MAGNET: A Mutation-Aware SE(3)  - Equivariant Graph Framework for 3D Binding in KRAS Variants

Overview

Annually, 20 million people are diagnosed with cancer, contributing to over 9.7 million deaths. Nearly 25% of these cases are linked to mutations in the KRAS gene, which normally functions as a molecular switch regulating cell growth and apoptosis. Oncogenic mutations hinder KRAS’s ability to hydrolyze GTP to GDP, leading to persistent activation of downstream pathways resulting in cancer-causing tumors. Current in-vitro methods are resource-intensive and slow, while computational models are unable to accurately represent structurally distinct KRAS mutations. Thus, this project introduces MAGNET, a mutation-aware computational framework for precise representation of distinct KRAS mutations. The model integrates three main layers: a FiLM conditioning layer to encode mutation-specific effects, an equivariant message passing layer to preserve structural differences, and a graph transformer layer for a global context. After extensive testing, MAGNET demonstrated excellent predictive performance across main metrics (MAE: 0.0301±0.002, MSE: 0.0026±0.00003, RMSE: 0.051±0.0004, MedAE: 0.007±0.002). Predicted ligands were further validated through Molecular Docking and molecular dynamics simulations, with top ligands reaching -10.1kcal/mol and -7.2kcal/mol, RMSD values < 1.5Å and RMSF values consistently < 1Å and 1.5Å for G12D and G12V respectively. Ultimately, this proves mutation-conditioned equivariant modeling is a powerful tool for accurate representations of structurally distinct mutations. MAGNET provides the first scalable, non-invasive mutation-aware framework for precise modeling of structurally distinct KRAS mutations, offering researchers a powerful tool to accelerate precision oncology and targeted drug discovery.

Competition history

• CSEF 2026 Medicine & Physiology (Track 2) (Senior Division) · Entry S-21-15

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Biomedical and Health Sciences projects
• CSEF Biomedical and Health Sciences projects
• Every CSEF 2026 project