Integrating Genomic Data and Biological Knowledge to Learn Context-Specific Gene

描述（由申请人提供）：癌症是一种相当复杂的疾病。癌症起源于并受多种基因组变化的支持，这些变化常常导致细胞控制电路的剧烈变化，从而在癌细胞中产生大量不同的分子机制。当试图从多样化的人群中进行推断时，网络监管连接或操作规则的异质性模糊了这些关系，并迅速降低了准确识别监管相互作用的能力。这对那些试图确定在所有可以看到的变化中哪些是结果性的人提出了相当大的挑战。然而，了解这些调控变化将有助于发现预后标志物，并提供强有力的候选药物靶点。目前的分析方法很少考虑这种复杂性和异质性。此外，目前对调控机制进行逆向工程的方法主要涉及一组几乎同质的组件，这些组件几乎同质地共同调控。本应用程序建议开发计算方法，可以搜索异质样本集，以识别样本子集，其中协同执行特定病理功能的基因集是均匀调节的。然后，该方法利用具有较高同质性的识别样本子集来学习上下文特定的调节机制。该方法将使用多种分子和临床特征的同时分析，以及在高度异质性背景下检测有限的行为同质性的分析策略。这些方法将通过随后对三个NIH资助项目（包括多发性骨髓瘤、胶质母细胞瘤和胰腺癌）的基因组数据、临床信息和现有生物学知识的综合分析来验证。开发的算法将作为一套具有图形用户界面的计算机软件实施，这些软件将公开提供。

DESCRIPTION (provided by applicant): Cancer is a disease of considerable complexity. Cancer originates from and is supported by a wide variety of alterations in the genome that often lead to drastic alterations of the cell's control circuitry, producing a great deal of diversity in the molecular mechanisms operating in cancer cells. When attempting to make inferences from a diverse population, heterogeneity in either the network regulatory connections or operating rules blurs the relationships and rapidly reduces the ability to accurately discern regulatory interactions. This provides a considerable challenge to those attempting to determine which, of all the changes that can be seen, are consequential. However, knowledge of these regulatory changes will facilitate the discovery of prognostic markers and provide strong candidate drug targets. Current analytic methods rarely consider such complexity and heterogeneity. In addition, current methods to reverse-engineer regulatory mechanisms mostly concern a nearly homogeneous set of components that are nearly homogeneously co-regulated. This application proposes to develop computational methods that can search through heterogeneous sample sets to identify subsets of samples in which sets of genes that collaborate to carry out particular pathologic functions are homogeneously regulated. The method then utilizes identified subsets of samples with higher homogeneity to learn context-specific regulatory mechanisms. The method will use simultaneous analysis of a variety of molecular and clinical characterizations, and an analytic strategy that detects limited homogeneity of behavior against a background of high heterogeneity. The methods will be validated via subsequent analysis of combined genomic data, clinical information and available biological knowledge from three NIH funded projects which include multiple myeloma, glioblastoma and pancreatic cancers. Developed algorithms will be implemented as a set of computer software with graphical user-interface, which will be publicly available. This project is intended to produce types of analysis that are specifically designed to identify collaborative molecular behaviors in cancer via simultaneous analysis of multiple types of biomedical data, and to make them available to biomedical researchers, in the form of easily utilized software so that anyone with these types of biomedical data can use the methods developed in this project. Even a modest improvement in the ability to predict what patients would benefit from what treatments would significantly improve patient care. Understandings that would identify sets of synergistic drugs could have even higher impact.