FateDriver: Weakly supervised dual-view graph learning for transition-specific prioritization of candidate regulators in cellular reprogramming

FateDriver, a weakly supervised dual-view graph neural network, ranked first or second among six methods on three single-cell reprogramming datasets, recovering curated regulators and pluripotency chemical protein targets.

• Why it matters: Endogenous regulators for a specific source-to-target transition are still unclear, and prior ranking approaches ignore the experimentally delivered factors that induce reprogramming.
• What they did: FateDriver jointly models cells and genes from single-cell transcriptomes, uses delivered factors as sparse gene-level labels, and applies a positive–unlabeled objective to rank remaining genes for a chosen transition across OSKM and GMT datasets.
• The result: It outperformed a supervision-matched non-graph model despite weaker co-expression with delivered factors, achieved the top recovery of pluripotency-inducing chemical cocktail protein targets in both OSKM datasets, and identified transition-specific candidate programs including a DDX58–interferon axis linked to cells lacking endogenous pluripotency markers.

AI summary of the abstract · may contain errors · not medical advice · read the paper.