Fatecode: Cell fate regulator prediction using classification autoencoder perturbation

Abstract

Cell reprogramming, which guides the conversion between cell states, is a promising technology for tissue repair and regeneration. Typically, a group of key genes, or master regulators, are manipulated to control cell fate, with the ultimate goal of accelerating recovery from diseases or injuries. Of importance is the ability to correctly identify the master regulators from single-cell transcriptomics datasets. To accomplish that goal, we propose Fatecode, a computational method that combines in silico perturbation experiments with cell trajectory modeling using deep learning to predict master regulators and key pathways controlling cell fate. Fatecode uses only scRNA-seq data from wild-type samples to learn and predict how cell type distribution changes following a perturbation. We assessed Fatecodes performance using simulations from a mechanistic gene regulatory network model and diverse gene expression profiles covering blood and brain development. Our results suggest that Fatecode can detect known master regulators of cell fate from single-cell transcriptomics datasets. That capability points to Fatecodes potential in accelerating the discovery of cell fate regulators that can be used to engineer and grow cells for therapeutic use in regenerative medicine applications.