Multi-sample deconvolution and cell-type-specific inference



报告地点:国家天元数学西南中心516报告厅  腾讯会议 334-799-481

报告摘要:Existing tools for cell abundance quantification in bulk omics data either underestimate the between-subject heterogeneity of reference panel or rarely incorporate the multimodalities of molecules. Here, we employ the support vector machine and the linear mixed effect model with Restricted Maximum Likelihood or EM algorithm to recover individualized reference panel, improving the quantified cellular composition and cell-type-specific statistical inference. On the other hand, for tissue-matched transcriptome-proteome data, we integrate the bulk multi-omics profiles by assuming cell counts fractions shared across molecular modalities and estimating the high-dimensional parameters with Joint Nonnegative Matrix Factorization and Projected Gradient Descent. We rigorously benchmarked the performance of our methods and competing methods via extensive simulation studies and applications to different diseases, with implementation in Bioconductor packages ISELT and MICSQTL.

专家简介:黎倩,助理教授,博士后导师,美国圣裘德儿童研究医院生物统计系。生物统计与计算生物学者。2009 年于四川大学数学学院 (基础数学)获学士学位,2015 年于密苏里大学堪萨斯城数学统计系获博士学位。曾在美国堪萨斯大学医学院和墨菲特癌症中心从事博士后研究,以及在南佛罗里达大学医学院担任研究助理教授。主要研究方向包括纵向数据、高维数据的统计模型与方法在多组学(转录组,蛋白质组,甲基化,代谢组,微生物),流行病学以及癌症研究中的应用。发表了 30 篇学术论文,其中 13 篇为第一作者或通讯/高级作者并包括高影响力国际学术期刊:Bioinformatics, Genome Biology, Genome  Medicine, Journal of National Cancer Institute, Communications Biology, Diabetes。曾获得美国国立卫生院(NIH)科研资助。