Directed cell-fate conversion (including differentiation, reprogramming, transdifferentiation, and dedifferentiation) holds immense potential for regenerative medicine but remains limited by low efficiency and incomplete mechanistic understanding. Identifying chemical compounds that precisely modulate gene-regulatory networks (GRNs) governing cell identity is essential to improve the efficiency, specificity, and scalability of these processes. We developed Refate [1], a computational framework that integrates large-scale multimodal single-cell atlas data to predict small molecules capable of driving targeted cell-fate transitions. Refate quantifies each gene’s contribution to cell identity through a Cell Identity Score (CIS) derived from integrative analysis of single-cell multi-omic atlases. By combining CIS with differential gene expression between source and target cell types, Refate then identifies key driver genes and transcription factors defining each transition. GRNs reconstructed from curated protein–protein and transcription factor–target interaction databases are then cross-referenced with six drug and perturbation databases to prioritise compounds predicted to modulate these networks. Across 18 benchmarked conversions, Refate outperformed existing tools in recovering experimentally validated transcription factors and compounds. Proof-of-concept assays confirmed that several predicted molecules significantly enhanced the differentiation of human embryonic stem cells into cranial neural crest cells, validating its predictive power. Collectively, these findings establish Refate as a robust, interpretable, and scalable framework for discovering and repurposing small molecules that target GRNs to drive cell-fate conversion. By integrating transcriptional and epigenetic layers, it achieves greater predictive accuracy and expands the accessible chemical space for regenerative medicine, disease modelling, and cell engineering.