Fig. 4

Discovery of potential drug candidates derived from sorafenib. a Comparison of joint distributions over binding affinity to ABCG2 (x-axis) and Tanimoto similarity to sorafenib (y-axis) using kernel density estimation between COMA and UGMMT. The success rate is defined as the affinity score against ABCG2 being less than 4.7, and the Tanimoto coefficient being greater than 0.4 simultaneously. b Comparison of generated molecules achieving weak ABCG2 affinity and high similarity with sorafenib. The affinity scores were evaluated by DeepPurpose. c Comparison of binding energies of sorafenib and 19 molecules satisfying the constraints. The binding energy scores to ABCG2 were evaluated using AutoDock Vina. For each box, the center line and box limits shows the quartiles of binding energy to ABCG2, and whiskers represent 1.5x interquartile range. The 15 of 19 molecules have higher binding energies with ABCG2 compared to sorafenib, which makes them potential drug candidates for alleviating sorafenib resistance. d Docking simulation results of sorafenib and the identified hit molecule COMA018 against BRAF. (Left) 3D visualization of ABCG2 and ligand complexes drawn by Chimera to show a binding pocket and contact regions. (Right) 2D visualization of complexes drawn by LigPlot Plus to show binding sites and residues