Abstract：High-dimensional compositional data are frequently encountered nowadays in scientific research of many disciplines, such as in high-throughput sequencing experiments widely used in modern biological and biomedical studies. Statistical analyses with a single compositional dataset have been well studied in the past a few decades in different application contexts, such as regression analysis, clustering analysis, hypothesis testing and so on. However, the inventory of statistical analysis tools for multiple compositional datasets is surprisingly limited, especially in a high-dimensional setting. To fill this research gap, we focus on statistical integrative analysis of multiple compositional datasets in this talk. We first discuss a horizontal integrative analysis, where both predictors and responses are compositional. To investigate associations between two high-dimensional compositional vectors, we propose a Composition-On-Composition regression analysis framework. Then, we introduce our recent vertical integration analysis method that takes two compositional vectors as predictors. Comparing to individual analysis with a single set of compositional predictors, our vertical integration analysis significantly boost the statistical power of association testing between a scalar response variable and two sets of compositional predictors. Superior performance of both integrative analysis methods for multiple compositions are demonstrated through comprehensive numerical simulations studies and real data application examples.

Bio：占翔，东南大学统计与数据科学学院教授、博士生导师。近年来一直从事高维复杂分子生物组学数据的统计分析理论方法研究。先后主持1项美国国家科学基金委 (NSF) 科研基金项目，1项美国国家卫生研究院 (NIH) 科研基金项目，1项国家自然科学基金面上项目，在生物统计学相关领域国际权威期刊JASA, Annals of Applied Statistics, Biometrics, Bioinformatics等发表科研论文50余篇，其中27篇为第一作者或通讯作者。