Learning protein ubiquitination from 3D structures: patterns of cancer somatic mutations and deep learning models for prediction

This dissertation presents a structure-informed deep learning framework for predicting protein ubiquitination sites using AlphaFold2-derived 3D structural features and graph neural networks. It integrates spatial modeling with cancer genomics data to analyze the enrichment and functional impact of somatic mutations near ubiquitination sites. By combining structural bioinformatics, machine learning, and cancer mutation analysis, the work advances both predictive performance and biological interpretability. The findings highlight new mechanistic insights into ubiquitin signaling and its dysregulation in cancer.