Deciphering cellular signaling system by deep mining a comprehensive genomic compendium

 描述（由申请人提供）： 细胞信号转导系统（Cellular Signal Transduction System，CSTS）在维持细胞内环境稳定中起着重要的作用，CSTS的紊乱会导致癌症、糖尿病等疾病的发生。大多数细胞信号通路最终调节基因表达，因此后者可以作为反映CSTS状态的通用生物测定。然而，来自细胞的基因表达谱反映了对所有活性信号传导途径的反应的混合物，因此对嵌入在基因表达数据中的信号进行去卷积是一个挑战。随着高通量生物技术的出现，主要的公共数据库托管了数百万次基因表达测定的结果，这些结果是在不同疾病以及实验生物学家设计的实验条件下收集的，用于探索CSTS的几乎所有方面。这些丰富的大数据为研究生理和病理条件下的CSTS提供了前所未有的机会，但也带来了前所未有的挑战：如何从错综复杂的数据中揭示信号，并将大数据转化为知识。我们假设，给定在不同条件下收集的足够大的基因表达数据概要，其中CSTS的不同部分受到干扰（无论是通过设计的实验还是疾病），嵌入基因表达数据中的细胞信号可以被揭示，并且可以使用当前最先进的“深度学习”模型推断它们的组织。在这个项目中，我们将编译一个人类基因表达数据的综合纲要，然后使用现代深度学习算法和超级计算机来挖掘数据。我们的目标是揭示在生理和病理条件下调节基因表达的主要细胞信号，并推断人类CSTS中的信号组织。将识别的信号与基因组改变数据和药物反应数据相结合，我们的目标是进一步识别疾病（如癌症）的潜在途径，使用基因组数据预测癌细胞系的药物敏感性，并以途径为中心的方式预测患者的临床结果。

 DESCRIPTION (provided by applicant): Cellular signal transduction system (CSTS) plays a fundamental role in maintaining homeostasis of a cell, and perturbations of CSTS lead to diseases such as cancers and diabetes. Most of cellular signaling pathways eventually regulate gene expression, thus the latter can be used as a universal bioassay reflecting the state of CSTS. However, the gene expression profile from a cell reflects a mixture of responses to all active signaling pathways, thus it is a challenge to de-convolute the signals embedded in the gene expression data. With advent of high throughput biotechnology, major public databases host the results of millions of gene expression assays, collected under diverse diseases as well as the experimental conditions designed by experimental biologists to probe almost all aspects of CSTS. The wealth of these big data provides unprecedented opportunities to investigate CSTS under physiological and pathological conditions, but it also poses unprecedented challenges: how to reveal signals from convoluted data and turn big data into knowledge. We hypothesize that given a sufficiently large compendium of gene expression data collected under diverse conditions, in which different parts of CSTS are perturbed (either by designed experiments or by diseases), the cellular signals embedded in the gene expression data can be revealed and their organization can inferred using current state-of-the-art "deep learning" models. In this project, we will compil a comprehensive compendium of human gene expression data and then employ modern deep-learning algorithms and supercomputers to mine the data. We aim to reveal major cellular signals that regulate gene expression under physiological and pathological conditions and to infer the organization of signals in human CSTS. Combined the identified signals with genomic alteration data and drug response data, we aim to further identify pathways underlying disease such as cancers, to use the genomic data to predict drug sensitivity of cancer cell lines, and to predict patient clinical outcomes, in a pathway-centered manner.