Early multi-omic signatures and machine learning models predict cardiomyocyte differentiation efficiency and enable robust hPSC differentiation to cardiomyocytes

Protocols for generating cardiomyocytes (CMs) from human pluripotent stem cells (hPSCs) have existed for nearly two decades, yet manufacturing variability in terminal cell identity continues to limit clinical translation. To uncover the origin of fate divergence during hPSC-CM differentiation, we performed temporal transcriptomics, proteomics, and metabolomics of high and low efficiency differentiations. We identified significant early multi-omic divergence between differentiation batches and key pathways underlying fate divergence at critical differentiation stages included Wnt, MAPK, and glucose metabolism. Machine learning models trained on early candidate gene markers predicted hPSC-CM purity better than models using canonical cardiac development markers. Lastly, multi-omic insights informed perturbations, including Wnt and MAPK inhibition, which produced higher CM purities and yields. Our results showcase multi-omic analysis coupled with machine learning models as a powerful tool to identify cell fate determinants and enable robust manufacturing of complex cell products such as hPSC-derived cell therapies.