Title: "Low-Rank Approximation and Its Applications in Bioinformatics"
Speaker: Can Yang; Department of Biostatistics & Psychiatry, Yale School of Public Health & Medicine

Place: Rawls (RAWL) Hall 1086
Date: October 22, 2013; Tuesday
Time: 4:30pm

Abstract:

Recently, low-rank approximation has arisen as an effective tool in the areas of statistics and machine learning (e.g., robust principal component analysis, matrix completion, etc.). In this talk, I am going to introduce low-rank approximation and demonstrate its applications in Bioinformatics. Specifically, accounting for confounding effects in expression quantitative trait loci (eQTL) mapping, and detecting copy-number variation (CNV).

In eQTL studies, the relationship between gene expression levels and genotypes is under systematic investigation. A major issue in inferring eQTL is that a few number of factors, such as unobserved covariates, experimental artifacts, and unknown environmental perturbations, may confound the observed expression levels. This may both mask real associations, and lead to spurious associations. The key challenge accounting for the confounding effects is that these factors may not be directly and completely observable, and thus remain hidden. Noticing that the effects of a few of unobserved confounding variables can be captured by a low-rank structure, this problem can be formulated as low-rank approximation in the setting of sparse regression. An efficient algorithm is available for large-scale data analysis and its performance has been demonstrated in both simulated and real data.

Several other applications of low-rank approximation, such as joint detection of CNV in multiple samples, and moving-object segmentation in biomedical image analysis, will be presented.