An artificial intelligence optimized hepatic differentiation unveils NR5A2 and AP-1 transcriptional regulation in...

The generation of hepatocyte-like cells (HLCs) from human pluripotent stem cells (hPSCs) holds great promise for drug discovery and cell-based therapy for liver disease. However, current differentiation protocols are complicated and unstable, and the underlying gene regulatory mechanisms of hepatic differentiation remain incompletely defined. Here, we developed a machine learning-based artificial intelligence (AI) tool using phase-contrast images of hepatic progenitor cells (HPCs), which are essential for generating HLCs. The AI tool significantly improves the success rate of hepatic differentiation without the need for immunostaining or lineage tracing. By optimizing the methodology, we achieved an impressive purity of 90 to 95% for HLCs derived from hPSCs, aided by the AI algorithm. Through further investigating transcriptomes and epigenomic changes, we discovered the pivotal roles of nuclear receptor subfamily 5 group A member 2 and activator protein-1 transcription factors in regulating the maturation of hepatocytes. Single-cell RNA sequencing demonstrated the upregulation of nuclear receptor subfamily 5 group A member 2 and activator protein-1 during hepatic differentiation. Importantly, mutation analysis and tumorigenesis assays confirmed the safety of this modified hepatic differentiation protocol. This work highlights the potential of combining AI algorithm and computational genomics to facilitate development of lineage differentiation and molecular mechanism study.