AI-Driven discovery and experimental validation of covalent FGFR4 inhibitors for hepatocellular carcinoma

FGFR4 signaling is an essential driver in hepatocellular carcinoma. However, traditional screening is often time-consuming, highlighting a need for efficient strategies to identify covalent chemotypes and accelerate FGFR4 drug discovery. We established an integrated AI-driven virtual screening framework to discover FGFR4 covalent inhibitors. The theoretical predictions were evaluated through a biochemical pipeline, encompassing in vitro FGFR4 kinase assays, immunoblotting of intracellular signaling cascades, and bottom-up LC-MS/MS peptide mapping. Biological validation of the computational predictions identified five distinct chemical scaffolds (hits 1, 2, 5, 7, and 8) exhibiting antiproliferative activity. The two most active candidates, hit 1 and hit 2, were selected for further mechanistic profiling. These compounds demonstrated dose-dependent FGFR4 kinase inhibition with IC 50 values of 1.06 μM and 3.57 μM, respectively. Cellular assays revealed that both compounds attenuate FGFR4 autophosphorylation and its downstream FRS2/ERK1/2 signaling cascade without inducing non-specific protein degradation. Furthermore, bottom-up LC-MS/MS peptide mapping provided direct structural evidence that hit 1 and hit 2 engage the target cysteine residue via a Michael addition mechanism. Our AI-guided computational workflow identified multiple covalent FGFR4 inhibitors with measurable biological activity. Hit 1 and hit 2 represent structurally characterized covalent scaffolds. This study provides chemical starting points for targeted HCC therapy and demonstrates the integration of theoretical prediction and experimental validation in covalent drug discovery.