Finding important associations in genetic and genomic data

点击翻译按钮获取中文摘要

The upcoming era of research in genetics and genomics can be characterized as a data integration era. By combining various kinds of data, measured possibly in different ways, at different times or tissues, or from several sources, the hope is to better understand what are the relevant factors influencing disease or traits, and how they are linked together. This ongoing program of research has the long-term objective of developing statistical methods applicable to analysis of high-throughput genetic and genomic data of various types, with the over-arching theme of better using information from multiple sources. Recent progress from my previous NSERC Discovery Grant (such as developing and implementing a stratified false discovery rate approach for interpreting the results of genome-wide association studies, or investigating pathway methods for rare genetic variants), has led to the main objectives proposed here over the next five years. The two main objectives are: (1) When performing a large number of tests of significance, methods are proposed that will improve the ranking of the results by incorporating external information. From a ranked list of results, the most interesting ones are selected for further investigation in subsequent studies. Therefore obtaining improved rankings implies not only a better power to find true signals, but also a more effective use of resources for the follow-up research. I propose methodological improvements through effective understanding of the correlations between the tests, and using external genomic annotation information. (2) Since the number of quantities that can be measured in genomics is so much larger than the number of individuals, it is easy to develop a predictive model that is optimized for one single data set, but performs poorly on new data. In this second objective, the goal is to address this concern in two ways. Firstly, I propose to work on development of data-type-specific methods for defining summaries that capture the intrinsic signals in the data in a much smaller number of measures. Then secondly, I plan to develop ways of combining these summary signals across different types of data, to give better accuracy of the prediction models. Whenever possible, the proposed methods will be programmed and released as publicly available software packages. Hence, the body of work developed through this proposal will be beneficial to collaborators who are generating massive genetic and genomic data sets, and who are searching for effective methods of analysis.