Computational Discovery of Selective KRAS G12D Inhibitors Using a Consensus Docking Pipeline A Proof-of-Concept Structure-Based Virtual Screening Study Independent Research Computational Biology / Machine Learning

KRAS G12D is one of the most prevalent oncogenic mutations in pancreatic, colorectal, and lung cancers, yet it remained historically difficult to target due to the shallow topology of the KRAS binding surface. Recent advances in mutant-selective inhibitors have renewed interest in computational approaches capable of identifying candidate molecules with mutation-specific affinity. In this study, a proof-of-concept consensus docking pipeline was developed to identify potential KRAS G12D inhibitors using structure-based virtual screening. A small filtered chemical library was evaluated using three docking engines (AutoDock Vina, Smina, and Gnina) with weighted consensus scoring. Binding poses were validated using protein-ligand interaction profiling to confirm the presence of a salt bridge with Asp12, a key structural feature of the G12D mutant pocket. Cross-docking against wild-type KRAS was performed to evaluate mutation selectivity. From an initial screen of ten molecules, four candidates demonstrated favorable docking scores and interaction geometry consistent with Asp12 engagement. Three compounds exhibited near-zero predicted affinity for wild-type KRAS while maintaining strong G12D binding, yielding a selectivity gap exceeding −8.3 kcal/mol in the top candidates. Although preliminary, these results suggest that consensus docking combined with structural interaction validation can identify mutation-selective ligands without requiring GPU resources. Experimental validation and larger-scale screening will be required to confirm these computational predictions.