 1000 Genomes Project, dbSNP, Exome Aggregation Consortium (ExAC), DNA Data Bank of Japan (DDBJ), gnomAD browser, and UK biobank.

[0082] The genomics dataset may be generated using one or more sequencing methods selected from the group consisting of Sanger sequencing, next-generation sequencing, nanopore sequencing, shotgun sequencing, pyrosequencing, and singlemolecule real-time sequencing.

[0083] The transcriptomic dataset may be generated using one or more sequencing methods selected from the group consisting of poly(A) selection sequencing, total RNA sequencing, strand-specific RNA sequencing, ribosome profiling sequencing, small RNA sequencing, long read RNA sequencing, capture-based RNA sequencing, and chimeric RNA sequencing.

[0084] The epigenomic dataset may be generated using one or more methods selected from the group consisting of DNA methylation profiling, histone modification profiling, and RNA methylation profiling. Nonlimiting examples of DNA methylation profiling include bisulfite sequencing, methylated DNA immunoprecipitation (MeDIP), MethylC-seq, and a chomatin accessibility assay.

[0085] The spatial genomics dataset may be generated using one or more methods selected from the group consisting of Hi-throughput chromosome conformation capture (Hi-C) sequencing, Capture Hi-C sequencing, Chromosome Conformation Capture (3C) sequencing, Circular Chromosome Conformation Capture (4C) sequencing, Chromosome Conformation Capture Carbon Copy (5C) sequencing, Hi-C with Chromatin Immunoprecipitation (HiChIP) sequencing, Capture-C sequencing, Targeted Chromosome Conformation Capture (T2C) sequencing, and Promoter Capture Hi-C sequencing.

[0086] The chromatin accessibility dataset may be generated using one or more sequencing methods selected from the group consisting of Assay for Transposase- Accessible Chromatin (ATAC) sequencing, DNase sequencing, MNase sequencing, Formaldehyde-Assisted Isolation of Regulatory Elements (FAIRE) sequencing, Cleavage Under Targets and Release Using Nuclease (CUT&RUN), and Chromatin Immunoprecipitation (ChIP) sequencing, Cleavage Under Targets and Tag mentation (CUT&Tag) sequencing.

[0087] The proteomic dataset may be generated using one or more methods selected from the group consisting of mass spectrometry, tandem mass spectrometry, label-free quantification, isotope labeling, a protein-protein interaction analysis, and structural proteomics.E. Categorizing Differential Features

[0088] In general, the one or more differential features represent the distinctive characteristics that set apart different classes or categories within the dataset, whether it’s in a machine learning context, data analysis, or any other domain where data manipulation is required. Identifying the one or more differential features helps categorize the augmented single-cell dataset such that the final transformed bulk-cell dataset is an accurate reflection of the sample source.

[0089] The one or more differential features may comprise a differentially expressed gene (DEG) relative to a control, a differentially abundant protein relative to a control, or a genetic variant relative to a control.

[0090] Some subsets may be regrouped at any point to improve accuracy of the cell subset label prediction and DEG identification. For example, FIGs. 8A-B illustrate predicted DEGs within PBMC samples based on identified DEGs using actual subset labels. Cell type subset labeling was regrouped according to Table II.Table II: Regrouping Metrics

[0091] The term “precision” refers to the fraction of differential feature-predicted cell type subsets that are actual cell type subsets before regrouping (FIG. 8A), and the term “recall” refers to the fraction of differential feature-predicted cell type subsets that are actual cell type subsets after grouping (FIG. 8B).

[0092] The term “F1 score” refers to the value of Fi, whereFi = 2 / (recall1+ precision-1) =2([(precision)(recall)] / [(precision + recall)]).

[0093] Different regrouping strategies are shown in FIGs. 9A-O, where predicted cell type subsets were determined. Recall, precision, and F1 scores are shown without regrouping (FIGs. 9A-C), Regrouping Strategy I (FIGs. 9D-F), Regrouping Strategy II (FIGs. 9G-I), Regrouping Strategy III (FIGs. 9J-L), Regrouping strategy IV (FIGs. 9M- O) and Regrouping strategy V (FIGs. 9P-R), shown in Table III. This is compared to the results shown in FIG. 9P, where no regrouping has occurred. As shown in FIG. 9R, Pearson correlation between transformed bulk-cell data samples with no regrouping based on predicted cell type labels compared to actual cell type labels revealed that 19 of 24 cell type subsets had strong correlation, with 7 cell type subsets having perfect DEG overlap.Table III: Regrouping Strategies

[0094] Measurement or selection of the one or more differentially expressed features may vary, as variance calculations may depend on continuous exact expression levels that are noisy in single-cell RNA sequencing. As an example, FIG. 10 and FIG. 1 1 illustrate exemplary DEG selection methods. As shown in FIG. 10, singlecell gene expression data was divided based on average expression (left) and genes having higher than expected variance were selected (right). Alternatively, FIG. 1 1 illustrates an alternative to highly variable gene selection, where single-cell gene expression data was divided based on average expression and genes having a higher- than-expected dropout rate were selected for.F. Associated Methods of the Transformation Module

[0095] The transformation module of the present technology may be used for a plurality of applications. Such applications include, but are not limited to, identification of disease, a drug target, a risk factor, a cell type subset, and a biomarker in a subject. In some embodiments, the transformation module is applied to a single-cell dataset derived from a biological sample of the subject.

[0096] The biological sample may comprise a urine sample, a saliva sample, a circulatory fluid sample, or a tissue sample. In some embodiments, the tissue sampleis a solid tissue sample. The solid tissue sample may comprise a fresh tissue, a frozen tissue, or a fixed tissue. The solid tissue sample may comprise one or more tissues selected from the group consisting of an epithelial tissue, a connective tissue, a nervous tissue, an adipose tissue, a cartilage, a bone tissue, a skin tissue, a mucous membrane tissue, a glandular tissue, a vascular tissue, a cardiac tissue, a smooth muscle tissue, a skeletal muscle tissue, a neural tissue, a fibrous tissue, a bone marrow tissue, a liver tissue, a kidney tissue, a pancreatic tissue, a pulmonary tissue, and a brain tissue. The circulatory fluid sample comprises a blood sample or a lymph sample. The blood sample may comprise PBMCs. i. Identifying Cell Type Subsets

[0097] The transformation module of the present technology may also be used to identify one or more cell type subsets from a single-cell dataset. Identifying different cell type subsets aids the understanding of tissues, cell microenvironments, and disease mechanisms, and may provide information for diagnostic use. Additionally, identification of cell type subsets may comprise identification of subpopulations with classical cell type subsets and resolve changes in population composition.

[0098] In some embodiments, the cell-type subset is an immune subset. Nonlimiting examples of immune subsets include myeloid cell subsets, T cell subsets, B cell subsets. The myeloid cell subset may comprise a dendritic cell (DC) subset (e.g., conventional DC or a plasmacytoid DC) or a monocyte subset (e.g., a classical monocyte, an intermediate monocyte, or a nonclassical monocyte). The T cell subset may comprise a CD4+a T cell subset (e.g., a naive CD4+T cell, a central memory CD4+T cell, an effector memory CD4+T cell, a CD45RA+ effector memory CD4+T cell, or a CD4+T regulatory cell), a CD8+ap T cell subset (e.g., comprises a naive CD8+T cell, a central memory CD8+T cell, an effector memory CD8+T cell, or a CD45RA+effector memory CD8+T cell), a yd T cell subset (e.g., a gammadelta T cell), or a natural killer (NK) cell subset (e.g., a CD56hiNK cell or a CD56lowNK cell). The B cell subset comprises a naive B cell (e.g., an unswitched naive B cell, a class switched naive B cell, or a transitional B cell) or an effector B cell (e.g., a class switched classical memory B cell, an lgM+Ig D- classical memory B cell, an atypical memory B cell, or a class switch plasmablast).ii. Identifying Disease or Factors Associated with Disease

[0099] The transformation module of the present technology may be used in the context of a plurality of diseases. Such uses include, but are not limited to, identifying a disease or a biomarker or risk factor associated with disease in a subject.

[0100] The disease of any one of the methods of the present technology may comprise an immune disease or a cancer. In some embodiments, the immune disease comprises an autoimmune disorder, such as RA or multiple sclerosis (MS). ill. Identifying Risk Factors and Drug Targets

[0101] The transformation module of the present technology may be used to identify one or more risk factors or drug targets associated with disease in a subject. The risk factor may comprise a genetic variation relative to a control, a differentially expressed gene relative to a control, or a differentially abundant biomarker relative to a control. The genetic variation may comprise a single nucleotide variation (SNV), a single nucleotide polymorphism (SNP), a haplotype difference, an insertion-deletion (INDEL), a mitochondrial DNA (mtDNA) variation, a chromosomal abnormality, or a ploidy difference. The drug target may comprise a receptor, such as a transmembrane receptor. In some embodiments, the transmembrane receptor comprises one or more G-protein coupled receptor (GPCR) subunits. iv. Supplementing Single-cell Seguencing

[0102] The transformation module of the present technology may also be used to supplement one or more factors associated with single-cell sequencing. For example, single-cell sequencing may be accompanied by high dropout rates, high false discovery rates, high noise, and reduced gene, drug target, and cell type subset identification. The transformation module may be used or modified to recover, supplement, or correct one or more of these factors resulting from single-cell sequencing. Additionally, the transformation module may be used to augment single-cell datasets, such as by increasing an amount or depth of data relative to the original single-cell dataset. v. Multi-omics Integration

[0103] Multi-omics strategies encompass a plurality of distinct methods (e.g., subset single-cell sequencing, bulk RNA sequencing, Xseq, etc.), each valuable in itsindividual application while simultaneously compensating for the limitations of others. Methods like single-cell sequencing or bulk sequencing have their inherent pitfalls, such as limited spatial resolution or issues with data comprehensiveness. However, when integrated into a multi-omics framework, these methods together provide valuable insight that cannot be achieved with the individual methods alone. Any one of the associated methods of the present technology may comprise a multi-omics module or multi-omics analysis platform that integrates the transformation module of the present technology alongside other sequencing methods and datasets.G. Applying the Module to the OneK1 K Cohort

[0104] FIG. 13A-F illustrates an example applying the transformation module to the OneK1 K Cohort. As shown in FIG. 12, use of the transformation module on singlecell RNA sequencing data from PBMCs in the OneK1 K cohort performed well for 9 out of 15 examined cell type subsets, where more than 90% of predicted cells in the T8nv, plDC, pBcs, NKl0, NKhi, MoCI, CD4T Memory, and CD16+Monocytes were labeled with the correct cell type subsets. Additionally, 79% of coDC cells labeled with the correct cell type subsets.

[0105] Next, cell type subset marker genes were selected from the literature based on immunology domain knowledge. These cell type subset marker genes were used to verify that cells predicted to be a subset within the OneKI K cohort expressed appropriate marker genes. This was conducted for cells predicted to be a T cell subset (FIG. 13A), a B cell subset (FIG. 13B), a myeloid subset (FIG. 13C), and an NK subset (FIG. 13D).

[0106] Cells from each predicted subset of each donor were then aggregated into a transformed bulk-cell dataset and analyzed to determine clustering of predicted cell type subsets (FIG. 13E). This was also repeated for the predicted T cell and NK subsets alone (FIG. 13F). These analyses revealed that the transformed bulk-cell data from the OneKI K cohort single-cell RNA sequencing express correct markers and cluster according to cell type subsets.H. Applying the Module to Identify Potential Drug Targets i. Subject Selection

[0107] To identify novel drug targets for treating autoimmune disease, control subjects (n=18) and RA subjects (n=19) were selected based on the presence or absence of candidate genetic risk factors associated with rheumatoid arthritis. FIG. 14A illustrates an example of selecting the candidate risk factors, where cell type subsetspecific epigenetic signals concordant with genetic risk were identified using ATAC-seq and RNA sequencing methods. Subset single-cell RNA sequencing in subject samples revealed concordance of cell-type specificity with risk-associated epigenetic signal (FIG. 14B). Further examination through cell-type subset-level resolution identified thousands of novel single nucleotide polymorphism (SNP) expression associations (FIG. 14C). Additionally, subset single-cell RNA sequencing revealed novel eQTLs that are linked to known autoimmune disease genetic risk factors (FIG. 14D). ii. Gene K Case Study a. Gene K is a Differentially Expressed Genes in T Cell Subsets of RA Subjects

[0108] To assess potential gene expression changes in cell type subsets in RA subjects compared to control subjects, T cell subsets were examined for DEGs. This revealed that come CD4+T cell subsets possess DEGs in the RA subjects (“Autoimmune”) relative to control subjects (“healthy”) (FIG. 15). DEGs included Gene K. b. Gene K Stimulates Inflammatory Signaling

[0109] To determine whether Gene K is implicated in inflammation, PBMC samples from control and RA subjects were treated for 24 hours with recombinant Gene K or lipopolysaccharide (LPS) plus interferon gamma (INF-g). Supernatants were collected from each sample and macrophages were isolated. Inflammatory factors such as TNFo and cytokines (including IL-6 and IL-1 b) were quantified (FIG. 16). This revealed that Gene K alone can stimulate inflammatory cytokines from macrophages.c. CRISPR Knockout of Gene K in CD4+T Cells Reduced TNFa and IL-6 Production

[0110] To determine if knockout of Gene K could reduce inflammatory signaling, Gene K was deleted from CD4+t cells (non-Treg) and verified by Next Generation Sequencing (NGS) and ELISA of Gene K in lysates (FIG. 17A and B). Supernatants were collected after 7 days of inflammatory activation and culturing of CD4+T cells. TNFa and IL-6 were measured by Luminex® technologies (FIG. 17C and D). d. Gene K is a Potential Drug Target

[0111] Variant analysis of Gene K revealed a Gene K SNP associated with genetic risk for RA, where the Gene K SNP was most commonly linked to T cell subsets (FIG. 18). These data suggest that Gene K is a potential drug target for treating inflammatory conditions, such as RA, where reducing Gene K activity with a therapeutic may mitigate inflammatory responses (FIG. 19).Hi. Gene X Case Study a. Gene X Expression is Increased in Tregs

[0112] Subset RNA sequencing in RA subject samples revealed that Gene X expression is increased in Tregs relative to other immune cell type subsets (FIG. 20A). ATAC-seq analysis on these same samples revealed an RA-associated SNP in accessible regions of Gene X. b. Gene X Deletion in Tregs Increases Suppression Function

[0113] Because Tregs may maintain immune tolerance and are able to ameliorate inflammatory and autoimmune responses, consequences of Gene X deletion was next examined. CRISPR deletion of Gene X (FIG. 21 A and B), which consequentially reduced Gene X mRNA (FIG. 21 C) and protein levels (FIG. 21 D), effectively increased Treg suppression function (FIG. 22B and C) relative to controls when treated with leukopaks (FIG. 22A). Gene X knockout also increased presence of Teff cells relative to wild type Treg controls.

[0114] Gene X knockout Treg cells were next profiled with RNA sequencing and ATAC-seq, which revealed DEGs relative to wild type Treg controls. These DEGsincluded those with known immune modulating functions, such as IL-13, IL-10, and IL- 7R (FIG. 23). Each of these DEGs displayed reduced expression in Gene X knockout Treg cells relative to wild type Treg controls.

[0115] These data suggest that Gene X is a potential drug target for treating inflammatory conditions, such as RA, where reducing Gene X activity with a therapeutic may increase Treg suppression of Teff cells (FIG. 24).

[0116] Given the increased depth, total gene identification, and drug target identification provided by the transformation module of the present technology relative to single-cell datasets, the transformation module of the present technology may be used in applications such as these, where novel drug targets may be identified through enhanced cell type subset characterization.Additional Embodiments

[0117] The present technology includes, but is not limited to, the follow specific embodiments set forth herein below in paragraphs

[0118] -

[0182] :

[0118] 1. A non-transitory medium with instructions stored thereon that, when executed by a processor of a computing device, causes the computing device to perform steps for transforming a single-cell dataset into a transformed bulk-cell dataset, comprising the steps of:(i) applying, to the single-cell dataset, a machine learning model that is configured to predict a cell type subset label for each cell, so as to create an augmented single-cell dataset;(ii) normalizing the augmented single-cell dataset from (i) and performing a negative binomial distribution test to identify one or more differential features from the single-cell dataset relative to a control bulk-cell dataset;(Hi) imputing additional values for the augmented single-cell dataset by employing an algorithm that utilizes matrix regularization;(iv) compiling the augmented single-cell dataset from (I) and the additional values from (iii) into a data structure that is representative of the bulk-cell dataset; and(v) storing the data structure in a memory.

[0119] 2. The non-transitory medium of embodiment 1 , further comprising a step of performing a batch correction operation based on the one or more differential features from (ii).

[0120] 3. The non-transitory medium of embodiment 1 or 2, wherein the steps further comprise (vi) training a machine learning model using a training dataset that includes the bulk-cell dataset.

[0121] 4. The non-transitory medium of any one of embodiments 1 -3, further comprising combining the transformed bulk-cell dataset with a second bulk-cell dataset to generate a combined bulk-cell dataset.

[0122] 5. The non-transitory medium of embodiment 4, wherein the second bulk-cell data set is generated by a method comprising: a) contacting a cell-containing sample with a plurality of antigen-binding molecules under conditions that promote specific binding of the antigen-binding molecules to target antigens of the cells in the sample; wherein each antigenbinding molecule in the plurality binds to a unique target antigen in one or more cells of the sample, wherein each antigen-binding molecule is conjugated to an oligonucleotide, and wherein the oligonucleotide comprises (i) a unique molecular identifier sequence, (ii) an antigen-binding molecule identifier sequence corresponding to the unique target antigen, and (iii) an assay molecular identifier sequence; b) tagmenting genomic DNA in the cells to produce a plurality of double-stranded genomic DNA fragments comprising an oligonucleotide adaptor sequence at the 5' end of each genomic DNA fragment; c) lysing the cells in the sample; d) isolating mRNA transcripts from the cell lysate to produce a first fraction comprising the mRNA transcripts and a second fraction comprising tagmented genomic DNA fragments and the antigen-binding molecule-bound antigens from the cell lysate; e) generating a cDNA library by reverse transcribing the mRNA transcripts in the first fraction;f) amplifying (i) the unique molecular identifier sequences; (ii) the antigen-binding molecule identifier sequences; (iii) the assay molecular identifier sequences, and the (iv) tagmented genomic DNA fragments in the second fraction; g) analyzing the amplified unique molecular identifier sequences, the assay molecular identifier sequences, and the antigen-binding molecule identifier sequences to identify at least one target antigen in the second fraction to identify the expression profile of the sample; h) analyzing the tagmented genomic DNA fragments in the second fraction to identify the epigenetic profile of the sample; and i) analyzing the cDNA library generated from the first fraction to identify the transcriptional profile of the sample.

[0123] 6. The non-transitory medium of embodiment 4, wherein the second bulk-cell data set comprises data from one or more of a genome wide association study (GWAS), a database of transcription factor binding profiles, a quantitative trait loci (QTL) mapping, and an expression QTL (eQTL) mapping.

[0124] 7. The non-transitory medium of any one of the preceding embodiments, wherein the single-cell dataset, the transformed bulk-cell dataset, the control bulk-cell dataset, or the second bulk-cell dataset is selected from the group consisting of a genomic dataset, a transcriptomic dataset, an epigenomic dataset, a spatial genomics dataset, a chromatin accessibility dataset, a proteomics dataset, and a metabolomics dataset.

[0125] 8. The non-transitory medium of embodiment 7, wherein the genomic dataset is generated using one or more sequencing methods selected from the group consisting of Sanger sequencing, next-generation sequencing, nanopore sequencing, shotgun sequencing, pyrosequencing, and single-molecule real-time sequencing.

[0126] 9. The non-transitory medium of embodiment 7, wherein the transcriptomic dataset is generated using one or more sequencing methods selected from the group consisting of poly(A) selection sequencing, total RNA sequencing, strand-specific RNA sequencing, ribosome profiling sequencing, small RNA sequencing, long read RNA sequencing, capture-based RNA sequencing, and chimeric RNA sequencing.

[0127] 10. The non-transitory medium of embodiment 7, wherein the epigenomic dataset is generated using one or more methods selected from the group consisting of DNA methylation profiling, histone modification profiling, and RNA methylation profiling.

[0128] 11. The non-transitory medium of embodiment 7, wherein the spatial genomics dataset is generated using one or more methods selected from the group consisting of Hi-throughput chromosome conformation capture (Hi-C) sequencing, Capture Hi-C sequencing, Chromosome Conformation Capture (3C) sequencing, Circular Chromosome Conformation Capture (4C) sequencing, Chromosome Conformation Capture Carbon Copy (5C) sequencing, Hi-C with Chromatin Immunoprecipitation (HiChIP) sequencing, Capture-C sequencing, Targeted Chromosome Conformation Capture (T2C) sequencing, and Promoter Capture Hi-C sequencing.

[0129] 12. The non-transitory medium of embodiment 7, wherein the chromatin accessibility dataset is generated using one or more sequencing methods selected from the group consisting of Assay for Transposase-Accessible Chromatin (ATAC) sequencing, DNase sequencing, MNase sequencing, Formaldehyde-Assisted Isolation of Regulatory Elements (FAIRE) sequencing, Cleavage Under Targets and Release Using Nuclease (CUT&RUN), and Chromatin Immunoprecipitation (ChIP) sequencing, Cleavage Under Targets and Tag mentation (CUT&Tag) sequencing.

[0130] 13. The non-transitory medium of embodiment 7, wherein the proteomic dataset is generated using one or more methods selected from the group consisting of mass spectrometry, tandem mass spectrometry, label-free quantification, isotope labeling, a protein-protein interaction analysis, and structural proteomics.

[0131] 14. The non-transitory medium of any one of the preceding embodiments, wherein the machine learning model is developed using a Ridge logistic regression algorithm, a random forest algorithm, an extreme Gradient Boosting (XGBoost) algorithm, or a Single-cell Prediction of Population-level Effects from Singlecell Data (scPred) algorithm.

[0132] 15. The non-transitory medium of any one of the preceding embodiments, wherein the differential feature comprises a differentially expressed generelative to a control, a differentially abundant protein relative to a control, or a genetic variant relative to a control.

[0133] 16. The non-transitory medium of any one of the preceding embodiments, wherein the batch correction comprises one or more approaches selected from the group consisting of identifying pairs of mutual nearest neighbors (MNN), canonical correlation analysis (CCA), negative binomial regression, surrogate variable analysis, distance-based correlation, and principal component analysis (PCA) correction.

[0134] 17. The non-transitory medium of any one of the preceding embodiments, wherein the single-cell dataset is derived from a human sample.

[0135] 18. The non-transitory medium of embodiment 17, wherein the human sample is derived from a human having a disease.

[0136] 19. The non-transitory medium of embodiment 18, wherein the disease comprises a cancer or an immune disease.

[0137] 20. The non-transitory medium of any one of embodiments 17-19, wherein the human sample comprises a urine sample, a saliva sample, a circulatory fluid sample, or a tissue sample.

[0138] 21 . The non-transitory medium of embodiment 20, wherein the tissue sample comprises one or more tissues selected from the group consisting of an epithelial tissue, a connective tissue, a nervous tissue, an adipose tissue, a cartilage, a bone tissue, a skin tissue, a mucous membrane tissue, a glandular tissue, a vascular tissue, a cardiac tissue, a smooth muscle tissue, a skeletal muscle tissue, a neural tissue, a fibrous tissue, a bone marrow tissue, a liver tissue, a kidney tissue, a pancreatic tissue, a pulmonary tissue, and a brain tissue.

[0139] 22. The non-transitory medium of embodiment 21 , wherein the circulatory fluid sample comprises a blood sample or a lymph sample.

[0140] 23. The non-transitory medium of embodiment 22, wherein the blood sample comprises peripheral blood mononuclear cells (PBMC).

[0141] 24. The non-transitory medium of any one of the preceding embodiments, wherein the transformed bulk-cell dataset comprises one or more of a recovered sequencing dropout event from the singe-cell dataset, an increased amountof data relative to the single-cell dataset or Xseq data, or an increased depth of data relative to the single-cell dataset.

[0142] 25. A method for identifying a subject as having a disease, the method comprising:(i) obtaining a single-cell dataset derived from a biological sample of the subject;(ii) applying, to the single-cell dataset, a machine learning model that is configured to predict a cell type subset label for each cell, so as to create an augmented single-cell dataset;(iii) normalizing the augmented single-cell dataset from (i) and performing a negative binomial distribution test to identify one or more differential features from the single-cell dataset relative to a control bulk-cell dataset;(iv) imputing additional values for the augmented single-cell dataset by employing an algorithm that utilizes matrix regularization;(v) compiling the augmented single-cell dataset from (i) and the additional values from (iii) into a data structure that is representative of the bulk-cell dataset; and(vi) storing the data structure in a memory.

[0143] 26. A method for identifying one or more of a disease drug target, a risk factor for developing a disease, a cell type subset, or a biomarker in a subject, the method comprising:(i) obtaining a single-cell dataset derived from a biological sample of the subject,(ii) applying, to the single-cell dataset, a machine learning model that is configured to predict a cell type subset label for each cell, so as to create an augmented single-cell dataset;(iii) normalizing the augmented single-cell dataset from (i) and performing a negative binomial distribution test to identify one or more differential features from the single-cell dataset relative to a control bulk-cell dataset;(iv) imputing additional values for the augmented single-cell dataset by employing an algorithm that utilizes matrix regularization;(v) compiling the augmented single-cell dataset from (i) and the additional values from (iii) into a data structure that is representative of the bulk-cell dataset; and(vi) storing the data structure in a memory.

[0144] 27. The method of embodiment 25 or 26, wherein the disease comprises an immune disease or a cancer.

[0145] 28. The method of embodiment 27, wherein the immune disease comprises an autoimmune disorder.

[0146] 29. The method of embodiment 28, wherein the autoimmune disorder comprises rheumatoid arthritis (RA) or multiple sclerosis (MS).

[0147] 30. The method of any one of embodiments 25-29, wherein the biological sample comprises a urine sample, a saliva sample, a circulatory fluid sample, or a tissue sample.

[0148] 31 . The method of embodiment 30, wherein the solid tissue sample comprises one or more tissues selected from the group consisting of an epithelial tissue, a connective tissue, a nervous tissue, an adipose tissue, a cartilage, a bone tissue, a skin tissue, a mucous membrane tissue, a glandular tissue, a vascular tissue, a cardiac tissue, a smooth muscle tissue, a skeletal muscle tissue, a neural tissue, a fibrous tissue, a bone marrow tissue, a liver tissue, a kidney tissue, a pancreatic tissue, a pulmonary tissue, and a brain tissue.

[0149] 32. The method of embodiment 31 , wherein the circulatory fluid sample comprises a blood sample or a lymph sample.

[0150] 33. The method of embodiment 32, wherein the blood sample comprises peripheral blood mononuclear cells (PBMC).

[0151] 34. The method of embodiment 26, wherein the drug target comprises a G-protein coupled receptor (GPCR) subunit.

[0152] 35. The method of embodiment 26, wherein the cell type subset is an immune subset.

[0153] 36. The method of embodiment 35, wherein the immune subset is a myeloid cell subset, a T cell subset, or a B cell subset.

[0154] 37. The method of embodiment 36, wherein the myeloid cell subset comprises a dendritic cell (DC) subset or a monocyte subset.

[0155] 38. The method of embodiment 37, wherein the DC subset comprises a conventional DC or a plasmacytoid DC.

[0156] 39. The method of embodiment 37, wherein the monocyte subset comprises a classical monocyte, an intermediate monocyte, or a nonclassical monocyte.

[0157] 40. The method of embodiment 36, wherein the T cell subset comprises a CD4+a T cell subset, a CD8+ap T cell subset, a yb T cell subset, or a natural killer (NK) cell subset.

[0158] 41 . The method of embodiment 40, wherein the CD4+op T cell subset comprises a naive CD4+T cell, a central memory CD4+T cell, an effector memory CD4+T cell, a CD45RA+effector memory CD4+T cell, or a CD4+T regulatory cell.

[0159] 42. The method of embodiment 40, wherein the CD8+op T cell subset comprises a naive CD8+T cell, a central memory CD8+T cell, an effector memory CD8+T cell, or a CD45RA+ effector memory CD8+T cell.

[0160] 43. The method of embodiment 40, wherein the yb T cell subset comprises a gammadelta T cell.

[0161] 44. The method of embodiment 40, wherein the NK cell subset comprises a CD56hiNK cell or a CD56lowNK cell.

[0162] 45. The method of embodiment 36, wherein the B cell subset comprises a naive B cell or an effector B cell.

[0163] 46. The method of embodiment 45, wherein the naive B cell comprises an unswitched naive B cell, a class switched naive B cell, or a transitional B cell.

[0164] 47. The method of embodiment 45, wherein the effector B cell comprises a class switched classical memory B cell, an lgM+IgD" classical memory B cell, an atypical memory B cell, or a class switch plasmablast.

[0165] 48. The method of embodiment 26, wherein the risk factor is a genetic variation relative to a control, a differentially expressed gene relative to a control, or a differentially abundant biomarker relative to a control.

[0166] 49. The method of embodiment 48, wherein the genetic variation comprises a single nucleotide variation (SNV), a single nucleotide polymorphism (SNP), a haplotype difference, an insertion-deletion (INDEL), a mitochondrial DNA (mtDNA) variation, a chromosomal abnormality, or a ploidy difference.

[0167] 50. The method of any one of embodiments 25-49, further comprising a step of performing a batch correction operation based on the one or more differential features from (ii);

[0168] 51 . The method of any one of embodiments 25-50, wherein the steps further comprise (vii) training a machine learning model using a training dataset that includes the bulk-cell dataset.

[0169] 52. The method of any one of embodiments 25-51 , further comprising combining the transformed bulk-cell dataset with a second bulk-cell dataset to generate a combined bulk-cell dataset.

[0170] 53. The method of embodiment 52, wherein the second bulk-cell data set is generated by a second method comprising: a) contacting a cell-containing sample with a plurality of antigen-binding molecules under conditions that promote specific binding of the antigen-binding molecules to target antigens of the cells in the sample; wherein each antigenbinding molecule in the plurality binds to a unique target antigen in one or more cells of the sample, wherein each antigen-binding molecule is conjugated to an oligonucleotide, and wherein the oligonucleotide comprises (i) a unique molecular identifier sequence, (ii) an antigen-binding molecule identifier sequence corresponding to the unique target antigen, and (iii) an assay molecular identifier sequence; b) tagmenting genomic DNA in the cells to produce a plurality of double-stranded genomic DNA fragments comprising an oligonucleotide adaptor sequence at the 5' end of each genomic DNA fragment; c) lysing the cells in the sample;d) isolating mRNA transcripts from the cell lysate to produce a first fraction comprising the mRNA transcripts and a second fraction comprising tagmented genomic DNA fragments and the antigen-binding molecule-bound antigens from the cell lysate; e) generating a cDNA library by reverse transcribing the mRNA transcripts in the first fraction; f) amplifying (i) the unique molecular identifier sequences; (ii) the antigen-binding molecule identifier sequences; (iii) the assay molecular identifier sequences, and the (iv) tagmented genomic DNA fragments in the second fraction; g) analyzing the amplified unique molecular identifier sequences, the assay molecular identifier sequences, and the antigen-binding molecule identifier sequences to identify at least one target antigen in the second fraction to identify the expression profile of the sample; h) analyzing the tagmented genomic DNA fragments in the second fraction to identify the epigenetic profile of the sample; and i) analyzing the cDNA library generated from the first fraction to identify the transcriptional profile of the sample.

[0171] 54. The method of embodiment 52, wherein the second bulk-cell data set comprises data from one or more of a GWAS, a database of transcription factor binding profiles, a QTL mapping, and an eQTL mapping.

[0172] 55. The method of any one of embodiments 25-54, wherein the singlecell dataset, the transformed bulk-cell dataset, the control bulk-cell dataset, or the second bulk-cell dataset is selected from the group consisting of a genomic dataset, a transcriptomic dataset, an epigenomic dataset, a spatial genomics dataset, a chromatin accessibility dataset, a proteomics dataset, and a metabolomics dataset.

[0173] 56. The method of embodiment 55, wherein the genomic dataset is generated using one or more sequencing methods selected from the group consisting of Sanger sequencing, next-generation sequencing, nanopore sequencing, shotgun sequencing, pyrosequencing, and single-molecule real-time sequencing.

[0174] 57. The method of embodiment 55, wherein the transcriptomic dataset is generated using one or more sequencing methods selected from the group consistingof poly(A) selection sequencing, total RNA sequencing, strand-specific RNA sequencing, ribosome profiling sequencing, small RNA sequencing, long read RNA sequencing, capture-based RNA sequencing, and chimeric RNA sequencing.

[0175] 58. The method of embodiment 55, wherein the epigenomic dataset is generated using one or more methods selected from the group consisting of DNA methylation profiling, histone modification profiling, and RNA methylation profiling.

[0176] 59. The method of embodiment 55, wherein the spatial genomics dataset is generated using one or more methods selected from the group consisting of Hi-throughput chromosome conformation capture (Hi-C) sequencing, Capture Hi-C sequencing, Chromosome Conformation Capture (3C) sequencing, Circular Chromosome Conformation Capture (4C) sequencing, Chromosome Conformation Capture Carbon Copy (5C) sequencing, Hi-C with Chromatin Immunoprecipitation (HiChIP) sequencing, Capture-C sequencing, Targeted Chromosome Conformation Capture (T2C) sequencing, and Promoter Capture Hi-C sequencing.

[0177] 60. The method of embodiment 55, wherein the chromatin accessibility dataset is generated using one or more sequencing methods selected from the group consisting of Assay forTransposase-Accessible Chromatin (ATAC) sequencing, DNase sequencing, MNase sequencing, Formaldehyde-Assisted Isolation of Regulatory Elements (FAIRE) sequencing, Cleavage Under Targets and Release Using Nuclease (CUT&RUN), and Chromatin Immunoprecipitation (ChIP) sequencing, Cleavage Under Targets and Tag mentation (CUT&Tag) sequencing.

[0178] 61. The method of embodiment 55, wherein the proteomic dataset is generated using one or more methods selected from the group consisting of mass spectrometry, tandem mass spectrometry, label-free quantification, isotope labeling, a protein-protein interaction analysis, and structural proteomics.

[0179] 62. The method of any one of embodiments 25-61 , wherein the machine learning model is developed using a Ridge logistic regression algorithm, a random forest algorithm, an extreme Gradient Boosting (XGBoost) algorithm, or a Single-cell Prediction of Population-level Effects from Single-cell Data (scPred) algorithm.

[0180] 63. The method of any one of embodiments 25-62, wherein the differential feature comprises a differentially expressed gene relative to a control, a differentially abundant protein relative to a control, or a genetic variant relative to a control.

[0181] 64. The method of any one of embodiments 25-63, wherein the batch correction comprises one or more approaches selected from the group consisting of identifying pairs of mutual nearest neighbors (MNN), canonical correlation analysis (CCA), negative binomial regression, surrogate variable analysis, distance-based correlation, principal component analysis (PCA) correction.

[0182] 65. The method of any one of embodiments 25-64, wherein the transformed bulk-cell dataset comprises one or more or a recovered sequencing dropout event from the singe-cell dataset, an increased amount of data relative to the single-cell dataset, or an increased depth of data relative to the single-cell dataset.Remarks

[0183] The foregoing description of various embodiments of the claimed subject matter has been provided for the purposes of illustration and description. It is not intended to be exhaustive or to limit the claimed subject matter to the precise forms disclosed. Many modifications and variations will be apparent to one skilled in the art. Embodiments were chosen and described in order to best describe the principles of the invention and its practical applications, thereby enabling those skilled in the relevant art to understand the claimed subject matter, the various embodiments, and the various modifications that are suited to the particular uses contemplated.

[0184] Although the Detailed Description describes certain embodiments and the best mode contemplated, the technology can be practiced in many ways no matter how detailed the Detailed Description appears. Embodiments may vary considerably in their implementation details, while still being encompassed by the specification. Particular terminology used when describing certain features or aspects of various embodiments should not be taken to imply that the terminology is being redefined herein to be restricted to any specific characteristics, features, or aspects of the technology with which that terminology is associated. In general, the terms used in the following claims should not be construed to limit the technology to the specific embodiments disclosedin the specification, unless those terms are explicitly defined herein. Accordingly, the actual scope of the technology encompasses not only the disclosed embodiments, but also all equivalent ways of practicing or implementing the embodiments.

[0185] The language used in the specification has been principally selected for readability and instructional purposes. It may not have been selected to delineate or circumscribe the subject matter. It is therefore intended that the scope of the technology be limited not by this Detailed Description, but rather by any claims that issue on an application based hereon. Accordingly, the disclosure of various embodiments is intended to be illustrative, but not limiting, of the scope of the technology as set forth in the following claims.

Claims

CLAIMSI / We claim:

1. A non-transitory medium with instructions stored thereon that, when executed by a processor of a computing device, causes the computing device to perform steps for transforming a single-cell dataset into a transformed bulk-cell dataset, comprising the steps of:(i) applying, to the single-cell dataset, a machine learning model that is configured to predict a cell type subset label for each cell, so as to create an augmented single-cell dataset;(ii) normalizing the augmented single-cell dataset from (i) and performing a negative binomial distribution test to identify one or more differential features from the single-cell dataset relative to a control bulk-cell dataset;(iii) imputing additional values for the augmented single-cell dataset by employing an algorithm that utilizes matrix regularization;(iv) compiling the augmented single-cell dataset from (i) and the additional values from (iii) into a data structure that is representative of the bulk-cell dataset; and(v) storing the data structure in a memory.

2. The non-transitory medium of claim 1 , further comprising a step of performing a batch correction operation based on the one or more differential features from (ii).

3. The non-transitory medium of claim 1 or 2, wherein the steps further comprise (vi) training a machine learning model using a training dataset that includes the bulk-cell dataset.

4. The non-transitory medium of any one of claims 1 -3, further comprising combining the transformed bulk-cell dataset with a second bulk-cell dataset to generate a combined bulk-cell dataset.

5. The non-transitory medium of claim 4, wherein the second bulk-cell data set is generated by a method comprising: a) contacting a cell-containing sample with a plurality of antigen-binding molecules under conditions that promote specific binding of the antigenbinding molecules to target antigens of the cells in the sample; wherein each antigen-binding molecule in the plurality binds to a unique target antigen in one or more cells of the sample, wherein each antigen-binding molecule is conjugated to an oligonucleotide, and wherein the oligonucleotide comprises (i) a unique molecular identifier sequence, (ii) an antigen-binding molecule identifier sequence corresponding to the unique target antigen, and (iii) an assay molecular identifier sequence; b) tagmenting genomic DNA in the cells to produce a plurality of double-stranded genomic DNA fragments comprising an oligonucleotide adaptor sequence at the 5' end of each genomic DNA fragment; c) lysing the cells in the sample; d) isolating mRNA transcripts from the cell lysate to produce a first fraction comprising the mRNA transcripts and a second fraction comprising tagmented genomic DNA fragments and the antigen-binding moleculebound antigens from the cell lysate; e) generating a cDNA library by reverse transcribing the mRNA transcripts in the first fraction; f) amplifying (i) the unique molecular identifier sequences; (ii) the antigen-binding molecule identifier sequences; (iii) the assay molecular identifier sequences, and the (iv) tagmented genomic DNA fragments in the second fraction; g) analyzing the amplified unique molecular identifier sequences, the assay molecular identifier sequences, and the antigen-binding molecule identifier sequences to identify at least one target antigen in the second fraction to identify the expression profile of the sample; h) analyzing the tagmented genomic DNA fragments in the second fraction to identify the epigenetic profile of the sample; and i) analyzing the cDNA library generated from the first fraction to identify the transcriptional profile of the sample.

6. The non-transitory medium of claim 4, wherein the second bulk-cell data set comprises data from one or more of a genome wide association study (GWAS), a database of transcription factor binding profiles, a quantitative trait loci (QTL) mapping, and an expression QTL (eQTL) mapping.

7. The non-transitory medium of any one of the preceding claims, wherein the single-cell dataset, the transformed bulk-cell dataset, the control bulk-cell dataset, or the second bulk-cell dataset is selected from the group consisting of a genomic dataset, a transcriptomic dataset, an epigenomic dataset, a spatial genomics dataset, a chromatin accessibility dataset, a proteomics dataset, and a metabolomics dataset.

8. The non-transitory medium of claim 7, wherein the genomic dataset is generated using one or more sequencing methods selected from the group consisting of Sanger sequencing, next-generation sequencing, nanopore sequencing, shotgun sequencing, pyrosequencing, and single-molecule real-time sequencing.

9. The non-transitory medium of claim 7, wherein the transcriptomic dataset is generated using one or more sequencing methods selected from the group consisting of poly(A) selection sequencing, total RNA sequencing, strand-specific RNA sequencing, ribosome profiling sequencing, small RNA sequencing, long read RNA sequencing, capture-based RNA sequencing, and chimeric RNA sequencing.

10. The non-transitory medium of claim 7, wherein the epigenomic dataset is generated using one or more methods selected from the group consisting of DNA methylation profiling, histone modification profiling, and RNA methylation profiling.1 1 . The non-transitory medium of claim 7, wherein the spatial genomics dataset is generated using one or more methods selected from the group consisting of Hi-throughput chromosome conformation capture (Hi-C) sequencing, Capture Hi-C sequencing, Chromosome Conformation Capture (3C) sequencing, Circular Chromosome Conformation Capture (4C) sequencing, Chromosome Conformation Capture Carbon Copy (5C) sequencing, Hi-C with Chromatin Immunoprecipitation(HiChIP) sequencing, Capture-C sequencing, Targeted Chromosome Conformation Capture (T2C) sequencing, and Promoter Capture Hi-C sequencing.

12. The non-transitory medium of claim 7, wherein the chromatin accessibility dataset is generated using one or more sequencing methods selected from the group consisting of Assay forTransposase-Accessible Chromatin (ATAC) sequencing, DNase sequencing, MNase sequencing, Formaldehyde-Assisted Isolation of Regulatory Elements (FAIRE) sequencing, Cleavage Under Targets and Release Using Nuclease (CUT&RUN), and Chromatin Immunoprecipitation (ChIP) sequencing, Cleavage Under Targets and Tag mentation (CUT&Tag) sequencing.

13. The non-transitory medium of claim 7, wherein the proteomic dataset is generated using one or more methods selected from the group consisting of mass spectrometry, tandem mass spectrometry, label-free quantification, isotope labeling, a protein-protein interaction analysis, and structural proteomics.

14. The non-transitory medium of any one of the preceding claims, wherein the machine learning model is developed using a Ridge logistic regression algorithm, a random forest algorithm, an extreme Gradient Boosting (XGBoost) algorithm, or a Single-cell Prediction of Population-level Effects from Single-cell Data (scPred) algorithm.

15. The non-transitory medium of any one of the preceding claims, wherein the differential feature comprises a differentially expressed gene relative to a control, a differentially abundant protein relative to a control, or a genetic variant relative to a control.

16. The non-transitory medium of any one of the preceding claims, wherein the batch correction comprises one or more approaches selected from the group consisting of identifying pairs of mutual nearest neighbors (MNN), canonical correlation analysis (CCA), negative binomial regression, surrogate variable analysis, distancebased correlation, and principal component analysis (PCA) correction.

17. The non-transitory medium of any one of the preceding claims, wherein the single-cell dataset is derived from a human sample.

18. The non-transitory medium of claim 17, wherein the human sample is derived from a human having a disease.

19. The non-transitory medium of claim 18, wherein the disease comprises a cancer or an immune disease.

20. The non-transitory medium of any one of claims 17-19, wherein the human sample comprises a urine sample, a saliva sample, a circulatory fluid sample, or a tissue sample.

21. The non-transitory medium of claim 20, wherein the tissue sample comprises one or more tissues selected from the group consisting of an epithelial tissue, a connective tissue, a nervous tissue, an adipose tissue, a cartilage, a bone tissue, a skin tissue, a mucous membrane tissue, a glandular tissue, a vascular tissue, a cardiac tissue, a smooth muscle tissue, a skeletal muscle tissue, a neural tissue, a fibrous tissue, a bone marrow tissue, a liver tissue, a kidney tissue, a pancreatic tissue, a pulmonary tissue, and a brain tissue.

22. The non-transitory medium of claim 21 , wherein the circulatory fluid sample comprises a blood sample or a lymph sample.

23. The non-transitory medium of claim 22, wherein the blood sample comprises peripheral blood mononuclear cells (PBMC).

24. The non-transitory medium of any one of the preceding claims, wherein the transformed bulk-cell dataset comprises one or more of a recovered sequencing dropout event from the singe-cell dataset, an increased amount of data relative to the single-cell dataset or Xseq data, or an increased depth of data relative to the single-cell dataset.

25. A method for identifying a subject as having a disease, the method comprising:(i) obtaining a single-cell dataset derived from a biological sample of the subject;(ii) applying, to the single-cell dataset, a machine learning model that is configured to predict a cell type subset label for each cell, so as to create an augmented single-cell dataset;(iii) normalizing the augmented single-cell dataset from (i) and performing a negative binomial distribution test to identify one or more differential features from the single-cell dataset relative to a control bulk-cell dataset;(iv) imputing additional values for the augmented single-cell dataset by employing an algorithm that utilizes matrix regularization;(v) compiling the augmented single-cell dataset from (i) and the additional values from (iii) into a data structure that is representative of the bulk-cell dataset; and(vi) storing the data structure in a memory.

26. A method for identifying one or more of a disease drug target, a risk factor for developing a disease, a cell type subset, or a biomarker in a subject, the method comprising:(i) obtaining a single-cell dataset derived from a biological sample of the subject,(ii) applying, to the single-cell dataset, a machine learning model that is configured to predict a cell type subset label for each cell, so as to create an augmented single-cell dataset;(iii) normalizing the augmented single-cell dataset from (i) and performing a negative binomial distribution test to identify one or more differential features from the single-cell dataset relative to a control bulk-cell dataset;(iv) imputing additional values for the augmented single-cell dataset by employing an algorithm that utilizes matrix regularization;(v) compiling the augmented single-cell dataset from (i) and the additional values from (iii) into a data structure that is representative of the bulk-cell dataset; and(vi) storing the data structure in a memory.

27. The method of claim 25 or 26, wherein the disease comprises an immune disease or a cancer.

28. The method of claim 27, wherein the immune disease comprises an autoimmune disorder.

29. The method of claim 28, wherein the autoimmune disorder comprises rheumatoid arthritis (RA) or multiple sclerosis (MS).

30. The method of any one of claims 25-29, wherein the biological sample comprises a urine sample, a saliva sample, a circulatory fluid sample, or a tissue sample.31 . The method of claim 30, wherein the solid tissue sample comprises one or more tissues selected from the group consisting of an epithelial tissue, a connective tissue, a nervous tissue, an adipose tissue, a cartilage, a bone tissue, a skin tissue, a mucous membrane tissue, a glandular tissue, a vascular tissue, a cardiac tissue, a smooth muscle tissue, a skeletal muscle tissue, a neural tissue, a fibrous tissue, a bone marrow tissue, a liver tissue, a kidney tissue, a pancreatic tissue, a pulmonary tissue, and a brain tissue.

32. The method of claim 31 , wherein the circulatory fluid sample comprises a blood sample or a lymph sample.

33. The method of claim 32, wherein the blood sample comprises peripheral blood mononuclear cells (PBMC).

34. The method of claim 26, wherein the drug target comprises a G-protein coupled receptor (GPCR) subunit.

35. The method of claim 26, wherein the cell type subset is an immune subset.

36. The method of claim 35, wherein the immune subset is a myeloid cell subset, a T cell subset, or a B cell subset.

37. The method of claim 36, wherein the myeloid cell subset comprises a dendritic cell (DC) subset or a monocyte subset.

38. The method of claim 37, wherein the DC subset comprises a conventional DC or a plasmacytoid DC.

39. The method of claim 37, wherein the monocyte subset comprises a classical monocyte, an intermediate monocyte, or a nonclassical monocyte.

40. The method of claim 36, wherein the T cell subset comprises a CD4+a T cell subset, a CD8+ap T cell subset, a yb T cell subset, or a natural killer (NK) cell subset.41 . The method of claim 40, wherein the CD4+ap T cell subset comprises a naive CD4+T cell, a central memory CD4+T cell, an effector memory CD4+T cell, a CD45RA+ effector memory CD4+T cell, or a CD4+T regulatory cell.

42. The method of claim 40, wherein the CD8+ap T cell subset comprises a naive CD8+T cell, a central memory CD8+T cell, an effector memory CD8+T cell, or a CD45RA+effector memory CD8+T cell.

43. The method of claim 40, wherein the yb T cell subset comprises a gammadelta T cell.

44. The method of claim 40, wherein the NK cell subset comprises a CD56hiNK cell or a CD56lowNK cell.

45. The method of claim 36, wherein the B cell subset comprises a naive B cell or an effector B cell.-SO-46. The method of claim 45, wherein the naive B cell comprises an unswitched naive B cell, a class switched naive B cell, or a transitional B cell.

47. The method of claim 45, wherein the effector B cell comprises a class switched classical memory B cell, an lgM+IgD classical memory B cell, an atypical memory B cell, or a class switch plasmablast.

48. The method of claim 26, wherein the risk factor is a genetic variation relative to a control, a differentially expressed gene relative to a control, or a differentially abundant biomarker relative to a control.

49. The method of claim 48, wherein the genetic variation comprises a single nucleotide variation (SNV), a single nucleotide polymorphism (SNP), a haplotype difference, an insertion-deletion (INDEL), a mitochondrial DNA (mtDNA) variation, a chromosomal abnormality, or a ploidy difference.

50. The method of any one of claims 25-49, further comprising a step of performing a batch correction operation based on the one or more differential features from (ii).51 . The method of any one of claims 25-50, wherein the steps further comprise (vii) training a machine learning model using a training dataset that includes the bulk-cell dataset.

52. The method of any one of claims 25-51 , further comprising combining the transformed bulk-cell dataset with a second bulk-cell dataset to generate a combined bulk-cell dataset.

53. The method of claim 52, wherein the second bulk-cell data set is generated by a second method comprising: a) contacting a cell-containing sample with a plurality of antigen-binding molecules under conditions that promote specific binding of the antigenbinding molecules to target antigens of the cells in the sample; whereineach antigen-binding molecule in the plurality binds to a unique target antigen in one or more cells of the sample, wherein each antigen-binding molecule is conjugated to an oligonucleotide, and wherein the oligonucleotide comprises (i) a unique molecular identifier sequence, (ii) an antigen-binding molecule identifier sequence corresponding to the unique target antigen, and (iii) an assay molecular identifier sequence; b) tagmenting genomic DNA in the cells to produce a plurality of double-stranded genomic DNA fragments comprising an oligonucleotide adaptor sequence at the 5' end of each genomic DNA fragment; c) lysing the cells in the sample; d) isolating mRNA transcripts from the cell lysate to produce a first fraction comprising the mRNA transcripts and a second fraction comprising tagmented genomic DNA fragments and the antigen-binding moleculebound antigens from the cell lysate; e) generating a cDNA library by reverse transcribing the mRNA transcripts in the first fraction; f) amplifying (i) the unique molecular identifier sequences; (ii) the antigen-binding molecule identifier sequences; (iii) the assay molecular identifier sequences, and the (iv) tagmented genomic DNA fragments in the second fraction; g) analyzing the amplified unique molecular identifier sequences, the assay molecular identifier sequences, and the antigen-binding molecule identifier sequences to identify at least one target antigen in the second fraction to identify the expression profile of the sample; h) analyzing the tagmented genomic DNA fragments in the second fraction to identify the epigenetic profile of the sample; and i) analyzing the cDNA library generated from the first fraction to identify the transcriptional profile of the sample.

54. The method of claim 52, wherein the second bulk-cell data set comprises data from one or more of a GWAS, a database of transcription factor binding profiles, a QTL mapping, and an eQTL mapping.

55. The method of any one of claims 25-54, wherein the single-cell dataset, the transformed bulk-cell dataset, the control bulk-cell dataset, or the second bulk-cell dataset is selected from the group consisting of a genomic dataset, a transcriptomic dataset, an epigenomic dataset, a spatial genomics dataset, a chromatin accessibility dataset, a proteomics dataset, and a metabolomics dataset.

56. The method of claim 55, wherein the genomic dataset is generated using one or more sequencing methods selected from the group consisting of Sanger sequencing, next-generation sequencing, nanopore sequencing, shotgun sequencing, pyrosequencing, and single-molecule real-time sequencing.

57. The method of claim 55, wherein the transcriptomic dataset is generated using one or more sequencing methods selected from the group consisting of poly(A) selection sequencing, total RNA sequencing, strand-specific RNA sequencing, ribosome profiling sequencing, small RNA sequencing, long read RNA sequencing, capture-based RNA sequencing, and chimeric RNA sequencing.

58. The method of claim 55, wherein the epigenomic dataset is generated using one or more methods selected from the group consisting of DNA methylation profiling, histone modification profiling, and RNA methylation profiling.

59. The method of claim 55, wherein the spatial genomics dataset is generated using one or more methods selected from the group consisting of Hi- throughput chromosome conformation capture (Hi-C) sequencing, Capture Hi-C sequencing, Chromosome Conformation Capture (3C) sequencing, Circular Chromosome Conformation Capture (4C) sequencing, Chromosome Conformation Capture Carbon Copy (5C) sequencing, Hi-C with Chromatin Immunoprecipitation (HiChIP) sequencing, Capture-C sequencing, Targeted Chromosome Conformation Capture (T2C) sequencing, and Promoter Capture Hi-C sequencing.

60. The method of claim 55, wherein the chromatin accessibility dataset is generated using one or more sequencing methods selected from the group consisting of Assay for Transposase-Accessible Chromatin (ATAC) sequencing, DNasesequencing, MNase sequencing, Formaldehyde-Assisted Isolation of Regulatory Elements (FAIRE) sequencing, Cleavage Under Targets and Release Using Nuclease (CUT&RUN), and Chromatin Immunoprecipitation (ChIP) sequencing, Cleavage Under Targets and Tag mentation (CUT&Tag) sequencing.61 . The method of claim 55, wherein the proteomic dataset is generated using one or more methods selected from the group consisting of mass spectrometry, tandem mass spectrometry, label-free quantification, isotope labeling, a protein-protein interaction analysis, and structural proteomics.

62. The method of any one of claims 25-61 , wherein the machine learning model is developed using a Ridge logistic regression algorithm, a random forest algorithm, an extreme Gradient Boosting (XGBoost) algorithm, or a Single-cell Prediction of Population-level Effects from Single-cell Data (scPred) algorithm.

63. The method of any one of claims 25-62, wherein the differential feature comprises a differentially expressed gene relative to a control, a differentially abundant protein relative to a control, or a genetic variant relative to a control.

64. The method of any one of claims 25-63, wherein the batch correction comprises one or more approaches selected from the group consisting of identifying pairs of mutual nearest neighbors (MNN), canonical correlation analysis (CCA), negative binomial regression, surrogate variable analysis, distance-based correlation, principal component analysis (PCA) correction.

65. The method of any one of claims 25-64, wherein the transformed bulk-cell dataset comprises one or more or a recovered sequencing dropout event from the singe-cell dataset, an increased amount of data relative to the single-cell dataset, or an increased depth of data relative to the single-cell dataset.

Images