Integrating epigenomic maps to predict regulatory functions of genetic variants

精神疾病是影响人类的最具破坏性的疾病之一，给个人、家庭和社会带来巨大负担。全基因组关联研究（GWAS）已经确定了数十种与精神疾病相关的常见单核苷酸多态性（SNP），但这些SNP中的大多数已被定位于基因间或内含子区域，并且在功能上未分类。现有的软件或算法只能查询多个数据库，并生成命中列表，而没有智能集成，忽略了许多有价值的监管信息。该提案的总体目标是整合所有可用的遗传、基因组和表观基因组数据， 生成关于SNP对脑中基因表达水平的影响的基于概率的预测。我们以前的研究表明，精神病GWAS信号富含脑eQTL SNPs（eSNPs），这些脑eSNPs可能是功能性的，并有助于疾病易感性。我们将使用eQTL中的SNP来锚证据链，包括组蛋白标记、保守序列、转录因子结合位点、DNA甲基化、可接近的染色质、非编码RNA和其他数据。我们将使用机器学习方法来预测基于这些表观遗传标记与其靶基因之间的已知关系以及它们在基因组中的独特模式的调控SNP。我们还将使用我们新颖的无监督去卷积算法来提取细胞类型（即，神经元与非神经元）特异性测量来改进我们的预测。我们将使用统计和实验方法来验证预测。将在诱导多能细胞系上使用定量PCR和CRISPR-cas9来比较预测的功能SNP的等位基因的基因表达水平。算法和预测的功能变体都将通过网站和独立应用程序公开。新算法将通过揭示疾病相关的非编码SNP的基因调控功能，显着提高我们对精神疾病遗传学的理解。

DESCRIPTION (provided by applicant): Mental illnesses are some of the most devastating diseases affecting human populations, placing a huge burden on individuals, families and society. Genome-wide association studies (GWAS) have identified dozens of common single nucleotide polymorphisms (SNPs) that are associated with psychiatric diseases, but a majority of those SNPs have been mapped to intergenic or intronic regions and are functionally unclassified. Existing software or algorithms only query multiple databases and produce lists of hits without intelligent integration and ignore much of the valuable regulatory information. The overall goal of this proposal is to integrate all available genetic, genomic and epigenomic data to generate a probability-based prediction about a SNP's influence on gene expression level in brain. Our previous studies have shown that psychiatric GWAS signals are enriched with brain eQTL SNPs (eSNPs), and these brain eSNPs are likely to be functional and contribute to disease susceptibilities. We will use SNPs in eQTLs to anchor a chain of evidence incorporating histone marks, conserved sequences, transcription factor binding sites, DNA methylation, accessible chromatins, non-coding RNA, and other data. We will use a machine learning method to predict regulatory SNPs based on known relationships between these epigenetic marks and their target genes, as well as their distinct patterns in genome. We will also use our novel unsupervised deconvolution algorithm to extract cell-type (i.e., neuron vs. non-neuron) specific measures from heterogeneous brain tissue data to improve our predictions. We will use both statistical and experimental methods to validate the predictions. Quantitative PCR and CRISPR-cas9 will be used on induced pluripotent cell lines to compare gene expression levels of alleles of predicted functional SNPs. Both algorithm and predicted functional variants will made public via a website and standalone application. The novel algorithm will significantly improve our understanding of psychiatric disease genetics by uncovering the gene-regulatory functions for disease-associated, non-coding SNPs.