Event Title

Transcriptional Regulatory Network Inference from Single-Cell RNA Measurements in Embryonic Stem Cells

Presenter Information

Diep Nguyen, Oberlin College

Location

Science Center, Bent Corridor

Start Date

10-27-2017 6:40 PM

End Date

10-27-2017 7:20 PM

Research Program

Cincinnati Children's Hospital Summer Undergraduate Research Fellowship

Poster Number

48

Abstract

Single-cell RNA-seq (scRNA-seq) allows transcriptome-wide analyses of individual cells and quantifies intra-population heterogeneity. Here we develop computational tools to learn how these gene expression patterns are regulated, modeling these patterns as functions of transcription factor activities (TFA). However, due to low capture rate, scRNA-seq data matrices are sparse and contain many ambiguous zeros, in which sampling zeros, “dropout genes”, and true zeros are not well distinguished. For downstream analyses, biological variations, the trends of interest, should be resolved from technical variations. Four statistical approaches to normalize scRNA data were benchmarked: Reads Per Million (RPM), Markov Affinity-based Graph Imputation of Cells (MAGIC), Bayesian Inference for Single-cell Clustering and Imputing (BISCUIT), and Robust Principal Component Analysis (RPCA). Regulatory networks are inferred via the Inferelator algorithm, Bayesian Best Subset Regression, with the incorporation of ATAC-seq priors. TFA can be estimated with mRNA levels or with the expression levels of its known targets using priors. Post-Inferelator networks are then compared with gold-standard and bulk-RNA networks via precision-recall analyses. Of four methods, MAGIC and RPCA demonstrate promising ability to impute “dropout genes” while preserving cell-cell variations and recovering cluster-specific gene expression. Preliminary network data show that using mRNA levels have higher predictive power of the models than estimated TFA. The next step is to continue finessing methods that enable inference of TRNs from scRNA-seq with robust predictive power, increasing the impact of single-cell genomics from qualitative descriptive to quantitatively predictive.

Major

Biology; Computer Science

Project Mentor(s)

Emily Miraldi, Immunobiology and Biomedical Informatics, Cincinnati Children's Hospital

Document Type

Poster

This document is currently not available here.

Share

COinS
 
Oct 27th, 6:40 PM Oct 27th, 7:20 PM

Transcriptional Regulatory Network Inference from Single-Cell RNA Measurements in Embryonic Stem Cells

Science Center, Bent Corridor

Single-cell RNA-seq (scRNA-seq) allows transcriptome-wide analyses of individual cells and quantifies intra-population heterogeneity. Here we develop computational tools to learn how these gene expression patterns are regulated, modeling these patterns as functions of transcription factor activities (TFA). However, due to low capture rate, scRNA-seq data matrices are sparse and contain many ambiguous zeros, in which sampling zeros, “dropout genes”, and true zeros are not well distinguished. For downstream analyses, biological variations, the trends of interest, should be resolved from technical variations. Four statistical approaches to normalize scRNA data were benchmarked: Reads Per Million (RPM), Markov Affinity-based Graph Imputation of Cells (MAGIC), Bayesian Inference for Single-cell Clustering and Imputing (BISCUIT), and Robust Principal Component Analysis (RPCA). Regulatory networks are inferred via the Inferelator algorithm, Bayesian Best Subset Regression, with the incorporation of ATAC-seq priors. TFA can be estimated with mRNA levels or with the expression levels of its known targets using priors. Post-Inferelator networks are then compared with gold-standard and bulk-RNA networks via precision-recall analyses. Of four methods, MAGIC and RPCA demonstrate promising ability to impute “dropout genes” while preserving cell-cell variations and recovering cluster-specific gene expression. Preliminary network data show that using mRNA levels have higher predictive power of the models than estimated TFA. The next step is to continue finessing methods that enable inference of TRNs from scRNA-seq with robust predictive power, increasing the impact of single-cell genomics from qualitative descriptive to quantitatively predictive.