Method and system for analyzing biological samples

Abstract

A method for analyzing a biological sample comprising a plurality of targets is provided. The method comprises the steps of staining a target of the biological sample with a plurality of detectable markers; providing affinity information specifying the affinity of each marker for at least some of the targets; generating a readout of the biological sample with a marker to generate raw readout data; the raw readout data is unmixed by applying an unmixing algorithm and the affinity information to generate unmixed readout data with respect to the target. In another aspect, a system for analyzing a biological sample is provided.

Description

Technical Field

[0001] The present invention relates to a method for analyzing a biological sample comprising a plurality of targets. In another aspect, a system is provided, the system comprising an apparatus configured to perform the method. Background Art

[0002] The determination of proteomic, genomic and transcriptomic data from biological samples remains a major focus in the life sciences. Typically, optical readout is used to identify target structures that are labeled with fluorescent markers and that are specifically attached to the target structure with the aid of affinity reagents. This enables, for example, the identification of proteins. In combination with imaging of the biological sample, the light emitted by the marker at a specific location in the biological sample can be used to determine the identified protein. Many years of scientific effort have generated an enormous volume of data generated in this way.

[0003] The basic assumption behind these efforts is that affinity reagents can be used for these purposes because they bind to their exclusive targets with sufficient affinity and specificity. However, increasing evidence suggests that this basic principle does not stand up to scrutiny and that most affinity reagents are cross-reactive, i.e., they do not bind exclusively to their targets.

[0004] As fields like flow cytometry, plasma proteomics, and microscopy move toward higher levels of complexity, the issue of cross-reactivity (CR) has received increasing attention and has been recognized by the scientific community as a key component of the reproducibility crisis. In a position paper published in Nature (518, 27-29 (2015)) by Bradbury and Pluckthin, with 110 co-signers, it is stated that "…of the approximately 6,000 commercial antibodies in routine use, less than half recognize only their designated targets". A recent study by Schwenk et al. analyzed 11,000 affinity-purified monoclonal antibodies and found that only 531 monoclonal antibodies produced a single band on a Western blot.

[0005] With the advent of more multiplexed analytical techniques such as microsphere-based assays or protein microarrays, there is a growing awareness that the cross-reactivity (CR) of affinity reagents, i.e. their binding to non-specific or so-called off-target targets, poses a fundamental limitation to this approach, with significant impacts on life science research and society. Summary of the invention

[0006] The object is to provide a method and a system for analyzing a biological sample with a marker that is cross-reactive with several targets.

[0007] The aforementioned objects are achieved by the subject-matter of the independent claims. Advantageous embodiments are defined in the dependent claims and in the following description.

[0008] A method for analyzing a biological sample including a plurality of targets is provided. The method comprises the following steps: staining or contacting or linking the targets of the biological sample with a plurality of detectable markers; providing affinity information specifying the affinity of each marker for at least some of the targets; generating a readout of the biological sample with the markers to generate raw readout data; unmixing the raw readout data by applying an unmixing algorithm and the affinity information to generate unmixed readout data for the targets.

[0009] Each of the plurality of detectable markers comprises an affinity agent and a detectable label that is bound to the affinity agent for detecting the marker. The binding of the detectable label to the determined affinity agent can be performed before staining or contacting or connecting the target (pre-made label; e.g., direct antibody label conjugate) of the biological sample. Alternatively, the binding of the detectable label can be performed simultaneously or after the affinity agent is introduced into the sample (e.g., a barcoded antibody that is subsequently bound by a barcoded label).

[0010] The affinity reagent can include, for example, an amino acid structure or a nucleic acid structure. The amino acid structure can be, for example, an antibody or an antibody fragment. Further, the amino acid structure can be an aptamer based on a peptide. The nucleic acid structure can be, for example, a nucleic acid aptamer. Usually, affinity reagents, especially aptamer affinity reagents have complex secondary, tertiary and / or quaternary structures. This complex secondary, tertiary and / or quaternary structure usually promotes the combination of any affinity reagent with any specific target. Specifically, this combination based on complex secondary, tertiary and / or quaternary structure does not include a combination based on complementary sequence hybridization, which is determined only by the linear order of nucleotides in the nucleic acid sequence.

[0011] Generally, when it is intended to identify a target in a biological sample with a detectable marker, the affinity reagent of a detectable marker in a plurality of detectable markers is configured to specifically bind to a target in a plurality of targets. Therefore, compared with the remaining targets, the affinity of the affinity reagent to the one target is very high, so that the marker preferentially binds to the one target. This makes it possible to identify the target by detecting the detectable marker. By providing at least a specific and distinguishable marker equal in quantity to the number of targets, all targets can be identified. Adding a marker to a biological sample so that the marker is combined with its corresponding target is often referred to as biological sample staining. For example, the affinity of an antibody is generally considered to be the intensity of the binding action between an antigen and an antibody. However, an affinity reagent often has a measurable and considerable affinity for more than one target, and therefore may also be combined with more than one target. This is often referred to as the cross-reactivity of an affinity reagent. For example, in the case where more than one of these targets are present in a biological sample, an affinity reagent and its associated marker may bind to several targets, wherein the target to which the affinity reagent is configured to specifically bind is only the primary binding partner. Targets that are not primary binding partners (or targets) may also be referred to as off-target binding partners, and the corresponding binding may be referred to as off-target binding. For example, relative to the primary target binding, the binding to the off-target binding partner may be in the range of 1% to 20%. In some cases, the binding to the off-target binding partner may be in the range of 1% to 5%. In other cases, the binding to the off-target binding partner may be comparable to the binding to the primary binding partner. This off-target binding often reduces the ability to distinguish between targets, especially to distinguish between targets and off-target targets.

[0012] The target of a detectable marker in a plurality of detectable markers can be a specific portion or region of a molecule to which the marker binds, such as an epitope of a specific protein (such as an antigen). In particular, each target of a plurality of targets is different, at least in part, from the remaining targets of the plurality of targets. For example, affinity information can be provided as a matrix that includes the affinity or binding behavior of each detectable marker for each of the targets, or at least for some of the targets that are the primary targets of other affinity reagents for analyzing biological samples. For example, as a measure of affinity, the matrix can include a dissociation constant (K D The biological sample may be a tissue section, which, for example, includes a large number of proteins of interest as targets.

[0013] Thus, the raw readout data generated from the readout of a biological sample using a marker can be unmixed using an unmixing algorithm and affinity information. This enables discrimination between targets despite the presence of off-target target binding by the marker (particularly its corresponding affinity reagent).

[0014] Preferably, the target is located on at least one target molecule. For example, each target can correspond to a target molecule. In this case, each target molecule has a target, which is a part of the target molecule to which a marker (especially an affinity reagent) is bound. This makes it possible to analyze biological samples with a large number of targets. Alternatively, the target molecule can include several targets. In this case, several markers, especially affinity reagents, are bound to the corresponding parts of the target molecule. For example, a protein can include several epitopes as targets, and each epitope can be bound to a corresponding marker. This makes it possible to analyze biological samples with improved accuracy.

[0015] Preferably, the detectable marker for staining the target is selected from a detectable marker library based on affinity information, which specifies the affinity of each detectable marker in the detectable marker library for at least the target of the biological sample. This enables the selection of a marker suitable for staining the target. In particular, this selection is performed before the step of staining the target of the biological sample. The selection may also include consideration of the type of biological sample to be analyzed, in particular the protein in the biological sample. The affinity information can be determined using a language model, in particular the corresponding affinity reagent or its (amino acid) sequence having the desired cross-reactivity characteristics can be determined. An example of a (large) language model is AlphaFold.

[0016] It is particularly preferred that the detectable markers for staining the target are selected, in particular from a library of detectable markers, such that the affinity of each detectable marker to more than one target is below a predetermined threshold. A preferred predetermined threshold for cross-reactivity may have a value of, for example, 0.2 or 0.1, 0.05 or less. This enables the selection of markers with reduced cross-reactivity. In particular, a marker with (high) affinity to only one target in a biological sample or a marker with (high) affinity to as few targets as possible in a library of detectable markers may be selected.

[0017] The method can also utilize affinity reagents that are known to be strongly bound to a group of targets (i.e., targets and off-target targets). For example, in the determination involving other affinity reagents, affinity reagents that are strongly bound to its main binding partner and two off-target binding partners can be used. The contribution of a certain affinity reagent with a specific cross-reactivity spectrum to the determination using the method depends on the cross-reactivity spectrum of other affinity reagents in the determination. In a particularly preferred embodiment, a user-defined target group is analyzed using a minimum affinity reagent group in the determination, which is both necessary and fully generates a well-posed linear equation system. For simplicity, the noise derived from nonspecific background binding, dye labeling and readout is ignored because the noise can be experimentally processed well enough by various means (such as blocking, washing, using bright labels, longer acquisition time). This means that the method makes affinity reagents that are bound to multiple targets work. For example, in a standard immunofluorescence experiment involving only one marker including an antibody, an antibody that is strongly bound to 5 targets and has considerable affinity may not be very useful because the signal cannot be distinguished into 5 target-specific channels. Similarly, even if more markers are used, if cross-reactivity information is not used to unmix using the strategy proposed in this article, then the antibody may not be very useful, because the result produced in this way will be a convolution of the signal. However, if the method described in this article is used, the antibody may be very useful as part of the marker group for analyzing the target group. In other words, more important than the single cross-reactivity of the affinity reagent is the nature and structure of the cross-reactivity matrix of the affinity information, which describes the cross-reactivity of all affinity reagents used in the determination using the method. In this sense, modulation is a more important aspect, or in other words, a sharp cross-reactivity spectrum is preferred over a spectrum with a lower degree of modulation. In other words, an affinity reagent that is comparable in affinity but binds to a large number of targets, whether its affinity is high, medium or low, may not be very useful.

[0018] In a further particularly preferred embodiment, the detectable markers for staining the target are selected, in particular selected from a library of detectable markers, such that at least some of the detectable markers have affinity for more than one target. Specifically, at least some of the detectable markers have affinity for different targets in more than one target. Thus, at least some of the detectable markers may have affinity for a group of targets, respectively. Since cross-reactivity is taken into account in the form of affinity information in the unmixing step, this intentional cross-reactivity of the markers enables the analysis of the biological sample with increased accuracy.

[0019] Preferably, the raw readout data is generated during the step of generating the readout by spectral unmixing relative to the detectable label of the marker. For example, where the label is a fluorophore, the fluorescence emission spectra of the fluorophores may overlap. When generating the readout of the biological sample with the marker, the spectral data can be unmixed using knowledge of the specific fluorescence emission spectrum. This enables analysis of the biological sample with improved accuracy.

[0020] Preferably, prior to the step of providing affinity information, affinity information specifying the affinity of each marker for at least some of the targets is generated by experimentation or modeling. This enables accurate affinity information to be generated. In particular, affinity information can be generated for a library of detectable markers suitable for a particular biological sample based on the expected target or target molecules in the particular biological sample. Modeling can be performed on a computer and can be performed remotely by a person or system performing the method. For example, modeling can be performed on a remote server and the generated affinity information can be provided to a person or system performing the method.

[0021] It is particularly preferred that the affinity information specifying the affinity of each marker to at least some targets is generated by means of a language model. This enables accurate affinity information to be generated. In particular, the language model can be a large (protein) language model constructed using, for example, a machine learning method (such as an artificial neural network). The training data for the language model can include paired or grouped targets or epitopes and affinity reagents or paratopes that have a specific degree of affinity to each other. The training data can be generated experimentally by modeling. A general example of such a method is the AlphaFold model. Such language models can be further used to generate markers with desired cross-reactivity characteristics, especially corresponding affinity reagents or their (amino acid) sequences.

[0022] Preferably, during the step of unmixing the raw read data, information about the biological sample is additionally applied to generate unmixed read data about the target. This enables the biological sample to be analyzed with improved accuracy. For example, if a target is known to be absent in a particular biological sample, any cross-reactivity with the target can be ignored. The information about the biological sample can be genomic data, transcriptomic data, proteomic data of the biological sample. This information about the biological sample can be part of the affinity information and is applied together with the unmixing algorithm.

[0023] Preferably, during the step of unmixing the raw read data, the unmixing algorithm also provides a measure of the deviation relative to the target. This enables a measure of the confidence associated with the unmixed read data to be provided relative to the target. For example, the measure of the deviation can be a residual, a confidence interval, or an error value. The measure of the deviation can be further used to generate a reliability score for a specific experiment. In the case where the measure of the deviation is very high, different marker groups that are more suitable in terms of cross-reactivity and affinity information can be suggested to the user.

[0024] Preferably, the steps of dyeing the biological sample and generating the readout of the biological sample are repeated iteratively, and wherein, during each iteration, the biological sample is dyed with different detectable marker groups. This makes it possible to analyze the biological sample with improved accuracy. For example, the difference between the different marker groups and the markers of the previous iteration can only be the association of the label with the affinity reagent. Then, a specific target can also be identified by the order of the labels of the target in several iterations. In particular, in this way, the number of labels and affinity reagents can be less than the number of targets. Before dyeing with the next marker group, each iteration can also include removing, dissolving and / or disabling the markers of the previous iteration. The markers in the different marker groups can be selected, for example, from a detectable marker library based on the affinity information of the markers.

[0025] Preferably, the raw readout data is generated during the step of generating a readout based on all readouts of the iteratively generated readouts of the biological sample, and the unmixing of markers with respect to different detectable marker groups is performed based on the raw readout data, the raw readout data comprising all readouts in the iteratively generated readouts. This enables analysis of the biological sample with improved accuracy.

[0026] Preferably, the marker comprises a metal tag, and the readout of the biological sample generated with the marker is based on the mass of the metal tag. For example, the readout can be generated with the aid of a mass spectrometer. This enables detection of a variety of different targets.

[0027] Preferably, the marker comprises an oligonucleotide tag, in particular a barcoded oligonucleotide tag, and the readout of the biological sample generated using the marker is based on the sequence of the oligonucleotide. For example, the readout can be generated by sequencing the oligonucleotide tag. This enables detection of a variety of different targets.

[0028] Preferably, the marker comprises an optically detectable label, such as a fluorophore, and the readout of the biological sample generated with the marker is based on the light emitted by the optically detectable label. For example, the readout can be generated with the aid of a cytometer, a plate reader, or a microscope. In particular, the readout can be at least one of fluorescence intensity, fluorescence lifetime, or photon count. This enables detection of a variety of different targets.

[0029] On the other hand, a detectable marker kit suitable for performing a method for analyzing a biological sample is provided, wherein the detectable marker kit comprises a plurality of detectable markers, and in particular, the detectable markers can be selected from a detectable marker library based on their affinity information.

[0030] On the other hand, a system for analyzing a biological sample comprising multiple targets is provided. The system comprises a device configured to perform a method according to any one of the preceding claims. In particular, the system may include at least one controller configured at least to perform a step of unmixing the raw readout data. The controller may include an interface for inputting (additional) affinity information by a user and / or for receiving (additional) affinity information from a database or a cloud service. The controller may also be configured to instruct the stainer to perform a step of target staining of the biological sample, and to instruct the readout device to generate a (preferably iterative) readout of the biological sample with a marker.

[0031] The system has the same advantages as the method described above. In particular, the system can be combined with the features of the dependent claims relating to the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The specific embodiments are described below with reference to the accompanying drawings, in which:

[0033] Figure 1 is a flow chart of a method for analyzing a biological sample comprising a plurality of targets,

[0034] Figure 2 The unmixing algorithm is schematically shown as a system of linear equations for each pixel,

[0035] Figure 3 The affinity information of three affinity reagents is schematically shown.

[0036] Figure 4 is a schematic summary of the method,

[0037] Figure 5 The method is schematically shown in combination with dye unmixing, and

[0038] Figure 6Schematic diagram of how the cell type or cell state specific affinity information matrix is ​​derived by weighting the cell type or cell state specific affinity information matrix with the cell type or cell state specific weight. DETAILED DESCRIPTION

[0039] Figure 1 The invention relates to a flow chart of a method for analyzing a biological sample comprising a plurality of targets. The biological sample may be, for example, a tissue section. The target of the biological sample is a portion or region of the biological sample or a component thereof to which a marker can bind. For example, a target may be an epitope of a protein of the biological sample to which a marker, in particular an antibody to the marker, can bind. Thus, a target may be a portion of at least one target molecule of the biological sample. A target molecule may include a single or several targets.

[0040] The method starts from step S100. In step S102, the target of the biological sample is stained with multiple detectable markers. Target staining involves introducing markers into the biological sample so that the markers can bind to the corresponding targets. Each marker includes an affinity reagent and a detectable label. The affinity reagent is configured to bind to the target and the label is detectable so that the entire marker can be detected. Several copies of each marker can be introduced, which bind to multiple targets, especially copies of the target, present in the biological sample.

[0041] Examples of affinity agents are antibodies or aptamers. For example, the label can be a fluorophore. Although affinity agents such as antibodies are usually selected to specifically and primarily bind to one target, these affinity agents often exhibit affinity for several additional off-target targets and can therefore often bind to these several additional off-target targets with reduced affinity. Therefore, when staining a target of a biological sample with a detectable marker, each detectable marker can bind to the target specific to the corresponding marker, but these markers can bind to other targets, especially off-target targets.

[0042] In step S104, affinity information for each marker used to stain the biological sample is provided. The affinity information includes a measure of the affinity of each marker for at least some of the targets. Such an affinity measure may be, for example, a dissociation constant (K) for a pair of a particular marker and a target. D ). Affinity information about each marker can preferably be provided for all targets, and in particular, all targets can include off-target targets of each marker.

[0043] This affinity information can be generated by large protein language models.

[0044] The affinity information may include, for example, experimental data from biochemical assays such as immunoprecipitation coupled to mass spectrometry or phage display. In addition, the affinity information may include prediction data generated by analyzing sequence homology or structural similarity, molecular docking or machine learning, deep learning, and AI-based protein modeling and docking, which may be further derived from literature mining using natural language processing. Similarly, protein structure predictions using methods such as AlphaFold and molecular docking (e.g., LZerD, ZDOCK, HADDOCK, ClusPro, RosettaDock, HEX, SwarmDock, ATTRACT, SymmDock, PSI-BLAST, or FASTA) can be used to generate affinity information in computers.

[0045] For example, this affinity information can be derived from a large protein language model trained with epitope and paratope information, and these epitopes and paratope information can be derived or predicted by experiment, and then used to predict epitope: paratope interaction, to generate affinity information and / or generate a sequence that can be used for affinity reagent to generate from scratch. In this way, affinity reagent with required affinity characteristics can be generated from scratch, and then affinity reagent can be verified by a method in the above-mentioned method (for example, protein microarray, phage display). This is particularly useful for predicting the aptamer sequence used as affinity reagent. Further, this aspect of the method can be used for predicting the aptamer modification that allows the formation of the paratope with required affinity characteristics. The above-mentioned prediction method can be used for generating training data that can be used for training corresponding models.

[0046] In step S106, a readout of a biological sample stained with a marker is generated to generate raw readout data. The specific type of the readout depends on the type of label included in the marker. For example, the label of the marker can be optically detectable and include an optically detectable label, such as a fluorescent dye. In this case, an optical readout can be generated with the aid of a microscope or a flow cytometer. In particular, the fluorescent dye can be excited with excitation light, and the fluorescent emission light emitted by the fluorescent dye is captured and detected by a microscope with a suitable detector. All markers that are specific or primarily specific to a particular target in the target typically include labels with the same optical properties, for example, they have the same excitation wavelength and emission wavelength. Markers that are specific or primarily specific to different targets typically include labels with different optical properties so that the markers can be optically distinguished, thereby distinguishing the targets. For example, the generated optical readout can include the fluorescence brightness or intensity value and / or fluorescence lifetime based on pixel by pixel of the detected marker, especially its label.

[0047] Alternatively, the tags may comprise metals, and the readout may be generated by a mass spectrometer. In this case, the readout may include the mass of the tag. In a further alternative, the tags may comprise oligonucleotides, and the readout may be generated by sequencing the oligonucleotides. The oligonucleotide sequence of a particular marker may be specific to the target to which the marker specifically or predominantly binds. Thus, markers may be distinguished by their sequence.

[0048] Therefore, the raw readout data may include a relative or absolute value of the abundance or presence of the marker in the biological sample. For example, the fluorescence intensity value of the tag may be used as a measure of the relative or absolute value of the abundance or presence of the marker in the biological sample.

[0049] Additionally, the raw readout data generated in step S106 may optionally be generated by unmixing of the detectable labels of the markers. For example, in the case of fluorescent labels, the raw readout data may be generated by spectral unmixing of the optically detectable labels of the markers.

[0050] In step S108, the raw read data is unmixed with respect to the target by applying the unmixing algorithm and the affinity information to generate unmixed read data with respect to the target. Thus, the tendency of a specific marker to bind only to the corresponding target or to bind mainly to the corresponding target and to bind to other targets or off-target targets is taken into account.

[0051] For example, the raw readout data associated with a specific pixel or region read out can be represented as a signal S of a marker (m) that stains a biological sample. Each marker includes an affinity reagent that has a known affinity or cross-reactivity to a target in the biological sample. For example, each target is part of a different protein of the biological sample. The affinity information of the marker m, especially the associated affinity reagent, to the target (t) can be represented by a matrix A. The unmixed readout data can be represented by the abundance X of each specific target t in the pixel or region read out. In short, the signal S in the pixel or region read out is determined by the abundance X of the target t in the pixel or region read out and the affinity A of the marker m to the target t:

[0052]

[0053] By applying the unmixing algorithm, the abundance X of the target can be determined based on the signal S and the affinity information A:

[0054]

[0055] The method ends at step S110.

[0056] As an alternative to the above method, the detectable marker used for staining in step S102 can be selected from the detectable marker library before step S102. The selection can include considering the affinity information of the markers of the detectable marker library and the target of the biological sample. For example, the marker used for staining can be selected so that it has affinity only for one target, and has no affinity for the remaining targets in the multiple targets.

[0057] Alternatively or additionally, the marker for staining can be selected based on the affinity information including experimental conditions, which can include the details of the staining scheme, such as the temperature of which buffer or staining process is being used. Preferably, based on the affinity information, other experimental conditions are suggested or predicted, which improve the results of the read data generated through unmixing. Alternatively, the marker can be selected so that its affinity to targets other than the main target or the target of missing the target is lower than a predetermined threshold. Therefore, the cross-reactivity of the marker can be minimized.

[0058] Alternatively, the markers can be selected so that the smallest marker set can be used in the method that allows the generation of a well-posed linear equation system. Therefore, the maximization of the assay information content of each marker, the minimization of the cost of each marker assay, and the minimization of the acquisition time of each marker readout data are achieved.

[0059] In another alternative, steps S102 and S106 can be repeated iteratively. For example, the biological sample can be stained with different marker groups (S102), and after each staining step, the readout of the biological sample can be generated with the corresponding marker group (S106), and the raw readout data can be generated. In step S104, the affinity information of all markers in the marker group is provided. Then, step S108 uses the raw readout data of all iteratively generated readouts of the biological sample to generate the unmixed readout data about the target. For each staining step iteratively performed, the corresponding marker can be selected for specific staining.

[0060] Figure 2The unmixing algorithm is schematically shown as a system of linear equations for each pixel as an example of an unmixing algorithm. The system of linear equations can be completely determined or well-posed, in which case there is a unique solution that can be found using a Gaussian elimination algorithm. Alternatively, the system of linear equations may have no solution, or it can be under-determined or ill-posed and have infinite solutions. When a solution cannot be found, the method can be used to notify the user, which is a value in itself because it indicates that a meaningful measurement cannot be performed. Similarly, in the case where the system of linear equations is ill-posed, the method can be used to notify the user, who can then decide to use compressed sensing as a method and / or can decide or not to include prior information in the measurement. Prior information can include known data about target expression, biological sample cell type, cell state, pathway, protein-protein interaction. For example, the range of possible solutions may be limited by gene expression information about the cell type being studied, which can also be determined based on experimental information.

[0061] Figure 3 The affinity information of three affinity reagents, especially aptamers, with reference to their primary target is schematically shown, wherein the binding to the target is normalized to 1. In this example, for simplicity, the differences between the affinities of the above three aptamers are ignored, and these aptamers are considered to have substantially the same affinity for their corresponding targets. Alternatively, the actual affinity, i.e., K D The cross-reactivity profile contains information about the binding of the above aptamers to their off-target targets. Although aptamers are used in this example, similar effects are also found for other affinity reagents such as antibodies or nanobodies.

[0062] By cross-referencing different experiments that employed overlapping sets of affinity reagents, the method can learn which cross-reactivity profiles produce the least residuals retrospectively, and optimize the cross-reactivity profiles and target unmixing algorithms accordingly. Similarly, as more and more epitope-paratope surface structures become available, it is possible to (A) predict high-quality cross-reactivity information and (B) design specific paratopes that exhibit the desired cross-reactivity profiles. This is especially true for aptamers, which are often nucleic acid-based and can be easily synthesized. In summary, there are many methods to generate a priori information about the cross-reactivity of affinity reagents. Such affinity information can then be stored on a cloud server, which can also be used to perform target unmixing on user-uploaded data.

[0063] Further, it is known that the interaction between the target and the affinity reagent depends on all conditions that affect the epitope-paratope interface ("relevant conditions"), including but not limited to ionic strength (e.g., salt), detergents (e.g., Tween-20, Triton X-100), pH (affecting the charge of the epitope or paratope), reducing agents (e.g., dithiothreitol, β-mercaptoethanol), temperature, certain alcohols, chaotropic agents (e.g., urea, guanidine hydrochloride), and possible fixation conditions (e.g., paraformaldehyde fixation with different concentrations and durations versus other forms of fixation, such as glutaraldehyde, methanol fixation, cryofixation, or non-fixed native samples). For certain applications (e.g., immunofluorescence staining), affinity information can be generated that is specific to, for example, denatured epitopes or paraformaldehyde-fixed epitopes that may be partially masked. Affinity information can also be modified based on measured or predicted results. For example, it may be found that higher ionic strength reduces the affinity of some affinity reagents for some of the targets. This can be determined or predicted experimentally.

[0064] The system can assist in making this decision by providing an estimate of the quality of the resulting target unmixing, i.e., the expected residual and statistical confidence in the unmixing results, for a defined set of targets and a given set of affinity reagents and a given set of assays and associated conditions, or in order to define a suitable set of affinity reagents.

[0065] This affinity information can also include information about binding to different isoforms of the same protein. This is particularly important because up to 95% of human multi-exon genes are alternatively spliced. In fact, the increase in complexity of metazoans, while accompanied by a general increase in genome size, is not primarily due to an increase in gene number, but rather to an increase in regulatory complexity. Alternative splicing is one such mechanism, and it is estimated that approximately 15% of human genetic diseases and cancers are associated with alternative splicing. It is estimated that the number of human genes is about 20,000, while the number of transcript isoforms is about 150,000 (Jiang and Chen 2021; Alternative splicing: Human disease and quantitative analysis from high-throughput sequencing, Computational and Structural Biotechnology Journal, Vol. 19, 2021, pp. 183-195, ISSN 2001-0370, https: / / doi.org / 10.1016 / j.csbj.2020.12.009). Therefore, reliable identification of protein isoforms is of great significance. Similarly, post-translational modifications (PTMs), such as phosphorylation, ubiquitination, sumoylation, glycosylation and other PTMs, are key to signal transduction, protein complex formation and regulation of protein degradation. The Cell Atlas initiative has generated large datasets, including, for example, single-cell RNA-seq data, which can be used to derive cell-type / cell-state-specific consensus expression profiles that include protein isoforms.

[0066] Figure 4 is a schematic summary of the method: the user can optionally provide information about the experiment, such as research questions, targets of interest, sample types, sample sources, developmental stages, physiological conditions, electronic health record data, drug treatments, affinity reagents planned to be used in the assay, and assay types. The method can then optionally analyze this information and generate predictions that can adjust the best quality of target unmixing, the fastest time to obtain results, or, for example, the lowest cost of the assay. At this point, the method can already predict whether the assay has a high, medium, low, or current probability of generating target unmixing results that exceed user-defined quality indicators (e.g., maximum error, statistical confidence level). The user can then perform the experiment and analyze it, or historical data can be reanalyzed, preferably on a cloud server that hosts cross-reactivity information, experimental data, and target unmixing software. Target unmixing can then obtain decomposed results.

[0067] Figure 5 The method is schematically shown in combination with dye unmixing. In this example, three dyes are used, and their emission spectra 500a, 500b, 500c overlap with the tag antibodies AR220, AR3 and AR6, respectively, and each dye has its own target (thick dashed line) and off-target target (thinner dashed line). In the first step, dye unmixing is performed to obtain a "clean channel of the dye" and a "residual". Then target unmixing is performed to obtain a "clean value of the analyte", which includes the target value and the off-target target value. Cyclic staining, imaging and inactivation processes can be used to increase the complexity of the assay, that is, another affinity reagent group with the same tag group can be used in each round.

[0068] There are also other strategies for increasing complexity that are compatible with this method, such as using combination tags with, for example, multiple colors, or a mixture of fluorescent dyes and Raman tags. Such combination tags can be used in combination with a single round of readout, or in combination with a periodic staining, imaging, and inactivation process. In particular, such combination tags can be used in combination with the methods and devices described in the following patent applications: WO 2022 / 242887 A1, European patent applications with application numbers EP23178065, EP23155584, EP23183879 and EP23168397, the entire contents of which are incorporated herein by reference.

[0069] Figure 6 is a schematic diagram of how a cell type or cell state specific affinity information matrix can be derived by weighting it with cell type or cell state specific weights. Cell Atlas Project and user generated single cell RNA-seq data can be used, for example, to derive cell type specific gene expression signatures. In the simplest case, these weights can be "1" and "0" for genes expressed in a given cell type, or for genes known not to be expressed in a particular cell type. The same applies to splice isoforms, i.e. protein N isoform K may be expressed in cell type A, but isoform L of protein N may not be expressed. Using such profiles, a cell type specific affinity information matrix can be generated and used to decompose the signal using cell type specific affinity information. To do this, an image analysis step must be included in the workflow that includes at least one optical readout, which segments individual cells and assigns them specific cell types / states. This can greatly facilitate the solution of the linear system of equations.

[0070] In all figures, the same or similar acting elements are indicated by the same reference numerals. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items and may be abbreviated as " / ".

[0071] Although some aspects have been described in the context of an apparatus, it is apparent that these aspects also represent a description of a corresponding method, wherein a block or device corresponds to a method step or a feature of a method step. Similarly, aspects described in the context of a method step also represent a description of a corresponding block or item or feature of a corresponding device.

[0072] Reference numerals

[0073] Emission spectra of 500a, 500b, and 500c

Claims

1. A method for analyzing a biological sample comprising a plurality of targets, comprising the following steps: staining a target in a biological sample with multiple detectable markers, providing affinity information specifying the affinity of each marker for at least some of the targets, generating a readout of the biological sample using the marker to generate raw readout data, and The raw read data is unmixed by applying an unmixing algorithm and the affinity information to generate unmixed read data for the target.

2. The method of claim 1, wherein the target is located on at least one target molecule.

3. A method according to any of the preceding claims, wherein the detectable marker used to stain the target is selected from a detectable marker library based on affinity information, wherein the affinity information specifies the affinity of each of the detectable markers in the detectable marker library for the target in the biological sample.

4. A method according to any one of the preceding claims, wherein the detectable markers used to stain the targets are selected such that the affinity of each detectable marker for more than one of the targets is below a predetermined threshold.

5. A method according to any one of the preceding claims, wherein the detectable markers used to stain the targets are selected such that at least some of the detectable markers have affinity for more than one of the targets.

6. A method according to any one of the preceding claims, wherein the raw readout data is generated during the step of generating a readout by spectral unmixing with respect to the detectable label of the marker.

7. The method according to any one of the preceding claims, wherein prior to the step of providing the affinity information, the affinity information specifying the affinity of each marker for at least some of the targets is generated by experiments or modelling.

8. The method of claim 7, wherein the affinity information specifying the affinity of each marker for at least some of the targets is generated with the aid of a language model.

9. The method according to one of the preceding claims, wherein during the step of unmixing the raw readout data, information about the biological sample is additionally applied to generate the unmixed readout data with respect to the target.

10. A method according to any one of the preceding claims, wherein during the step of unmixing the raw readout data, the unmixing algorithm also provides a measure of the deviation with respect to the target.

11. The method according to any of the preceding claims, wherein the steps of staining the biological sample and generating the readout of the biological sample are repeated iteratively, and wherein, During each iteration, the biological sample is stained with a different set of detectable markers.

12. A method according to claim 11, wherein the raw readout data is generated during a step of generating readouts, which step of generating readouts is based on all readouts in an iterative generation of readouts of the biological sample, and the unmixing of the markers with respect to different detectable marker groups is performed based on the raw readout data.

13. The method according to any of the preceding claims, wherein the marker comprises a metal tag and the readout of the biological sample generated with the marker is based on the mass of the metal tag.

14. The method according to any of the preceding claims, wherein the marker comprises an oligonucleotide tag and the readout of the biological sample generated with the marker is based on the sequence of the oligonucleotide.

15. The method according to any of the preceding claims, wherein the marker comprises an optically detectable label and the readout of the biological sample generated with the marker is based on light emitted by the optically detectable label.

16. A detectable marker kit suitable for performing the method of one of the preceding claims, wherein the detectable marker kit comprises a plurality of detectable markers having the features of any one of claims 3 to 5.

17. A system for analyzing a biological sample comprising a plurality of targets, the system comprising means configured to perform the method of one of the preceding claims.

Images