Combinatorial Optimization for Predicting Optimal Cell-state Conversion Paths

Cellular reprogramming constitutes a crucial, yet unexplored field that studies the manipulation of cell states through transcriptomic changes. Our study aims to expand cellular reprogramming and combat two complex problems within the field: disease-to-healthy cell state conversions and transdifferentiation through computational modeling. We use optimization and machine learning algorithms to predict the optimal combinations of perturbations needed to be made on transcription factors, genes that regulate gene expression in a cell, to cause erythroleukemia-retinal pigment epithelial cell state conversion. Our best optimization algorithm achieves a conversion rate of 90%, hence outcompeting that of benchmarks. This is the first step towards generalizing to any-to-any cell state conversions and realizing cellular reprogramming.