Time-Varying Stochastic Block Models via Kernel Smoothing, with Application to RNA-Seq Data and Cell Development

• Activity Number: 44 - Statistical Methods in Gene Expression Data Analysis I
• Type: Contributed
• Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Kevin Lin* and Jing Lei and Kathryn Roeder
• Companies: Carnegie Mellon University and Carnegie Mellon University and Carnegie Mellon University
• Keywords: stochastic block model; time varying; kernel smoothing; gene networks; cell development; gene clutsers

Abstract

Gene networks typically change as cells develop, implying the necessity to account for time-varying behaviors when we analyze genomic data. In this work, we develop a kernel-smoothing method to estimate time-varying stochastic block models. We prove theoretical results for estimating the memberships as well as the connectivity matrix among the clusters across time. To do this, similar to other works, we use techniques from nonparametric regression to analyze our estimator, but we weaken the existing assumptions and obtain sharper rates by building upon the latest developments in tensor concentration results and spectral perturbation bounds. We apply our method to RNA-seq data to estimate the connectivity among cell clusters across development, and provide scientifically-principled ways to tune and validate our procedure.