3D Structural Modelling for Enhanced Peptide-MHC and TCR Specificity predictions скачать в хорошем качестве

3D Structural Modelling for Enhanced Peptide-MHC and TCR Specificity predictions

Prof Li Xue from Radboud University Medical Center (Radboudumc). Abstract Cancer immunotherapies hold tremendous promise, yet their progress is often constrained by high costs, long development timelines, and off-target toxicities. A major challenge lies in accurately identifying tumor-specific mutated peptides that bind strongly to MHC molecules and elicit targeted T cell responses. Existing predictive models have achieved limited success, largely due to incomplete coverage of the diverse TCR-peptide-MHC interaction space and a reliance on one-dimensional (1D) sequence-based representations. To address these limitations, we introduce 3D structure-based AI methods, which capture rich spatial and physicochemical features that enhance predictive accuracy and generalizability. In this talk, I will present our recent advances in peptide–MHC binding and TCR specificity prediction enabled by these 3D approaches. Specifically, I will introduce SwiftMHC, our ultra-fast AI simulator capable of generating 3D pMHC structures and predicting binding affinities within milliseconds (excluding file writing), and SwiftTCR, a physics-based modeling tool that constructs TCR–pMHC complexes within minutes using only CPUs. I will conclude with a discussion on future directions for integrating structural modeling and deep learning to decode the principles of TCR recognition and guide next-generation immunotherapy design.