Cancer Immunotherapy Design with Equivariant Diffusion Models

This thesis investigates the use of artificial intelligence in cancer immunotherapy, focusing on the prediction of peptide-MHC (Major Histocompatibility Complex) structures, essential for effective treatment design. We develop a specialized SE(3) equivariant diffusion model utilizing Geometric Deep Learning principles to accurately predict the 3D structures of peptide-MHC complexes. Validated on the Pandora dataset, our model achieves state-of-the-art results, enhancing both prediction accuracy and speed. These advancements are vital for developing new cancer immunotherapies, enabling more precise targeting of cancer cells by the immune system and paving the way for personalized treatments.