整合序列与染色质状态大数据解析基因调控模式的深度学习方法

解析组织特异的基因调控模式是研究细胞分化、个体发育、疾病发生的基础。近年来转录组和表观遗传学实验数据大量积累，但研究方法以单类数据为主，定性分析居多，尚缺乏智能化深入挖掘和多种数据整合分析，更忽略了基因组序列这一基础信息来源。针对这些问题，本项目研究整合序列与染色质状态大数据解析基因调控模式的深度学习方法。首先整合基因组序列、染色质状态、基因表达等数据建立深度学习模型，实现细胞系特异的调控元件预测。其次整合调控元件与基因的序列和实验数据，使用自然语言处理技术和深度学习模型实现细胞系特异的调控元件目标基因预测。最后发展基于基因调控图谱的统计遗传学模型，建立全基因组遗传学数据解读新方法。项目预期成果将汇总为三个数据库和三套分析软件，通过肺动脉高压全基因组测序数据开展综合示范应用研究，从而推动深度学习在生物医学大数据分析中的应用实践，为运用信息科学理论方法解决生命科学关键问题提供有益借鉴。

Deciphering mechanisms of gene regulation specific to a tissue is of fundamental importance to the understanding of cell differentiation, individual development and disease mechanism. In recent years, experimental big data have been rapidly accumulated for both transcriptomics and epigenetics. However, most studies thus far focused on a single type of data and produced qualitative results, not only lacking intelligent deep mining and integrative analysis of multiple types of data, but also overlooking the basic information in the sequence of a genome. Aiming at solving these shortcomings, we propose to develop deep learning approaches for deciphering gene regulation patterns via the integration of sequence and chromatin accessibility big data. We first predict regulatory elements specific to a tissue or cell line by deep learning models that integrate sequence, chromatin status and gene expression data. Then, we integrate sequence and experiment data to predict target genes of these elements by using natural language processing techniques and deep learning models. Finally, we develop novel statistical models to understand whole genome genetic data based on tissue specific regulatory networks composed of genes and regulatory elements. We will summarize research results as three databases and three sets of software, and demonstrate their applications via the analysis of a real whole genome sequencing data set of pulmonary artery hypertension (PAH), thereby promoting applications of deep learning in biological and medical big data, and providing a reference to applications of theories and techniques in information sciences to solving important scientific questions in life sciences.