AGEAS: Automated Machine Learning based Genetic Regulatory Element Extraction System

Abstract

With rapid progress of robust single-cell transcriptome sequencing since last decade, numerous complex mechanisms underlying cell development has been revealed. Single-cell RNA sequencing (scRNA-seq) analysis is widely accepted as the main approach to define cell stages and phenotypes. As conversion of somatic cells into induced pluripotency cells succeeded, identification key genetic factors(GFs) with scRNA-seq for cell reprogramming in biological research and regenerative medicine fields gained increasing attention. Herein, we describe Odysseia, an interpretable machine learning classifier based single-cell gene expression profile(scGEP) analysis system, that assesses importances of genetic regulatory features in differentiating cell states(CSs). Furthermore, extracted factors, when combining with regulatory network analysis, can help to find key GFs in classifying CSs and possibly inducing CS conversions. Analyzed three published scRNA-seq datasets used to study divergent cell types, Odysseia correctly extracted GFs acclaimed to be capable of inducing CS conversions. Overall, Odysseia provides an automated alternative to obtain guidance information while explicating mechanism to engineer cellular phenotypes.