Abstract A016: A computational chemistry and AI-driven framework for structure-based drug design informed by underlying factors of mutation-induced drug resistance: A study of KRAS

Abstract Mutation-induced drug resistance is a major obstacle in effective cancer treatment. We present a framework that integrates computational chemistry and AI for structure-based drug design targeting drug resistant mutations. To demonstrate the application of our framework, we use Kirsten Rat Sarcoma (KRAS) oncogene as a proof of concept. KRAS is one of the most mutated oncogenes in pancreatic, colorectal, and lung cancers. Mutations in KRAS cause its prolonged activation and excessive cell growth. While the primary mutant KRAS G12C responds to approved covalent inhibitors, several secondary mutations in the binding site (e.g., G12C/Y96C, G12C/Y96S, G12C/Y96D) lead to drug resistance. To understand conformational differences between treatment-sensitive and treatment-resistant populations and to enable structure-based drug design, we conducted molecular dynamics simulations on several drug-free KRAS mutants. Each simulation was performed in triplicate, and trajectory clustering was applied to extract the most populated conformations. Molecular features were calculated for the representative structures. The resulting data were analyzed using three supervised machine learning (ML) models: logistic regression, random forest, and support vector machine. Distinct structural differences in protein dynamics were observed between the two groups, particularly in the switch II binding site region, where covalent inhibitors bind. Variations were detected in residue conformations and the spatial arrangement of molecular features such as hydrogen bond donors and acceptors, as well as aromatic and aliphatic groups. Using ML, we identified that the molecular features of the most populated protein conformations differed significantly between treatment-sensitive and treatment-resistant systems. Notably, solvent exposure and conformational flexibility of residues G10, E62, and H95 within the switch II binding site emerged as the most predictive features of treatment sensitivity, alongside other features such as Lennard-Jones 1-4 energy and backbone mean square displacement. Given these differences, pharmacophores describing the physicochemical and spatial properties of switch II binding site conformations have been extracted for resistant and sensitive systems and will serve as input conditions for generative molecular design. To achieve that, we will utilize existing string- and graph-based generative ML models to design ligands within the binding site of the target. In this approach, the protein binding pocket will be represented by the coordinates of the pharmacophore, guiding the construction of molecular graphs for ligands. Generative ML applied to structure-based drug design will enable the discovery of potential bioactive compounds at an increased rate by accessing vast chemical space. Incorporating protein dynamics into this process provides deeper insights into mutation-induced drug resistance by revealing critical molecular determinants. These insights are essential for guiding the design of selective small molecules against difficult-to-target proteins. Citation Format: Katarzyna Mizgalska, Denis J. Imbody, Eric B. Haura, Wayne C. Guida, Aleksandra Karolak. A computational chemistry and AI-driven framework for structure-based drug design informed by underlying factors of mutation-induced drug resistance: A study of KRAS [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr A016.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.