学术动态

12月9日 | 林伟:Absolute or Relative: Basis Regression with Compositional Data

时  间:2021年12月9日(周四)13:00-16:00

地  点:腾讯会议:961 561 656

题  目:Absolute or Relative: Basis Regression with Compositional Data

主讲人:林伟 北京大学数学科学学院研究员

主持人:方方 教授

摘  要:

Regression with compositional data arises in applications such as associating the microbiome with host phenotypes. Existing methods tend to relate microbial relative abundances to the response, which impose strong restrictions on the effect signs and sizes. Motivated by recent efforts to quantify microbial absolute abundances, we propose a basis regression methodology for linking the absolute abundances with the response using only the compositional data, thereby allowing the effects of the basis variables to be directly and individually interpreted. By decomposing the influence of the basis into the effect of the composition and that of the total abundance, we show that the bias due to not observing the total abundance vanishes as the dimensionality increases and that a surprisingly simple lasso estimator based on the centered log-ratio variables achieves estimation and variable selection guarantees. We demonstrate through simulations the superiority of our method over two alternative procedures for confounding adjustment, and contrast it with compositional regression in gut microbiome data.

报告人简介:

林伟,北京大学数学科学学院概率统计系、统计科学中心长聘副教授、研究员,统计学教研室主任。2011年获南加州大学应用数学博士学位,2011至2014年在宾夕法尼亚大学做博士后研究,2014年加入北京大学。主要研究方向为高维统计、统计机器学习、成分数据分析、因果推断、生存分析等,以及在基因组学、宏碁因组学和环境科学等领域的应用,代表性成果发表在JASA、Biometrika、Biometrics、IEEE TIT、Operations Research、Environmental Science & Technology、《中国科学:数学》等统计学及相关领域顶级期刊上。2015年入选国家高层次人才计划青年项目,主持国家重点研发计划课题、北京市自然科学基金重点研究专题项目、国家自然科学基金面上项目等。


发布者:张瑛发布时间:2021-12-02浏览次数:13