Applied Math Seminar: Umang Varma “Marker Gene Selection in scRNA-seq via Mutual Information”

Single cell RNA sequencing (scRNA-seq) gives biologists a unique insight into the heterogeneity of cells. However, the analysis of scRNA-seq data poses novel biological and algorithmic challenges. One such problem is that of marker selection: given a gene expression matrix and cluster labels for each cell, determine a relatively small subset of genes that is most informative about the given clustering. In the machine learning literature, this problem is known as feature selection. The focus of this talk is on information theoretic approaches to marker selection. We will discuss algorithms in the literature, examine their shortcomings, and present a principled approach to addressing these shortcomings.