Systems and methods for predicting high resolution cell-type-specific gene transcription and applications thereof

Abstract

Systems and methods for predicting cell-type-specific gene expression from genetic sequence are described. A computational framework can comprise a computational model for predicting epigenetic regulatory markers using genetic sequence as input. A computational framework can additionally, or alternatively, comprise a computational model for predicting cell-type-specific gene expression utilizing epigenetic regulatory markers spaced along sequence of genes. The computational model for predicting cell-type-specific gene expression can be trained with single-cell genetic expression profiles, such as scRNA-seq or snRNA-seq. The computational framework can be utilized to assess polymorphic sequences, such as polymorphisms related to medical disorders. Various methods further develop research tools, perform diagnostics, and treat individuals based on cell-type-specific gene expression levels, which can be further based on differential gene expression patterns among polymorphic sequences.

Description

SYSTEMS AND METHODS FOR PREDICTING HIGH RESOLUTION CELL-TYPE- SPECIFIC GENE TRANSCRIPTION AND APPLICATIONS THEREOFCROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Application Ser. No. 63 / 581 ,560 entitled “Systems and Methods for Assessing Single-Cell Gene Expression and Applications Thereof,” filed September 8, 2023, the disclosure of which is hereby incorporated by reference in its entirety for all purposes.STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

[0002] This invention was made with Government support under Grants Nos. HG005998, HL117798, and GM071966. awarded by the National Institutes of Health. The Government has certain rights in the invention.TECHNOLOGICAL FIELD

[0003] The invention is generally directed to systems and methods to predict cell-type- specific gene expression in with high resolution using single-cell expression data.BACKGROUND

[0004] Eukaryotic cells utilize complexes of DNA, RNA and protein, referred to as chromatin, to regulate gene expression and thus control its cellular function. When a DNA strand is in a packed, condensed chromatin state, genes on the DNA strand are repressed from expressing. On the contrary, when a DNA strand is in an unpacked, open chromatin state, genes on the DNA are expressed to produce RNA and subsequent protein products. A cell will modulate its chromatin status along DNA strands to control gene expression.

[0005] Within the sequence of a cell’s DNA are elements that recruit various factors to modulate chromatin and gene expression. Various gene expression elements include transcription factor binding sites, operators, enhancers, silencers, promoters, transcriptional start sites, and insulators. Single nucleotide variations, nucleotideinsertions, and nucleotide deletions in genetic sequences that affect chromatin formation can produce various effects on gene expression, resulting in heterogeneity in gene expression between different cell types.SUMMARY OF THE INVENTION

[0006] Several embodiments of the disclosure are directed towards systems and methods for cell-type-specific gene expression with high resolution. A computational framework can comprise one or more machine-learned computational models. A first computational model can be trained to predict epigenetic regulatory markers within proximity to a gene transcription start site, using the sequences proximal to the gene transcription start site as input. A second computational model can be trained to predict cell-type-specific gene expression using epigenetic regulatory markers, which in some implementations is predicted from the first computational model. In some implementations, the computational framework comprises a plurality of computational models trained to predict cell-type-specific gene expression, each model tailored to a specific cell type . Utilizing the framework, sequences of a gene is utilized to predict whether the gene is expressed and / or the relative expression level within a cell type. Gene sequences comprising one or more polymorphisms can be utilized as input to assess the effect of the polymorphisms on gene transcription within specific cell types. Polymorphisms include any multi-allelic possibilities of a genetic sequence, including (but not limited to) nucleotide variants, insertions, deletions, inversions, translocations, and repeat elements (e.g., microsatellites, retrotransposons, Alus, and LINEs). In some implementations, multiple cell-type-specific computational models are utilized to deconvolute which cell types are affected by the polymorphisms. Further actions can be performed, such as genetically modifying a cell of the specific cell type by inducing sequence mutagenesis at the genetic locus where the one or more polymorphisms reside in order to alter gene transcription in that cell. The effect of altered cell-type gene expression due to one or more polymorphisms on various phenotypes can be assessed on many biological platforms, such as cells, biological systems, tissues, organs, and / or organisms.

[0007] In some aspects, the techniques described herein relate to a computer- implemented method to evaluate gene expression in a specific cell type, including: obtaining, using a computational processing system, a first genetic sequence, wherein the first genetic sequence includes one or more genes, wherein the sequence of each gene includes: a sequence structure, and adjacent sequence upstream and downstream the sequence structure; predicting, using the computational processing system and a computational framework, a cell-type-specific expression characteristic of each gene for one or more cell types to yield a predicted cell-type-specific expression characteristic of each gene for each cell type, wherein the computational framework includes a first computational model to predict epigenetic regulatory markers spatially along the sequence of each gene, wherein the first computational model utilizes gene sequence as input, wherein the computational framework includes one or more cell-type-specific computational models to predict cell-type-specific expression characteristics of each gene, wherein each cell-type-specific computational model utilizes the epigenetic regulatory markers along the gene sequence as input; and based on the predicted celltype-specific expression characteristic of the one or more genes, performing a biochemical assay on a first cell to empirically assess an effect of the cell's genetic sequence on the cell-type-specific expression characteristic of at least one gene of the one or more genes, wherein the cell's genetic sequence of the at least one gene is the same as the first genetic sequence and the cell type of the first cell is the same cell type as one of the cell-type-specific computational models.

[0008] In some aspects, the techniques described herein relate to a computer- implemented diagnostic method for determining a treatment option for an individual for a medical disorder that arises due to gene expression levels in a particular cell type, including: obtaining, using a computational processing system, a first genetic sequence, wherein the first genetic sequence includes one or more genes, wherein the sequence of each gene includes: a sequence structure, and adjacent sequence upstream and downstream the sequence structure; predicting, using the computational processing system and a computational framework, a cell-type-specific expression characteristic of each gene for one or more cell types to yield a predicted cell-type-specific expression characteristic of each gene for each cell type, wherein the computational frameworkincludes a first computational model to predict epigenetic regulatory markers spatially along the sequence of each gene, wherein the first computational model utilizes gene sequence as input, wherein the computational framework includes one or more cell-type- specific computational models to predict cell-type-specific expression characteristics of each gene, wherein each cell-type-specific computational model utilizes the epigenetic regulatory markers along the gene sequence as input; and based on the determination cell-type-specific expression level of the gene, detecting that a cell type has misregulated expression of the gene; wherein the misregulated expression provides an indication of at least one of: diagnosis of the medical disorder, early detection of the medical disorder before identification of symptoms, personalized medicine diagnosis, a treatment of the disorder to ameliorate a symptom associated with a misregulated pathway in the one of more cell types.

[0009] In some aspects, the techniques described herein relate to a computer- implemented method to perform site-directed mutagenesis on a particular cell type, the method including: obtaining, using a computational processing system, a first genetic sequence, wherein the first genetic sequence includes one or more genes, wherein the sequence of each gene includes: a sequence structure, and adjacent sequence upstream and downstream the sequence structure; predicting, using the computational processing system and a computational framework, a cell-type-specific expression characteristic of each gene for one or more cell types to yield a predicted cell-type-specific expression characteristic of each gene for each cell type, wherein the computational framework includes a first computational model to predict epigenetic regulatory markers spatially along the sequence of each gene, wherein the first computational model utilizes gene sequence as input, wherein the computational framework includes one or more cell-type- specific computational models to predict cell-type-specific expression characteristics of each gene, wherein each cell-type-specific computational model utilizes the epigenetic regulatory markers along the gene sequence as input; determining using the computational processing system and the computational framework, a set of one or more genetic polymorphisms that alter expression level of the gene in the particular cell type, wherein the set of polymorphisms that alter the expression level of the gene are not present in the first genetic sequence; and based on the determination of the set ofpolymorphisms that alter the expression level of the one or more genes, performing site- directed mutagenesis on DNA of a cell of the particular cell type to introduce the set of polymorphisms, wherein the cell's DNA includes genetic the same set of polymorphisms as the first genetic sequence and the cell type of the first cell is the same cell type as one of the cell-type-specific computational models.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The description and claims will be more fully understood with reference to the following figures and data graphs, which are presented as exemplary embodiments of the invention and should not be construed as a complete recitation of the scope of the invention.

[0011] Fig. 1 provides a schematic of an example of an overview process to predict cell-type expression of genes.

[0012] Fig. 2A provides a schematic of an example of a computer-implemented method to train a computational model to predict epigenetic regulatory markers along a genetic sequence.

[0013] Fig. 2B provides a schematic of an example of a computer-implemented method to train a computational model to predict cell-type gene expression.

[0014] Fig. 3A provides a schematic of an example of a computer-implemented method to predict epigenetic regulatory markers of one or more genetic sequences.

[0015] Fig. 3B provides a schematic of an example of a computer-implemented method to predict cell-type expression using cell-type-specific computational models.

[0016] Fig. 4 provides a schematic of an example of a computer system to perform computer-implemented methods.

[0017] Fig. 5 provides a schematic of the architecture of the ExPectoSC framework and application to defining cell-type-specific effects of genetic variation in human health and disease. The ExPectoSC framework is a specific example of a computer- implemented method that was constructed to provide proof of concept. ExPectoSC implements a modular approach to predict cell-type-specific expression directly from sequence using Module 1 : a convolutional neural network (CNN) module to learn biochemical regulatory features, followed by Module 2: regularized linear models withclustered single cell RNA-seq data to provide information about cell-type-specific expression during training. The outputs can be applied to interrogate the effect of any genomic variant list at cell type resolution.

[0018] Figs. 6A-6C provide schematics and data results that show ExPectoSC accurately models single cell gene expression for over 100 cell types in seven organs. Fig. 6A: Schematic overview of the number of cell types in organ systems that were utilized in this example. Fig 6B: Distribution of AUC ROC scores for predicting high and low expressed genes (top and bottom 25th percentile) in each cell type, grouped by organ. Fig. 6C: Comparing AUC ROC for the cell type specific and whole organ models for the cell type specific genes. Cell type specific models achieve considerably better results, with 10 models more than doubling the AUC score and 14 models moving up from under 0.5 AUC.

[0019] Fig. 7 provides a data table that summarizes the datasets utilized to train the cell-type specific computational models.

[0020] Fig. 8 provides AUC ROC curves for the variably expressed genes genes in the test set (25% cutoff) in all the cell types, stratified by the organ system.

[0021] Fig. 9 provides data results of the ExPectoSC framework. On the top left is a data graph of the distribution of AUC ROC scores for predicting high and low expressed genes (top and bottom 25th percentile) in each cell type, grouped by organ for the cross- validation dataset. ON the top right is a data graph on the gain in AUC coming from cell type specific models compared to whole organ models on the cross-validation dataset. On the bottom is a data graph of the gain in AUC coming from cell type specific models compared to whole organ models for all genes for the test set.

[0022] Figs. 10A-10C provide assessment data of the ability of ExPectoSC to partition risk for heritability of traits and diseases to specific and appropriate cell types. Fig. 10A: Enrichment of immune cell types across all organs for the SURE autoimmune traits GWAS, only cell-types with significant enrichment and sorted by the p-value produced by the sLDSC. Fig. 10B: Differential sLDSC enrichment of the cell types for BMI and waisthip ratio (WHR) adjusted for BMI, sorted by the relative p-value and only showing significant cell types. BMI is enriched for brain cell types but WHR (conditioned on BMI, which accounts for distribution of adipose tissues) shows highest enrichment of cell typesoutside of the brain, as would be expected. Fig. 10C: Brain cell types enrichments for BMI and adjusted WHR. Size of the dot shows relative enrichment, while the shading represents p-value. GABAergic neurons have the greatest enrichment.

[0023] Figs. 11A-11 B provide additional information for the autoimmune disease analysis. Fig. 11 A: Two subsets of autoimmune diseases GWAS. Data taken from Finucane et al 2015. Fig. 11 B: Significant cell types for the autoimmune traits GWAS. Results show enrichment for the immune cell types.

[0024] Figs. 12A-12B provide experimental data that assesses the ability of ExPectoSC to predict cell-type specific gene expression alteration from polymorphic sequences. ExPectoSC-predicted and experimentally confirmed cell type specific effects of ClinVar variants in the PTEN promoter Fig. 12A: The plot shows the predicted impact of ClinVar variants in the region, showing variants of high predicted ExPectoSC impact near each other. Module 1 epigenetic predictions for these variants pointed to high impact on YY1 binding. Experimentally tested variants are additionally labeled for clarity. Fig. 12B: The PTEN promoter shows robust transcriptional activity in BE(2)-C neuroblastoma cells using a luciferase assay. The top three highest predicted impact ClinVar variants in the region (89623405G>C, 89623406C>G, and 89623408A>G) and one of the low predicted effect variants at 89623412T>C were tested. The y axis shows the magnitude of transcriptional activation normalized to the activity for the empty vector and the reference allele. Significance levels were computed on the basis of a t test (two sided with unequal variance ****p < 0.0001 ). Central values of the boxplot represent the median of normalized magnitude, the box extends from the 25th to the 75th percentile, and whiskers extend to the maximum and minimum values.DETAILED DESCRIPTION

[0025] The various embodiments of the disclosure are directed towards systems and methods to predict cell-type gene expression from genetic sequences. Prediction of gene expression can be used in a number of applications, including (but not limited to) gene expression analysis, assessment of polymorphic sequences on phenotypes, clinical diagnostics, drug development, and clinical treatments. Numerous embodiments are directed towards the use of a computational framework and computer-implementedmethods for predicting expression of genes based on sequence. In several embodiments, computational models utilize gene sequence, inclusive of noncoding sequence and especially promoter-like sequences, as input to determine gene-specific epigenetic regulatory markers ( / .e., local chromatin architecture of a gene) that influence gene expression. Various sequences can be utilized as input. In one implementation, overlapping sliding sequence windows bidirectionally extending from a gene’s transcription start site (TSS) for a certain distance (e.g., + / - 20kb from TSS) are utilized as sequence input. And in many embodiments, computational models utilize epigenetic regulatory markers as input to predict cell-type expression of genes. In several embodiments, cell-type-specific computational models are trained utilizing single-cell genetic sequencing results, such as single-cell RNA sequencing (scRNA-seq) and singlenucleus RNA sequencing (scRNA-seq).

[0026] Throughout the disclosure and in accordance with many of the embodiments, cell types are sub-organ level, meaning the cell types are more particular than organ- related cells (e.g., more particular cardiac cells or neural cells). And in accordance with many of the embodiments, cell types are sub-tissue level, meaning the cell types are more particular than tissue-related cells (e.g., more particular epithelial cells or mesenchymal cells). And in accordance with many of the embodiments, cell types are primary cell types, meaning the cell types are of living tissue and organs (e.g., cells isolated from patient blood or donor tissue, and not immortalized cell lines grown in vitro). In many embodiments, single-cell genetic sequencing results can be utilized to cluster and identify single cells to yield specific and particular cell-type populations, having shared epigenetic and expression patterns. In some embodiments, clustered cell populations are sub-organ cell types, sub-tissue cell types, or primary cell types. And in several embodiments, clustered cell populations are sub-organ, sub-tissue, and primary cell types. And in some embodiments, a clustered cell population is a population of a specific cell type that further specifically responded to one or more stimuli (e.g., activated macrophages), which can be differentiated from a clustered cell population that did not or has not responded to the stimuli (e.g., mature but inactivate macrophages). In many embodiments, single-cell genetic sequencing results of clustered cell-type populations are utilized to train cell-type-specific computational models. Based on experimentationsperformed, cell-type-specific computational models can greatly improve the accuracy of these model to predict gene expression that use epigenetic regulatory markers as input.

[0027] In several embodiments, various applications can be performed based on prediction of cell-type gene expression of genetic sequences. It has been found that celltype-specific computational models correctly predict that genes will be differentially expressed in different cell types, as is determined by the genetic sequence and cell-type- specific patterns of epigenetic regulatory markers. Furthermore, cell-type-specific computational models can predict that polymorphic sequences of a gene will yield different expression characteristics. In many instances, the cell-type-specific computational models can identify and / or rank specific sequence polymorphisms on their effect on gene transcription in each cell type assessed. And often, the effect of specific sequence polymorphisms on gene expression is different among various cell types. In many embodiments, biochemical analysis is performed on one more specific cell types to assess and / or to confirm the computationally predicted gene expression characteristics. Accordingly, a biochemical experiment can be performed to empirically determine celltype-specific gene expression of reference sequences and / or the effect polymorphisms on gene expression within a cell type or between two or more cell types. These applications can be performed on many biological platforms, such as cells, biological systems, tissues, organs, and / or organisms. A biological system is any system comprising two or more cell types that interact and / or form structures having more complexity than a single cell-type culture. Examples of biological systems include (but are not limited to) a mixed cell culture system, a cell co-culture system, an embryoid body system, an organoid system, a tissue system, an organ system, and an organismal system.

[0028] Beyond mere assessment of gene expression, in some embodiments, genetic material is manipulated to incorporate one or more polymorphisms that have been computationally predicted to affect gene expression characteristics in one or more cell types. In some embodiments, polymorphic genetic sequences (or one or more specific polymorphisms) are incorporated into transgene plasmids (e.g., luciferase) to determine their effect on expression level. In some embodiments, CRISPR mutagenesis, Zinc-finger mutagenesis, and / or TALEN mutagenesis is performed on genetic material, such as agenome of a biological cell. Biological cells with mutagenized genetic material can be assessed via a biochemical experiment to empirically analyze the effects of one or more polymorphisms on endogenous gene expression within particular cell types. Accordingly, in some embodiments, genetic material of cell lines, embryonic stem cell lines, primary cells, induced pluripotent cells of primary cells, primary tissue, ex vivo tissue, embryonic tissue and / or in vivo tissue is mutagenized with one or more polymorphisms to create biological models to better understand the polymorphisms on cell-type-specific gene expression. When a biological system comprises two or more cell types, in some embodiments, a specific cell type is targeted for mutagenesis, which can be achieved utilizing cell-type targeting aptamers, cell-type specific promoters, viral vectors with celltype-specific tropism, or any other cell-type targeting technique.

[0029] Inspiration of polymorphisms to be assess can be from any source. In some embodiments, one or more polymorphisms are inspired by naturally occurring polymorphisms. In some embodiments, one or more polymorphisms are inspired by polymorphisms identified within genetic material that is associated with a particular phenotype, such as a medical phenotype. In some embodiments, one or more polymorphisms are inspired by in silico analysis, which can be a random assessment of polymorphisms or a methodical assessment of polymorphisms, and can be performed utilizing the computational framework described herein utilizing one or more cell-type- specific modules. The computational framework inclusive of cell-type-specific modules can be utilized to computationally predict the effect of any singular polymorphism and / or any combination of two or more concurrent polymorphisms on gene expression within specific cell types. Based computational prediction of expression characteristics, polymorphic sequences can be selected, and then generated by polymeric nucleotide synthesis and / or site-directed mutagenesis. In some embodiments, CRISPR mutagenesis, Zinc-finger mutagenesis, and / or TALEN mutagenesis is performed to specifically engineer a polymorphic sequence that has desired effects on gene expression within specific cell types. Accordingly, in some embodiments, genetic material of cell lines, embryonic stem cell lines, primary cells, induced pluripotent cells of primary cells, primary tissue, ex vivo tissue, embryonic tissue and / or in vivo tissue is mutagenized with one or more polymorphisms to yield altered cells, tissues, organs, and / or organisms having adesired effect one or more gene expression characteristics. When a system comprises two or more cell types, in some embodiments, a specific cell type is targeted for mutagenesis, which can be achieved utilizing cell-type targeting aptamers, cell-type specific promoters, viral vectors with cell-type-specific tropism, or any other cell-type targeting technique. In some embodiments, altered cells, tissues, or organs are transplanted into a patient or other subject, which may be useful as treatment (e.g., immune cell therapy), vaccination (e.g., cancer vaccination), organismal model generation (e.g., mouse models).Overview of Cell-Type Gene Expression as Determined by Sequence

[0030] An example of a process to predict cell-type gene expression based on its spatial chromatin profile as determined from its genetic sequence is provided in Fig. 1 . In some embodiments, the process is utilized to identify sets of polymorphisms that alter cell-type gene expression, as determined by their effect on local chromatin profile. Polymorphisms and their effects on gene expression can be used in various applications downstream, such as (for example) performing various biochemical assays, generating polymorphic cells, and developing clinical diagnostics and treatments.

[0031] Process 100 can begin with obtaining (101 ) one or more genetic sequences. Genetic sequences can be full or partial genome. Genetic sequences can be sequences related a singular gene, sequences related to one or more genes, sequences related to two or more genes, sequences related to five or more genes, sequences related to ten or more genes, sequences related to twenty or more genes, sequences related to fifty or more genes, sequences related to one hundred or more genes, sequences related to two hundred or more genes, sequences related to five hundred or more genes, sequences related to one thousand or more genes, sequences related to two thousand or more genes, sequences related to five thousand or more genes, sequences related to ten thousand or more genes, sequences related to fifteen thousand or more genes, sequences related to nineteen thousand or more genes, or about all genes of a genome. In several embodiments, each gene analyzed includes the transcription start site of the sequence. In several embodiments, each gene analyzed includes the transcription start site and immediately adjacent sequences (e.g., + / - 20kb).

[0032] In some embodiments, the one or more sequences are derived from biological source. In some embodiments, the biological sample has a particular phenotype, such as (for example) a medical phenotype (e.g., presence of cancer).

[0033] Genetic sequences can be derived from a number of sources. In some instances, genetic sequences are obtained de novo by extracting the DNA from a biological source and sequencing it. Alternatively, genetic sequence data can be obtained from publicly or privately available databases. Many databases exist that store datasets of sequences from which a user can extract the data to perform computational analysis upon.

[0034] Once genetic sequences are obtained, Process 100 can predict (103) a spatial chromatin profile along one or more genes using a first computational model. In many embodiments, a chromatin profile is a spatial pattern of epigenetic regulatory markers along the genetic sequence, which can be determined from its sequence ab initio. In some embodiments, the genetic sequence is entered into a trained computational model that is trained to predict a spatial pattern of regulatory elements based on sequence. The computational model can predict epigenetic regulatory marker patterns based on sequence, polymorphisms within the sequence, and context of proximal sequence. In some embodiments, the trained computational model predicts the spatial chromatin profile on sequences that include the promoter and / or the TSS of each gene analyzed.

[0035] In several embodiments, the first computational model is trained utilizing chromatin regulatory element profiles, which can be derived experimentally utilizing various epigenetic assays including (but not limited to) chromatin immunoprecipitation sequencing (ChlP-seq), DNAse I hypersensitivity sequencing (DNase-seq), Assay for Transposase-Accessible Chromatin sequencing (ATAC-seq), Formaldehyde- Assisted Isolation of Regulatory Elements (FAIRE-seq), Hi-C capture sequencing, bisulfite sequencing (BS-seq), and methyl array. In some embodiments, chromatin profiles are retrieved from database such as (for example) Encyclopedia of DNA Elements (ENCODE), NIH Roadmap Epigenomics Mapping Consortium, and the International Human Epigenome Consortium (IHEC). These data sets can inform genomic loci that are important in regulating transcription and / or posttranscriptional processing.

[0036] In several embodiments, the computational model for predicting a spatial chromatin profile is a neural network. In some embodiments, the computational model is a deep neural network. In some embodiments, the computational model is a convolutional neural network. In some embodiments, after predicting a chromatin profile for a sequence, the profile is spatially transformed to reduce the complexity of chromatin features along a gene’s sequence.

[0037] Based on the spatial chromatin profile, Process 100 predicts (105) cell-type gene expression characteristics of the one or more genes using a cell-type specific computational model. Cell-type gene expression characteristics that can predicted can be presence of gene expression, expression levels of a gene, other expression characteristics, or any combination thereof. In some embodiments, the expression level of at least one-fourth of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least one-third of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least one-half of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least three-fourths of the genes within the input genetic sequences is determined. In some embodiments, the expression level of each gene within the input genetic sequences is determined.

[0038] In some embodiments, the second computational model is trained using singlecell expression data sets with chromatin profiles. In some embodiments, the chromatin profiles are predicted by the first computational model.

[0039] In some embodiments, the second computational model is a regressor. In some embodiments, the second computational model is a classifier. The second computational model can be any appropriate regressor or classifier, such as (for example) a linear regressor, logistic regressor, Lasso regressor, support vector regressor, nonlinear regressor, polynomial regressor, naive Bayes, K-nearest neighbors, support vector machine, etc. In some embodiments, the second computational model is a linear regressor. In some embodiments, the second computational model incorporates ridge regression.

[0040] In several embodiments, processes to determine expression levels of genes in a biological sample is utilized in various downstream applications, including (but notlimited to) performing various biochemical assays, generating polymorphic cells, and developing clinical diagnostics and treatments.Computational Model Training

[0041] An example of a computer-implemented method to train a computational model to predict epigenetic regulatory markers from genetic sequence is provided in Fig. 2A. Computer-Implemented Method 200 can begin by obtaining (201 ) epigenetic regulatory marker profiles and epigenetic regulatory marker library. An epigenetic regulatory markers profile is a collection of data indicating where various epigenetic markers that can affect transcription interact with DNA along a genetic sequence. In many embodiments, a library of epigenetic regulatory markers will include robust set of markers, and ideally all markers that affect gene transcription, including cell-type and / or tissue specific markers. Examples of epigenetic regulatory markers include (but are not limited to) sites of chromatin accessibility (e.g., DNase I hypersensitivity), chromatin marks (e.g., histone code), transcription factor binding sites, and other epigenetic factors. A library of 2,002 epigenetic regulatory markers was utilized in computational processes described within the Examples and Data section below.

[0042] Generally, epigenetic regulatory markers profiles can be determined utilizing various epigenetic assays including (but not limited to) chromatin immunoprecipitation sequencing (ChlP-seq), DNAse I hypersensitivity sequencing (DNase-seq), Assay for Transposase-Accessible Chromatin sequencing (ATAC-seq), Formaldehyde- Assisted Isolation of Regulatory Elements (FAIRE-seq), Hi-C capture sequencing, bisulfite sequencing (BS-seq), and methyl array. Several databases store chromatin and RBP / RNA-element profiles which can be used, including (but not limited to) Encyclopedia of DNA Elements (ENCODE), NIH Roadmap Epigenomics Mapping Consortium, and the International Human Epigenome Consortium (IHEC).

[0043] Utilizing epigenetic regulatory markers profiles and concordant sequences as input training data, Computer-Implemented Method 200 trains (203) a computational model to predict epigenetic regulatory markers on a genetic sequence. In several embodiments, the computational model is a neural network. In some embodiments, the computational model is a deep neural network. In some embodiments, the computationalmodel is a convolutional neural network. Generally, the computational model can be trained using any sequence with epigenetic regulatory markers profiles, yield a generalized model that can generate a comprehensive epigenetic regulatory marker profile inclusive of both generalized epigenetic markers and specialized (e.g., tissue specific) epigenetic markers.

[0044] In several embodiments, epigenetic regulatory markers are determined on large sets of genes, including all Pol II transcribed genes of a whole organismal genome. In some embodiments, for each gene that is analyzed, epigenetic regulatory markers are determined in relation to a sequence structure of the gene, such as (for example) the TSS, known promoter region, or similar. In some embodiments, epigenetic features are determined for some distance upstream and / or downstream from a gene sequence structure, such as (for example) 20 kilobases (kb) upstream and downstream of the genes TSS. It should be understood that the length of sequence for which epigenetic features are determined in relation to a gene sequence structure can vary and does not have to be equal upstream and downstream. In some embodiments, epigenetic features are determined 1 kb, 2 kb, 3 kb, 4, kb, 5 kb, 7.5 kb, 10 kb, 15 kb, 20 kb, 25 kb, or 50 kb upstream of a gene sequence structure. Similarly, in some embodiments, epigenetic features are determined 1 kb, 2 kb, 3 kb, 4, kb, 5 kb, 7.5 kb, 10 kb, 15 kb, 20 kb, 25 kb, or 50 kb downstream of a gene sequence structure. It should be further noted that epigenetic features can be determined on any sequence that would affect gene expression and doesn’t have to be in relation to the TSS.

[0045] In some embodiments, prediction of epigenetic features along a sequence is performed via sliding windows of sequence. For example, if epigenetic features are to be determined for 20kb upstream and downstream the TSS of a gene, the determination of epigenetic features can be performed in increments such as (for example) 200 bp windows. In some embodiments, sequence windows of 50 bp, 100 bp, 150 bp, 200 bp, 250 bp, 300 bp, 400 bp, or 500bp are utilized to determine epigenetic features along a sequence. In some embodiments, sliding windows are overlapped, which may help gain better resolution.

[0046] In some embodiments, determination of epigenetic features within a window also considers surrounding sequence context, which may better provide context of thelocal epigenetic architecture of the binned sequence. For example, when determining epigenetic features of a 200 bp window, the surrounding 1800 bp (900 upstream and 900 downstream) can be considered, resulting in 2000bp of sequence considered for each 200 bp window ( / .e., 10x more sequence considered per window). In some embodiments, 1x, 2x, 3x, 5x, 7.5x, 10x, 15x, 20x, or 30x of surrounding sequence is considered to determine epigenetic factors for each window.

[0047] Computer-Implemented Method 200 also assesses (205) the ability of the computational model to predict epigenetic regulatory markers on gene sequence using the computational model. Input data can be split into a training portion, a validation portion, and a testing portion. The training portion trains the model. The validation portion validates the training. And the testing portion is utilized to assess the predictive ability of the model. Alternative, or in addition, a leave-one-out approach can be utilized to train and cross validate the model.

[0048] Computer-Implemented Method 200 optionally spatially transforms (207) predicted epigenetic regulatory markers, which may reduce the dimensionality of the features due to the use of sliding overlapping windows along the sequence. In some embodiments, a spatial transformation module is utilized. In some embodiments, a spatial transformation model is an exponential decay function. In some embodiments, spatial transformation put weights on windows utilized to determine epigenetic features, and the weights can be determined as appropriate. In some embodiments, windows closer to a selected gene structure (e.g., TSS) are weighted greater than windows further away to the selected gene structure. It should be understood that windows upstream from a selected gene structure may be weighted differently than windows downstream, as appropriate to the task at hand. For example, 2,000 epigenetic features can be determined for a 40kb sequence (e.g., 20 kb upstream and 20 kb downstream a TSS) of a nearby gene in 200 bp windows (i.e., 200 total bins), resulting in 400,000 epigenetic features for the nearby single gene. Reducing the epigenetic features 20-fold to 20,000 features may be beneficial, as determined by application.

[0049] Computer-Implemented Method 200 can also store and / or report (209) the trained computational model. While a specific example of a computer-implemented method for training a computational model to predict epigenetic regulatory markers fromgenetic sequence is described above, one of ordinary skill in the art can appreciate that various steps of the method can be performed in different orders and that certain steps may be optional according to some embodiments. As such, it should be clear that the various steps of the method could be used in accordance with the requirements of specific applications.

[0050] An example of a computer-implemented method to train a computational model to predict cell-type gene expression from epigenetic regulatory markers is provided in Fig. 2B. The method utilizes the epigenetic regulatory markers along a sequence of a gene and single-cell gene expression profiles within a specific cell type to train the model. The method can be utilized to train the model to predict cell-type gene expression of up to every Pol II transcribed gene in an organismal genome.

[0051] Computer-Implemented Method 250 can begin by obtaining (251 ) single-cell expression data with concordant spatial epigenetic regulatory markers along a sequence of one or more genes. The obtained expression data can be for any number of genes; the greater the number of genes, the more comprehensive training of the model will be. Accordingly, although expression data obtained can minimally relate to one or more genes, it is ideal to obtain expression data of many more sequences. In some implementations, single-cell expression of one thousand or more genes is obtained, single-cell expression of two thousand or more genes is obtained, single-cell expression of five thousand or more genes is obtained, single-cell expression of ten thousand or more genes is obtained, single-cell expression of fifteen thousand or more genes is obtained, single-cell expression of nineteen thousand or more genes is obtained, or single-cell expression of about all Pol II transcribed genes in an organismal genome is obtained.

[0052] An issue with prior models that were trained using primary tissue for predicting gene expression from epigenetic regulatory markers is that the predictive ability of these models was not as good as desired. It was hypothesized that the heterogeneity of cell types within the primary tissues that were used to generate gene expression profiles may contribute to the underachieving predictive ability. Because of the heterogeneity of primary tissues, the gene expression profiles are variable and many genes have fairly high standard deviations. Further, because the prevalence of some cell types is high in atissue and others are low, it can be difficult to detect expression profiles of minority cell types within the tissue. It was hypothesized that utilizing sequence results that were derived from sources with less cell-type heterogeneity would improve the ability to resolve detection of minority cell-type gene expression and could potentially also improve predictive power of a trained model. Accordingly, it is a goal of the current disclosure to reduce gene expression variability by reducing the heterogeneity of the source material utilized for RNA sequencing.

[0053] Several different trials were performed to try reduce gene expression variability and eventually it was discovered that single-cell expression data yielded models able to detect minority cell types and also provided much great predictive ability. Specifically, scRNA-Seq data of primary tissue samples was delineated into specific cell types. This scRNA-seq data was then utilized to train the computational models for predicting gene expression from epigenetic regulatory markers

[0054] In many embodiments, in addition to obtaining gene sequencing results, profiles of spatial epigenetic regulatory markers are also obtained. Generally, any mechanism for obtaining profiles of spatial epigenetic regulatory markers can be utilized. In some embodiments, the spatial epigenetic regulatory markers are obtained from the results of the computational model that predicts epigenetic regulatory markers from genetic sequence, such as the computer-implemented method example described in if Fig. 2A.

[0055] Computer-Implemented Method 250 can optionally cluster and identify (253) single cells based on single cell expression data. Generally, any means for yield a more homogenic cell-type gene expression result can be utilized. For example, in some embodiments, cells (especially primary cells) are sorted into more homogenous populations and then sequences. Sorting can be done by any appropriate method, such as fluorescence-activated cell sorting. In some embodiments, single-cell expression data is generated utilizing scRNA-seq or snRNA-seq. The single-cell expression data results are utilized to cluster the cells the cells via a computational clustering application. Any appropriate computational clustering application can be utilized, such as (for example) K- means clustering, hierarchical clustering, spectral clustering, Gaussian mixtures models (GMM), and density-based spatial clustering of applications with noise (DBSCAN). Uponclustering, the specific cell-type of the cluster can be identified. In the experiments described in the Examples and Data section below, over 100 computational models were trained, each utilizing a scRNA-seq data of a single clustered cell type derived from seven diverse organ systems: kidney, liver, lung, heart, pancreas, brain, and retina.

[0056] Computer-Implemented Method 250 trains (255) a computational model using the single-cell expression data and spatial epigenetic regulatory markers to predict celltype gene expression characteristics. Cell-type gene expression characteristics that can predicted can be presence of gene expression, expression levels of a gene, other expression characteristics, or any combination thereof. In some embodiments, the expression level of at least one-fourth of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least one-third of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least one-half of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least three-fourths of the genes within the input genetic sequences is determined. In some embodiments, the expression level of each gene within the input genetic sequences is determined.

[0057] In several embodiments, single-cell expression data are used as labels for gene expression outputs, without any further imputation. In some embodiments, a computational model is trained utilizing scRNA-Seq data of only one clustered and identified cell type, which can reduce gene expression variability in the training data. In some embodiments, using the single-cell expression data and spatial epigenetic regulatory markers to predict cell-type gene expression. Other qualities ca also be utilized, such as (for example), requiring that clusters have a minimum number cells to be utilized in training or removing clusters with unidentified cells or removing doublet cells.

[0058] In some embodiments, the impact of non-cell-type genes are reduced. For example, weights can be applied to genes in which genes with high standard deviation across cell-types are weighted higher (e.g., standard deviation > 70thpercentile is weighted 2X) and genes with low standard deviation across cell-types are weighted lower (e.g., standard deviation < 10thpercentile is weighted 0.5X).

[0059] In some embodiments, the computational model for predicting gene expression is a regressor. In some embodiments, the computational model is a classifier. Thecomputational model can be any appropriate regressor or classifier, such as (for example) a linear regressor, logistic regressor, Lasso regressor, support vector regressor, nonlinear regressor, polynomial regressor, naive Bayes, K-nearest neighbors, support vector machine, etc. In some embodiments, the computational model is a linear regressor. In some embodiments, the computational model incorporates ridge regression.

[0060] In a number of embodiments, epigenetic regulatory markers are spaced along a genetic sequence. Epigenetic regulatory markers are various factors and elements that affect transcription interact with DNA along a genomic sequence. In many embodiments, epigenetic regulatory markers are cell-type and / or tissue specific and include (but are not limited to) sites of chromatin accessibility (e.g., DNase I hypersensitivity), chromatin marks (e.g., histone code), transcription factor binding sites, and other epigenetic factors. An example of 2,002 epigenetic regulatory markers is described in the Examples and Data section. In some instances, epigenetic regulatory variants and their spacing along a gene are determined in accordance with Computer-Implemented Method 200 as described in relationship to Fig. 2A.

[0061] Computer-Implemented Method 250 also assesses (257) the ability of the trained computational model to predict cell-type expression of one or more genes utilizing the based upon the regulatory features along the sequence Input data can be split into a training portion, a validation portion, and a testing portion. The training portion trains the model. The validation portion validates the training. And the testing portion is utilized to assess the predictive ability of the model. Alternative, or in addition, a leave-one-out approach can be utilized to train and cross validate the model.

[0062] Computer-Implemented Method 250 can also store and / or report (259) the trained computational model. While a specific example of a computer-implemented method for training a computational model to predict gene expression characteristics from genetic sequence is described above, one of ordinary skill in the art can appreciate that various steps of the method can be performed in different orders and that certain steps may be optional according to some embodiments. As such, it should be clear that the various steps of the method could be used in accordance with the requirements of specific applications.Cell-Type Gene Expression Prediction

[0063] An example of a computer-implemented method to predict the epigenetic regulatory markers along the sequence one or more genes is provided in Figure 3A. Computer-Implemented Method 300 can begin by obtaining (301 ) one or more genetic sequences for assessment. Genetic sequences can be full or partial genome. Genetic sequences can be sequences related a singular gene, sequences related to one or more genes, sequences related to two or more genes, sequences related to five or more genes, sequences related to ten or more genes, sequences related to twenty or more genes, sequences related to fifty or more genes, sequences related to one hundred or more genes, sequences related to two hundred or more genes, sequences related to five hundred or more genes, sequences related to one thousand or more genes, sequences related to two thousand or more genes, sequences related to five thousand or more genes, sequences related to ten thousand or more genes, sequences related to fifteen thousand or more genes, sequences related to nineteen thousand or more genes, or about all genes of a genome. In several embodiments, each gene analyzed includes the transcription start site of the sequence. In several embodiments, each gene analyzed includes the transcription start site and immediately adjacent sequences (e.g., + / - 20kb).

[0064] Genetic sequences can be derived from a number of sources. In some instances, genetic sequences are obtained de novo by extracting the DNA from a biological source and sequencing it. In some embodiments, the one or more sequences are derived from biological source. In some embodiments, the biological sample has a particular phenotype, such as (for example) a medical phenotype (e.g., presence of cancer). In many embodiments, the genetic sequences are derived from a collection or biopsy of an individual. In some embodiments, DNA can be derived from common research sources, such as in vitro tissue culture cell lines or research mouse models. In some embodiments, a biological sample is any collection of eukaryotic cells in which epigenetic regulatory markers of a set of genes is to be determined. Alternatively, genetic sequence data can be obtained from publicly or privately available databases. Many databases exist that store datasets of sequences from which a user can extract the data to perform computational analysis upon.

[0065] In some embodiments, genetic sequences comprise one or more polymorphisms, which can be derived from any source. In some embodiments, one or more polymorphisms are derived from naturally occurring polymorphisms (e.g., as detected in the sequence of an organism). In some embodiments, one or more polymorphisms are derived from polymorphisms identified within genetic material that is associated with a particular phenotype, such as a medical phenotype. In some embodiments, one or more polymorphisms are derived from in silico analysis, which can be a random assessment of polymorphisms or a methodical assessment of polymorphisms.

[0066] Computer-Implemented Method 300 can process (303) the one or more genetic sequences to identify gene structures therein, such as a gene’s TSS or promoter.

[0067] Computer-Implemented Method 300 can obtain (305) a computational model that predicts epigenetic regulatory markers based on gene sequence. In some embodiments, a trained computational model is trained as shown in and described in relation to Fig. 2A, however, any methodology for assigning epigenetic regulatory markers onto sequences of genes may be used.

[0068] Computer-Implemented Method 300 can predict (307) epigenetic regulatory markers along the sequence one or more genes of the obtained one or more genetic sequences, using the computational model. In addition, Computer-Implemented Method 300 can optionally spatially transform (309) the predicted epigenetic regulatory markers, which may reduce the dimensionality of the spatial epigenetic regulatory markers.

[0069] Computer-Implemented Method 300 can also store and / or report the predicted epigenetic regulatory markers of the one or more genes. In numerous embodiments, the epigenetic regulatory markers can be used in a number of downstream applications, which may include (but is not limited to) determining cell-type expression characteristics of the set of genes.

[0070] While a specific example of a computer-implemented method for determining the epigenetic regulatory markers of a set of genes of one or more genetic sequences is described above, one of ordinary skill in the art can appreciate that various steps of the method can be performed in different orders and that certain steps may be optionalaccording to some embodiments. As such, it should be clear that the various steps of the method could be used in accordance with the requirements of specific applications.

[0071] An example of a computer-implemented method to predict the epigenetic regulatory markers along the sequence one or more genes is provided in Figure 3B. Computer-Implemented Method 350 can begin by obtaining (351 ) epigenetic regulatory markers spaced along sequence of one or more genes of one or more genetic sequences to be assessed. Epigenetic regulatory markers are various factors and elements that affect transcription interact with DNA along a genomic sequence. In many embodiments, epigenetic regulatory markers are cell-type and / or tissue specific and include (but are not limited to) sites of chromatin accessibility (e.g., DNase I hypersensitivity), chromatin marks (e.g., histone code), transcription factor binding sites, and other epigenetic factors. An example of 2,002 epigenetic regulatory markers is described in the Examples and DAta section. In some instances, epigenetic regulatory variants and their spacing along a gene are determined in accordance with Computer-Implemented Method 300 as shown in and described in reference to Fig. 3A. It should be noted however, that any appropriate epigenetic regulatory markers and any appropriate method to determine epigenetic regulatory markers may be used.

[0072] Computer-Implemented Method 350 obtains (353) one or more cell-type specific computational models that predict expression characteristics of genes based on epigenetic regulatory markers spaced along the sequence of the genes. In some embodiments, a trained computational model is trained as shown in and described in reference to Fig. 2B, however any computational model capable of determining gene expression levels based on spatial epigenetic regulatory markers may be used.

[0073] Cell-type gene expression characteristics that can predicted can be presence of gene expression, expression levels of a gene, other expression characteristics, or any combination thereof. In some embodiments, the expression level of at least one-fourth of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least one-third of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least one-half of the genes within the input genetic sequences is determined. In some embodiments, the expression level of at least three-fourths of the genes within the input genetic sequences isdetermined. In some embodiments, the expression level of each gene within the input genetic sequences is determined.

[0074] In several embodiments, each of the one or more cell-type specific computational models that predict expression characteristics can be trained using singlecell expression data as labels for gene expression outputs. In some embodiments, the single-cell expression data was derived from primary tissue samples. In some embodiments, the single-cell expression data was generated utilizing scRNA-seq or snRNA-seq. The single-cell expression data results are utilized to cluster the cells the cells via a computational clustering application. Upon clustering, the specific cell-type of the cluster can be identified. In some embodiments, clustered cell populations are suborgan cell types, sub-tissue cell types, or primary cell types. And in several embodiments, clustered cell populations are sub-organ, sub-tissue, and primary cell types. And in some embodiments, a clustered cell population is a population of a specific cell type that further specifically responded to one or more stimuli (e.g., activated macrophages), which can be differentiated from a clustered cell population that did not or has not responded to the stimuli (e.g., mature but inactivate macrophages). In the experiments described in the Examples and Data section below, over 100 computational models were trained, each utilizing a scRNA-seq data of a single clustered cell type derived from seven diverse organ systems: kidney, liver, lung, heart, pancreas, brain, and retina.

[0075] Computer-Implemented Method 350 predicts (355) expression of the one or more genes in the one or more specific cell types using the cell-type specific computational models and the spatial epigenetic regulatory markers as input. For each specific cell type assessed for gene expression, the computational model is a matching cell-type specific computational model that was trained utilizing single-cell expression data derived from the cell type. In various embodiments, gene expression is predicted for at least one cell type, for at least two cell types, for at least five cell types, for at least ten cell types, for at least twenty cell types, for at least fifty cell types, or for at least one hundred cell types. In some embodiments, gene expression is predicted for at least two cell types of an organ system or a tissue system, for at least five cell types of an organ system or a tissue system, for at least ten cell types of an organ system or a tissue system, or for at least twenty cell types of an organ system or a tissue system.

[0076] Computer-Implemented Method 350 stores and / or reports (357) the expression characteristics of the one or more genes for the one or more single-cell genetic sequences in the one or more specific cell types. In several embodiments, cell-type gene expression characteristics are used in a number of downstream applications, including (but not limited to) performing various biochemical assays, generating polymorphic cells, and developing clinical diagnostics and treatments.

[0077] While a specific example of a process for determining cell-type expression levels of a set of genes is described above, one of ordinary skill in the art can appreciate that various steps of the method can be performed in different orders and that certain steps may be optional according to some embodiments. As such, it should be clear that the various steps of the method could be used in accordance with the requirements of specific applications.Computer Systems for Genetic Sequence Analysis

[0078] Provided in Fig. 4 is an example of a computer system 401 that can be implemented on one or more computing devices. The computing devices of computer systems 401 can comprise personal computers, laptop computers, remote servers, handheld computing devices, other computing devices, or any combination of devices and computers with sufficient processing power for the processes described herein. Computer systems 401 can include a processor system 403, which may include one or more processors within one or more computing devices. Processor system 403 can be configured to perform computations via machine readable instructions stored within a memory 407 of computer systems 401. The processor system may include one or more microprocessors (CPUs), one or more graphics processing units (GPUs), and / or one or more digital signal processors (DSPs).

[0079] Memory 407 can contain one or more computational applications that provide machine readable instructions to perform the various computer-implemented methods and / or processes described herein, such as shown within and described in reference to Figs. 1-3B. For example, memory 407 can comprise a Computational Framework for Cell-Type Expression Assessment 409 (e.g., see Fig. 1 ). Computational Framework for Cell-Type Expression Assessment 409 can comprise an Epigenetic Regulatory PredictionModel 409a (e.g., see Figs. 2A and 3A) or one or more Cell-Type Gene Expression Prediction Models 409b (e.g., see Figs 2B and 3B). Epigenetic Regulatory Prediction Model 409a and each of the one or more Cell-Type Gene Expression Prediction Models 409b, however, can each be their own independent computational application and are not required to be within a defined framework or co-associated in any way.

[0080] Memory 407 can contain input, intermediate, and / or output data or parameters associated with the various computer-implemented methods and / or processes, which can be transiently stored or archived. For example, memory 407 can comprise genetic sequence data 411 , which can be utilized as input for Epigenetic Regulatory Prediction Model 409a. Memory 407 can further comprise Predicted Epigenetic Regulatory Marker Data 413, which can be an output of Epigenetic Regulatory Prediction Model 409a and / or utilized as input for Cell-Type Gene Expression Prediction Models 409b. Memory 407 can also comprise Predicted Cell-Type Expression Data, which can be an output of Cell-Type Gene Expression Prediction Models 409b. In addition, memory 407 can store parameters, weights, and computational algorithms for the various computational models, including machine-learning models. Parameters, weights, and computational algorithms can be locked or available to be updated based on further machine-learning and / or user input.

[0081] Computer system 401 can include an input / output interface (605) for communicating with a variety of devices, including but not limited to other computing systems, internet or intranet communication, a visual display, user interface, and / or scientific, medical, or clinical machinery (e.g., nucleic acid sequencer). As can be readily appreciated, a variety of hardware and software architectures can be utilized to implement computer system 401 and the various applications to be executed therein.Cell-Type-Specific Biochemical Analysis and Site-directed Mutagenesis

[0082] A number of embodiments are directed towards performing biochemical assays based on the results of sequences and polymorphisms identified to affect cell-type gene expression characteristics. Accordingly, in several embodiments, computational methods are performed to predict the effect sequences and polymorphisms on gene expression in a variety of cell types. Based on computational predictions, a biochemical assay is performed to empirically assess the predictions. In various embodiments, computaitonalprediction of cell-type gene expression is performed any process or method (or portion thereof) described herein, such as (for exam pled) the processes and methods portrayed within and described in reference to Figs. 1 to 3B and 5

[0083] In many embodiments, biochemical assessments are performed as follows: a) select one or more genetic sequences for assessment b) computationally predict the effect of the one or more genetic sequences on cell-type expression for one or more genes in one or more cell types c) based the computationally predicted effect, perform a biochemical assay to empirically assess the effect on gene expression, cell function, or various phenotypes.In some embodiments, the biochemical assessment is performed on a cell type that is computationally predicted to have gene expression affected. In some embodiments, the biochemical assessment is performed on a cell type that is not computationally predicted to have gene expression affected but is within tissue system or organ system comprising one or more cell types computationally predicted to have gene expression affected. In some embodiments, the one or more genetic sequences for assessment are associated with a phenotype, and the effect on phenotype is assed via the biochemical assay. In some embodiments, the phenotype is a medical or clinical phenotype. In some embodiments, the medical or clinical phenotype is a clinical or medical disorder. In some embodiments, one or more genetic sequences for assessment comprise one or more polymorphisms, and the effect of the one or more polymorphisms on gene expression, cell function, or various phenotypes is assessed by the biochemical assay. In some embodiments, one or more polymorphisms are derived from a natural source. In some embodiments, one or more polymorphisms are derived from a biological sample associated with clinical or medical phenotype. In some embodiments, one or more polymorphisms are derived from in silico analysis. In some embodiments, one or more polymorphisms are predicted to effect gene expression that is further predicted to improve an undesirable phenotype (e.g., a medical or clinical disorder).

[0084] A number of biochemical assays can be performed to empirically assess computationally predict effects of the one or more genetic sequences on cell-type expression. Generally, biochemical assays will provide a greater in-depth assessment ofsequences and polymorphisms effects on various biological functions, which include effects on chromatin formation, chromatin binding, nearby gene transcription, RNA stability, RNA processing, translation, cellular function, tissue function, organ function, medical or clinical phenotypes, and organismal phenotypes. A number of biochemical assays are known in the art to assess local gene sequence (especially variants and indels within sequence) effect, including (but not limited to) chromatin immunoprecipitation sequencing (ChlP-seq), DNAse I hypersensitivity sequencing (DNase-seq), Assay for Transposase-Accessible Chromatin sequencing (ATAC-seq), Formaldehyde- Assisted Isolation of Regulatory Elements (FAIRE-seq), Hi-C capture sequencing, bisulfite sequencing (BS-seq), methyl array, transgene expression analysis (e.g., luciferase and eGFP), qPCR, RNA hybridization (e.g., ISH), cross-linking immunoprecipitation sequencing (CLIP-seq), RNA immunoprecipitation sequencing (RIP- seq), RNA-seq, western blot, immunodetection, flow cytometry, enzyme-linked immunosorbent assay (ELISA), and mass spectrometry.

[0085] Several embodiments are also directed towards editing genetic material in order to analyze sequences and polymorphisms, especially site-directed mutagenesis. In some embodiments, a sequence and / or one or more specific polymorphisms are incorporated into a DNA sequence construct for cell-type analysis (e.g., luciferase plasmid construct). In some embodiments, a sequence and / or one or more specific polymorphisms are introduced into at least one allele of the DNA of a biological cell, which can be of a particular cell type of interest (e.g., predicted to have gene expression affected by one or more polymorphisms). Several methods are well known to introduce polymorphic changes within an allele, including (but not limited to) CRISPR mutagenesis, Zinc-finger mutagenesis, and TALEN mutagenesis. In some embodiments, a common polymorphism is changed into rare polymorphism. In some embodiments, a rare polymorphism is changed into a common polymorphism. In some embodiments, a polymorphism is introduced to improve an undesired phenotype (e.g., clinical or medical disorder). In some embodiments, a polymorphism is introduced to introduce or worsen an undesired phenotype (e.g., clinical or medical disorder), which can be useful for generating disease research models.

[0086] Various embodiments are directed towards development of cells (e.g., primary cells) having a particular DNA sequences that have one or more polymorphisms affecting cell-type gene expression characteristics. In some embodiments, a cell can be manipulated by genetic engineering to harbor one or more polymorphisms. In some embodiments, the cell can be derived from an individual (e.g., from a biopsy or collection) which would harbor polymorphisms identified in the genetic sequence of that individual. In some embodiments, cells from an individual can be genetically manipulated to “correct” one or more polymorphisms. In some embodiments, cells having a set of undesired polymorphism and cells having a set of desired or “corrected” polymorphisms may be assessed to determine the cumulative effect of the set polymorphisms, especially when modeling a clinical or medical disorder that may be associated with the set of undesired polymorphisms. The cells of the individual can be of a specific cell type that is assessed via the systems and methods described herein.

[0087] A number of embodiments are directed to preliminary in silico analysis of polymorphisms that can affect cell-type gene expression in order to perform site-directed mutagenesis that modulates gene expression levels. Accordingly, utilizing computational models that can predict cell-type gene expression levels ab initio from sequence, a number of hypothetical and / or novel polymorphisms can be designed that increase and / or decrease gene expression. C / s-manipulation of gene expression by local mutagenesis can be a useful tool to understand a number of phenotypes that are based on stoichiometric levels of gene expression. In addition, a number of medical disorders are known to arise from improper gene expression stoichiometry (e.g., see A. M. Rice and A McLysaght BMC Biol. 15, 78 (2017), the disclosure of which is incorporated herein by reference).

[0088] In several embodiments, site-directed mutagenesis to introduce hypothetical polymorphism is performed as follows:(a) perform in silico analysis to select polymorphisms that affect cell-type gene expression characteristics(b) based on in silico analysis, perform site-directed mutagenesis to introduce selected polymorphisms into a specific cell-type(c) perform a biochemical assay to empirically assess the effect of polymorphisms on gene expression, cell function, or various phenotypes.In some embodiments, the site-directed mutagenesis is performed on a cell type that is computationally predicted to have gene expression affected. In some embodiments, the site-directed mutagenesis is performed on a cell type that is not computationally predicted to have gene expression affected but is within tissue system or organ system comprising one or more cell types computationally predicted to have gene expression affected. In some embodiments, the one or more introduced polymorphisms are associated with a phenotype, and the effect on phenotype is assed via the biochemical assay. In some embodiments, the phenotype is a medical or clinical phenotype. In some embodiments, the medical or clinical phenotype is a clinical or medical disorder. In some embodiments, the one or more introduced polymorphisms are inspired by a natural phenomenon. In some embodiments, the one or more introduced polymorphisms are identified from one or more genetic sequences associated with a clinical or a medical phenotype. In some embodiments, one or more polymorphisms are derived from in silico analysis. In some embodiments, the one or more polymorphisms are predicted to effect gene expression that is further predicted to improve an undesirable phenotype (e.g., a medical or clinical disorder).

[0089] In some embodiments, in silico analysis to determine effect of one or more polymorphisms on gene expression can be performed in accordance as described herein, such as the processes and methods described in Figs. 1 -3B and 5. In some embodiments, site-directed mutagenesis is performed utilizing CRISPR mutagenesis, Zinc-finger mutagenesis, and / or TALEN mutagenesis. In some embodiments, biochemical analysis to be performed is at least one of: chromatin immunoprecipitation sequencing (ChlP-seq), DNAse I hypersensitivity sequencing (DNase-seq), Assay for Transposase- Accessible Chromatin sequencing (ATAC-seq), Formaldehyde-Assisted Isolation of Regulatory Elements (FAIRE-seq), Hi-C capture sequencing, bisulfite sequencing (BS- seq), methyl array, transgene expression analysis (e.g., luciferase and eGFP), qPCR, RNA hybridization (e.g., ISH), cross-linking immunoprecipitation sequencing (CLIP-seq), RNA immunoprecipitation sequencing (RIP-seq), RNA-seq, western blot,immunodetection, flow cytometry, enzyme-linked immunosorbent assay (ELISA), and mass spectrometry.

[0090] In some embodiments, isolated cells of specific cell type, including in vivo, primary, and / or ex vivo cells, are mutagenized to alter sequence variants in order to modulate gene expression levels. Accordingly, cells are designed to have altered gene expression levels that may be useful for either understanding the consequence of one or more polymorphisms and / or “correcting” one or more polymorphisms. In some embodiments, a biopsy or collection is obtained from an individual to yield primary and / or ex vivo cells and one or more polymorphisms are introduced to modulate gene expression. In some embodiments, primary and / or ex vivo cells that have been introduced with one or more polymorphisms are transplanted back into the individual to provide a therapeutic benefit to the individual by the altered gene expression. In some embodiments, primary and / or ex vivo cells are of a specific cell type that have been assessed by the various processes or methods described herein.Diagnostics and Treatments Based on Cell-Type Gene Expression

[0091] Various embodiments are directed to development of diagnostics and treatments related to cell-type gene expression. As described herein, an individual may be diagnosed as having a particular gene expression status within a specific cell type, having an influence on a clinical or medical phenotype. In some embodiments, an individual is diagnosed as having a clinical or medical disorder or having a propensity to develop a clinical or medical disorder. Based on computationally predicted cell-type gene expression data, an individual can be treated with various medications and therapeutic regimens.Diagnostic Methods

[0092] A number of embodiments are directed towards diagnosing individuals by determining the effect of the individual’s genetic sequence on cell-type gene expression. In a number of embodiments, diagnostics can be performed as follows: a) obtain genetic sequence data of the individual to be diagnosedb) predict effect of the individual’s genetic sequence on cell-type gene expression related to a clinical or medical disorder c) diagnose the individual based on the gene expression levels.Diagnoses, in accordance with various embodiments, can be performed as portrayed in and described in reference to Figs. 3A and 3B. A diagnoses can be related to an indication of diagnosis of the medical disorder, early detection of the medical disorder before identification of symptoms, personalized medicine diagnosis, or a treatment of the disorder to ameliorate a symptom associated with the misregulated pathway in the one of more cell types.

[0093] Many embodiments of diagnostics improve on traditional diagnostic methods. Because genetic contribution to cell-type gene expression is difficult to determine from genetic sequence, traditional genetic tests of examining a single gene, variant, and / or locus have been unavailable. As described herein, however, in some embodiments, a diagnosis is performed ab initio utilizing genetic sequence to determine cell-type gene expression levels, such as described in Figs. 1 -3B and 5. In some embodiments, diagnoses are performed for congental disorders in which GWAS does not identify causal variants. In some embodiments, diagnoses are performed for congenital disorders that are determined arise at least in part by polymorphisms that affect cell-type gene expression levels.Diagnostic Kits

[0094] Embodiments are directed towards expression analysis (e.g., RT-PCR, microarray) genomic loci sequencing and / or single nucleotide polymorphism (SNP) array kits to be utilized within various methods as described herein. As described, various methods can diagnose an individual for a medical disorder by examining variants that affect gene expression or the expression levels directly. Accordingly, a number of embodiments are directed towards targeted genomic loci sequencing and SNP array kits that cover a set of genomic loci to diagnose a particular clinical or medical phenotype. In some instances, the set of genomic loci are computationally identified by a computational model, such as described in Figs. 2A and 2B. In addition, various embodiments aredirected towards expression analysis kits that cover a set of genes to diagnose a particular trait.

[0095] A number of targeted gene sequencing protocols can be utilized, including (but not limited to) partial genome sequencing, primer-directed sequencing, and capture sequencing. Generally, targeted sequencing involves selection step either by hybridization and / or amplification of the target sequences prior to sequencing. Therefore, embodiments are directed to sequencing kits that target genomic loci that are known to harbor variants that affect expression levels to diagnose a particular medical disorder.

[0096] Likewise, a number of SNP array protocols can be utilized. In general, chip arrays are set with oligo sequences having a particular SNP. Sample DNA derived from an individual can be processed and then applied to SNP array to determine sites of hybridization, indicating existence of a particular SNP. Thus, embodiments are directed to SNP array kits that target particular SNPs that known to be expression altering in order to diagnose a particular medical disorder.

[0097] The number of genomic loci and / or SNPs to include in a sequencing kit can vary, depending on the genomic loci and / or SNPs to examine for a particular trait and the computational model to be used. In some embodiments, the genomic loci and / or SNPs to be examined are identified by a computational model, such as the computational model described in Figs. 2A and 2B. In various embodiments, the number of genomic loci in a sequencing kit are approximately, 100, 1000, 5000, 10000, 20000, 30000, 40000, 50000, 60000, 70000, 80000, 90000, 100000, 150000, or 200000 loci. In various embodiments, the number of SNPs in an array kit are approximately, 1000, 10000, 50000, 100000, 200000, 300000, 400000, 500000, 600000, 700000, 800000, 900000, 1000000, 1500000, or 2000000 SNPs. In some embodiments, all identified loci are included in a kit. In some embodiments, only a subset of the loci are included. It should be understood that precise number and positions of loci can vary as the gene expression model can be updated with new data or recreated with a different data set (especially for biological attributes).

[0098] The number of probes or primer pairs to include in an expression analysis kit can vary, depending on the genes to examine for a particular trait and the computational model to be used. In some embodiments, the genes to be examined are identified by acomputational model, such as the computational model described in Figs. 2A and 2B. In various embodiments, the number of genes to analyze in an expression kit are approximately, 10, 20, 30, 40, 50, 100, 200, 300, 400, 500, 1000, 5000, or 10000 genes. Generally, the probes and / or primer pairs to be included in a kit are directed to cDNA and thus span exon-exon junctions. In some embodiments, all identified genes are included in a kit. In some embodiments, only a subset of the genes is included. It should be understood that precise number genes can vary as the gene expression model can be updated with new data or recreated with a different data set (especially for biological attributes).Medications and Supplements

[0099] Several embodiments are directed to the use of medications and / or dietary supplements to treat an individual based on their medical disorder diagnosis. In some embodiments, medications and / or dietary supplements are administered in a therapeutically effective amount as part of a course of treatment. As used in this context, to "treat" means to ameliorate at least one symptom of the disorder to be treated or to provide a beneficial physiological effect.

[0100] A therapeutically effective amount can be an amount sufficient to prevent reduce, ameliorate or eliminate symptoms of disorders or pathological conditions susceptible to such treatment. In some embodiments, a therapeutically effective amount is an amount sufficient to reduce the symptoms of a medical disorder.

[0101] Dosage, toxicity and therapeutic efficacy of the compounds can be determined, e.g., by standard pharmaceutical procedures in cell cultures or experimental animals, e.g., for determining the LDso (the dose lethal to 50% of the population) and the EDso (the dose therapeutically effective in 50% of the population). The dose ratio between toxic and therapeutic effects is the therapeutic index and it can be expressed as the ratio LD50 / ED50. Compounds that exhibit high therapeutic indices are preferred. While compounds that exhibit toxic side effects may be used, care should be taken to design a delivery system that targets such compounds to the site of affected tissue in order to minimize potential damage to other tissue and organs and, thereby, reduce side effects.

[0102] Data obtained from cell culture assays or animal studies can be used in formulating a range of dosage for use in humans. If the pharmaceutical is provided systemically, the dosage of such compounds lies preferably within a range of circulating concentrations that include the EDso with little or no toxicity. The dosage may vary within this range depending upon the dosage form employed and the route of administration utilized. For any compound used in the method of the invention, the therapeutically effective dose can be estimated initially from cell culture assays. A dose may be formulated in animal models to achieve a circulating plasma concentration or within the local environment to be treated in a range that includes the ICso ( / .e., the concentration of the test compound that achieves a half-maximal inhibition) as determined in cell culture. Such information can be used to more accurately determine useful doses in humans. Levels in plasma may be measured, for example, by liquid chromatography coupled to mass spectrometry.

[0103] An "effective amount" is an amount sufficient to effect beneficial or desired results. For example, a therapeutic amount is one that achieves the desired therapeutic effect. This amount can be the same or different from a prophylactically effective amount, which is an amount necessary to prevent onset of disease or disease symptoms. An effective amount can be administered in one or more administrations, applications or dosages. A therapeutically effective amount of a composition depends on the composition selected. The compositions can be administered one from one or more times per day to one or more times per week; including once every other day. The skilled artisan will appreciate that certain factors may influence the dosage and timing required to effectively treat a subject, including but not limited to the severity of the disease or disorder, previous treatments, the general health and / or age of the subject, and other diseases present. Moreover, treatment of a subject with a therapeutically effective amount of the compositions described herein can include a single treatment or a series of treatments. For example, several divided doses may be administered daily, one dose, or cyclic administration of the compounds to achieve the desired therapeutic result.EXAMPLES AND DATA

[0104] Bioinformatic and biological data support the systems and methods of predicting cell-type-specific expression of genes from sequence. In the ensuing sections, examples of computational methods and examples of applications that utilize cell-type- specific expression are provided. The examples show that the training methods employed yielded models that greatly improves the accuracy of determining cell-type-specific gene expression, especially for genes only express in certain cell-types. The example also describes the differential effect of variants on multiple cell types, further establishing that genetic variation in various cell-types can have differential effect on manifestation of various medical disorders. Further details and explanation on the supporting biological data is described in K. Sokolova, et al., Cell Rep Methods. 2023 Sep 25;3(9): 100580, the disclosure of which is hereby incorporated by reference.Atlas of Primary Cell-type-specific Sequence Models of Gene Expression and Variant EffectsSummary

[0105] Human biology is rooted in highly specialized cell types programmed by a common genome, 98% of which is outside of genes. Genetic variation in the enormous noncoding space is linked to the majority of disease risk. To address the problem of linking these variants to expression changes in primary human cells, we introduce ExPectoSC, an example of modular deep-learning-based models for predicting cell-type-specific gene expression directly from sequence. We provide models for 105 primary human cell types, covering 7 organ systems, demonstrate their accuracy and then apply them to prioritize relevant cell types for complex human diseases. The resulting atlas of sequence-based gene expression and variant effects is publicly available in a user-friendly interface and readily extensible to any primary cell types. We demonstrate the accuracy of our approach through systematic evaluations and apply the models to prioritize ClinVar clinical variants of uncertain significance, verifying our top predictions experimentally.Motivation

[0106] Our work addresses a critical challenge for personalized medicine and basic biology - that of interpreting the genome, focusing on the 98% of the human genome that encodes regulation. Linking this variation to the underlying functional consequences is extremely difficult because of the scale of the mutation space and the impossibility in culturing many primary human cell types in the lab. We developed a computational framework capable of predicting cell-type-specific expression directly from genomic sequence, which allows us to computationally interrogate expression effects of millions of variants across the diversity of cell types and organ systems.Introduction

[0107] Deciphering the regulatory code of gene expression includes understanding genomic signals for spatial, temporal, and cell-type-specificity in gene expression and elucidating the impact of genetic variation on expression dynamics. This regulatory code underpins fundamental biological processes, including development from a single cell into a complex organism and cellular specialization in function and morphology. Furthermore, precision medicine’s promise of improved, more precisely-tailored treatments relies on interpreting an individual's genome and mutation effects in the context of specialized cell types (e.g. specific neurons or heart fibroblasts). Because the human genome is three billion bases long, with four possible ‘letters’ at each position, experimentally determining regulatory architecture across every mutation, cell type, and environment is intractable and it is therefore necessary to develop computational methods for ab initio prediction of gene expression based on sequence.

[0108] Deep learning models have recently emerged as a promising approach for prediction of gene expression from sequence (see, e.g., J. Zhou, et al., Nat Genet. 2018 Aug;50(8):1171-1179, the disclosure of which is hereby incorporated by reference). However, existing models predict gene expression in whole tissues or in lab-derived cell lines from bulk RNA-Seq data. Whole-tissue resolution cannot identify cell types contributing to disease, and thus does not adequately model the etiology of human disease. Furthermore, cell lines represent a small portion of the emerging specialized cell types defined by recent single cell gene expression studies and often divergesubstantially in gene expression from cells in an organism. For example, brain tissue is composed of specialized neurons, immune cells, and structural support cells. Such primary cell types are difficult to study because the vast majority cannot be experimentally cultured. There is a critical need fora method capable of predicting sequence-based gene expression and variant effects in primary cell types (for example, to determine whether a given variant changes expression of a neural development gene in autism-relevant neurons or shuts down extracellular matrix proteins in heart cells, contributing to fibrosis).

[0109] Here we introduce ExPectoSC, an atlas consisting of 105 primary cell-type models across seven organ systems. We show that these models are capable of predicting gene expression in primary human cell types based only on the DNA sequence. The atlas can be readily extended to include any cell type characterized by single cell RNA-sequencing, including newly discovered human cell types. Each of the cell type models integrates a convolutional neural network with a regularized linear model to predict gene expression effects of variants. We demonstrate that ExPectoSC is accurate and can prioritize causal disease variants.ResultsA deep learning framework for building cell-type-specific models

[0110] To understand how genetic variation contributes to cell-type-specific gene expression and disease phenotypes, we developed a modular deep learning approach that is scalable to all genes of the human genome and extendable to every cell type captured in single cell experiments. Our approach, ExPectoSC, leverages the previous ExPecto framework, which previously provided only tissue-specific resolution available in the bulk RNAseq. ExPectoSC improves upon its predecessor to enable training of cell- type-specific models to yield a precise understanding of the biological effects for the full diversity of primary human cell types. The framework trains on the reference genome, without any variant information, to learn sequence rules predictive of chromatin patterns, and then to associate these patterns with single cell expression (Fig. 5). First, a neural network module learns interactions between DNA sequence and the genome-wide binding of biochemical regulators of gene expression. In this example, a convolutional neural network (CNN) serves as an encoder by accepting a DNA sequence windows of2 kb length and predicts 2002 chromatin profiling features from bulk tissues and cell lines, including histone marks, transcription factor binding, and DNA accessibility. Predictions are accumulated for the 40kbp region centered around the gene transcription start site (TSS) using a sliding window approach with 200bp overlap. Then, a linear models module uses these predictions to establish a regularized linear model for each cell type to infer how biochemical effects predictions at each TSS correspond to gene expression levels. After training, ExPectoSC utilizes a sequence + / - 20kbp at the transcription start site (TSS) to predict cell-type-specific expression. Moreover, any single cell dataset can be used, allowing users to train customizable models.

[0111] We trained ExPectoSC single cell gene expression models for over 100 cell types drawn from seven diverse organ systems: kidney, liver, lung, heart, pancreas, brain, and retina (Figs. 6A and 7) (for more on the cell-type data, see B. J. Stewart, et al., Science. 2019 Sep 27;365(6460):1461-1466; S. A. MacParland, et al., Nat Commun. 2018 Oct 22;9(1 ):4383; E. Madissoon, et al., Genome Biol. 2019 Dec 31 ;21 (1 ):1 ; M. Litvinukova, et al., Nature. 2020 Dec;588(7838):466-472; M. Baron, et al., Cell Syst. 2016 Oct 26;3(4):346-360.e4; N. Habib, et al., Nat Methods. 2017 Oct; 14(10): 955-958; and S. W. Lukowski, et al., EMBO J. 2019 Sep 16;38(18):e100811 ; the disclosures of which are each hereby incorporated by reference). ExPectoSC models are trained using cell clustering, cell identities, and gene expression values as defined in the respective cited study. ExPectoSC directly uses output data from single-cell RNA-seq experiments, and is therefore straightforward to extend to new datasets.ExpectoSC accurately models cell-type-specific gene expression

[0112] First, we evaluated ExPectoSC performance on cell-type-specific gene expression prediction using the test dataset, consisting of genes on chromosome 8 and 9, with a total of 1104 genes. Importantly, these chromosomes were left out of all steps of training, both for the neural network and for the regularized linear models. All cell-type- specific models achieve over 0.8 AUC ROC (Figs. 6B and 8) on the task of predicting variably expressed genes within the cell type, indicating that ExPectoSC models accurately predict single cell gene expression across the range of gene expressionobserved in diverse organ tissues in single cell experiments. These results are consistent with evaluation done using the 5-fold cross validation (Fig. 9).Cell-type-specific models are necessary for accurate single-cell gene expression prediction

[0113] To assess whether primary cell type resolution is necessary for prediction of cell-type-specific expression, we compare prediction performance of our cell-specific models to identical models aggregated to whole-organ resolution. We observe that for 93% of the models (98 / 105), cell-type-specific predictions have higher AUC ROC score than the whole organ models (Fig. 9) .

[0114] Importantly, the difference is strongest for cell-type-specific genes, those that are expressed in the top quartile in at least one cell type and bottom quartile of expression in a different cell type in the same organ (Figs. 6C and 9). These cell-type-specific genes may be important to cell type identity. For such cell-type-specific genes, cell-type- specificity of models is especially essential, with nearly all (103 / 105) models improving performance by 20% on average. For many cell types, whole organ predictions were inadequate (AUC < 0.5), while cell-type-specific models showed strong performance. For example, cell-type-specific expression in the microglial cells in the retina fails to be predicted by the whole organ model (AUC=0.36), whereas the cell-type-specific model demonstrates 0.79 AUC. In fact, the five models with lowest whole organ AUCs (< 0.4 AUC) all showed over 50% improvements. Importantly, ExPectoSC predictions are not confined to highly expressed genes and are accurate for genes expressed at both high and low levels, providing insight for genes at all levels of expression.ExpectoSC prediction of variant effects identifies disease-specific cell type dysregulation

[0115] We then applied the ExPectoSC framework to the study of genetic variation in disease from the single cell perspective. Our models can predict the effect of any TSS- proximal variant on single cell expression, thus enabling the study of variants that may be associated with human diseases or traits. Importantly, because ExPectoSC predicts the effect of a DNA sequence change, it enables inference of causality of gene variant impacts on gene expression. Additionally, since models were trained without any variantdata and the ExPectoSC approach is not limited to a preset list of mutations, it can be applied to common and rare variants, and even to variants that have not yet been observed in sequencing studies. In this section we consider applications of the model outputs: first using ExPectoSC scores with GWAS studies to infer the impact of common variants associated with human disease risk, and then applying ExPectoSC to mostly rare variants in ClinVar to identify variants with high cell-type-specific gene expression impact (for more on ClinVar, see M. J. Landrum, et al., Nucleic Acids Res. 2018 Jan 4;46(D1 ):D1062-D1067, the disclosure of which is hereby incorporated by reference).ExpectoSC-enabled identification of cell type contribution to disease risk

[0116] GWAS studies statistically associate a set of genomic loci with specific traits and diseases. Any one study can yield many candidate variants in each region due to linkage disequilibrium, and even with advanced fine mapping methods, identification of causal variants from other closely linked variants is uncertain. Since the associated loci are thought to act through dysregulation of gene expression, ExPectoSC can address the challenge by prioritizing candidate variants based on their impact on gene expression in relevant cell types. To enable this application, we applied ExPectoSC to a large, genomewide collection of over 2 million 1000 Genome Project variants within ±20kb of any TSSs for each of the 105 cell types in ExPectoSC models, generating a resource for the community.

[0117] The set of ExPectoSC predictions for the 1000G variants can be used to understand high-resolution cell-type-specific effects that contribute to inherited disease risk. To address this challenge, we leveraged stratified linkage disequilibrium regression (sLDSC) (H. K. Finucane, et al., Nat Genet. 2015 Nov;47(11 ):1228-35, the disclosure of which is hereby incorporated by reference). LD score regression models the LD structure between SNPs and enables comparisons of SNP effects across different cell types and GWAS studies. We applied sLDSC to the ExPectoSC predicted gene expression dysregulation across cell types to 102 GWAS studies to assess how the predicted SNP impacts (per cell type) contribute to disease risk heritability while also modeling the linkage disequilibrium of nearby SNPs. This analysis directly identified significantenrichment of inherited risk for cell types modeled in ExPectoSC for each of 102 GWAS diseases and traits.

[0118] In general, we found good correspondence between cell types and tissues known to be involved clinically in disease and the ExPectoSC cell type annotations enriched in disease risk. For example, we observed a significant (p-value < 2.2e-16) enrichment of immune cell types for the auto-immune diseases GWAS, independent of the subset of diseases included (Figs. 10A, 11 A, and 11 B).

[0119] Notably, the use of cell types from single cell data allows a higher resolution analysis of contribution to disease and traits. For example, use of ExPectoSC cell type annotations in sLDSC demonstrated specific and distinct enrichment differences between cell types in risk heritability for the BMI and BMI adjusted waist-hip ratio (WHR, a surrogate for abdominal fat measurement). BMI risk was enriched in brain cell types (p- value=1.84e-05), but BMI adjusted for WHR noncoding impacts are not (p-value= 0.0005) (Fig. 10B). Through examining gene-level associations, BMI is documented to be connected to the brain regions and development (see, e.g., M. Hamer and G. D. Batty, Neurology. 2019 Feb 5;92(6):e594-e600, the disclosure of which is hereby incorporated by reference), while adjusted WHR is associated with an enrichment of genes expressed in adipose tissue (D. Shungin, et al., Nature. 2015 Feb 12;518(7538): 187-196, the disclosure of which is hereby incorporated by reference). Within ExPectoSC predictions for brain cell types for BMI, GABAergic interneurons had the greatest enrichment (Fig. 10C). GABAergic neurons produce gamma-Aminobutyric acid (GABA), which is involved in metabolism and is associated with feeding behaviors and body weight. Therefore, the risks associated with ExPectoSC predicted cell types (based on non-coding variant impacts) are consistent with tissue signals from gene-level analyses, identifying similar tissue types with specific contributions to morphological traits. However, ExPectoSC goes further by enabling the disease risk to be partitioned to high resolution sub-tissue specific cell types.ExpectoSC inquiry of clinically relevant mutations

[0120] One of the major challenges of precision medicine is classifying variants from clinical genomes with respect to their impact on disease state. In the ClinVar database,over 40% of variants identified in the course of clinical genome sequencing are classified as Variants of Uncertain Significance (VUS). We used ExPectoSC to predict the cell-type gene expression impacts for >60,000 ClinVar non-coding variants in 105 cell types that are within + / -20kb region of every protein coding gene TSS. We first compared ExPectoSC predictions for those variants annotated as pathogenic versus benign, and observed that pathogenic variants had significantly stronger expression impacts (Mann- Whitney U-test p-value=2.921 e-07).

[0121] Then we examined ExPectoSC predictions to identify potentially disease- associated variants among the VUSs. Among the highest predicted gene expression altering variants were six variants located in the promoter of PTEN, a regulator of PI3K / Akt / mTORC1 signaling pathway, a tumor suppressor gene, and an autism risk gene. ExPectoSC predicted a strong decrease in expression of PTEN in several brain cell types, including neuronal stem cells - a cell type implicated in autism dysregulation during development and explicitly regulated by PTEN (Fig. 12A). We experimentally tested the transcriptional regulatory activity of top ClinVar variants in the characterized PTEN promoter with transcriptional reporter assays and found the region produced robust transcriptional activity in neuroblastoma cells (Fig. 12B). The alternate alleles at three positions with highest predicted impact decreased this activity while the alternate allele with low predicted impact did not. Based on the biochemical effects predicted by Module 1 of ExPectoSC, we found that these variants likely affect expression through disruption of YY1 binding. Together these predictions and validation experiments demonstrate how ExPectoSC can be used to identify potentially clinically relevant variants based on their impact on expression of disease-relevant genes in specific primary human cell types.Discussion

[0122] ExpectoSC cell-type-specific expression prediction offers an important insight into disease pathology at an unprecedented resolution, including prioritizing variants at the level of primary cell type effects. Much like RNA-seq set a foundation for the development of scRNAseq, our new method ExPectoSC builds on the previous work of predicting tissue gene expression to provide high-resolution primary cell-type resolved expression effects. The modular structure of the framework allows us to use the deeplearner to extract the transcriptional activity information and use it to predict the cell-type resolved gene expression. This also enables future innovations separately in each module. For example, updating Modulel , the genome activity encoder, with newer larger models.

[0123] ExPectoSC predicts expression disruptions for primary human cell types by leveraging the Module 1 predictions, which have cell line and tissue resolution. As demonstrated by accurate performance, ExPectoSC is able to link the relationship between the relevant epigenetic signals of tissues and cell lines predicted by Module 1 with primary cell-type expression predicted by Module 2. These relationships can be leveraged to gain biological insight. For example, the computational framework could be used to guide biomedical researchers in choosing appropriate and experimentally tractable cell lines for studying primary cell types, and possibly be used to learn new contextual relationships between epigenetic factors and alterations in gene expression. ExPectoSC enables biomedical researchers to create focused hypotheses for the role of variants in gene expression dysregulation and disease, meeting a critical need for guiding biomedical research into clinical mutation effects.Experimental MethodsCell culture

[0124] Human neuroblastoma BE(2)-C cells (ATCC CRL-2688) were used at low passage after purchase from ATCC and were cultured in 10cm plates with 1 :1 EMEM: Ham’s F10 and 10% FBS at 37°C, 5% CO2. Cells were passed at -70% confluence and diluted 1 :4.Promoter construct cloning

[0125] For the PTEN promoter, we cloned 605 nucleotides of genomic sequence reported as the “Sheng promoter” (X. Sheng, et al., Biochem Biophys Res Commun. 2002 Mar 29;292(2):422-6, the disclosure of which is hereby incorporated by reference). Variants of this promoter were synthesized and fragments were cut with Kpnl and Bglll restriction enzymes for cloning into pGL4.23 cut with the same enzymes.Luciferase reporter assays

[0126] Luciferase reporter assays were performed using 2x104 human neuroblastoma BE(2)-C cells / well plated in white 96-well plates with clear bottoms. Twenty-four hours post-plating, cells were transfected with each well receiving 0.2ul liposomes and 0.2ul PEI reagent, 75ng of pGL4.23 firefly luciferase vector containing the 605nt promoter region in Sul of OMEM. Forty-two hours after transfection, luminescence was detected with the luciferase assay system and a plate reader. Four replicates per variant were tested in each experiment and variants were tested in at least three separate experiments.Training models

[0127] The ExPectoSC models are composed of two modules, a deep learning module that was trained to predict epigenetic information from the genome and a linear regression module that translates these effects into gene expression predictions. The first deep learning module of ExPectoSC used the pre-trained BELUGA convolutional neural network, detailed in J. Zhou, et al., (Nat Genet. 2018 Aug;50(8):1171 -1179, the disclosure of which is hereby incorporated by reference) to encode epigenetic signals from more than 2000 ChlP-seq experiments with tissue and cell-line resolution. This module was run on a sequence within a region + / - 20kb around the TSS sites for every protein-coding gene. To capture sequence context, ExPectoSC used a sliding window with 200bp step, each step taking a 2000bp sequence as input. TSS for each gene was defined using CAGE peaks where possible; p1 peaks were used when multiple peaks were defined (N. Bertin, et al., Sci Data. 2017 Oct 3;4:170147, the disclosure of which is hereby incorporated by reference). If no CAGE peaks did not define a TSS, then Gencode TSS was utilized (Release 19, GRCh37; J. Harrow, et al., Genome Res. 2012 Sep;22(9):1760- 74, the disclosure of which is hereby incorporated by reference).

[0128] The resulting matrix U was of shape 2002x201 (2002 epigenetic markers, 201 sliding window steps) for each gene. We then reduced the matrix dimensions using an exponential function, downsampling from 2002x201 to 2002x10 input matrices. The exponential function weighted the middle of the window, giving higher weight to predictions closer to the TSS, by applying the weights of W / 201xl0, to each of the windows,Wi= e-^1with y = [0.01, 0.02, 0.05, 0.01,0.2, 0.2, 0.1, 0.05, 0.02, 0.01] and dtis the sliding window distance to the TSS. The resulting matrix was X = U x W. See Figure S4 for a visual example for a sample gene. The selection of the function was done in previous work (J. Zhou, et al., Nat Genet. 2018 Aug;50(8):1171-1179, the disclosure of which is hereby incorporated by reference).

[0129] The second module translated the deep learning epigenetic predictions to the cell-type-specific expression predictions using linear regression models (Ridge regression), with separate models for each of the cell types.Where gt jis i-th gene expression in j-th cell type, ft is the coefficient for j-th cell type, is the error residuals and wtis the weight assigned to the gene.

[0130] During training, the single-cell RNA-seq expression datasets were used as the labels g j for gene expression outputs. No additional imputation was performed. While we use scRNAseq to refer to all datasets used, ExpetoSC works with either scRNAseq or snRNAseq. The cell clustering, cell identities, scaling and gene expression values were kept as they were defined in the studies that generated the originating scRNA-seq data. Further quality control was applied, and cell clusters were required to have more than 100 cells in order to be used. Any clusters composed of “unclassified” cells or “doublets” were removed. During training, genes located on the X or Y chromosomes were excluded at all stages. For each of the cell-types and each gene, expression was averaged across cells belonging to the same cluster.

[0131] To reduce the impact of non-cell-type-specific genes, we assigned weights (w to genes during training based on the variation across cell types. Genes for which standard deviation across the cell types per organ was high (above 70th percentile) were weighted higher (weight of 2), while genes with low standard deviation across the cell types (std below 10th percentile) were weighted lower (weight of 0.5). Genes between those ranges had a weight of 1. The final prediction scores were then log-transformed, with a pseudocount of 0.01 .

[0132] The reference genome GRCH37 was used for the sequence information. Testing assessment of the models was performed using the genes on chromosomes 8 and 9. The rest of the chromosomal genes (excluding chromosomes X and Y) were used for training and validation. For each model, 5-fold cross validation was performed 5 times, each time leaving out different genes for validation. Separation into folds was random. This allowed us to obtain a large validation dataset without compromising the amount of available training data. Additionally, a grid search was performed over parameter space, which included regularization strength alpha and an option to shift the labels to have zero mean in the log space. Final models used in downstream analyses were trained on all chromosomes (excluding chromosomes X and Y).Test dataset evaluations

[0133] First, we evaluated the models on the test data. We defined two subgroups of genes: high / low expressed genes, and on / off genes. High expressed genes have expression above 75th percentile for the cell type, and low expressed genes were below the 25th percentile for the cell type in the ground truth dataset. Therefore, this definition was based on the relative gene expression within the cell type. When operating on the ground truth, the high expressed genes were labeled as 1 , and low expressed as 0. For the predicted expression values, we ranked and scaled them by the number of genes in total, thus converting ranks into the range (0, 1],

[0134] To further assess cell-type-specificity of gene expression, differentially expressed genes (on / off group) were defined for each of the organ systems. A gene was included in this subset when it was labeled as high in one cell type and low in a different cell type for each of the organs, and assigned a label of 1. The remaining genes were labeled as 0. This allowed us to evaluate the performance of the model on the differentially expressed genes. Variance explained scores were computed using the sklearn Python library, and spearman scores were computed using scipy (F. Pedregosa, et al. J. Mach. Learn. Res. 2011 ); 12:2825-2830; and P. Virtanen, et al., Nat Methods. 2020 Mar;17(3):261 -272; the disclosures of which are hereby incorporated by reference).ExpectoSC variant effect predictions

[0135] To infer the effect of genetic variation, we compared predicted gene expression values produced by the reference and alternative sequences. To stabilize the resulting scores, we exponentiated the predictions to obtain the predicted gene expression level, added a pseudocount of 1 to both predicted gene expression of reference and alternative sequence, as conventionally used in gene expression analysis, then took the log-fold change between them as the final ExPectoSC variant effect prediction, such that the final score was computed as follows:,zexp(predn,t)+1. s = log?( - —7 - ), ^exp(predref ) + where predrefis the prediction for the reference allele and predaltis the prediction for the alternate allele.

[0136] We ran the ExPectoSC framework on the 1000 Genomes data (~2 million variants used in sLDSC) and on Clinvar variants, for the variants that are within + / - 20kbp of any gene TSS (B. K. Bulik-Sullivan, et al., at Genet. 2015 Mar;47(3):291 -5; and M. J. Landrum, et al., Nucleic Acids Res. 2018 Jan 4;46(D1 ):D1062-D1067, the disclosures of which are hereby incorporated by reference). When more than one gene’s TSS was close, we selected the closest one. For Clinvar, we used variants that are labeled as 5 prime UTR variant, intron or 3 prime UTR. For the sLDSC analysis, the absolute value of the effect (s) for predictions for 1000 Genomes was used. The strand of the variant was based on the gene that is considered to be the closest. To compare the predictions between the cell-types, we normalized predictions of variant sets to those of 1000 Genomes variants by using the Z-scores computed per cell-type, such that the effect for the given variant is:(s - Tnean((51, ...,sm)) / std((51, ...,s7n)), where (s1(... , sin) are all 1000 Genomes variant predictions for the given cell type, and s is the unsealed variant effect prediction.Genome-wide heritability partitioning

[0137] To evaluate the predictive ability of ExPectoSC for noncoding genetic variants contributing to disease risk, we used stratified LD-score (s-LDSC) regression to partition the heritability of disease risk to genetic variants. We applied ExPectoSC to compute theeffect of all single-nucleotide variants in the 1000 Genomes data that are within + / - 20kb from any gene for each cell type. In case of a variant within range of more than one gene, the gene closest to the variant was chosen. The ExPectoSC predictions per cell type per SNP were used as SNP annotations in place of the LDSC baseline annotations. LDSC implementation was used for both LD score conversion and LDSC regression. Briefly, genome-wide variant effect predictions were converted to LD scores on a per-cell-type basis using the absolute value of the predicted effect (s), considering LD blocks of European ancestry. LDSC regression was performed in the multivariate mode by regressing all 105 cell-types’ LD scores to GWAS summary statistics. We performed the regression for 102 GWAS studies, where enrichment p-values were computed by LDSC and corrected using Benjamini-Hochberg FDR. GWAS studies for which none of the cell types were significantly enriched (corrected p-value cutoff of 0.05) or at least one of the cell types labeled as significant had enrichment’s standard deviation greater than half of the enrichment itself were removed, leaving 60 studies.Quantification and statistical analysis

[0138] All luciferase assay experiments were performed at least 3 separate times (three biological replicates with starting cells taken from separate cell culture passages). Four replicates per variant were tested in each experiment. For each sequence tested, the firefly luminescence was normalized to the empty vector (pGL4.23 with no insert). Statistics were calculated from the fold over empty vector values in each biological replicate by two-sided t-test, assuming unequal variance. Data from all three experiments were combined for the t-test. Sample size estimation power analyses were not performed before experiments. Researchers were not blinded to samples for experiments.

Claims

WHAT IS CLAIMED IS:1 . A computer-implemented method to evaluate gene expression in a specific cell type, comprising: obtaining, using a computational processing system, a first genetic sequence, wherein the first genetic sequence comprises one or more genes, wherein the sequence of each gene comprises: a sequence structure, and adjacent sequence upstream and downstream the sequence structure; predicting, using the computational processing system and a computational framework, a cell-type-specific expression characteristic of each gene for one or more cell types to yield a predicted cell-type-specific expression characteristic of each gene for each cell type, wherein the computational framework comprises a first computational model to predict epigenetic regulatory markers spatially along the sequence of each gene, wherein the first computational model utilizes gene sequence as input, wherein the computational framework comprises one or more cell-type- specific computational models to predict cell-type-specific expression characteristics of each gene, wherein each cell-type-specific computational model utilizes the epigenetic regulatory markers along the gene sequence as input; and based on the predicted cell-type-specific expression characteristic of the one or more genes, performing a biochemical assay on a first cell to empirically assess an effect of the cell’s genetic sequence on the cell-type-specific expression characteristic of at least one gene of the one or more genes, wherein the cell’s genetic sequence of the at least one gene is the same as the first genetic sequence and the cell type of the first cell is the same cell type as one of the cell-type-specific computational models.

2. The method of claim 1 , wherein the first genetic sequence is collected from a biological sample that is derived from an individual.

3. The method of claim 2, wherein the first genetic sequence has been obtained by sequencing DNA from the biological sample.

4. The method of claim 1 , wherein the first genetic sequence comprises one or more polymorphisms, wherein the computational framework predicts that the one or more polymorphisms first genetic sequence alters an expression characteristic of at least one gene of the one or more genes as compared to a reference sequence lacking the one or more polymorphisms.

5. The method of claim 1 , wherein the epigenetic regulatory markers include at least one of: sites of chromatin accessibility, chromatin marks, and transcription factor binding sites.

6. The method of claim 1 , wherein the first computational model comprises a deep convolutional neural network to predict the epigenetic regulatory markers spatially along the sequence of each gene.

7. The method of claim 6, wherein the deep convolutional neural network is trained on epigenetic regulatory data acquired by at least one epigenetic assay.

8. The method of claim 1 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence for all Pol II transcribed genes of an organismal genome.

9. The method of claim 1 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence in reference to the sequence structure.

10. The method of claim 9, wherein the first computational model predicts the epigenetic regulatory markers up to 1 kb, 2 kb, 3 kb, 4, kb, 5 kb, 7.5 kb, 10 kb, 15 kb, 20 kb, 25 kb, or 50 kb upstream and downstream of the gene sequence structure.11 . The method of claim 1 , wherein the sequence structure is a transcription start site or promoter sequence.

12. The method of claim 1 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence in overlapping sliding sequence windows, wherein the sequence windows are any length between 50 bp and 2000 bp in length, wherein the computational framework considers surrounding sequence context to determine the epigenetic regulatory markers for each sequence window.

13. The method of claim 1 , wherein the computational framework spatially transforms the epigenetic regulatory markers that are determined.

14. The method of claim 1 , wherein each cell-type-specific computational model comprises a regressor model to predict cell-type-specific expression characteristic of each gene.

15. The method of claim 14, wherein the regressor comprises ridge regression.

16. The method of claim 1 further comprising: predicting, using the computational processing system and a computational framework, a cell-type-specific expression characteristic of each gene for two or more cell types, wherein the computational framework comprises two or more cell-type- specific computational models to predict cell-type-specific expression characteristics of each gene, wherein the predicted cell-type-specific expression characteristic of the first cell type, wherein the predicted cell-type-specific expression characteristic of a first gene for a first cell type is a different result than the predicted cell-type-specific expression characteristic of a first gene for a second cell type; andperforming a biochemical assay on a first cell and a second cell to empirically assess the effect of each cell’s genetic sequence on the cell-type-specific expression characteristic of the first gene, wherein each cell’s genetic sequence is the same as the first genetic sequence, the cell type of the first cell is the same cell type as a first celltype-specific computational models, and the cell type of the second cell the same as the cell type of second cell-type-specific computational models, wherein the cell type of the first cell is different than the cell type of the second cell.

17. The method of claim 1 , wherein the first cell is a primary cell collected from a patient or a subject.

18. The method of claim 17, wherein the primary cell is within a system comprising two or more cell types.

19. The method of claim 1 further comprises training each cell-type-specific computational model, wherein the training comprises: generating single-cell expression data for the cell type of the cell-type-specific computational model, wherein generating single-cell expression data for the cell type comprises: performing scRNA-seq or snRNA-seq on a tissue or an organ to yield single-cell sequencing results, wherein the tissue of the organ comprises the cell type; clustering the single-cell sequencing results to yield clusters; identifying the cell type of the clusters, wherein one cluster is identified to be the cell type; and combining the single-cell sequencing results of the cell type cluster to yield the single-cell expression data for the cell type.

20. The method of claim 1 , wherein the biochemical assay is one of: chromatin immunoprecipitation sequencing (ChlP-seq), DNAse I hypersensitivity sequencing (DNase-seq), Assay for Transposase-Accessible Chromatin sequencing (ATAC-seq), Formaldehyde-Assisted Isolation of Regulatory Elements (FAIRE-seq), Hi-C capture sequencing, bisulfite sequencing (BS-seq), methyl array, transgene expression analysis (e.g., luciferase and eGFP), qPCR, RNA hybridization (e.g., ISH), cross-linking immunoprecipitation sequencing (CLIP-seq), RNA immunoprecipitation sequencing (RIP-seq), RNA-seq, western blot, immunodetection, flow cytometry, enzyme-linked immunosorbent assay (ELISA), or mass spectrometry.21 . A computer-implemented diagnostic method for determining a treatment option for an individual for a medical disorder that arises due to gene expression levels in a particular cell type, comprising: obtaining, using a computational processing system, a first genetic sequence, wherein the first genetic sequence comprises one or more genes, wherein the sequence of each gene comprises: a sequence structure, and adjacent sequence upstream and downstream the sequence structure; predicting, using the computational processing system and a computational framework, a cell-type-specific expression characteristic of each gene for one or more cell types to yield a predicted cell-type-specific expression characteristic of each gene for each cell type, wherein the computational framework comprises a first computational model to predict epigenetic regulatory markers spatially along the sequence of each gene, wherein the first computational model utilizes gene sequence as input, wherein the computational framework comprises one or more cell-type- specific computational models to predict cell-type-specific expression characteristics of each gene, wherein each cell-type-specific computational model utilizes the epigenetic regulatory markers along the gene sequence as input; andbased on the determination cell-type-specific expression level of the gene, detecting that a cell type has misregulated expression of the gene; wherein the misregulated expression provides an indication of at least one of: diagnosis of the medical disorder, early detection of the medical disorder before identification of symptoms, personalized medicine diagnosis, a treatment of the disorder to ameliorate a symptom associated with a misregulated pathway in the one of more cell types.

22. The method of claim 21 , wherein the first genetic sequence is collected from a biological sample that is derived from an individual.

23. The method of claim 22, wherein the first genetic sequence has been obtained by sequencing DNA from the biological sample.

24. The method of claim 21 , wherein the first genetic sequence comprises one or more polymorphisms, wherein the computational framework predicts that the one or more polymorphisms first genetic sequence alters an expression characteristic of at least one gene of the one or more genes as compared to a reference sequence lacking the one or more polymorphisms.

25. The method of claim 21 , wherein the epigenetic regulatory markers include at least one of: sites of chromatin accessibility, chromatin marks, and transcription factor binding sites.

26. The method of claim 21 , wherein the first computational model comprises a deep convolutional neural network to predict the epigenetic regulatory markers spatially along the sequence of each gene.

27. The method of claim 26, wherein the deep convolutional neural network is trained on epigenetic regulatory data acquired by at least one epigenetic assay.

28. The method of claim 21 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence for all Pol II transcribed genes of an organismal genome.

29. The method of claim 21 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence in reference to the sequence structure.

30. The method of claim 29, wherein the first computational model predicts the epigenetic regulatory markers up to 1 kb, 2 kb, 3 kb, 4, kb, 5 kb, 7.5 kb, 10 kb, 15 kb, 20 kb, 25 kb, or 50 kb upstream and downstream of the gene sequence structure.31 . The method of claim 21 , wherein the sequence structure is a transcription start site or promoter sequence.

32. The method of claim 21 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence in overlapping sliding sequence windows, wherein the sequence windows are any length between 50 bp and 2000 bp in length, wherein the computational framework considers surrounding sequence context to determine the epigenetic regulatory markers for each sequence window.

33. The method of claim 30, wherein the computational framework spatially transforms the epigenetic regulatory markers that are determined.

34. The method of claim 21 , wherein each cell-type-specific computational model comprises a regressor model to predict cell-type-specific expression characteristic of each gene.

35. The method of claim 34, wherein the regressor comprises ridge regression.

36. The method of claim 21 further comprising: predicting, using the computational processing system and a computational framework, a cell-type-specific expression characteristic of each gene for two or more cell types, wherein the computational framework comprises two or more cell-type- specific computational models to predict cell-type-specific expression characteristics of each gene, wherein the predicted cell-type-specific expression characteristic of the first cell type, wherein the predicted cell-type-specific expression characteristic of a first gene for a first cell type is a different result than the predicted cell-type-specific expression characteristic of a first gene for a second cell type; and detecting that a first cell type of the plurality of cell types has misregulated expression of the gene and a second cell type cell type of the plurality of cell types has proper expression; wherein the combination of misregulated expression of the first cell type and proper expression of a second cell type provides an indication of at least one of: diagnosis of the medical disorder, early detection of the medical disorder before identification of symptoms, personalized medicine diagnosis, a treatment of the disorder to ameliorate a symptom associated with the misregulated pathway in the one of more cell types.

37. The method of claim 21 further comprises training each cell-type-specific computational model, wherein the training comprises: generating single-cell expression data for the cell type of the cell-type-specific computational model, wherein generating single-cell expression data for the cell type comprises: performing scRNA-seq or snRNA-seq on a tissue or an organ to yield single-cell sequencing results, wherein the tissue of the organ comprises the cell type; clustering the single-cell sequencing results to yield clusters; identifying the cell type of the clusters, wherein one cluster is identified to be the cell type; and combining the single-cell sequencing results of the cell type cluster to yield the single-cell expression data for the cell type.

38. The method of claim 21 further comprising administering the treatment to the individual.

39. The method of claim 21 , wherein treating the individual comprises administering a therapeutic that ameliorates at least one symptom associated with the misregulated pathway in the one of more cell types.

40. The method of claim 39, wherein treating the individual comprises administering a therapeutic that modulates expression the misregulated gene.41 . A computer-implemented method to perform site-directed mutagenesis on a particular cell type, the method comprising: obtaining, using a computational processing system, a first genetic sequence, wherein the first genetic sequence comprises one or more genes, wherein the sequence of each gene comprises: a sequence structure, and adjacent sequence upstream and downstream the sequence structure; predicting, using the computational processing system and a computational framework, a cell-type-specific expression characteristic of each gene for one or more cell types to yield a predicted cell-type-specific expression characteristic of each gene for each cell type, wherein the computational framework comprises a first computational model to predict epigenetic regulatory markers spatially along the sequence of each gene, wherein the first computational model utilizes gene sequence as input, wherein the computational framework comprises one or more cell-type- specific computational models to predict cell-type-specific expression characteristics of each gene, wherein each cell-type-specific computational model utilizes the epigenetic regulatory markers along the gene sequence as input;determining using the computational processing system and the computational framework, a set of one or more genetic polymorphisms that alter expression level of the gene in the particular cell type, wherein the set of polymorphisms that alter the expression level of the gene are not present in the first genetic sequence; and based on the determination of the set of polymorphisms that alter the expression level of the one or more genes, performing site-directed mutagenesis on DNA of a cell of the particular cell type to introduce the set of polymorphisms, wherein the cell’s DNA comprises genetic the same set of polymorphisms as the first genetic sequence and the cell type of the first cell is the same cell type as one of the cell-type-specific computational models.

42. The method of claim 41 , wherein the first genetic sequence is collected from a biological sample that is derived from an individual or subject.

43. The method of claim 42, wherein the first genetic sequence has been obtained by sequencing DNA from the biological sample.

44. The method of claim 41 , wherein the first genetic sequence comprises two or more genes, wherein the first genetic sequence comprises two or more polymorphisms, wherein at least one polymorphism is within a sequence of a first gene and at least one polymorphism is within a sequence of a second gene.

45. The method of claim 41 , wherein the epigenetic regulatory markers include at least one of: sites of chromatin accessibility, chromatin marks, and transcription factor binding sites.

46. The method of claim 41 , wherein the first computational model comprises a deep convolutional neural network to predict the epigenetic regulatory markers spatially along the sequence of each gene.

47. The method of claim 46, wherein the deep convolutional neural network is trained on epigenetic regulatory data acquired by at least one epigenetic assay.

48. The method of claim 41 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence for all Pol II transcribed genes of an organismal genome.

49. The method of claim 41 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence in reference to the sequence structure.

50. The method of claim 49, wherein the sequence structure is a transcription start site or promoter sequence.51 . The method of claim 41 , wherein the sequence structure is a transcription start site or promoter sequence.

52. The method of claim 41 , wherein the first computational model determines the epigenetic regulatory markers spatially along the genetic sequence in overlapping sliding sequence windows, wherein the sequence windows are any length between 50 bp and 2000 bp in length, wherein the computational framework considers surrounding sequence context to determine the epigenetic regulatory markers for each sequence window.

53. The method of claim 41 , wherein the computational framework spatially transforms the epigenetic regulatory markers that are determined.

54. The method of claim 41 , wherein each cell-type-specific computational model comprises a regressor model to predict cell-type-specific expression characteristic of each gene.

55. The method of claim 54, wherein the regressor comprises ridge regression.

56. The method of claim 41 , wherein the site-directed mutagenesis is performed by one of: CRISPR mutagenesis, Zinc-finger mutagenesis, or TALEN mutagenesis.

57. The method of claim 41 , generating single-cell expression data for the cell type of the cell-type-specific computational model, wherein generating single-cell expression data for the cell type comprises: performing scRNA-seq or snRNA-seq on a tissue or an organ to yield single-cell sequencing results, wherein the tissue of the organ comprises the cell type; clustering the single-cell sequencing results to yield clusters; identifying the cell type of the clusters, wherein one cluster is identified to be the cell type; and combining the single-cell sequencing results of the cell type cluster to yield the single-cell expression data for the cell type.

58. The method of claim 41 , wherein the cell of the particular cell type is within one of: a primary cell derived from an individual or subject, an ex vivo cell derived from an individual or subject, or an in vivo cell of an individual or a subject.

59. The method of claim 41 , wherein the set of polymorphisms is identified from one or more genetic sequences associated with a clinical or a medical phenotype.

60. The method of claim 41 , wherein the cell of the particular cell type is a primary cell or an ex vivo cell derived from a patient with a medical disorder, wherein the method further comprises: transplanting the primary cell or the ex vivo cell back into the patient after site- directed mutagenesis is performed.

Images