Receptor tyrosine kinases (RTKs) constitute one of the most significant classes of oncologic therapeutic targets. For over a decade, all-atom molecular simulations have been employed to probe the determinants of tyrosine kinase inhibitor (TKI) selectivity. Today, we recognize that sequence and dynamical differences in key regions of kinases, such as the glycine-rich loop, DFG-motif, hinge-binding domain, and gatekeeper region, all contribute to the variability in TKI potency and selectivity across the kinome. This knowledge led to the structure-based rational design of the TKI osimertinib, which preferentially binds to the EGFR-T790M resistance mutant over wild-type EGFR. More recently, we demonstrated that a newly-identified EGFR osimertinib resistance mutation, G724S, develops in specific sequence variants of Ex19Del, but not in L858R, and we were able to attribute these differences to detailed changes in EGFR dynamics. Here, we present a strategy that leverages differences in protein dynamics between mutants to guide the design of mutant-selective inhibitors. Our algorithm employs machine learning-guided Monte Carlo-Metropolis (MCM) sampling of medicinal chemistry modifications to small-molecule scaffolds. The molecules are iteratively filtered through a pharmacophore map of the protein-ligand complex, where the softness of the edges scales with a molecular dynamics-derived measure of each residue’s flexibility with respect to the ligand.

Citation Format: Benjamin P. Brown, Jeffrey Mendenhall, Christine Lovly, Jens Meiler. Toward automated design of mutant-selective tyrosine kinase inhibitors using dynamic pharmacophore mapping and machine learning [abstract]. In: Proceedings of the AACR Special Conference on Advancing Precision Medicine Drug Development: Incorporation of Real-World Data and Other Novel Strategies; Jan 9-12, 2020; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2020;26(12_Suppl_1):Abstract nr 22.