Abstract

High-throughput sequencing of 16S gene or metagenomes provides an unprecedented opportunity to discover microbes associated with traits such as clinical outcomes or environmental factors. However, the microbial data are highly complex because they are compositional, sparse (50 to 90% zeros), high-dimensional, and in particular subject to ubiquitous experimental biases. Existing methods developed specifically for compositional analysis of the microbiome data cannot always control the false discovery rate and often require replacing zeros with a pseudocount. Our methods, based a logistic-regression model, always preserve the false discovery rate, has much improved sensitivity over existing methods, does not require pseudocounts, and thus can accelerate the search for microbial biomarkers for prognosis and diagnosis of diseases or microbial targets for drug discovery.

Biography

胡懿娟，北京大学博雅特聘教授，入选国家级人才计划。双聘于北京国际数学研究中心和北大医学部生物统计系。在北京大学数学科学学院获得学士学位（2005）和美国北卡教堂山大学获得生物统计学博士学位（2011）。在美国埃默里大学历任助理教授、副教授和教授。于2024年7月全职回国。致力于开发生物统计学中高维度、高噪声组学数据的统计理论和方法，特别针对微生物组数据和遗传数据中的高维假设检验、稳健推测、缺失/偏差数据等问题。代表工作发表于Journal of American Statistical Association (JASA) 、Proceedings of the National Academy of Sciences(PNAS) 、Microbiome、 American Journal of Human Genetics (AJHG) 等期刊。