Deep learning-enhanced morphological profiling predicts cell fate dynamics in real-time in hPSCs
Predicting how stem cells become patterned and differentiated into target tissues is key for optimising human tissue design. Here, we established DEEP-MAP - for deep learning-enhanced morphological profiling - an approach that integrates single-cell, multi-day, multi-colour microscopy phenomics with deep learning and allows to robustly map and predict cell fate dynamics in real-time without a need for cell state-specific reporters. Using human pluripotent stem cells (hPSCs) engineered to co-express the histone H2B and two-colour FUCCI cell cycle reporters, we used DEEP-MAP to capture hundreds of morphological- and proliferation-associated features for hundreds of thousands of cells and used this information to map and predict spatiotemporally single-cell fate dynamics across germ layer cell fates. We show that DEEP-MAP predicts fate changes as early or earlier than transcription factor-based fate reporters, reveals the timing and existence of intermediate cell fates invisible to fixed-cell technologies, and identifies proliferative properties predictive of cell fate transitions. DEEP-MAP provides a versatile, universal strategy to map tissue evolution and organisation across many developmental and tissue engineering contexts.