Zero-Inflated Random Forests for Genetic Regulatory Network Estimation in Single Cell RNA-Seq Data

• Activity Number: 179 - Statistical Methods in Single-Cell Transcriptomics
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Daniel Conn* and Kirby Johnson and Emery Bresnick and Sunduz Keles
• Companies: University of Wisconsin, Madison and University of Wisconsin, Madison and University of Wisconsin, Madison and University of Wisconsin, Madison
• Keywords: Random forests; Single Cell RNA Seq; genetic regulatory network

Abstract

We present a new random forests (RFs) algorithm for count valued outcomes, and we apply it to single-cell RNA-Sequencing (scRNA-seq) data in the context of genetic regulatory network (GRN) estimation. Specifically, we embed our algorithm into the single-cell GRN network method SCENIC to find direct targets for transcription factors. The algorithm accounts for zero-inflation via an iterative Monte-Carlo expectation-maximization (MC-EM) process. It is crucial that zero-inflation be incorporated into the algorithm because it is believed that the generative process for the excess zeroes is primarily driven by technical artifacts. As a result, we argue that our MC-EM RF algorithm targets biologically relevant parameters, in contrast to standard RFs (or other commonly used machine learning methods). We evaluate our method on 20 scRNA-seq data sets generated by the Tabula Muris Consortium.