Deep learning methods to predict the function of genetic variants in orofacial clefts

项目摘要 口面裂（OFC）在所有人群中（范围为100 - 200）占人类出生缺陷的很大比例。 1/500至1/2500活产），是一项重大的公共卫生挑战。出生时带有OFC的人 需要手术、营养、牙科、言语、医疗和行为干预， 经济和个人负担。有令人信服的证据表明，非综合征型OFC代表了 具有多因素病因学的人类复杂疾病，包括遗传风险因素，环境暴露， 以及它们之间复杂的相互作用到目前为止，已经有大约10个全基因组关联研究（GWAS） 对非综合征CL/P（NSCL/P）和>15个基因组位点进行了研究，报告了令人信服的统计学支持， 包括IRF 6、PAX 7、ABCA 4和8 q24基因座等基因。此外，下一代测序 (NGS)以及外显子组阵列已经进行了额外的基因分型深度，使检测罕见的 与OFC相关的变体。然而，在如何解释这些变体以及如何识别新变体方面存在差距 大量数据的变体，对“第二次”的新方法和新模型有很高的期望， 分析“全基因组数据”。在这一建议中，我们提出了两个相辅相成的目标， 对OFC的全基因组数据进行第二次分析。在目标1中，我们提出了一种深度学习方法， 这些模型可以在丰富的颅面表观基因组特征的背景下预测遗传变异的影响。与 将揭示序列模式的实质性精细图谱、特别基序以及干扰这些基序的变体 将提供关于OFC的机械见解。在目标2中，我们将重点转移到基因水平，并提出了一个网络 一种辅助方法，用于发现在空间和时间点上对口面至关重要的敏感组合基因 发展我们的目标是所有形式的离岸金融中心，特别是在非上市公司/P的兴趣。 在这个提议中，我们组建了一个多学科团队和配备GPU的本地计算基础设施， 实现这两个目标。我们的目标是不重叠的;相反，它们是一体化的，重点突出 关于我们感兴趣的基本问题：遗传变异如何导致OFC。圆满完成 我们的建议将导致对OFC遗传成分的深入了解。

Project Summary Orofacial clefts (OFCs) comprise a significant fraction of human birth defects in all populations (ranging between 1/500 to 1/2500 live births) and represent a major public health challenge. Individuals born with OFCs require surgical, nutritional, dental, speech, medical and behavioral interventions, imposing a substantial economic and personal burden. There has been convincing evidence that non-syndromic OFCs represent human complex disorders with a multifactorial etiology including genetic risk factors, environmental exposures, and their complex interactions. So far, there have been ~10 genome-wide association studies (GWAS) conducted for non-syndromic CL/P (NSCL/P) and >15 genomic loci reported with compelling statistical support, including genes such as IRF6, PAX7, and ABCA4 and the 8q24 locus. In addition, next-generation sequencing (NGS) as well as exome array have been conducted with extra depth of genotyping that enable detection of rare variants associated with OFCs. However, gaps exist in how to interpret these variants and how to identify novel variants from the large volume of data, with high expectations for new methods and new models for “second- analysis” of the genome-wide data. In this proposal, we propose two complementary aims to carry out deep and second-analysis of genome-wide data for OFCs. In Aim 1, we propose a deep learning method to build in silico models that can predict the effect of genetic variants in the context of rich craniofacial epigenomic features. With substantial fine map of sequence patterns, ad hoc motifs will be revealed and variants that disturb these motifs will provide mechanistic insights on OFCs. In Aim 2, we shift our focus to the gene level and propose a network assisted method to discover sensibly combined genes in spatial and temporal points that are critical to orofacial development. We target on all forms of OFCs, with particular interest in NSCL/P. To guarantee the success of this proposal, we form a multi-disciplinary team and local computational infrastructure equipped with GPUs for the implementation of both aims. Our aims are non-overlapping; rather, they are integrated and strongly focused on our fundamental question of interest: how genetic variants function to cause OFCs. The successful completion of our proposal will lead to deep understanding of genetic components in OFCs.