Machine-Guided Cell-Fate Engineering

Stem cells are the progenitor cells of all differentiating multi-cellular organisms. In principle, it is possible to differentiate these cells into any other type of cell, which can then be used for many different possible therapeutic or diagnostic applications. The creation of induced pluripotent stem cells (iPSCs) has enabled scientists to explore the derivation of many types of cells. While there are many general approaches for cell-fate engineering, one of the fastest and most efficient approaches is transcription factor (TF) over-expression. Over-expression of specific combinations of TFs is often a reliable method to differentiate stem cells, but since there are at least 1732 transcription factors in the human genome, selecting the right combination to differentiate iPSCs directly into other cell-types is a difficult task. Here were describe a machine-learning (ML) pipeline, called CellCartographer, for using chromatin accessibility next-generation sequencing (NGS) data to produce a multiplex TF pooled-screen for converting stem cells into other cell types. We then describe a barcoded bulk RNA-seq method for refining the set of TFs using iterative NGS experiments. We validate this method by differentiating stem cells into six medically-relevant cell types with the human TFome originating from all germ layers: cytotoxic T-cells, regulatory T-cells, B-cells, microglia, type II astrocytes, and hepatocytes. We demonstrate iterative improvement in differentiation efficiency and functionally characterize the cell lines to validate fast, robust, and functionally accurate differentiation of stem cells into cell types useful for downstream therapeutic and diagnostic pipelines.