Methods and systems for drug response and disease network reconstruction and uses thereof

Abstract

To provide a method that characterizes drug responses of a human being.SOLUTION: A method uses a characteristic of a functional topology of a three-dimensional architecture of a drug modulation spatial contact in a chromatin space, in which the use is conducted through a pharmacological genomics network using various data sources and metric on the basis of selection of candidates SNP by imputation, determination of a predicted causal relationship of the SNP using machine learning and deep layer learning, use of the SNP serving as a cause to probe a spatial genome determined by a chromosome conformation capture analysis combining a target gene controlled by the same cell and tissue-specific enhancer, and a result of a related research of an entire genome. The use is configured to decompose the pharmacological genomics network down to constituent effectiveness of the network, and an adverse event subnetwork using a segmentation method of a knowledge base; and apply to a clinical decision support, a diversion of the drug, and silico drug discovery.SELECTED DRAWING: Figure 12

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to and benefit of the filing date of (1) U.S. Provisional Patent Application No. 62 / 795,705, entitled "Methods and Systems to Reconstruct Drug Pharmacogenomic Networks from Pharmacogenomic Regulatory Interactions and Uses Thereof," filed January 23, 2019, and (2) U.S. Provisional Patent Application No. 62 / 795,710, entitled "Companion Diagnostic Assays for N-methyl-D-Aspartate Receptor Modulators," filed January 23, 2019, the entire disclosures of each of which are expressly incorporated herein by reference. [Technical Field]

[0002] The techniques described herein involve discovering gene network contacts in chromatin space that define the pharmacogenomic substrate of a particular drug, identifying functionally distinct gene sets within this drug's pharmacogenomic network, called subnetworks, and detecting regulatory genomic variants within the drug's subnetwork gene sets that influence treatment efficacy or adverse events. Methods are described for applying these results to characterize human drug response for clinical decision support, drug repurposing, including the development of novel companion therapeutics, and in silico drug target discovery. [Background technology]

[0003] Spatial pharmaco-epigenome, super-enhancers, and topologically associated domains New insights into the architecture and dynamics of noncoding regulatory genomes have transformed conventional views of pharmacodynamics and pharmacokinetics. The noncoding regulatory genome, whose variations influence human drug response, is hereafter referred to as the "pharmacoepigenome." The pharmacoepigenome can be defined as the active noncoding domain of the human genome, consisting of spatial, temporal, and mechanistic regulatory mechanisms of gene regulation in response to xenobiotic stimuli. It contains regulators of gene expression, including enhancers, promoters, and regulatory RNAs, and is characterized by a stereotypical hierarchy of transcriptional domains, whose mutations significantly impact human drug response. Transcriptional regulation is organized into canonical 3D structures, including enhancer-promoter pairs, super-enhancers, transcription hubs, mRNA splicing factors, topologically associated domains (TADs), and lamina-associated domains (LADs). Specific localized sets of canonical 3D structures are activated or repressed in a cell-type-specific manner. Because drug-disease networks are tightly coupled, genetic variants significantly associated with disease are often identical to or within the same regulatory network that determines the outcome of pharmaceutical-based treatments. Therefore, mutations that disrupt the spatial hierarchy of transcription within euchromatin convey not only disease risk but also the concomitant variability in drug response. The pathways encompassing variants of disease risk, drug response, and associated adverse events are fertile networks for discovering new drug targets using genotype / phenotype-based computational strategies. These insights will better inform patient treatment options based on the novel pharmacological basis of drug response and adverse drug events. Examples of future therapeutic strategies include combinatorial drug design targeted at pharmacogenomic networks, integrative multiscale analytical methods that draw from a diverse set of data types to enhance drug discovery based on stratification by molecular modifications per pharmaco-epigenomic environment, synthetic editing that alters non-coding regulatory elements that convey drug treatment resistance, and the development of transcription factor-like molecules for cellular reprogramming of tissue injury and atrophy.

[0004] Examining hundreds of thousands of humans from both genome-wide association studies (GWAS), phenome-wide association studies (PheWAS), and other biobanked patient data containing SNPs that convey disease risk, as well as individual responses to specific drugs, it is important to note that highly significant SNP-trait associations are found within non-coding genomic regulatory elements called enhancers. Enhancers often target gene promoters or regulatory RNAs within the same TAD and can be controlled by larger regulatory elements called super-enhancers.

[0005] Human genes that play important roles in health, disease, and drug response are often regulated by long DNA elements spanning two or more TADs, termed "super-enhancers" or "stretch enhancers," herein known as super-enhancers. Super-enhancers are clusters of enhancers that are populated by an unusually high density of interactors and activate differential transcription (also known as gene expression) at a higher frequency than that exhibited by typical enhancers. Super-enhancers are multimolecular assemblies that represent macromolecular condensates similar to nucleoli, centralizing and compartmentalizing transcriptional regulation within the nucleus of cells. Super-enhancers occupy known genomic locations that span multiple TADs and LADs in a cell- and developmentally specific manner.

[0006] Mutations that alter super-enhancers, disrupt gene and RNA regulation, abolish or alter chromatin loops between enhancer-promoter or promoter-promoter pairs, and / or disrupt TAD boundaries or disperse repressive subsets of TADs called LADs have serious consequences for variability in drug response and the occurrence of adverse drug events in the human population.

[0007] The largest pharmacogenomic effect sizes are seen in patients with SNPs, single-base changes that disrupt super-enhancers and cause life-threatening acute adverse drug events. Examples include clozapine-induced agranulocytosis / granulocytopenia and Stevens-Johnson syndrome, or toxic epidermal necrolysis caused by carbamazepine, lamotrigine, phenobarbital, allopurinol, nonsteroidal anti-inflammatory drugs, and certain other medications. These side effects are sufficiently severe that countries such as Singapore and Taiwan require patients to be tested for the presence of these SNPs before administering these medications.

[0008] Super-enhancers play a role in specifying the identity of various cell types during development and, in tissues such as the brain, serve as platforms for the binding of neural-specific transcription factors and mediator complexes. They represent nontraditional pharmacodynamic targets, and their involvement in differential neurogenesis in the adult brain is also a mechanism by which histone deacetylase inhibitors exert their actions in the CNS. Similarly, unconventional interpretation of drug response and remission single nucleotide polymorphisms (SNPs) from GWAS and PheWAS has significantly improved our understanding of how mutational perturbations of cellular molecular physiology lead to human pharmacogenomic variation.

[0009] The spatial hierarchy of transcriptional organization was first determined using chromatin conformation capture (CQC) techniques. Chromosomes fill most of the available volume of the nucleoplasm as chromosome regions (CTs), which contain localized A and B compartments composed of euchromatin and heterochromatin, respectively. Generally, compartment A contains euchromatin and more active gene transcription, while compartment B corresponds to heterochromatin and is gene-poor. Compartment B contains LADs located at the nucleus periphery. These appear to be largely invariant features of chromatin organization, as they are specific to chromosome regions and are not disrupted when TAD or LAD organization is disrupted using genome editing techniques. The A and B chromatin compartments of the CQC contain approximately 2,450 TADs, with an average linear sequence length ranging from 100 kbp (kilobase pairs) to 5 mbps (megabase pairs). TADs were first characterized using chromatin conformation capture (HCC) techniques, enabling high-resolution studies of enhancer-promoter loops within TADs and the organization of TAD boundary proteins.

[0010] Chromosomes are contained within the nucleoplasm of differentiated cells as large, rope-like coils of genomic DNA encapsulated in chromatin. CTs exist in 3D space, where spatial proximity and chromatin state, rather than distance measured by linear DNA sequence, determine regulatory interactions. Although CTs do not overlap significantly, there are multiple spatial interactions between different functional CTs. These include complex transcription hubs consisting of multiple genes, regulatory elements including enhancers and promoters, and functionally associated DNA-binding proteins such as transcription factors. Trans interactions include enhancer-promoter interactions or, in some cases, interchromosomal spatial contacts involving promoter-promoter pairs.

[0011] Drugs modify localized TADS in a cell-type-specific manner Most of the human genome is segregated into approximately 2,450 basic transcription units called TADs, yet approximately 5% of expressed genes and functional long non-coding RNAs are not found within these bounded 3D structures. TADs are delineated by localized boundaries, often contain multiple functionally related genes controlled by intra-TAD enhancers, and are invariant across all cell types studied to date. In most cases, few enhancers cross TAD boundaries unless the TAD boundaries are disrupted by SNPs or other genetic mutations. Differences in gene expression between different cell types are a function of the TADs activated or repressed in that cell type. TADs exhibit specific histone modifications and are units of DNA replication timing. Specific localized sets of TADs and their trans-interacting TADs constitute co-regulatory modules for drug and hormone responsiveness. TAD boundaries are relatively invariant across different human cell types. TAD boundary strength can be classified into five distinct domains based on the amount of CTCF bound to the boundaries and whether super-enhancers co-locate at TAD boundaries.

[0012] Regulatory pharmacogenomics determines drug pharmacogenomics networks Recent studies have revealed several fundamental principles of pharmacoepigenomics: (1) GWAS and PheWAS results indicate that over 90% of causative single nucleotide polymorphisms (SNPs) are located within regulatory enhancers, and approximately 5% are located within protein-coding exons; (2) in adults, chromatin contacts between enhancers and promoters or cooperating promoters always precede both gene transcription and alternative splicing of protein-coding mRNAs; (3) histone modifications indicate the regulatory state of a given genomic regulatory element or gene; and (4) in all cases studied to date, causative genetic variants exhibit allele specificity, regardless of whether the cells are diploid, tetraploid, or octaploid. Numerous studies have demonstrated that genetic mutations, such as SNPs located within enhancers, can be predicted as causative using machine learning algorithms trained on DNase I hypersensitivity, which indicates allele specificity and other features of the epigenome, including histone modifications associated with enhancers and promoters. The accuracy of the clinical utility of these machine learning applications has been validated using known causative SNPs and compared to the output of these software programs.

[0013] Recent studies from our laboratory and those of others have demonstrated the existence of a new class of pharmacodynamic and pharmacokinetic master regulator networks in chromatin that function to activate and repress large sets of interconnected genes that contact them in the chromatin space. These regulators of pharmacogenomic regulatory networks, distinct from previous generations of epigenetic drugs composed of writers, readers, and erasers, represent a new class of druggable targets in humans.

[0014] Enhancer and super-enhancer SNPs associated with disease risk and drug response are key to the discovery of drug pharmacogenomic networks. Causative mutations, such as SNPs found within enhancers, promoters, and splice sites, significantly alter the state of chromatin and can be used as "data probes" to discover drug networks within the 3D spatial environment of chromatin located in the cell's nucleus. While published literature exists describing methods for developing drug networks, current approaches to gene-gene and protein-protein regulatory pathways, even in complex tissues such as the human brain, are problematic because: (1) they are based on the assumption that mutations within protein-coding genes and protein-coding exons represent the majority of biologically relevant key mechanisms; and / or (2) they seek similarities in the structure and catalytic properties of new compounds that mimic those of FDA-approved drugs for a given indication, or match tissue-specific gene expression patterns with those of FDA-approved drugs for a given indication. Recent studies have shown that neither of these assumptions is particularly accurate for discovering new psychotropic drug candidates that offer superior efficacy and fewer adverse events than existing drugs. First, the most important SNP trait associations for disease risk and drug response alter the function of enhancers located within the non-coding genome, not proteins, and genetic variants within gene introns disrupt intragenic enhancers that may or may not regulate the expression of the gene in which they are located. Because SNPs located within protein-coding exons often disrupt alternative splicing of mRNA or may disrupt enhancers, methods for predicting a priori that missense SNPs will alter protein products are not accurate. Furthermore, many functional RNAs exist in the human genome, including long non-coding RNAs that are not translated into proteins. Second, the "guilt by association" approach used in programs such as the Integrated Network-Based Cellular Signature Library (LINCS) program is based entirely on gene expression profiles of cell lines as surrogates for discovering new drugs for human tissues such as the brain.This complexity of human tissues requires a more nuanced and comprehensive investigative approach than can be provided by surrogates using cell line-dependent "shotgun" expression profiling. Summary of the Invention

[0015] Methods and systems for detecting human regulatory drug networks using bioinformatics and computational techniques such as machine learning and deep learning. The foundation of these methods is the ability to uncover previously unrecognized drug-pharmacogenomic networks through the investigation of pharmacogenomic regulatory interactions embedded within the functional three-dimensional (3D) topology of the human genome, using mutations that stratify drug response in large human populations. These spatial regulatory interactions provide the architecture for the pharmacogenomic networks of most psychotropic and antitumor drugs.

[0016] There is now a vast amount of existing data that can be used to computationally map drug pathways without the use of sophisticated probabilistic inference methods, in lieu of additional experiments in animal and cell models. These knowledge-based methods described herein can be used to reconstruct drug pharmacogenomic networks acting in different cell types and tissues and decompose these networks into component parts that mediate different on-target and off-target mechanisms of drugs, which are subsequently validated using bioinformatics analysis.

[0017] These methods differ from studies that require experimental perturbation of cell or tissue biology following drug exposure, or that rely entirely on the centrality of learning machines for pathway mapping. A key part of the process of determining single nucleotide polymorphisms (SNPs, which can be single base pair changes or short insertions / deletions) that are significantly associated with a specific drug response is the use of various machine learning algorithms to determine possible mechanistic causal relationships. Nevertheless, the primary mapping methods are based on 3D genome structures and existing knowledge bases drawn from multiple public and / or experimental or proprietary data sources.

[0018] Figure 1D shows an exemplary model of how the system integrates and processes multiscale data using machine learning and deep learning for pharmacogenomic network reconstruction. This strategy for mapping drug networks provides insight into mechanistic on-target and off-target effects, laying the foundation for subsequent preclinical testing.

[0019] Figure 1E illustrates a method for detecting drug pharmacogenomic networks in humans that can be performed by a server device. The first step of this method involves extracting significant SNPs associated with specific drug responses. The majority of these SNPs have been published in genome-wide association studies (GWAS) and phenome-wide association studies (PheWAS), and there is a wealth of unbiased, peer-reviewed scientific publications from which such data can be obtained. To improve the accuracy of SNP location, this is processed by the server device using an automated pharmacoepigenomics informatics pipeline (PIP), as described with reference to Figures 4B, 4D, and 4E of U.S. Patent Application No. 15 / 977,347, filed May 11, 2018, which is incorporated herein by reference. Once imputation and annotation are performed to further characterize the SNPs in relation to tissues, multiple accurate and validated machine learning algorithms trained on causal disease SNPs are applied to determine possible mechanistic causal relationships. Additionally, missense SNPs, synonymous SNPs, and SNPs located within exons that may be splice site donors or acceptors are characterized using machine learning. The output of this pipeline is a set of "allowed" candidate SNPs that have been shown to stratify drug response within human populations to a specific drug of interest. The next method step, performed by the server device, involves performing spatial genomic classification using these causative SNPs to identify target genes within the same TAD as these SNPs in the case of enhancers. These enhancer SNPs are then used to identify statistically significant spatial contacts of the top-ranked (e.g., top three) resident TADs within the genome through analysis of datasets generated using chromosome conformation capture methods (most commonly generated from Hi-C methods). If the causative enhancer SNP is located in a TAD with an empirically determined strong boundary strength of III-V, characteristic of TAD boundaries containing genes involved in drug absorption, distribution, metabolism, and excretion (ADME), all genes within that TAD controlled by the same enhancer within that TAD are retained for further evaluation.Similarly, if the top ranking of statistically significant contacted “trans TADs” in the spatial genome contain genes controlled by enhancers active in the same cell type and / or tissue in which the drug acts, they are also kept for further evaluation.

[0020] The candidate gene set, including both intra-TAD and inter-TAD genes, is then evaluated for known network connections, using, for example, pathway analysis software from third-party software. Genes that form statistically significant interconnected pathways, most commonly determined using Fisher's exact test, and are expressed in the tissue of interest for the drug of interest, constitute a preliminary set of candidate spatial network genes. Genes that are not significantly interconnected with other genes are discarded. This constitutes a preliminary set of spatial network genes for the specific drug.

[0021] Semi-automated and automated curation of the knowledge base is then performed on this set, which contains the spatial network of a specific drug, to evaluate genes that should be added or removed. First, each member of the gene set is thoroughly examined in relation to its defined function, including those from major scientific publications that evaluated its function in relation to the specific drug of interest. Second, the entire set of known mutations within each gene, defined by linear distance ±10 kilobases (Kb) from its transcription start site and stop codon, is evaluated for their impact on the known efficacy and adverse event mechanisms of the specific drug of interest. Mutations include SNPs, variable number of tandem repeats, duplications, and large insertions or deletions. In this context, any functional relationship to physiological processes related to the efficacy or adverse events of the specific drug of interest is also included in the evaluation process. This is not limited to the influence of pharmacogenomics on specific drug responses. Third, in complex tissues such as the human brain, the expression pattern of each gene is compared to the neuroanatomical substrates on which the specific drug of interest is known to act from other studies. For example, in the reconstruction of the ketamine spatial network, datasets from 24 functional neuroimaging studies are examined to determine which brain regions are metabolically active after ketamine administration in humans. All genes in a preliminary set of ketamine spatial network genes are examined to see whether their expression in the human brain overlaps with a consensus neuromap derived from 24 functional neuroimaging studies detailing the neuroanatomical substrates of ketamine action in the human brain. To accomplish this task, microarray expression and in situ hybridization results from the Allen Brain Science Institute's Human Brain Atlas and RNA-seq results from the National Institutes of Health's GTEx program are interrogated against the neuroanatomical neuromap for each gene in the human brain. Genes whose expression patterns do not fit the consensus neuroanatomical neuromap are discarded.

[0022] The drug pharmacogenomics network gene set is then reassessed using pathway analysis, e.g., via third-party software, to determine whether each gene has a known network connection. Genes that form statistically significant interconnected pathways, most commonly determined using Fisher's exact test, and are expressed in the tissue of interest for the drug of interest constitute a preliminary set of candidate spatial network genes. Genes that are not significantly interconnected with other genes are discarded. This constitutes the final set of genes for the spatial network of a particular drug.

[0023] The next step of this method involves applying iterative gene set optimization tools and algorithms to organize the spatial network genes into functional subsets of genes, some of which include drug efficacy and adverse drug event subnetworks within the larger gene set. This involves measuring the similarity of input molecules from one or more diverse data sources involved in the mechanism of action of a particular drug, converting them to standardized human gene nomenclature, and comparing them with the genes of the drug pharmacogenomics network. The output of this process is the entire set of genes in the spatial network for a particular drug organized into its constituent subnetworks, including the efficacy and adverse event subnetworks.

[0024] The next step of this method involves providing scientific validation of the spatial network of a particular drug organized into constituent subnetworks, e.g., using third-party software applications for bioinformatics and biostatistics. These include examples of the spatial network and its subnetworks annotated using top-ranked (e.g., top 5) statistically significant terms from pharmaceutical databases such as Gene Ontology or MedDRA, top-ranked (e.g., top 5) canonical pathways determined by pathway analysis such as commercial or open-source pathway analysis software programs, top upstream xenobiotic regulators, and statistically significant SNP-trait associations from GWAS and PheWAS.

[0025] After validation is performed, the spatial network for a particular drug and its constituent sub-networks may be stored in a database and provided to a client device for display.

[0026] The drug spatial network and its constituent subnetworks can be applied in several contexts. For example, different embodiments are presented in pharmacogenomic decision support for drug selection, drug repurposing, and in silico drug target discovery. One embodiment of clinical decision support is a method for matching a reference drug pharmacogenomic network and its efficacy and adverse event subnetworks selected from a database of such spatial networks with a patient's specific drug efficacy and adverse event subnetworks. This comparison uses deep learning methods in which efficacy metrics are jointly trained between the reference and patient subnetworks and pattern matching scores are generated. The output is a separate drug efficacy similarity score and a drug adverse event similarity score. It should be noted that those skilled in the art will recognize that the reference drug pharmacogenomic network and its constituent efficacy and adverse event subnetworks do not represent optimal profiles. Instead, they reflect the entire mechanism of action of the drug, encompassing both the best and worst possible effects that the drug may have on an individual patient.

[0027] An example of in silico drug discovery is the selection of a gene member of the gene set within the ketamine spatial network, the PPP1R1B gene, which is regulated by the same enhancer that controls the NEUROD2 gene, a gene whose protein product is involved in neurogenesis and in critical spatial contacts with trans-TADs, including the genes DRD2 and ADORA2A. Using the methods described herein to map the gene set interconnected with the PPP1R1B gene and assessing gene ontology top terms and canonical pathways associated with the pathway, we demonstrated that the PPP1R1B gene is highly significantly involved in central nervous system (CNS) development, neural differentiation, and neurogenesis. Furthermore, the PPP1R1B gene is expressed in a restricted set of human brain regions, including the anterior caudate nucleus, nucleus accumbens, and putamen, and most of the 24 genes are significantly interconnected with genes that are neuroanatomical substrates involved in reward and addiction. Finally, PPP1R1B encodes a druggable phosphoprotein defined as a "bifunctional signaling molecule." Dopaminergic and glutamatergic receptor stimulation modulates its phosphorylation, allowing it to function as a kinase or phosphatase inhibitor. As a dopamine target, this gene may serve as a therapeutic target for neurological and psychiatric disorders. This represents a potentially druggable target identified using these methods.

[0028] The results of these methods include spatial networks for the drugs ketamine, valproate, lithium, lamotrigine, clozapine, and warfarin. Post-mortem validation of these pharmacogenomic networks using bioinformatics methods, as well as knowledge-based segmentation of their efficacy and adverse event subnetworks using the methods of the present disclosure, are provided. Details regarding specific efficacy and adverse event subnetworks are also provided to explain the output of the pharmacogenomic network identification system. [Brief explanation of the drawings]

[0029] [Figure 1A]FIG. 1 is a block diagram of a computer network and systems in which an exemplary pharmacogenomics network identification system can operate in accordance with embodiments described herein. [Figure 1B] FIG. 1B is a block diagram of an exemplary pharmacogenomics network server capable of operating in the system of FIG. 1A in accordance with embodiments described herein. [Figure 1C] 1B is a block diagram of an exemplary client device capable of operating in the system of FIG. 1A in accordance with embodiments described herein. [Figure 2] Figure 1 shows an exemplary model of how the system integrates and processes multiscale data using machine learning and deep learning for pharmacogenomic network reconstruction. This strategy for mapping drug networks provides insight into mechanistic on-target and off-target effects, laying the foundation for subsequent preclinical studies. [Figure 3] Figure 1 shows examples of gene promoters, enhancers, super-enhancers, and TADs containing structural proteins contained within the TAD boundaries, and subsequent chromatin looping of the promoter to different exons during alternative splicing of the gene. [Figure 4A-4B] FIG. 1 shows that the nature of drug expansion of adjacent TADs, acting through activation of enhancers and / or super-enhancers, results in differential gene expression. [Figure 4C] FIG. 1 shows that the TAD structure of the human genome provides more accurate information about the localization of target genes of enhancers and / or super-enhancers than traditional measurements of linkage disequilibrium within the human population. [Figure 5A] FIG. 1 shows a "ball of string" model of the chromatin organization of the human genome within the cell nucleus, including spatial interactions of chromatin. [Figure 5B]Figure 1 shows a simple drug network with three super-enhancers regulating six TADs and four TADs lacking super-enhancer regulation, as well as their trans-interactions in the spatial genome after exposing the "ball of thread" to drugs. [Figure 6] This figure shows a simple example in which a SNP located within an enhancer in a network can disrupt contact between the enhancer and one of its target gene promoters within a TAD, resulting in an adverse drug event in a patient within a drug-responsive cohort. Figure 6A illustrates how various experimental methods can be used to obtain measurements in three dimensions from the chromatin spatial interactome and analyze the data as a two-dimensional plot of enhancer-gene promoter interactions. Figure 6B illustrates how a SNP can disrupt a chromatin loop between the enhancer and one of the two gene promoters it regulates within the TAD. This disruption results in a loss of spatial connectivity between the enhancer and gene promoter 1, resulting in dysregulation of gene 1 and resulting in an adverse drug event in this patient and its cohort in response to administration of a particular drug of interest. [Figure 7] FIG. 1 shows spatial genomic features, including several enhancers in each TAD located within non-coding genomic DNA (i.e., intergenic or intronic) that selectively activate or repress specific functionally related genes within that TAD. [Figure 8A] FIG. 1 shows the nature of significant associations between ADME genes and human super-enhancers. [Figure 8B] FIG. 1 shows the association between non-coding mutations within super-enhancers that can significantly alter psychotropic drug response. [Figure 9] Figure 1 shows an example comparing the results of significance tests for SNP rs12967143-G, an intragenic enhancer located within the TCF4 gene, against other GWAS SNPs between various neuronal and non-neuronal cell types, illustrated using the numerical outputs from six different machine learning algorithms used in the analysis (*p≦0.05; **p≦0.01; ANOVA). [Figure 10]FIG. 1 shows that the TAD containing the PK and HLA gene clusters has strong TAD boundaries and is associated with important biological processes as determined by Gene Ontology. [Figure 11] FIG. 1 is a flow diagram representing a method for generating a reconstructed pharmacogenomic network and corresponding subnetworks for a drug of interest, including a human pharmacogenomics SNP input filter, a pharmacogenomics network reconstruction engine, and an iterative gene set optimization engine that outputs drug efficacy and adverse event subnetworks. [Figure 12] FIG. 1 is a flow diagram depicting an exemplary method of iterative gene set optimization for decomposing a pharmacogenomics network into sub-networks. [Figure 13] FIG. 1 is a flow diagram representing an exemplary method for post-mortem validation of drug pharmacogenomic networks and their constituent sub-networks using standardized bioinformatics analysis. [Figure 14] FIG. 1 is a flow diagram depicting an exemplary method for error correction of a pharmacogenomics network and its constituent efficacy and adverse event sub-networks utilizing individual patient response data. [Figure 15] FIG. 1 is a flow diagram representing an exemplary method for using similarity scores to match a patient's drug efficacy and adverse events with those of a reference pharmacogenomics network for optimizing medication selection in clinical decision support. [Figure 16A] FIG. 1 is a flow diagram depicting an exemplary method for in silico drug target identification and drug repurposing of druggable target PPP1R1B. [Figures 16B-16C] Figure 1 shows some characteristics of the druggable target PPP1R1B within neurodevelopmental and antidepressant mechanistic subnetwork 2 of the ketamine spatial network. Figure 2 shows the characteristics of the ketamine pharmacogenomics network determined from post-hoc validation of the ketamine pharmacogenomics network and its efficacy and adverse event subnetworks. [Figure 16D]FIG. 1 shows gene expression data of key pharmacogenomic efficacy genes in relevant brain tissue regions. [Figure 17A] FIG. 1 shows a general topological model of CNS and peripheral drug responses, including chromatin remodeling, PK / hormonal regulation, efficacy, adverse events (AEs), systemic PK, systemic AEs, and immune system responses. [Figure 17B]

[0013] Figure 1 illustrates four pharmacogenomics network topology models defining psychotropic and anti-tumor drug responses, an exemplary set of their constituent subnetworks, and exemplary pharmaceuticals that fit those topologies. These topologies are used by the systems described herein. [Figure 18] FIG. 1 shows a graphical representation of the valproic acid pharmacogenomic network and its constituent sub-networks, including chromatin remodeling, efficacy, adverse events, and hormonal regulation and pharmacokinetics in the human brain, using the methods and systems of the present invention. [Figure 19A] FIG. 1 shows the most significant disease annotations of the valproic acid pharmacogenomics network. [Figure 19B] FIG. 1 shows the top 10 drugs that are upstream regulators of the valproic acid pharmacogenomics network. [Figure 19C] FIG. 1 shows the topological model that most accurately fits the valproic acid pharmacogenomic network. [Figure 20] FIG. 1 shows an example of the valproic acid pharmacogenomic adverse event subnetwork. Post-hoc bioinformatics analysis indicates that the valproic acid pharmacogenomic adverse event subnetwork is significantly associated with cancer, severe psychiatric disorders, cognitive disorders, gastrointestinal disorders, lymphoproliferative disorders, movement disorders including tremor, and alopecia. [Figure 21]Diagram showing an example of the valproic acid pharmacogenomic neurogenesis subnetwork. Post-hoc bioinformatics analysis shows that valproic acid pharmacogenomic neurogenesis is significantly associated with neuronal mass, morphogenesis, neuronal proliferation, neuronal differentiation, embryonic tissue differentiation, epilepsy or neurodevelopmental disorders, cognitive impairment, mood disorders, Alzheimer's disease or frontotemporal dementia, and migraine. [Figure 22] 22A-22B show examples of disease risk and pharmacogenomic SNPs from GWAS that can be used to determine the propensity of an individual patient to experience an adverse event following valproic acid therapy, as shown in FIG. 22A, or the efficacy response, as shown in FIG. 22B. [Figure 23] Figure 1 shows the overlap between the output using this system and method and four other experiments and existing data sources containing significantly differentially expressed genes from pig (Sus scrofa) brain after peripheral administration of 150 mg / kg valproic acid, as well as drug databases including Ingenuity Pathway Analysis™, KEGG, DrugCentral, DrugBank, and LINCS. Note that this system outputs more shared valproic acid-induced genes than the other two comparisons. [Figure 24] FIG. 1 shows a list of genes included in the chromatin remodeling subnetwork of the valproic acid pharmacogenomics network. [Figure 25] FIG. 1 lists the genes included in the neuroplasticity and efficacy subnetworks of the valproic acid pharmacogenomics network. [Figure 26] FIG. 1 shows a list of genes included in the adverse event subnetwork of the valproic acid pharmacogenomics network. [Figure 27] FIG. 1 lists the genes included in the pharmacokinetic and hormone subnetworks of the valproic acid pharmacogenomics network. [Figures 28A-28I]Figure 1 shows selected chromatin spatial contacts of the valproic acid pharmacogenomics network and its functional network determined by chromosome conformation capture using Hi-C methods in human neurons. [Figure 29A] FIG. 1 shows the most significant disease annotations of the ketamine pharmacogenomics network. [Figure 29B] FIG. 1 shows the top five drugs that are upstream regulators of the ketamine pharmacogenomics network. [Figure 29C] FIG. 1 shows the topological model that most accurately fits the ketamine pharmacogenomic network. [Figure 30] Figure 30 shows an example of gene set enrichment in the output of the gene set optimization engine, which distinguished two significantly different subnetworks within the three subnetworks that comprise the ketamine pharmacogenomic network of the human brain. Figure 30A shows the ketamine pharmacogenomic glutamate receptor subnetwork responsible for drug- and neurotransmission-related adverse events. Figure 30B shows the ketamine pharmacogenomic neuroplasticity subnetwork mediating the antidepressant response of ketamine. [Figure 31] FIG. 1 shows an example of the ketamine pharmacogenomic glutamate receptor subnetwork. Post-hoc bioinformatics analysis indicates that the ketamine pharmacogenomic glutamate receptor subnetwork is significantly associated with the following adverse events (AEs): cognitive impairment, bipolar disorder, postoperative delirium, schizoaffective disorder, schizophrenia, non-cancer pain, postoperative pain, vomiting, nausea, and loss of consciousness. [Figure 32] Figure 1 shows an example of the ketamine pharmacogenomic neuroplasticity subnetwork. Post-hoc bioinformatics analysis indicates that the ketamine pharmacogenomic neuroplasticity subnetwork is significantly associated with emotional behavior, abnormal nervous system morphology, abnormal brain morphology, depression, anxiety, and abnormal neuronal morphology. [Figure 33]33A-33B show examples of GWAS disease risk SNPs that can be used to determine an individual patient's propensity to experience adverse events following ketamine therapy, as shown in FIG. 33A, or the effectiveness of an antidepressant response, as shown in FIG. 33B. [Figure 34] FIG. 1 lists genes included in the neuroplasticity and efficacy subnetworks of the ketamine pharmacogenomics network. [Figure 35] FIG. 1 lists genes included in the chromatin remodeling and adverse events subnetworks of the ketamine pharmacogenomics network. [Figure 36] FIG. 1 lists the genes included in the pharmacokinetic and hormone subnetworks of the ketamine pharmacogenomics network. [Figures 37A-37G] Figure 1 shows selected chromatin spatial contacts across the ketamine pharmacogenomics network as determined by chromosome conformation capture using Hi-C methods in human neurons. [Figure 38] Figure 1 shows the significant overlap between the neuroanatomical distribution of gene expression data within the ketamine pharmacogenomics network and localization results from a consensus brain map showing brain regions primarily affected by ketamine derived from 24 neuroimaging studies. [Figure 39] Figure 1 shows examples of beneficial combination mechanisms and therapies discovered using the methods of this system to combine neurogenesis and neural differentiation using valproic acid and ketamine for H3K9 acetylation and deacetylation, respectively. [Figure 40] 40A and 40B show complementary pharmacogenomic networks of valproic acid (FIG. 40A) and ketamine (FIG. 40B), showing neurogenesis and neuronal differentiation, respectively. [Figure 41] FIG. 1 shows the combinatorial effects of valproic acid and ketamine pharmacogenomic networks on neurogenesis, neuronal proliferation and terminal neuronal differentiation. [Figure 42A] FIG. 1 shows the most significant disease annotations of the lithium pharmacogenomics network. [Figure 42B] FIG. 1 shows the top five drugs that are upstream regulators of the lithium pharmacogenomics network. [Figure 42C] FIG. 1 shows the topological model that most accurately fits the lithium pharmacogenomic network. [Figure 43] FIG. 1 shows a high-resolution parcellation of gene set subnetworks as an example of one output from using this system on the lithium pharmacogenomics network. [Figure 44] Figure 1 shows a list of genes included in the chromatin remodeling subnetwork of the lithium pharmacogenomics network. [Figure 45] This figure lists the genes included in the neuroplasticity subnetwork of the lithium pharmacogenomics network. [Figure 46] FIG. 1 shows a list of genes included in the efficacy subnetwork of the lithium pharmacogenomics network. [Figure 47] Figure 1 shows a list of genes included in the drug-induced weight gain (adverse event) subnetwork of the lithium pharmacogenomics network. [Figure 48] FIG. 1 shows a list of genes included in the drug-induced tremor (adverse event) subnetwork of the lithium pharmacogenomics network. [Figure 49A] FIG. 1 shows the most significant disease annotations of the lamotrigine pharmacogenomics network. [Figure 49B] FIG. 1 shows the top five drugs that are upstream regulators of the lamotrigine pharmacogenomics network. [Figure 49C] FIG. 1 shows the topological model that most accurately fits the lamotrigine pharmacogenomic network. [Figure 50] FIG. 1 shows an example of a lamotrigine pharmacogenomic adverse event subnetwork output by the system. [Figure 51] FIG. 10 shows an example of a lamotrigine pharmacogenomics neuroplasticity and efficacy subnetwork output by the system. [Figure 52] FIG. 1 lists the genes included in the chromatin remodeling subnetwork of the lamotrigine pharmacogenomics network. [Figure 53] FIG. 1 shows a list of genes included in the neuroplasticity subnetwork of the lamotrigine pharmacogenomics network. [Figure 54] FIG. 1 shows a list of genes included in the adverse events subnetwork of the lamotrigine pharmacogenomics network. [Figure 55] FIG. 1 is a diagram listing the genes included in the pharmacokinetic subnetwork of the lamotrigine pharmacogenomics network. [Figure 56A] FIG. 1 shows the most significant disease annotations of the clozapine pharmacogenomics network. [Figure 56B] FIG. 1 shows the top five drugs that are upstream regulators of the clozapine pharmacogenomics network. [Figure 56C] FIG. 1 shows the topological model that most accurately fits the clozapine pharmacogenomic network. [Figure 57] FIG. 1 shows an example of a clozapine pharmacogenomic adverse event subnetwork output by the system. [Figure 58] FIG. 1 shows an example of a clozapine pharmacogenomics neuroplasticity and efficacy subnetwork output by the system. [Figure 59] Figure 1 shows a list of genes included in the chromatin remodeling subnetwork of the clozapine pharmacogenomics network. [Figure 60] Figure 1 shows a list of genes included in the neuroplasticity subnetwork of the clozapine pharmacogenomics network. [Figure 61] FIG. 1 shows a list of genes included in the adverse event subnetwork of the clozapine pharmacogenomics network. [Figure 62] FIG. 1 is a diagram listing the genes included in the pharmacokinetic subnetwork of the clozapine pharmacogenomics network. [Figure 63]Figure 63 shows a warfarin pharmacogenomics network that does not map to any of the psychotropic drug network topologies shown in Figure 11. The warfarin pharmacogenomics network is shown in Figure 63A, and the corresponding gene set enrichment profile is shown in Figure 63B. Figure 63C shows gene set enrichment for the warfarin anticoagulation subnetwork, and Figure 63D shows gene enrichment for the warfarin bleeding and vasoocclusion subnetwork. DETAILED DESCRIPTION OF THE INVENTION

[0030] While the following text sets forth detailed descriptions of many different embodiments, it should be understood that the legal scope of this specification is defined by the language of the claims set forth at the end of this disclosure. The detailed description should be construed as merely exemplary and does not describe every possible embodiment, as describing every possible embodiment would be impractical, if not impossible. Many alternative embodiments could be implemented using either current technology or technology developed after the filing date of this patent, and would still fall within the scope of the claims.

[0031] It should also be understood that unless a term is expressly defined in this patent using the sentence "As used herein, the term '______' is defined herein to mean..." or similar, there is no intention, either express or implied, to limit the meaning of the term beyond its plain or ordinary meaning, and such term should not be construed as limited in scope based on any statement made in any section of this patent (other than the claim language). If any term recited in the final claim of this patent is referred to in this patent in a manner consistent with a single meaning, this is done solely for clarity so as not to confuse the reader, and it is not intended that such claim term be limited, by implication or otherwise, to that single meaning. Finally, unless a claim element is defined by the word "means" and a recitation of a function, without any structural recitation, the scope of any claim element is not intended to be construed under application of 35 U.S.C. § 112, paragraph 6.

[0032] In this section, a detailed description of the drug pharmacogenomics network identification system and its application to pharmacophenomics decision support for drug selection and in silico discovery of pharmacodynamic drug targets within biological pathways is presented. A description of the methodology and its applications in clinical medicine and pharmaceutical research is presented first, followed by several exemplary illustrations of drug pharmacogenomics networks. The examples are non-limiting, and relevant variations of the method will be apparent to those skilled in the art and are intended to be encompassed within the scope of the appended claims.

[0033] The pharmacogenomics network identification system generates models of pharmacogenomic regulatory networks and their constituent subnetworks using a contemporary knowledge base that includes the functional topology of the pharmacogenomics genome architecture, the 3D molecular circuits within chromatin that control gene expression and mRNA splicing, and the drug-specific geometric expansion and contraction of TADs and their pharmacogenomic connections regulated by super-enhancers that influence enhancer-promoter and promoter-promoter interactions.

[0034] As shown in Figure 8, the nature of these interactions includes significant associations of genes encoding proteins involved in the absorption, distribution, metabolism, and excretion (ADME) of xenobiotic drugs; one example consists of a known mutation within the super-enhancer GH06J032184 that is responsible for the occurrence of an adverse drug event called neutropenia in certain individuals following treatment with the antipsychotic drug clozapine.

[0035] The reconstructed pharmacogenomic networks described herein are inseparable from those mediating disease pathogenesis and provide another avenue for investigating pharmacological mechanisms of action. Pharmacogenomic networks adapt over time to endogenous and exogenous stimuli based on the reactive chromatin plasticity in which they are embedded, explaining pharmacogenomic variation among humans. This determines individual patient responses to medications, including adverse drug events. Examples of this variation resulting from different proportional representations of subnetworks within a drug's pharmacogenomic network among patients are provided as examples of the output of this system and its methods.

[0036] Generally speaking, the technology for identifying a drug pharmacogenomics network can be implemented in a system including one or several client devices, one or several network servers, or a combination of these devices. For clarity, however, the following examples primarily focus on an embodiment in which the pharmacogenomics network server retrieves SNPs from human clinical trials that have been demonstrated to be significantly associated with response and adverse events for a particular drug of interest, or may include disease or trait risk SNPs. The pharmacogenomics network server compares the SNPs with SNPs reported from genome-wide association studies (GWAS), biobanks, phenome-wide association studies (PheWAS), and other candidate gene studies, and uses characteristics of the three-dimensional (3D) genome topology to identify additional SNPs linked to the SNPs and generate a set of allowed candidate variants.

[0037] The pharmacogenomics network server then performs bioinformatics analysis on each of the allowed candidate variants to filter the set of SNPs into a subset of intermediate candidate variants based on regulatory function, variant dependency, the presence of target gene relationships of the allowed candidate variants, and / or whether the allowed candidate variants are non-synonymous or synonymous coding variants that do not affect proteins but are involved in regulating gene expression. Further, the pharmacogenomics network server performs pathway analysis of target genes associated with the subset of intermediate candidate variants to filter the target genes and identify a set of genes causally related to a particular drug.

[0038] The pharmacogenomics network server then identifies a pharmacogenomics network for a particular drug of interest based on the set of identified genes and provides a representation of the pharmacogenomics network for display on the client device. For example, a representation of a drug pharmacogenomics network may include the name of the drug of interest, the name and / or graphical representation of each gene in the pharmacogenomics network, and the name and / or graphical representation of each sub-network in the pharmacogenomics network and the genes in each sub-network.

[0039] The Pharmacogenomics Network Server provides a wide range of algorithms, including regression algorithms (e.g., least squares regression, linear regression, logistic regression, stepwise regression, multivariate adaptive regression splines, local estimation scatterplot smoothing, etc.), instance-based algorithms (e.g., k-nearest neighbors, learning vector quantization, self-organizing maps, local weighted learning, etc.), regularization algorithms (e.g., ridge regression, least absolute shrinkage and selection operator, elastic net, least angle regression, etc.), and decision tree algorithms (e.g., classification and regression trees, iterative dichotomizer (ID3), C4, etc.).5, C5, chi-squared automatic interaction detection, decision strain, M5, conditional decision trees, etc.), clustering algorithms (e.g., k-means, k-medians, expectation maximization, hierarchical clustering, spectral clustering, mean shift, DBSCAN (density-based spatial clustering of applications with noise), OPTICS (ordering points to identify the clustering structure), etc.), association rule learning algorithms (e.g., Apriori algorithm, Eclat algorithm, etc.), Bayesian algorithms (e.g., Naive Bayes, Gaussian Naive Bayes, Multinomial Naive Bayes, Averaged One-Dependence estimators), Bayesian belief networks, Bayesian networks, etc.), artificial neural networks (e.g., perceptrons, Hopfield networks, radial basis function networks, etc.), deep learning algorithms (e.g., multi-layer perceptrons, deep Boltzmann machines, deep belief networks, convolutional neural networks, stacked autoencoders, generative adversarial networks, etc.), dimensionality reduction algorithms (e.g., principal component analysis, principal component regression, partial least squares regression, Sammon mapping, multidimensional scaling, projection pursuit, linear discriminant analysis, mixed discriminant analysis, quadratic discriminant analysis, flexible discriminant analysis, factor analysis, independent component analysis, non-negative matrix factorization, Various machine learning techniques may be used to analyze the data described herein, such as genomics data and spatial contact data, including, but not limited to, ensemble algorithms (e.g., boosting, bootstrap aggregating, Adaboost, stack generalization, gradient boosting machines, gradient boosted regression trees, random decision forests, etc.), reinforcement learning (e.g., differential learning, Q-learning, learning automata, SARSA (State-Action-Reward-State-Action), etc.), support vector machines, mixture models, evolutionary algorithms, probabilistic graphical models, etc.

[0040] 1A , an exemplary pharmacogenomics network identification system 100 identifies various drug pharmacogenomics networks. The pharmacogenomics network identification system 100 includes a pharmacogenomics network server 102 and multiple client devices 106-116, which may be communicatively connected through a network 130, as described below. In one embodiment, the pharmacogenomics network server 102 and the client devices 106-116 may communicate by wireless signal 120 over the communication network 130, which may be any suitable local or wide area network, including a WiFi network, a Bluetooth network, a cellular network such as 3G, 4G, Long Term Evolution (LTE), 5G, etc., the Internet, etc. In some examples, the client devices 106-116 may communicate with the communication network 130 through an intermediate wireless or wired device 118, which may be a wireless router, a wireless repeater, a base transceiver station of a cellular communication provider, etc. By way of example, the client devices 106-116 may include a tablet computer 106, a smart watch 107, a network-enabled mobile phone 108, a wearable computing device such as a Google Glass™ or Fitbit® 109, a personal digital assistant (PDA) 110, a mobile device such as a smartphone 112, also referred to herein as a “mobile device,” a laptop computer 114, a desktop computer 116, a wearable biosensor, a portable media player (not shown), a phablet, any device configured for wired or wireless RF (radio frequency) communication, etc. Additionally, any other suitable client device that records patient genomic data, receives pharmacogenomic datasets, or displays a representation of the pharmacogenomics network and / or sub-network may also communicate with the pharmacogenomics network server 102.

[0041] Each of the client devices 106-116 can interact with the pharmacogenomics network server 102 to identify a drug of interest to determine a corresponding pharmacogenomics network. Each client device 106-116 can also interact with the pharmacogenomics network server 102 to receive a representation of the pharmacogenomics network of the drug of interest and / or several pharmacogenomics subnetworks within the pharmacogenomics network. The client devices 106-116 can present a representation via a user interface for display to a medical professional or researcher, such as the one shown in FIG. 17 or a display showing the degree of overlap of drug-specific topology maps based on similarity scores shown in FIG. 15.

[0042] In an exemplary embodiment, the pharmacogenomics network server 102 may be a cloud-based server, an application server, a web server, or the like, and includes memory 150, one or more processors (CPUs) 142, such as microprocessors, coupled to the memory 150, a network interface unit 144, and an input / output module 148, which may be, for example, a keyboard or a touchscreen.

[0043] The pharmacogenomics network server 102 may also be communicatively connected to a database 154 of genomic data, including data from human clinical trials, biobanks, GWAS, and PheWAS studies.

[0044] Memory 150 may be tangible, non-volatile memory and may include any type of suitable memory module, including random access memory (RAM), read-only memory (ROM), flash memory, other types of persistent memory, etc. For example, memory 150 may store instructions executable by processor 142 for operating system (OS) 152, which may be any type of suitable operating system, such as, for example, a modern smartphone operating system. Memory 150 may also store instructions executable on processor 142 for, for example, network reconstruction engine 146A, pharmacogenomics network bandwidth regulator 146B, and gene set optimization engine 146C. Pharmacogenomics network server 102 is described in more detail below with reference to FIG. 1B. In some embodiments, the network reconfiguration engine 146A, the pharmacogenomics network bandwidth adjuster 146B, and the gene set optimization engine 146C may be part of one or more of the client devices 106-116, the pharmacogenomics network server 102, or a combination of the pharmacogenomics network server 102 and the client devices 106-116.

[0045] In either case, the network reconstruction engine 146A can receive a request from a client device 106-116, or from a database of existing reconstructed pharmacogenomic networks, to identify a pharmacogenomic network for a particular drug of interest. The client device 106-116 can also provide SNPs from human clinical trials, GWAS studies, PheWAS studies, etc., that have been demonstrated to have significant associations with response and adverse events for the particular drug of interest, or disease risk SNPs from GWAS that distinguish pharmacogenomic sub-networks for the drug. In another embodiment, the network reconstruction engine 146A can retrieve SNPs from the database 154. The network reconstruction engine 146A then generates a set of allowed candidate variants based on the retrieved SNPs and additional SNPs linked to the retrieved SNPs, according to TAD boundaries within adjacent TADs regulated by the super-enhancer or within distant trans-interactions determined from chromosome conformation capture. The network reconstruction engine 146A then performs bioinformatics analysis on each of the allowed candidate variants to filter the set of SNPs into a subset of intermediate candidate variants, and performs pathway analysis on target genes associated with the filtered set to identify a set of genes causally related to the specific drug. The network reconstruction engine 146A then identifies a pharmacogenomics network for the specific drug of interest based on the identified set of genes and provides a representation of the pharmacogenomics network for display via the user interface of the client device 106-116.

[0046] The pharmacogenomics network server 102 may communicate with the client devices 106-116 via a digital network 130. The digital network 130 may be a proprietary network, a secure public internet, a virtual private network, and / or some other type of network, such as, for example, a dedicated access line, an ordinary telephone line, a satellite link, or a combination thereof. When the digital network 130 includes the internet, data communication may occur over the digital network 130 using internet communication protocols.

[0047] 1B, the pharmacogenomics network server 102 may include a controller 224. The controller 224 may include a program memory 226, a microcontroller or microprocessor (MP) 228, a random access memory (RAM) 230, and / or input / output (I / O) circuitry 234, all of which may be interconnected via an address / data bus 232. In some embodiments, the controller 224 may also include or be communicatively connected to a database 239 or other data storage mechanism (e.g., one or more hard disk drives, optical storage drives, solid-state storage devices, etc.). The database 239 may include data such as genomic data, pharmacogenomics network display templates, web page templates, and / or web pages, and other data necessary to interact with users over the network 130. The database 239 may include data similar to the database 154 described above with reference to FIG. 1A.

[0048] 1B depicts only one microprocessor 228, it will be understood that the controller 224 may include multiple microprocessors 228. Similarly, the memory of the controller 224 may include multiple RAMs 230 and / or multiple program memories 226. While FIG. 1B depicts the I / O circuitry 234 as a single block, the I / O circuitry 234 may include many different types of I / O circuitry. The controller 224 may implement the RAM 230 and / or program memory 226 as, for example, semiconductor memory, magnetically readable memory, and / or optically readable memory.

[0049] As shown in FIG. 1B , the program memory 226 and / or RAM 230 may store various applications for execution by the microprocessor 228. For example, a user interface application 236 may provide a user interface to the pharmacogenomics network server 102, which may allow, for example, a system administrator to configure, troubleshoot, or test various aspects of the server's operation. The server application 238 may operate to identify a pharmacogenomics network for a particular drug of interest, identify the pharmacogenomics network and pharmacogenomics subnetwork for a particular drug, and receive requests to transmit a representation of the pharmacogenomics network to the client devices 106-116. The server application 238 may be a single module 238 or multiple modules 238A, 238B, such as a network reconstruction engine 146A, a pharmacogenomics network bandwidth adjuster 146B, and a gene set optimization engine 146C.

[0050] 1B as including two modules 238A and 238B, server application 238 may include any number of modules that accomplish tasks associated with implementing pharmacogenomics network server 102. Additionally, while only one pharmacogenomics network server 102 is depicted in FIG. 1B, it will be understood that multiple pharmacogenomics network servers 102 may be provided to distribute server load, serve different web pages, etc. These multiple pharmacogenomics network servers 102 may include web servers, enterprise-specific servers (e.g., Apple® servers, etc.), servers located in retail or proprietary networks, etc.

[0051] 1C, laptop computer 114 (or any client device 106-116) may include a display 240, a communications unit 258, user input devices (not shown), and, like pharmacogenomics network server 102, a controller 242. Like controller 224, controller 242 may include program memory 246, a microcontroller or microprocessor (MP) 248 (random access memory (RAM) 250), and / or input / output (I / O) circuitry 254, all of which may be interconnected via address / data bus 252. Program memory 246 may include an operating system 260, data storage 262, multiple software applications 264, and / or multiple software routines 268. Operating system 260 may include, for example, Microsoft Windows, OS X, Linux, Unix, etc. Data storage 262 may include data such as application data 264 for multiple applications, routine data 268 for multiple routines, and / or other data necessary to interact with pharmacogenomics network server 102 over digital network 130. In some embodiments, controller 242 may also include or be communicatively connected to other data storage mechanisms present within laptop computer 114 (e.g., one or more hard disk drives, optical storage drives, solid-state storage devices, etc.).

[0052] The communications unit 258 may communicate with the pharmacogenomics network server 102 over any suitable wireless communication protocol network, such as a wireless telephone network (e.g., GSM, CDMA, LTE, etc.), a WiFi network (802.11 standard), a WiMAX network, a Bluetooth network, etc. The user input device (not shown) may include a "soft" keyboard displayed on the display 240 of the laptop computer 114, an external hardware keyboard (e.g., a Bluetooth keyboard) communicating via a wired or wireless connection, an external mouse, a microphone for receiving voice input, or any other suitable user input device. As noted with respect to the controller 224, while FIG. 1C depicts only one microprocessor 248, it should be understood that the controller 242 may include multiple microprocessors 248. Similarly, the memory of the controller 242 may include multiple RAMs 250 and / or multiple program memories 246. While FIG. 1C depicts the I / O circuitry 254 as a single block, the I / O circuitry 254 may include many different types of I / O circuitry. The controller 242 may implement the RAM 250 and / or the program memory 246 as, for example, semiconductor memory, magnetically readable memory, and / or optically readable memory.

[0053] The one or more processors 248 may be adapted and configured to execute any one or more of a plurality of software applications 264 and / or any one or more of a plurality of software routines 268, as well as other software applications, present in the program memory 246. One of the plurality of applications 264 may be a client application 266, which may be embodied as a series of machine-readable instructions for performing various tasks related to receiving information at, displaying information at, and / or transmitting information from the laptop computer 114.

[0054] One of the plurality of applications 264 may be a native application and / or web browser 270, such as Apple's Safari®, Google Chrome™, Microsoft Internet Explorer®, and Mozilla Firefox®, which may be implemented as a set of machine-readable instructions for receiving, interpreting, and / or displaying web page information from the pharmacogenomics network server 102 while also receiving input from a user, such as a medical professional or researcher. Another of the plurality of applications may include an embedded web browser 276, which may be implemented as a set of machine-readable instructions for receiving, interpreting, and / or displaying web page information from the pharmacogenomics network server 102.

[0055] One of the plurality of routines may include a pharmacogenomics network display routine 272 that obtains a display of the pharmacogenomics network including the name of the drug of interest, the name and / or graphical representation of each gene in the pharmacogenomics network, and the name and / or graphical representation of each sub-network in the pharmacogenomics network and the genes in each sub-network, and presents the display on the display 240. Another of the plurality of routines may include a pharmacogenomics network request routine 274 that obtains a request identifying the pharmacogenomics network of a particular drug of interest and sends the request to the pharmacogenomics network server 102.

[0056] Preferably, a user may launch a client application 266 from a client device, such as one of client devices 106-116, to communicate with the pharmacogenomics network server 102 and implement the pharmacogenomics network identification system 100. Additionally, a user may also launch or instantiate any other suitable user interface application (e.g., a native application or web browser 270, or any other one of the software applications 264) to access the pharmacogenomics network server 102 and implement the pharmacogenomics network identification system 100.

[0057] To identify the pharmacogenomics network for a particular drug of interest, the pharmacogenomics network server 102 performs a general method 180, as shown in Figure 11. In some embodiments, the method 180 may be implemented in a set of instructions stored in non-transitory computer-readable memory and executable on one or more processors in the pharmacogenomics network server 102. For example, the method 180 may be performed by the network reconstruction engine 146A, the pharmacogenomics network bandwidth adjuster 146B, and the gene set optimization engine 146C.

[0058] FIG. 2 illustrates the integration of multi-scale data and processing by the system 100. The methods and systems are based on the properties of pharmacogenomic networks, which consist of enhancer and super-enhancer networks that are activated or repressed in the same human cell types where a drug initially acts. This can be achieved by matching gene expression changes caused by a drug or other therapeutic agent to the higher-order structures where the drug initially acts, compared to a placebo or control, determining whether SNPs associated with the drug's pharmacogenomic network and subnetwork act in these human tissues, and assessing whether enhancer and super-enhancer regulatory elements are localized to specific target anatomical substrates. Machine learning, deep learning, reinforcement learning, and other artificial intelligence methods can be used, where applicable, to perform machine-executable steps in the system.

[0059] A detailed overview of one embodiment of system 100 is shown in Figure 11. The system combines automated executables and semi-manual curation, including a human pharmacogenomics SNP filter that can accept SNPs that both influence drug response or disease risk. This is followed by a drug-specific pharmacogenomics network reconstruction engine with a drug-space network bandwidth adjuster, and this is followed by iterative gene set optimization to decompose the drug-specific pharmacogenomic network into component functional subnetworks.

[0060] Genetic variant selection using the human pharmacogenomics SNP input filter In block 181, the pharmacogenomics network server 102 retrieves SNPs from human clinical trials that showed significant associations with response and adverse events to the drug of interest, or disease-risk SNPs that can be used to distinguish the relative weights of expression, efficacy, and adverse event subnetworks within a patient's regulatory genome. Because the location of SNPs associated with the trait under study is most often imprecisely assigned to the nearest gene or nearby candidate gene in published literature and GWAS per linear sequence of the reference human genome assembly, precise localization and annotation techniques using imputation are used to determine the actual location of the reported SNPs.

[0061] The new research has several important implications for drug pharmacogenomics network identification systems. First, by collecting pharmacogenomics network outputs in a training set using computer vision-based modifications of the 3D genome architecture triggered by permissive SNPs using deep learning (machine learning) and validation using correspondence to known drug-induced genome-wide changes in the genome, new drug target mechanisms may be identified. Second, clustering of new drug target mechanisms in previously defined but incompletely informed biological pathways will increase the likelihood of success. Third, insights gained using the 3D genome architecture to determine drug targets from drug response and disease risk SNPs will lead to the next generation of drug candidates, significantly improving the accuracy of pharmacogenomic diagnostics.

[0062] Figure 3 shows the nature of TAD organization, including the key enhancer and promoter transcription factors CREBBP (CEB-binding protein), EP300 (E1A-binding protein P300), POLR2A (RNA polymerase II subunit A), and YY1 (Yin Yang transcription factor 1), as well as the aggregation of TAD boundary proteins and chromatin loop-binding proteins, including CTCF (CCCTC-binding protein). In addition to enhancer-promoter and promoter-promoter pairs, adjacent TADs also contain super-enhancers. As shown in Figure 3, the chromatin loop protein CTCF is also involved in pre-mRNA splicing.

[0063] Figure 4 illustrates the nature of drug effects on TADs regulated by super-enhancers, suggesting that TAD-based localization of causal SNP targets provides greater accuracy than measurements of linkage disequilibrium in human populations. Figures 4A and 4B show two adjacent TADs regulated by a super-enhancer. In Figure 4A, in the absence of drug, the super-enhancer is silent, and differences in gene expression within adjacent TADs are minimal. Figure 4B shows activation of the super-enhancer controlling two adjacent TADs in the presence of drug, causing geometric expansion of the TADs and a concomitant increase in expression of genes located within these expanded TADs. Figure 4C demonstrates that TAD organization in spatial regulatory genomes provides a more accurate method for identifying gene promoter targets of causal enhancer SNPs from GWAS and other studies than those provided by traditional measurements of linkage disequilibrium.

[0064] Figure 5 shows that the human genome is organized in 3D like a "ball of string." This 3D organization changes dynamically over time, and regulatory interactions can be understood by examining the location of TADs and their regulatory super-enhancers following drug-induced changes.

[0065] In block 182, the pharmacogenomics network server 102 evaluates candidate causal SNPs using a pharmacogenomics informatics pipeline. The pharmacogenomics informatics pipeline uses lead SNPs reported from GWAS, Biobank, PheWAS, and other candidate gene studies to find genetically linked, permissive candidate SNPs using TAD boundaries instead of linkage disequilibrium measurements, as shown in Figure 4C. The enhancer-regulatory SNP workflow evaluates permissive candidate SNPs in disease-associated tissues for DNA methylation, transcription factor binding, histone marks, DNase I hypersensitivity, chromatin state, quantitative trait loci (QTL), chromatin loop-based contacts determined using chromosome conformation capture techniques such as Hi-C, and disruption of transcription factor binding sites using tissue-specific omics datasets. As shown in Figure 11, the pharmacogenomics network server 102 then evaluates the final output SNPs using an open-source machine learning algorithm to determine whether the SNPs are causal (block 183), and the causal variants are retained for further analysis in the workflow (block 184). Exonic SNPs are also evaluated as splice donors or splice acceptors using the Altrans algorithm. If found to be involved in alternative splicing, they are retained as such.

[0066] Figure 9 shows an example of SNP selection for predictive causal associations using six different machine learning and deep learning algorithms based on tissue-specific distributions. It shows a candidate SNP, rs12967143-G, located within an intragenic enhancer located in the transcription factor 4 (TCF4) gene, relative to other GWAS SNPs explained using numerical output from the machine learning algorithms used in the analysis. * p<0.05, **p<0.01. A numerical score from each algorithm was generated for each GWAS SNP, and a SNP was retained in further analyses only if each output scored the SNP as predicted to be causal in SK-N-SH and H1 cells but not in HepG2 cells and PBMCs. The scores of all predicted causal SNPs were independently tested to determine whether they were significantly different from scores generated using 10 randomly selected GWAS SNPs for all human traits listed in the EBI-NHGRI GWAS catalog with p<5E-08 using ANOVA. Only if a SNP met this significance criterion was it selected for further analysis by the system.

[0067] Using casual enhancer SNPs to interrogate drug pharmacogenomic networks In block 185, enhancer SNPs are used as probes to determine target genes as cis interactions within the same TAD or adjacent TADs controlled by the same super-enhancer. Pharmacogenomic trans interactions with other TADs are determined using Hi-C chromosome conformation capture and ChIA-PET datasets. 3D pharmacogenomic connections (block 187) are then mapped from the cell types and tissues in which the drug of interest acts. Genes containing other functional elements, such as long non-coding RNAs, are located within the same TAD or adjacent TADs controlled by the same super-enhancer. For trans interactions, if the TADs involved in the cis and trans interactions have strong boundaries as predicted by the amount of bound CTCF and / or significant association with the super-enhancer, then those that are targets of enhancers that significantly alter drug response in human populations are selected for the pharmacogenomic network (block 186). For trans interactions, TADs containing neighboring TADs regulated by the same super-enhancer that contain the top three statistically significant pharmacogenomic contacts of a first set of pharmacogenomic TADs within the same cell and / or tissue type in which the drug of interest acts are then evaluated, and genes within these "trans TADs" are selected if they are also regulated by the same cell and / or tissue-specific enhancers in which the drug of interest acts (block 188).

[0068] Figure 7 shows the distribution of TAD characteristics in one cell type of the human genome. 98% of TADs contain known or predicted enhancers, and 40% of TADs have known super-enhancers that span adjacent TADs in the genome.

[0069] In block 189, the pharmacogenomics network server 102 evaluates the combined set of genes for interconnectivity, where the combined set of genes is selected from genes selected from a "trans TAD" that includes genes coordinately regulated with the first set of TADs having pharmacogenomic SNPs and the first set of cis-interacting genes. For example, the pharmacogenomics network server 102 may use 、 Third-party software such as Ingenuity Pathway Analysis™ can be utilized. Using Fisher's right-sided exact test, if the drug pharmacogenomics network server 102 determines that significant interconnectivity exists within the combined set of genes based on published literature, the genes are placed within a preliminary set of genes that constitute the pharmacogenomics network for the drug of interest. Any genes that do not form a connected network are discarded as non-candidate genes for the pharmacogenomics network (block 190).

[0070] Knowledge base revision using a pharmacogenomics network bandwidth regulator Then, in block 191, manual, semi-automated, or automated curation, or a combination thereof, is performed on each gene of this set of genes that constitute the preliminary drug pharmacogenomics network to remove genes whose function is not related to the drug of interest in the cell and / or tissue type in which it acts, or to add other genes that are not part of this preliminary set of drug pharmacogenomics network if they are determined to be particularly affected by the drug of interest in the cell and / or tissue type in which it acts. The research steps include defining the function of each individual gene, the phenotypic consequences of the disorder due to mutations in the gene, and the human cells and tissues in which the gene is expressed to see if it can be a candidate for membership in the pharmacogenomics network for the particular drug of interest.

[0071] In one embodiment, these determinations can be made using manual and semi-automated strategies, combining manual curation of each gene, its mutation profile, and the localization of its expression within human tissues. These are achieved through various web-based search tools, including gene definitions, genome browser annotations, GWAS catalogs, and other bioinformatics resources. For example, the pharmacogenomics network server 102 can invoke an application programming interface (API) with executable files written in R, Python, PERL, or other programming languages ​​to facilitate data access, data cleansing, and data analysis. While this embodiment is an enhanced model of manual curation, it can be time-limiting when there are many genes within a gene set of a pharmacogenomics network or within a gene subset of a subnetwork, especially when functional genomic elements may include functional RNAs such as regulatory RNAs or long non-coding RNAs, or when the function of the gene is not well understood. Listing and analyzing the mutational landscape of a given gene (±110 kb upstream and downstream) is the easiest of the three research steps to perform, as these databases are the most comprehensive. Other resources exist for analyzing the tissue distribution of gene expression patterns. Where these patterns are compared to the site of action of a particular drug of interest, the pharmacogenomics network identification system 100 may leverage results from imaging modalities, including radiological studies, light microscopic analysis in pathology, and more advanced methods. In some embodiments, the pharmacogenomics network server 102 performs this analysis using machine learning techniques, such as neural networks.

[0072] In another embodiment, the pharmacogenomics network server 102 can use a Bayesian probabilistic classifier based on machine learning or using Bayesian probability calculations. Automated methods can be used to reduce the complexity of data analyzed from heterogeneous data resources, where a gene's functional knowledge profile, its mutation landscape, and its tissue expression mapping are input to a learning machine trained on many such instances and individually tested on a separate set of instances to determine accuracy. Predictive features selected by the trained neural network can be implemented in a support vector machine classifier to build a gene function and mutation prediction model, and the subsequent machine state determines the adequacy of the statistical fit to the pharmacogenomics network.

[0073] In some scenarios, machine learning is subject to overfitting, producing false positives or false negatives. In another embodiment, the pharmacogenomics network server 102 can perform semi-automated naive Bayes classification using machine learning in parallel to increase the accuracy of the final output.

[0074] The pharmacogenomics network server 102, and more specifically, the pharmacogenomics network bandwidth regulator 146B, can perform knowledge base curation in the following steps. First, the pharmacogenomics network server 102 examines gene definitions from multiple databases to understand whether they are specifically (but not generally) affected by the drug of interest. Furthermore, published literature containing gene names, precursor gene names, or equivalent protein names, as well as text word strings containing any functions related to the drug of interest, are evaluated following exhaustive internet searches, for example, using Google Scholar™ and / or PubMed. These may include binding affinity studies that have reproducibly found molecules that bind with affinities within 10-fold of the affinity that the drug of interest binds to the same pharmacodynamic target. Second, the pharmacogenomics network server 102 examines each gene for all mutations, including SNPs, variable numbers of tandem repeats, duplications, and all other known mutational alterations, spanning the linear sequence plus ±10 kb from the gene's transcription start site and stop codon, as examined in a genome browser such as the UCSC Genome Browser or the Ensembl Genome Browser. If any of these mutations are found in either the published literature or sources such as unpublished clinical trial data and are involved in the drug's action, including efficacy, adverse events, or first-pass metabolism, they are added to a preliminary set of genes that comprise the pharmacogenomics network (block 192). Third, particularly for complex tissues such as the brain, skin, and cardiovascular system, the pharmacogenomics network server 102 qualitatively performs match mapping to compare the expression of all genes in this final set with known cases in which the drug of interest exerts its action. Genes whose expression does not match the pharmacodynamic substrate of the drug of interest are discarded (block 192). Finally, third-party software such as Ingenuity Pathway Analysis™ is used to examine the connectivity of this gene set (block 193).If, using Fisher's right-sided exact test, the drug pharmacogenomics network server 102 determines that significant interconnectivity exists based on the published literature, they are placed within a preliminary set of genes that comprise the pharmacogenomics network for the drug of interest. Any genes that do not form a connected network are discarded as non-candidate genes for the pharmacogenomics network (block 194).

[0075] The drug pharmacogenomics network server 102 can perform knowledge base curation using gene expression patterns where overlap suggests functional correspondence. Such an example is shown in Figure 38, which may exist when genes in the ketamine pharmacogenomics network show statistically significant overlap (P<1E-56; Fisher's exact test) and the drug exerts rapid action in the human brain, including the anterior cingulate cortex (ACC) and frontal cortex (FC), but not in the somatosensory cortex (SSC), occipital cortex (OC), or corpus callosum (CC).

[0076] Pharmacogenomics Network Reconstruction Engine Figure 11 shows the configuration of the pharmacogenomics network reconstruction engine 146A, which uses proprietary public knowledge of the 3D human genome, previously defined TADs, super-enhancers, and other features of the regulatory genome, provided in spreadsheet lookup tables from either the original version of Human Genome Build 19 (hg19) or the more recent sparse Human Genome Build 38 (hg38). This is a key component of the network reconstruction engine 146A. As shown in Figures 22 and 33, or from public or other private sources, all experimental data generated from chromosome conformation capture methods undergo evaluation, and chromosome conformation capture methods can be performed in vivo and generated from causal SNP probing of chromatin datasets of cell types in which the drug of interest is active. Parsimonious models of drug-induced changes in the 3D regulatory genome, including TAD matrices, enhancer-promoter pairs, promoter-promoter pairs, and super-enhancers, are developed in 2D or 3D using the SNPs or candidate variants identified above based on selected methods, such as those shown in Figure 6, including 3D modeling of the human genome architecture in Euclidean space, high-resolution light microscopy using FISH, and / or a combination of gene expression measurements (RNA-seq, promoter capture Hi-C, etc.). Following assessment of the drug's pharmacogenomic interactome in chromatin, the resulting pharmacogenomic network is defined in a preliminary manner. Significance scores are provided using third-party pathway analysis software to determine whether network elements are significantly interconnected based on existing biomedical knowledge. Commonly used programs for gene pathway analysis include Ingenuity Pathway Analysis™, Panther Gene Ontology pathway mapping, and KEGG (Kyoto Encyclopedia of Genes and Genomes).In this manner, the network reconstruction engine 146A determines interconnections between SNPs and target genes that correlate with drug response or adverse events for a particular drug of interest.

[0077] Iterative gene set optimization engine In block 195, the drug pharmacogenomics network server 102, more specifically, the gene set optimization engine 146C, performs iterative gene set optimization on the set of identified candidate genes in the pharmacogenomics network for a particular drug of interest. An example of a method for iterative gene set optimization for decomposing a drug pharmacogenomics network into subnetworks is shown in the flow diagram of FIG. 12. Iterative gene set optimization can be performed to identify subnetworks of the pharmacogenomics network. More specifically, iterative gene set optimization involves converting all input molecular terms into gene or long non-coding RNA names from Human Gene Nomenclature Committee (HGNC) names, for example, using an API. Iterative gene set optimization differs from gene set enrichment methods in that it not only combines various statistical methods but also does not rank genes hierarchically like threshold-dependent methods; iterative gene set optimization does not rely on comparison of experimental results, such as full distribution testing. Instead, iterative gene set optimization first measures the similarity between two genes or long non-coding RNAs based on the dissimilarity of a user-selected term, using Jaccard distance to group genes or long non-coding RNAs, where Jaccard distance is the ratio of the size of the symmetric difference gene A Δgene B = A∩B∪B to the union. This can be extended to clusters of related distinct gene names. The pharmacogenomics network server 102 then sorts these sets automatically, or using a user-defined number of clusters, into subsets of clustered subsets of functionally related genes using a minimum entropy sorting algorithm such as the COOLCAT algorithm. Following gene subset optimization using entropy minimization, the pharmacogenomics network identification system 100 can use manual curation to assign efficacy, adverse event, or functional mechanism subnetworks based on known attributes of the drug's mechanism of action.

[0078] Post-mortem validation using third-party bioinformatics toolsFor scientific validation of the decomposition of the pharmacogenomics network into mechanistic subnetworks based on functional gene subset optimization, the pharmacogenomics network server 102 evaluates each pharmacogenomics network subnetwork a posteriori for top Gene Ontology terms (molecular functions and biological processes), top terms from pharmaceutical databases, top canonical pathways determined using other proprietary or open-source pathway analysis software, for example, disease risk gene variant analysis determined using other proprietary or open-source pathway analysis software, and upstream xenobiotic regulators determined using different bioinformatics resources (block 196). The upstream xenobiotic regulators are compared with a specific drug of interest to confirm that the specific drug of interest is the drug most significantly associated with the pharmacogenomics network. More specifically, the upstream xenobiotic regulators can be ranked according to their respective associations with the pharmacogenomics network using different bioinformatics resources. For example, the upstream xenobiotic regulator with the lowest p-value for the pharmacogenomics network may have the strongest association. The drug pharmacogenomics network server 102 can then determine whether a particular drug of interest is a highly ranked upstream xenobiotic regulator or ranks above a threshold rank (e.g., ranked in the top three or top five). Additionally, the GWAS catalogs of the European Bioinformatics Institute, the National Human Genome Research Institute, and the National Institutes of Health can be searched to find significant SNP-characteristic associations for each gene in each subnetwork's gene set. Providing examples of statistically significant SNPs from GWAS can provide additional evidence that mutational disorders in genes included in each subnetwork offer insight into the normal, unimpaired function of the subnetwork. An example method for performing post-mortem validation of a pharmacogenomics network and its constituent subnetworks is shown in the flow diagram of FIG. 13.

[0079] In some embodiments, after post-validation is performed, the resulting pharmacogenomics network and constituent subnetworks of the particular drug of interest are stored in database 154, for example, as shown in FIG. 1A. In some embodiments, the pharmacogenomics network server 102 can provide a representation of the pharmacogenomics network and constituent subnetworks to the client devices 106-116 for display to a medical professional or researcher. The client devices 106-116 can then present the pharmacogenomics network and constituent subnetworks in a graphical representation.

[0080] Error correction for pharmacogenomic networks and subnetworks In some embodiments, the pharmacogenomics network server 102 can adjust or tune specific drug pharmacogenomic networks and constituent subnetworks to provide accurate models for measuring human drug response phenotypes for use in real-world clinical applications. Studies of population structure using principal component analysis, allele sharing distance, and other measures suggest that the distribution of pharmacogenomic phenotypes can be modeled using a normal distribution, albeit with some outliers. For example, it was previously thought that variation in cytochrome P450 genes, which leads to differences in CYP450 isoform activity, is the primary determinant of variability in drug response among humans.

[0081] One embodiment of the present disclosure involves SNPs within enhancer networks that regulate PK gene expression, as well as other genes located within the same TAD. Depending on the drug and patient, variations within these networks can affect tissue-specific metabolism beyond missense codons. As shown in the example of the PK gene, in patients where the TAD boundaries of the PK gene may be compromised, trans-interactions of enhancers with less constraints on the TAD in which they are located can result in adverse drug events.

[0082] Figure 10 shows that drug metabolism genes and human leukocyte antigen (HLA) genes involved in immune-related adverse drug responses have strong boundaries. Of the 13 gene clusters shown here, 12 have the strongest TAD boundaries in the human genome (Class V). These include genes encoding cytochrome P450 enzymes (CYP genes), genes in the glucuronosyltransferase (UGT) superfamily, genes in the sulfotransferase (SULT) superfamily, genes in the N-acetyltransferase (NAT) family, and most HLA genes. Mutations, such as SNPs, located at the TAD boundaries of these genes have deleterious effects on variability in drug metabolism and drug response, including the occurrence of adverse drug events in the human population.

[0083] It is also recognized that additional variables play a role in human drug response, including the influence of sociological status and other environmental factors, and these are often variables that are difficult to measure.

[0084] Figure 17 shows pharmacogenomic network topologies that can be used by the system to model the effects of most psychotropic drugs. In Figure 17A, a template model of the central nervous system (CNS) includes boxes containing the drug and (1) chromatin remodeling, (2) efficacy (EFF) and / or neuroplasticity (NP), (3) adverse CNS events (AE), and (4) centrally active pharmacokinetic enzymes (PK) and hormones (H). For some drugs, systemic pharmacokinetics (SPK) is the primary determinant of variability in human drug response, while peripheral adverse drug events involving the immune system (IAE) are problematic. In Figure 17B, various pharmacogenomic network topologies for psychotropic drugs are shown, along with examples of various biological profiles that fit the various two-dimensional topologies. Thus, a drug pharmacogenomics network can be decomposed into constituent subnetworks using the topology of subnetwork types shown in Figure 17, where the subnetwork types for most psychotropic drugs can include two or more of: (1) chromatin remodeling, (2) efficacy (EFF) and / or neuroplasticity (NP), (3) adverse CNS events (AE), (4) centrally active pharmacokinetic enzymes (PK) and hormones (H), (5) systemic pharmacokinetics (SPK), and (6) peripheral adverse drug events involving the immune system (IAE).

[0085] FIG. 14 illustrates a machine-learning-based method 600 whereby computationally predicted efficacy and adverse event subnetwork measures of a pharmacogenomics network are calibrated on a human population, including a training set and a test set, to obtain an accurate discretization of response phenotypes. In some embodiments, the pharmacogenomics network server 102 performs the method 600 shown in FIG. 14 . Also, in some embodiments, the method 600 may be implemented in a set of instructions stored in non-transitory computer-readable memory or executable on one or more processors in the pharmacogenomics network server 102. In any event, the method 600 refines the estimated distribution of human response phenotypes for a particular drug developed from the computational analysis by training such drug-specific subnetworks using the pharmacogenomics network identification system 100, and the human drug response phenotypes are obtained from the efficacy and adverse event subnetworks. In this way, the pharmacogenomics network server 102 improves the utility of the distribution of human response phenotypes for use in real-world clinical applications, which in turn makes them useful for any reference-based comparative measurements performed in medicine or the life sciences.

[0086] Matching of reference drugs to patients in a pharmacogenomics network The learning architecture for training the pattern matching sub-network includes pre-training a reference set (reference numeral 710), as shown in FIG. This is further described with reference to method 700 shown in FIG. 15, which may also be performed by the pharmacogenomics network server 102. More specifically, in block 704, the pharmacogenomics network server 102 develops a pattern-matching subnetwork for the patient obtained from the patient input biological sample, and simultaneously develops separate trained pattern metrics that include features of the efficacy and adverse event subnetworks into a joint functional expression metric (block 712). To determine similarity with the reference set (blocks 706, 708), two different pairs of reference patient metrics are used to determine the accuracy of the similarity, and similarity scores are output for efficacy and adverse events, respectively (blocks 714, 716). In block 702, a biological sample (cheek swab, blood, or urine sample) obtained from the patient undergoes targeted enhancer SNP genotyping and a combination of chromosome conformation capture and RNA-seq. Then, in block 704, the pharmacogenomics network server 102 performs the necessary analyses to construct input patient-specific maps of efficacy and adverse event subnetworks for the particular drug of interest. These patient-specific drug-induced subnetwork patterns can be further processed using Bayesian probability calculations to fill in sparse or missing data. As new patients are entered as input, the pre-trained reference set of drug-specific efficacy and adverse event subnetworks for pattern matching is re-optimized for subsequent patients, generating more accurate measures of inter-person pharmacogenomic variation with enhanced clinical utility. This matching task assumes that patches are passed through the same functional encoding before calculating and outputting similarity scores, significantly improving efficiency while reducing computational requirements.

[0087] Therefore, each set of inputs (reference set (reference number 710) and patient set (reference number 720)) is constructed differently by extracting feature sets and inferring sparse data using Bayesian distribution-based probabilistic calculations to increase the accuracy of the reference and patient maps. The trained feature network is based on a "siamese" network approach, with the constraint that the two sets must share the same parameters. Once completed, the patient's drug-induced trained pattern network is combined with a network obtained from the reference database, resulting in paired efficacy feature sets and paired adverse event feature sets. These provide the basis for the development of a trained efficacy metric and a trained adverse event metric, which attempt to match all features from the patient and reference set for the drug of interest. These pairwise matching scores result in distinct efficacy and adverse event similarity scores between the reference and patient.

[0088] A further elaboration of this embodiment is to develop a reference pattern matching set for each patient, which can be used to create a patient-specific database of such reference maps, which can be updated periodically as additional biological samples are obtained from the patient in a longitudinal manner, either at the clinical site or at an outpatient pharmacy over time.

[0089] Methods for in silico drug target identification 16A and 16B show another method 800 that utilizes a pharmacogenomics network and efficacy and adverse event sub-networks for a particular drug of interest to develop a molecule as a druggable pharmacodynamic target. In some embodiments, the pharmacogenomics network server 102 performs the method 800 shown in FIG. 16A. In some embodiments, the method 800 may be implemented in a set of instructions stored in a non-transitory computer-readable memory and executable on one or more processors in the pharmacogenomics network server 102.

[0090] In any case, previously unrecognized genes encoding druggable pharmacodynamic targets can be linked at the pharmacogenomics regulatory level to the efficacy subnetwork of a specific drug's pharmacogenomic network with minimal connectivity to multiple genes in the adverse event subnetwork of the specific drug's pharmacogenomic network. In the example shown in Figures 16A and 16B, the ketamine pharmacogenomic network is used, and the druggable target is PPP1R1B (protein phosphatase 1 regulatory inhibitor subunit 1B) (reference number 804), a bidirectional signaling molecule regulated by the neurotransmitter dopamine. The PPP1R1B gene 804 is located within the same TAD as the NEUROD2 (neural differentiation 2) gene 802, and the same enhancer regulates both genes in both neuronal and astrocyte cell lines. Furthermore, TADs containing these genes interact with TADs containing the DRD2 (dopamine receptor D2) 806 and ADORA2A (adenosine A2a receptor) genes 808, which are also regulated by the same neuronal and astrocytic enhancers in their respective TADs. Following the use of the methods described herein, the PPP1R1B pathway, although not a known drug pharmacogenomics network, was significantly interconnected in the human brain (p=1E-88). Seven of these genes were included in the ketamine pharmacogenomics network, including BDNF, DRD2, GRIA1, GRIN1, GRIN2A, and KLF6, as well as PPP1R1B. As shown in Figure 29 and Figure 16C, four genes were included in the ketamine neuroplasticity subnetwork of the ketamine pharmacogenomics network, and the other four genes were included in the glutamate receptor subnetwork of the ketamine pharmacogenomics network. Comparing the location of gene expression among different human brain regions for different genes in the PPPR1B pathway, only 14 are expressed at detectable levels in the human brain. With the exception of GRIA1, GRIN1, and GRIN2A, which are more widely expressed in the human brain, the expression of the remaining 11 genes in this pathway shows a significantly more restricted pattern of gene expression, limited to the anterior caudate nucleus, nucleus accumbens, and putamen, as determined by human RNA-seq data (Figure 16D).

[0091] In block 810, the pharmacogenomics network server 102 performs a bioinformatics analysis of the PPP1R1B pathway limited to these 14 genes expressed in the human brain. The bioinformatics analysis indicates that the 14 genes are significantly associated with neuronal differentiation, neuronal development, regulation of neurogenesis and CNS development, and opioid drug signaling. These characteristics are shared with the neuroplasticity subnetwork of the valproic acid pharmacogenomics network shown in Figures 18 and 21, the neuroplasticity subnetwork of the ketamine pharmacogenomics network shown in Figure 32, and the lithium pharmacogenomics network shown in Figure 43.

[0092] As shown in Figures 39, 40, and 41, the pharmacogenomics network identification system 100 can reveal complementary properties of existing medications that can be combined together as new drug compounds or administered sequentially, or similar combinations of drugs from the same drug class can be identified to provide more comprehensive treatment for a given clinical indication. Figure 41 shows that the valproic acid neurogenesis pharmacogenomics subnetwork is enriched for stimulating early neurogenesis, while the ketamine neuroplasticity subnetwork is responsible for late neurogenesis. Figure 39 shows that one mechanism by which this therapeutic combination works is through the sequential acetylation and deacetylation of the histone lysine 9 (H3K9) moiety. Thus, valproic acid converts multipotent neural progenitors into committed neural progenitors by combining histone deacetylation inhibition and induction of the neural progenitor BAF chromatin remodeling complex, as shown in Figure 39 (top), and ketamine converts committed neural progenitors into terminally differentiated neurons by activating HUSH (human silencing complex) and H3K9 methylation (Figure 39, bottom).

[0093] Figure 40 shows gene set enrichment of nuclear genes contained within the valproic acid pharmacogenomics network (Figure 40A) and the ketamine pharmacogenomics network (Figure 40B). Figure 40A shows that analysis of the valproic acid pharmacogenomics network is enriched for both H3K9 histone deacetylase activity and neurogenesis, and Figure 40B shows that the ketamine pharmacogenomics network is enriched for both H3K9 histone methyltransferase activity and neuronal differentiation.

[0094] Therefore, as shown in Figure 41, these approved drugs may be combined in clinical use to provide a comprehensive solution to provide both early and mid-to-late stages of neurogenesis, including mechanisms of neurogenesis, neural proliferation, neural differentiation, and synaptic integration. For example, a first drug, such as valproic acid, can be administered to a patient at a first time point, and then a second drug, such as ketamine, can be administered to a patient at a second time point later than the first time point. Thus, this combination of FDA-approved therapies can be used not only for conditions characterized by neuronal cell loss, but also for the aging human brain to maintain the integrity of gray matter. Conditions can include neurodegenerative disorders such as frontotemporal dementia, Alzheimer's disease and Parkinson's disease, as well as neuropsychiatric disorders such as bipolar disorder and schizophrenia, and acute brain injury.

[0095] For example, a method for treating a patient with a neurodegenerative disorder may include administering valproic acid to the patient and administering ketamine to the patient. In some embodiments, the method may include obtaining a biological sample from the patient and comparing or comparing the biological sample to one or more SNPs in the valproic acid pharmacogenomics network associated with neurogenesis. The method may also include comparing or comparing the biological sample to one or more SNPs in the ketamine pharmacogenomics network associated with neural differentiation. In response to determining that the patient's biological sample contains both SNPs in the valproic acid pharmacogenomics network associated with neurogenesis and SNPs in the ketamine pharmacogenomics network associated with neural differentiation, valproic acid and ketamine can be administered to the patient to treat the patient's neurogenesis disorder.

[0096] More generally, the pharmacogenomics network identification system 100 may identify pharmacogenomic networks for any number of drugs. Then, for a first and a second drug, the pharmacogenomics network identification system 100 may compare characteristics (e.g., drug response phenotypes) associated with genes in the pharmacogenomics network and / or constituent subnetworks of the first drug with characteristics (e.g., drug response phenotypes) associated with genes in the pharmacogenomics network and / or constituent subnetworks of the second drug to identify complementary characteristics between the first and second drugs. If complementary characteristics of a set of drugs, such as early-stage neurogenesis and late-stage neurogenesis, are identified, the set of drugs may be repurposed for testing as treatments for a particular disease or condition.

[0097] Figure 18 shows the valproic acid pharmacogenomic network of the human central nervous system as the system output, and its constituent subnetworks. The gene subnetworks consist of (1) chromatin remodeling and H3K9 acetylation, (2) neurogenesis and antiepileptic, antimanic, and antimigraine properties, (3) adverse events, and (4) hormone regulation and pharmacokinetics.

[0098] Figure 19 shows the results of a post-hoc bioinformatics analysis of the valproic acid pharmacogenomic network and its associated network topology model. Figure 19A shows that the most significant disease annotations of genes included in the valproic acid pharmacogenomic network are epilepsy or neurodevelopmental disorders, cognitive disorders, mood disorders, migraine, and mania. It should be noted that valproic acid is indicated as a monotherapy and adjunctive therapy in the treatment of simple and complex absence seizures, as an adjunctive therapy in patients with multiple seizure types, including absence seizures, for the treatment of mania in bipolar disorder and other mood disorders, and for the prevention and alleviation of migraine. A common adverse event of valproic acid is cognitive clouding (cognitive impairment).

[0099] Figure 19B shows the most significant drugs acting as upstream regulators of the valproic acid pharmacogenomic network, including valproic acid (p-value = 5.20E-114; Fisher's exact test), the HDAC inhibitor trichostatin A (p = 3.21E-35), and nicotine (p = 5.57E-21).

[0100] Figure 19C shows that the valproic acid pharmacogenomics network, along with decomposed gene set subnetworks including chromatin remodeling (CR), neuroplasticity and drug efficacy (NP, EFF), adverse events and neurotransmission (AE, NT), and pharmacokinetics and hormone regulation (PK, H), fits the model network topology labeled 1. The non-CNS, peripheral systemic pharmacokinetic (SPK) subnetwork of the valproic acid pharmacogenomics network cannot be determined from the output of this system.

[0101] Figure 28 shows examples of trans interactions in the 3D chromatin space of the valproic acid pharmacogenomics network as output of the methods and systems described herein. Genome-wide Hi-C data mapping was performed using SNPs as data probes. The probes included SNPs contained within the valproic acid subnetwork, including those significantly associated with disease risk and valproic acid response and dissociation, retrieved from the GWAS catalog. These results were used to detect both cis and trans interactions with other members of the valproic acid pharmacogenomics pathway in human neurons. Figure 28A shows a genome-wide plot that is key to understanding the gene-gene interactions shown in Figures 28B-28I, determined by Hi-C methods. Figure 28B shows Hi-C contacts between GABBR1, a gene encoding a receptor for gamma-aminobutyric acid (GABA), the major inhibitory neurotransmitter in the human CNS (enhancer mutations in this gene underlie brain disorders such as epilepsy), and CRHR1 and CRHR1-IT1, genes important for neural progenitor cell differentiation under the control of several super-enhancers, corticotropin hormone binding in the adult brain, and mutations associated with anxiety and mania. Figure 28C shows spatial Hi-C contacts in human neurons between GABRG2, a gene encoding a GABA receptor whose mutations are significantly associated with febrile and infantile epilepsy, and KCNJ3, a gene encoding a potassium channel in the human CNS whose mutations are significantly associated with cognition and epilepsy. Figure 28C shows spatial Hi-C contacts in human neurons between the transcription factor RUNX1 and HDAC9, a member of the histone deacetylase superfamily. Mutations in the enhancers and super-enhancers of both genes are significantly associated with alopecia and mild hair loss, adverse events associated with valproic acid therapy.Figure 28E shows spatial Hi-C contacts in human neurons between GABRB2 and GABRG2, genes encoding GABA receptors whose mutations are associated with ataxia and epilepsy, and KCNQ5, a potassium channel gene whose super-enhancer mutations are significantly associated with autosomal mental retardation and intellectual disability. Figure 28F shows spatial Hi-C contacts in human neurons between NEUROD1, a master transcription factor involved in neurogenesis and the valproic acid response in humans, and NEUROG3, a transcription factor involved in the lineage commitment of neural progenitor cells to neurons. Figure 28G shows spatial Hi-C contacts in human neurons between GRIN2A, a gene encoding a member of the N-methyl-D-aspartate (NMDA) receptor family whose mutations are significantly associated with schizophrenia, bipolar disorder, and mania, and ANK3, a gene encoding a member of the synaptic cytoskeleton whose mutations are significantly associated with bipolar disorder, chronotype, and schizophrenia. Figure 28H shows spatial Hi-C contacts in human neurons between GRIN2B, a gene encoding a member of the NMDA receptor family, and SNCA, a presynaptic protein whose super-enhancer mutations are significantly associated with late-onset Parkinson's disease. Figure 28I shows spatial Hi-C contacts in human neurons between PAX6, a gene encoding a master transcription factor involved in early development of the human central nervous system, eyes, and nose, and SOX2, a gene related to PAX6 that regulates the neural progenitor BAF remodeling complex involved in neurogenesis.

[0102] Figure 20 shows the valproic acid pharmacogenomic adverse event subnetwork, and post-hoc bioinformatics analysis indicates that the valproic acid adverse event subnetwork is significantly associated with cancer, severe psychiatric disorders, gastrointestinal disorders, tremor, and alopecia.

[0103] Figure 21 shows the valproic acid pharmacogenomic neuroplasticity subnetwork, and post-hoc bioinformatics analysis indicates that the glutamate receptor subnetwork is significantly associated with neurogenesis, neuronal differentiation, and neuronal proliferation, as well as pathologies including epilepsy, mood disorders, and migraine.

[0104] Figure 22 shows examples of 237 unique GWAS disease risk and pharmacogenomic SNPs that can be used to differentiate individual patient efficacy and adverse event profiles according to the antiepileptic, antimanic, and antimigraine properties of valproic acid. Figure 22A shows that multiple GWAS disease risk SNPs within the valproic acid adverse event subnetwork can be annotated as enhancers and super-enhancers associated with alopecia, CNS-mediated gastrointestinal disorders, and reduced efficacy of the combined antidepressant bupropion in bipolar depression. Figure 22B shows that the valproic acid pharmacogenomic neuroplasticity subnetwork contains multiple GWAS disease risk and pharmacogenomic SNPs that can be annotated as enhancers and super-enhancers significantly associated with valproic acid efficacy in chronic migraine, epilepsy, bipolar disorder (International Classification of Diseases (ICD) codes F31.0–F31.64), and bipolar mania.

[0105] Figure 23 shows validation of the system by comparing experimental results with those of publicly available data contained in the most widely used open-source and commercially available drug databases. The Venn diagram graphically displays the output of the present invention of the valproic acid pharmacogenomics network genes in human neural tissues compared with differentially regulated genes by valproic acid after drug administration to control pig (Sus scrofa) brains, and the finite set of genes thought to be regulated by valproic acid in all human tissues using Ingenuity Pathway Analysis™ (IPA; Qiagen; GmbH) and the Kyoto Encyclopedia of Genes (KEGG), as well as the LINCS database from DrugBank, the National Institutes of Health, and DrugCentral. The set of genes in the valproic acid pharmacogenomics network as output from the system described herein shows the highest degree of overlap in all head-to-head comparisons of results and data from all these sources.

[0106] FIG. 24 is a list of all genes included in the chromatin remodeling (CR) sub-network of the valproic acid pharmacogenomic network in the human brain.

[0107] Figure 25 is a list of all genes included in the neuroplasticity and drug efficacy (NP, EFF) sub-network of the valproic acid pharmacogenomics network in the human brain.

[0108] Figure 26 is a list of all genes included in the adverse events and neurotransmission (AE, NT) sub-network of the valproic acid pharmacogenomics network in the human brain.

[0109] Figure 27 is a list of all genes included in the pharmacokinetic and hormone regulation (PK, H) sub-network of the human brain valproic acid pharmacogenomics network.

[0110] Figure 29 shows the results of a post-hoc bioinformatics analysis of the ketamine pharmacogenomics network and its associated network topology model. Figure 29A shows that the most significant disease annotations of genes included in the ketamine pharmacogenomics network are schizophrenia, treatment-resistant depression, bipolar disorder, postoperative delirium, and postoperative pain.

[0111] Figure 29B shows the most significant drugs acting as upstream regulators of the ketamine pharmacogenomics network, including ketamine (p-value = 6.26E-33 ketamine; Fisher's exact test), morphine (p = 1.97E-17), and nicotine (p = 6.62E-17).

[0112] Figure 29C shows that the ketamine pharmacogenomics network, along with decomposed gene set subnetworks including chromatin remodeling (CR), adverse events (AE), neurotransmission (NT), neuroplasticity and drug efficacy (NP, EFF), and pharmacokinetics and hormone regulation (PK, H), fits the model network topology labeled 2. The non-CNS, peripheral systemic pharmacokinetic (SPK) subnetwork of the ketamine pharmacogenomics network cannot be determined from the output of this system.

[0113] Figure 37 shows an example of interactions in the 3D chromatin space of the ketamine pharmacogenomics network as an output of the methods and systems described herein. Genome-wide Hi-C data mapping was performed using SNPs as data probes. The probes included SNPs contained within the ketamine subnetwork and retrieved from the GWAS catalog, including those significantly associated with disease risk and ketamine antidepressant response and dissociation. These results validated the pathway analysis and demonstrated both cis and trans interactions with other members of the ketamine pharmacogenomics pathway in human neurons. These pharmacogenomic contacts were significantly enriched for associations with specific super-enhancers from the cingulate cortex and frontal cortex. Figure 37A shows a whole-genome plot that is key to understanding the gene-gene interactions shown in Figures 37B-37G. Figure 37B shows Hi-C contacts between RASGRF2, a gene associated with synaptic plasticity and alcoholism, and co-localized nicotinic receptor genes CHRNA3 and CHRNA5, which contain SNPs significantly associated with smoking status in GWAS. Figure 37C shows interactions between the ROBO2 gene, which contains numerous SNPs associated with unipolar depression and both dissociative and antidepressant responses to ketamine in GWAS, and both the GRIN2B and ATF7IP genes. The ATF7IP gene encodes a chromatin remodeling protein involved in HUSH-mediated heterochromatin formation and gene silencing as part of the SETDB1 complex stabilization required for histone 3 lysine 9 (H3K9me3) methylation. Figure 37D shows pharmacogenomic contacts between the TCF4 and GRM5 genes, which encode members of the glutamate metabotropic receptor family and contain enhancers significantly associated with depression in GWAS. Figure 37E shows a Hi-C map of interactions between the CACNA1C, GRIN2A, and ATF7IP2 genes. Figure 37F shows Hi-C pharmacogenomic contacts obtained from human glutamatergic neurons showing trans-interactions between the CAMK2A gene, located on chromosome 5, and the genes GRIN1 and ANAPC2, located on chromosome 9.The ANAPC2 protein is part of a complex that controls the formation of synaptic vesicle clusters at the active zone onto the presynaptic membrane of postmitotic neurons, and this complex also degrades NEUROD2 as a key component of presynaptic differentiation during neuronal differentiation. FIG. 37G shows a neuronal pharmacogenomic contact between the DRD2 gene and the RHOA gene, which encodes a signaling protein that regulates the cytoskeleton during neuronal synaptic transmission.

[0114] Figure 38 shows overlap between genes in the ketamine pharmacogenomics network in postmortem human brains and in the human brain regions where ketamine first exerts a rapid antidepressant response. This provides additional evidence supporting the ketamine pharmacogenomics network and can be demonstrated by comparing the neuroanatomical distribution of gene expression data within the ketamine subnetwork with localization results from a consensus brain map showing the brain regions first affected by ketamine obtained from 24 neuroimaging studies. The consensus map consistently highlights the anterior cingulate cortex (ACC), dorsolateral and dorsolateral prefrontal cortex (PFC), and supplementary motor area (SMA) as the first human brain regions activated by the drug. However, other CNS regions have been reported in neuroimaging studies to be rapidly affected by ketamine in humans after drug administration. While these are shown in black in Figure 12A, they did not constitute a clear majority of the brain regions reported to be first affected by ketamine in the neuroimaging studies examined during our study. To serve as controls, adjacent human brain regions unaffected by ketamine in neuroimaging studies were selected, including the corpus callosum (CC), occipital cortex (OC), and somatosensory cortex (SS). As shown in Figure 12B, genes in the ketamine network were expressed at significantly higher levels in the ACC and PFC than in the adjacent CC, SS, and OC, suggesting no evidence that ketamine exerts a rapid antidepressant effect. The ACC is part of the cingulate cortex, and the PFC is part of the frontal cortex.

[0115] Figure 30 shows the output of an iterative gene set optimization analysis of the ketamine pharmacogenomics network in the human brain. The larger ketamine pharmacogenomics network contains three subnetworks, generating two distinct subnetworks. The glutamate receptor subnetwork is enriched for synaptic signaling, glutamate receptor signaling, glutamate pathway regulation, and chromatin organization. The top xenobiotic (chemical) upregulator of the glutamate receptor subnetwork is ketamine with p=2.1E-09 (Figure 13A). In contrast, the neuroplasticity subnetwork is enriched for regulation of nervous system development, regulation of neurogenesis, regulation of neural differentiation, neurogenesis, and nervous system development (Figure 13B). The neuroplasticity subnetwork shows significant overlap with the "Cardiovascular, Neurological and Biological Injury Disorders" network category as determined by Ingenuity Pathway Analysis™ with P=1E-59, and its top xenobiotic upregulator is also ketamine with p=6E-12.

[0116] Figure 31 shows the ketamine pharmacogenomic glutamate receptor subnetwork, and post-hoc bioinformatics analysis indicates that the glutamate receptor subnetwork is significantly associated with cognitive impairment, bipolar disorder, postoperative delirium, schizoaffective disorder, schizophrenia, non-cancer pain, postoperative pain, vomiting, nausea, and loss of consciousness.

[0117] Figure 32 shows the ketamine pharmacogenomic neuroplasticity subnetwork, and post-hoc bioinformatics analysis indicates that the neuroplasticity subnetwork is significantly associated with emotional behavior, abnormal nervous system morphology, abnormal brain morphology, depression, anxiety, and abnormal neuronal morphology.

[0118] Figure 33 shows examples of 108 unique GWAS disease risk and pharmacogenomic SNPs that can be used to differentiate individual patient efficacy and adverse event profiles depending on the antidepressant and other actions of ketamine and its analogs. Figure 33A shows that multiple GWAS disease risk SNPs located within the glutamate receptor subnetwork can be annotated as enhancers associated with smoking status, chronic schizophrenia (ICD diagnosis code F20), and type 1 bipolar disorder (ICD diagnosis codes F31.0–F31.64). Figure 33B shows that the ketamine neuroplasticity subnetwork contains multiple GWAS disease risk SNPs that can be annotated as enhancers associated with recurrent depression (ICD code F33), alcoholism, and response to ketamine.

[0119] Figure 34 is a list of all genes included in the neuroplasticity and drug efficacy (NP, EFF) sub-network of the ketamine pharmacogenomics network in the human brain.

[0120] Figure 35 is a list of all genes included in the chromatin remodeling (CR), adverse event (AE), and neurotransmission (NT) subnetworks of the ketamine pharmacogenomics network in the human brain.

[0121] Figure 36 is a list of all genes included in the pharmacokinetic and hormone regulation (PK, H) sub-network of the human brain ketamine pharmacogenomics network.

[0122] Figure 42 shows the results of a post-hoc bioinformatics analysis of the lithium pharmacogenomic network and its associated network topology model. Figure 42A shows that the most significant disease annotations of genes included in the lithium pharmacogenomic network are cognitive impairment, mood disorder, drug-induced tremor, drug-induced weight gain, and schizophrenia.

[0123] Figure 42B shows the most significant drugs acting as upregulators of the lithium pharmacogenomic network, including lithium chloride (p-value = 2.23E-23; Fisher's exact test), lithium (p = 4.20E-19), and fluoxetine (p = 2.94E-14).

[0124] Figure 42C shows that the lithium pharmacogenomics network, along with decomposed gene set subnetworks including chromatin remodeling (CR), neuroplasticity (NP), efficacy (EFF), adverse events (AE), and adverse events (AE), conforms to the model network topology label 4. The non-CNS, peripheral systemic pharmacokinetic (SPK) subnetwork of the lithium pharmacogenomics network cannot be determined from the output of this system.

[0125] Figure 43 shows a high-resolution partitioning of gene set subnetworks as an example of an embodiment of detailed output using this system for the lithium pharmacogenomics network.

[0126] Figure 44 is a list of all genes included in the chromatin remodeling (CR) sub-network of the human brain lithium pharmacogenomic network.

[0127] Figure 45 is a list of all genes included in the neuroplasticity (NP) sub-network of the human brain lithium pharmacogenomic network.

[0128] Figure 46 is a list of all genes included in the efficacy sub-network (EFF) of the human brain lithium pharmacogenomic network.

[0129] Figure 47 is a list of all genes included in the drug-induced tremor, adverse event (AE) sub-network of the human brain lithium pharmacogenomics network.

[0130] Figure 48 is a list of all genes included in the drug-induced weight gain, adverse event (AE) sub-network of the human brain lithium pharmacogenomic network.

[0131] Figure 49 shows the results of a post-hoc bioinformatics analysis of the lamotrigine pharmacogenomics network and its associated network topology model. Figure 49A shows that the most significant disease annotations of genes included in the lamotrigine pharmacogenomics network are epilepsy, fibromyalgia, bipolar disorder type 1, mania, and treatment-resistant schizophrenia.

[0132] Figure 49B shows the most significant drugs acting as upstream regulators of the lamotrigine pharmacogenomic network, including lamotrigine (p-value=1.08E-10; Fisher's exact test), carbamazepine (p=3.51E-08), and mirtazapine (p=1.06E-07).

[0133] Figure 49C shows that the lamotrigine pharmacogenomics network, along with decomposed gene set subnetworks including chromatin remodeling (CR), adverse events (AE), neurotransmission (NT), neuroplasticity and drug efficacy (NP, EFF), and pharmacokinetics and hormone regulation (PK, H), fits the model network topology label 2. The non-CNS, peripheral systemic pharmacokinetic (SPK) subnetwork of the ketamine pharmacogenomics network cannot be determined from the output of this system.

[0134] Figure 50 shows the lamotrigine pharmacogenomic adverse event subnetwork, and post-hoc bioinformatics analysis indicates that the lamotrigine adverse event subnetwork is significantly associated with Stevens-Johnson syndrome, drug-induced hypersensitivity, progressive cognitive impairment, choroidal motion, and headache with dizziness.

[0135] Figure 51 shows the lamotrigine pharmacogenomic neuroplasticity subnetwork, and post-hoc bioinformatics analysis indicates that the lamotrigine pharmacogenomic neuroplasticity subnetwork is significantly associated with neuronal development and neuritogenesis, as well as pathologies such as epilepsy, fibromyalgia, bipolar disorder type 1, and mania.

[0136] Figure 52 is a list of all genes included in the chromatin remodeling (CR) sub-network of the lamotrigine pharmacogenomic network in the human brain.

[0137] Figure 53 is a list of all genes included in the neuroplasticity and efficacy (NP, EFF) sub-network of the lamotrigine pharmacogenomics network in the human brain.

[0138] Figure 54 is a list of all genes included in the adverse event (AE) sub-network of the lamotrigine pharmacogenomics network in the human brain.

[0139] Figure 55 is a list of all genes included in the pharmacokinetic (PK) sub-network of the lamotrigine pharmacogenomic network in the human brain.

[0140] Figure 56 shows the results of a post-hoc bioinformatics analysis of the clozapine pharmacogenomic network and its associated network topology model. Figure 56A shows that the most significant disease annotations of genes included in the clozapine pharmacogenomic network are psychosis, agitation, bipolar spectrum disorder, non-affective psychosis, and treatment-resistant schizophrenia.

[0141] Figure 56B shows the most significant drugs acting as upstream regulators of the clozapine pharmacogenomic network, including clozapine (p-value=8.85E-110; Fisher's exact test), haloperidol (p=1.45E-42), and chlorpromazine (p=6.95E-20).

[0142] Figure 56C shows that the clozapine pharmacogenomics network, along with decomposed gene set subnetworks including chromatin remodeling (CR), adverse events, central nervous system (AE), pharmacokinetics (PK), and adverse events, peripheral immune system (IAE), conforms to the model network topology label 3. The non-CNS, peripheral systemic pharmacokinetic (SPK) subnetwork of the clozapine pharmacogenomics network cannot be determined from the output of this system.

[0143] Figure 57 shows the clozapine pharmacogenomic adverse events CNS and peripheral immune system adverse events subnetwork (AE, IAE), and post-hoc bioinformatics analysis indicates that the clozapine adverse events subnetwork is significantly associated with glucose metabolism disorders, systemic autoimmune disorders, weight gain, drug-induced neutropenia, and neuronal apoptosis.

[0144] Figure 58 shows the clozapine pharmacogenomic efficacy subnetwork (EFF), and post-hoc bioinformatics analysis indicates that the clozapine efficacy subnetwork is significantly associated with psychosis, treatment-resistant schizophrenia, non-affective psychosis, manic bipolar disorder, recurrent schizophrenia, mania, bipolar disorder, treatment-refractory schizophrenia, and schizophrenia.

[0145] Figure 59 is a list of all genes included in the chromatin remodeling (CR) sub-network of the clozapine pharmacogenomic network in the human brain.

[0146] Figure 60 is a list of all genes included in the Efficacy of Clozapine (EFF) subnetwork of the clozapine pharmacogenomics network in the human brain.

[0147] Figure 61 is a list of all genes included in the adverse event (AE) sub-network of the clozapine pharmacogenomics network of the human brain and peripheral immune system.

[0148] Figure 62 is a list of all genes included in the pharmacokinetic (PK) sub-network of the clozapine pharmacogenomic network in the human brain.

[0149] Another application of these methods is the identification of genes encoding novel druggable molecules whose functions are known but which are not known to be part of the drug efficacy subnetwork for a class of drugs of interest. As shown in Figure 63, the warfarin pharmacogenomics network includes several candidate drug targets not previously known to be members of this drug's anticoagulant pharmacogenomics network. Pharmacogenomic network mapping methods using SNPs as part of drug pathway reconstruction allowed the addition of the following genes as part of subnetwork 1 mediating warfarin's anticoagulant effect: AXL (AXL receptor tyrosine kinase), F9 (clotting factor IX), MERTK (MER proto-oncogene tyrosine kinase), PDGFB (platelet-derived growth factor subunit B), PROC (protein C, inactivator of clotting factors Va and VIIIa), PROCR (protein C receptor), PROS1 (protein S), and PROZ (protein Z, vitamin K-dependent). Some of these novel genes may encode druggable products for use as anticoagulants.

[0150] As shown in Figure 63, the identified pharmacogenomic network of warfarin includes the ABO, alpha 1-3-N-acetylgalactosaminyltransferase and alpha 1-3-galactosyltransferase (ABO) genes, aldo-ketoreductase family 1 member C3 (AKR1C3) gene, AXL receptor tyrosine kinase (AXL) gene, complement factor H-related 5 (CFHR5) gene, cytochrome P450 family 2 subfamily C member 19 (CYP2C19) gene, and cytochrome P450 family 2 subfamily C member 19 (CYP2C19) gene. C member 8 (CYP2C8) gene, cytochrome P450 family 2 subfamily C member 9 (CYP2C9) gene, cytochrome P450 family 3 subfamily A member 4 (CYP3A4) gene, cytochrome P450 family 4 subfamily F member 2 (CYP4F2) gene, erythropoietin (EPO) gene, coagulation factor V (F5) gene, coagulation factor VII (F7) gene, coagulation factor IX (F9) gene, coagulation factor X (F10) gene, coagulation factor XI (F11) gene, coagulation factor XII (F12) gene, and coagulation factor XIII A chain (F13A1) gene, fibrinogen alpha chain (FGA) gene, fibrinogen gamma chain (FGG) gene, growth arrest specific 6 (GAS6) gene, histidine-rich glycoprotein (HRG) gene, kininogen 1 (KNG1) gene, lysozyme (LYZ) gene, MER proto-oncogene tyrosine kinase (MERTK) gene, matrix Gla protein (MGP) gene, orosomucoid 1 (ORM1) gene, polycomb group ring finger 3 (PCGF3) gene, platelet-derived growth factor receptor 3 (PDGF3) gene, and factor subunit B (PDGFB) gene, protein C, inactivator of coagulation factors Va and VIIIa (PROC) gene, protein C receptor (PROCR) gene, protein S (PROS1) gene, protein Z, vitamin K-dependent plasma glycoprotein (PROZ) gene, serine protease 8 (PRSS8) gene, serine protease 53 (PRSS53) gene, sphingosine kinase 1 (SPHK1) gene, signal transducer and activator of transcription 3 (STAT3) gene, and Syntaxin4 (STX4) gene, Surfeit 4 (SURF4) gene, transient receptor potential cation channel subfamily C member 4-related protein (TRPC4AP) gene, ubiquitin-specific peptidase 7 (USP7) gene, vitamin K epoxide reductase complex subunit 1 (VKORC1) gene, vitamin K epoxide reductase complex subunit 1-like 1 (VKORC1L1) gene, or von Willebrand factor (VWF) gene.

[0151] Throughout this specification, multiple instances may implement components, operations, or structures described as a single instance. While individual operations of one or more methods are illustrated and described as separate operations, one or more of the individual operations may be performed simultaneously, and the operations need not be performed in the order illustrated. Structures and functions presented as separate components in example configurations may be implemented as combined structures or components. Similarly, structures and functions presented as single components may be implemented as separate components. These and other variations, modifications, additions, and improvements are within the scope of the subject matter of this specification.

[0152] Furthermore, certain embodiments are described herein as including logic or a number of routines, subroutines, applications, or instructions. These can constitute either software (e.g., code embodied on a machine-readable medium or in a transmission signal) or hardware. In hardware, routines, etc., are tangible units capable of performing particular operations and can be configured or arranged in a particular way. In exemplary embodiments, one or more computer systems (e.g., standalone, client, or server computer systems), or one or more hardware modules of a computer system (e.g., a processor or processors), can be configured as hardware modules that operate by software (e.g., an application or portion of an application) to perform particular operations described herein.

[0153] In various embodiments, a hardware module can be implemented mechanically or electronically. For example, a hardware module may include dedicated circuitry or logic that is permanently configured to perform specific operations (e.g., a special-purpose processor such as a field programmable gate array (FPGA) or an application-specific integrated circuit (ASIC)). A hardware module may also include programmable logic or circuitry that is temporarily configured by software to perform specific operations (e.g., implemented in a general-purpose processor or other programmable processor). It will be appreciated that the decision as to whether a hardware module is implemented mechanically, with dedicated, permanently configured circuitry, or with temporarily configured circuitry (e.g., configured by software) may be based on cost and time considerations.

[0154] Thus, the term "hardware module" should be understood to encompass a tangible entity that is physically constructed, permanently configured (e.g., physically embedded), or temporarily configured (e.g., programmed) to operate in a particular manner or to perform particular operations as described herein. When considering embodiments in which hardware modules are temporarily configured (e.g., programmed), each of the hardware modules need not be configured or instantiated at any one time. For example, if a hardware module includes a general-purpose processor configured using software, the general-purpose processor can be configured as different hardware modules at different times. Thus, the software may configure the processor, for example, to configure a particular hardware module at one time and a different hardware module at another time.

[0155] Hardware modules can provide information to and receive information from other hardware modules. Thus, the hardware modules described herein can be understood as communicatively coupled. When such hardware modules coexist, communication can be achieved via signal transmission (through appropriate circuits and buses) connecting the hardware modules. In embodiments in which multiple hardware modules are configured or instantiated at different times, communication between such hardware modules can be achieved, for example, through the storage and retrieval of information in a memory structure to which the multiple hardware modules have access. For example, a hardware module can perform an operation and store the output of that operation in a memory device to which the hardware module is communicatively coupled. An additional hardware module can then later access the memory device to retrieve and process the stored output. Hardware modules can also initiate communication with input or output devices to operate on resources (e.g., collect information).

[0156] The various operations of the example methods described herein may be performed, at least in part, by one or more processors that are temporarily (e.g., by software) or permanently configured to perform the associated operations. Whether temporarily or permanently configured, such processors may constitute processor-implemented modules that operate to perform one or more operations or functions. Modules referred to herein may, in some example embodiments, include processor-implemented modules.

[0157] Similarly, methods or routines described herein may be at least partially processor-implemented. For example, at least some of the operations of a method may be performed by one or more processors or processor-implemented hardware modules. Certain performance of an operation may reside within a single machine, but may also be distributed among one or more processors deployed across several machines. In some embodiments, one or more processors may reside in a single location (e.g., in a home environment, in a work environment, or as a server farm), while in other embodiments, the processors may be distributed across multiple locations.

[0158] The assured performance of the operations can reside not only in a single machine but also be distributed among one or more processors deployed across several machines. In some exemplary embodiments, one or more processors or processor-implemented modules may reside in a single location (e.g., in a home environment, in a work environment, or as a server farm). In other exemplary embodiments, one or more processors or processor-implemented modules may be distributed across multiple locations.

[0159] Unless otherwise indicated, descriptions herein using words such as "processing," "computing," "calculating," "determining," "presenting," and "displaying" may refer to the operations or processes of a machine (e.g., a computer) that manipulates or transforms data represented as physical (e.g., electronic, magnetic, or optical) quantities in one or more memories (e.g., volatile memory, non-volatile memory, or a combination thereof), registers, or other machine components that receive, store, transmit, or display information.

[0160] As used herein, any reference to "one embodiment" or "an embodiment" means that a particular element, feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment.

[0161] Some embodiments may be described using the terms "coupled" and "connected," along with their variations. For example, some embodiments may be described using the term "coupled" to indicate that two or more elements are in direct physical or electrical contact. However, the term "coupled" may also mean that two or more elements are not in direct contact with each other, but yet still cooperate or interact with each other. The embodiments are not limited in this context.

[0162] As used herein, the terms "comprises," "comprising," "includes," "including," "has," or any other variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, article, or apparatus that includes a list of elements is not necessarily limited to only those elements and may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Furthermore, unless expressly stated to the contrary, "or" refers to an inclusive or, not an exclusive or. For example, condition A or B is satisfied by any one of A being true (or present) and B being false (or absent), A being false (or absent) and B being true (or present), and both A and B being true (or present).

[0163] Additionally, the use of "a" or "an" is used to describe elements and components of embodiments herein. This is done merely for convenience and to give a general gist of the description. This description should be read to include one or at least one, and the singular also includes the plural unless it is clear that otherwise is meant.

[0164] The detailed description should be construed as merely providing examples and does not describe every possible embodiment, as describing every possible embodiment would be impractical, if not impossible. Numerous alternative embodiments could be implemented using either current technology or technology developed after the filing date of this application.

Claims

1. 1. A method for identifying a pharmacogenomics network of a drug, comprising: obtaining, by one or more processors, a plurality of single nucleotide polymorphisms (SNPs) correlated with psychotropic drug response or adverse events; performing, by the one or more processors, 3D spatial connectivity mapping using the plurality of SNPs as probes in chromatin data to determine interconnections between target genes that correlate with drug response or adverse events of the psychotropic drug; identifying, by the one or more processors, a pharmacogenomic network of the psychotropic drug based on the target gene; decomposing, by the one or more processors, the pharmacogenomics network into a plurality of pharmacogenomics sub-networks; providing, by the one or more processors, a representation of the pharmacogenomics network and a representation of the plurality of pharmacogenomics subnetworks for display.

2. identifying a pharmacogenomic network of the psychotropic drug based on the set of target genes; comparing, by the one or more processors, the plurality of SNPs to a SNP database to identify additional SNPs linked to the plurality of SNPs using topological association domain (TAD) boundaries within a chromosomal region, wherein the plurality of SNPs and the additional SNPs are included in a set of candidate variants, and 3D spatial connectivity mapping is performed using the candidate variants; performing, by the one or more processors, a pathway analysis on the target genes associated with the set of candidate variants to filter the target genes and identify a set of genes causally related to the psychotropic drug; and identifying, by the one or more processors, the pharmacogenomic network for the psychotropic drug based on the identified set of genes.

3. 3. The method of claim 2, further comprising: performing, by the one or more processors, a bioinformatics analysis on the pharmacogenomics network to identify the drug most significantly associated with the plurality of pharmacogenomics subnetworks as the psychotropic drug.

4. performing a bioinformatics analysis on the pharmacogenomics network; 4. The method of claim 3, comprising validating the pharmaco-genomic network and the plurality of pharmaco-genomic subnetworks of the psychotropic drug via bioinformatics.

5. validating the pharmacogenomic network and the plurality of pharmacogenomic sub-networks for the psychotropic drug, the pharmacogenomics network of the psychotropic drug and the plurality of pharmacogenomics sub-networks, Terms from gene ontology or pharmaceutical databases; a standard biological pathway in the cell or tissue in which the psychotropic drug acts; or 5. The method of claim 4, comprising comparing the psychotropic drug with one or more upstream regulators of xenobiotics in the cell or tissue in which the psychotropic drug acts.

6. 3. The method of claim 2, wherein performing pathway analysis on target genes associated with a set of candidate variants to filter the target genes comprises identifying, by the one or more processors, a set of candidate pharmacogenomics network genes by identifying a subset of the target genes that form statistically significant interconnected pathways expressed in tissues associated with the psychotropic drug.

7. analyzing each gene in the set of candidate pharmacogenomics network genes according to at least one of the function of the gene relative to the psychotropic drug, the set of mutations within the gene, or the expression pattern of the gene relative to neuroanatomical substrates, as defined by RNA expression data, functional imaging, or other integrative multi-scale data indicative of where the psychotropic drug is known to act; analyzing functional genomic elements, including other genes and long non-coding RNAs, based on each gene in the set of candidate pharmacogenomics network genes; 7. The method of claim 6, further comprising adding or removing genes from said set of candidate drug pharmacogenomics network genes based on said analysis.

8. analyzing each gene in the set of candidate pharmacogenomics network genes according to an expression pattern of the gene relative to a neuroanatomical substrate; comparing each gene in the set of candidate pharmacogenomic network genes to a neuromap representing the neuroanatomical substrates of the psychotropic drug; and filtering genes from the set of candidate pharmacogenomics network genes that are not expressed in the same neuroanatomical region as the psychotropic drug.

9. 2. The method of claim 1, further comprising determining, by the one or more processors, one or more spatial contacts comprising a TAD based on the plurality of single nucleotide polymorphisms (SNPs) that correlate with drug response or adverse events of the psychotropic drug, wherein the one or more spatial contacts are differentially expanded and suppressed.

10. The pharmacogenomics network comprises: Mutations affecting enhancer-promoter pairs, promoter-promoter pairs, super-enhancer-promoter pairs, regulatory RNA, euchromatic or heterochromatic state, Topologically Associated Domains (TADs), or 10. The method of claim 1, comprising one of the more distinct features of functional spatial genomics, including at least one of lamina association domains (LADs).

11. 2. The method of claim 1, further comprising storing, by the one or more processors, the representation of the pharmacogenomic network for the psychotropic drug and the representations of the plurality of pharmacogenomic subnetworks in a database as a reference set for comparing and scoring input patient pharmacogenomic networks for the same psychotropic drug to determine efficacy and adverse events when treating a patient for a condition or disease.

12. 10. The method of claim 1, further comprising identifying, by the one or more processors, a set of molecular pharmacodynamic targets within one of the plurality of pharmacogenomic subnetworks for the same or similar clinical indication as the psychotropic drug.

13. 13. The method of claim 12, wherein the set of molecular pharmacodynamic targets and the psychotropic drug within one of the plurality of pharmacogenomic sub-networks act in the human brain, and a majority of members of the pathways of the molecular pharmacodynamic targets are expressed in the same region of the human brain as the psychotropic drug.

14. 10. The method of claim 1, further comprising repurposing drugs based on loci underlying pharmacogenomic enhancer SNPs associated with the plurality of pharmacogenomic subnetworks within the subnetwork type topology.

15. 10. The method of claim 1, further comprising repurposing the drug combination for testing as a therapeutic agent after physiological mechanisms demonstrate that complementarity between subnetwork properties of two or more drugs provides a more optimal treatment for a particular disease or condition than a single psychotropic drug.

16. 16. The method of claim 15, wherein repurposing a drug combination comprises repurposing valproic acid and ketamine as a combined treatment for neurological and neuropsychiatric disorders.

17. 10. The method of claim 1, wherein the psychotropic drug is one of valproic acid, ketamine, lithium, lamotrigine, or clozapine.

Images