Targeting the untargetable: accelerated discovery of KRAS G12D inhibitors through a deep learning-enhanced in silico pipeline

KRAS mutations represent some of the most prevalent oncogenic alterations in human cancers, especially in pancreatic, colorectal, and non-small cell lung carcinomas. Among these, the KRAS G12D mutation is notably aggressive, producing a constitutively active KRAS protein that drives unchecked cell proliferation and therapeutic resistance. Developing effective inhibitors against this mutation remains a significant challenge and a critical objective in precision oncology. Traditional drug discovery approaches, while valuable, are often time-intensive and financially demanding. In this context, computational drug repurposing offers a strategic advantage by accelerating the identification of novel therapeutic applications for FDA-approved drugs with established safety profiles. This study presents a robust, integrated in silico workflow designed to screen FDA-approved compounds and commercial libraries for potential inhibitors of the KRAS G12D mutation. The pipeline combines pharmacophore-based filtering, GNINA deep learning-augmented molecular docking, and molecular dynamics (MD) simulations to prioritize candidates with strong binding affinity and structural stability. Application of this multi-tiered approach yielded three high-confidence compounds exhibiting favorable and persistent interactions within the KRAS G12D switch-II pocket. These findings lay the groundwork for subsequent experimental validation and provide a promising avenue for the rapid development of therapeutic interventions against KRAS-driven cancers.