Fig. 6: KRASG12D candidate selection and validation of unknown molecules.
From: Optimizing drug design by merging generative AI with a physics-based active learning framework
A Scatter plot of KRASG12D unknown molecules with a Glide gscore below −8 kcal·mol−1 and a maximum similarity to the initial-specific set below 0.30. Red dotted lines indicate the stringent thresholds of −9 kcal·mol−1 for gscore and 0.25 for similarity used during candidate selection. B Linear regression and correlation coefficient between Glide gscore and ABFE ΔG (left) and between PELE BFE and ABFE ΔG (right) for the 19 candidate drugs selected after PELE rescoring. C Table showing MM affinity predictions and maximum similarity to the unknown initial-specific set molecules for molecules with potential activity against KRAS. D Structures of molecules with potential activity against KRAS by ABFE simulations. E MRTX1133 Mirati’s KRASG12D inhibitor currently in clinical trials.
