Illuminating the Dark Kinome through Protein Language Models, Symbolic Regression, and Structure-Based Prioritisation

Computational prioritisation framework for 143 Tdark kinases integrating ESM-2 embeddings, AlphaFold2 structures, KIBA bioactivity data, symbolic regression (PySR), and AutoDock Vina virtual screening. 18/20 top-ranked candidates de-orphanised post-2018 (OR ~19, p < 10⁻¹⁰). Full reproducible code available at: https://www.kaggle.com/code/igormerlinicomposer/dark-kinome-unveiled