Algebraic and Graph-Based Machine Learning Methods for Classifying Pathogenic Repeat Expansion Sequences in Neurodegenerative Disease

This project classifies pathogenic DNA repeat expansions, the mutations behind diseases like Huntington's, myotonic dystrophy, and C9orf72-related ALS/FTD, using algebraic and graph-based sequence representations (Chaos Game Representation and de Bruijn graphs) instead of the hand-engineered features current tools rely on. The core hypothesis is that encoding the repeat's geometric and structural properties captures signal that standard feature-based models miss, since these repeats are known to fold into unusual DNA structures like G-quadruplexes. Work involves building these representations, training CNNs and GNNs on them, and benchmarking against existing pathogenicity predictors.