From Cryo-EM Structures to AI-Powered Predictions of Amyloid Formation

The propensity of proteins to form amyloid fibrils is encoded in their amino acid sequences, motivating the development of computational methods to predict amyloidogenicity directly from sequence. Two recent advances offer new opportunities in this field: cryo-electron microscopy, which has revealed the 3D atomic structures of numerous amyloid fibrils, and artificial intelligence (AI), which is transforming computational approaches across science and technology. In this talk, I will present our recent predictors that integrate AI approaches with the rapidly expanding structural repertoire of amyloid fibrils, and show how human expertise and AI can work synergistically to improve the prediction of amyloidogenicity.

I will also show how structural data can improve the prediction of protein co-aggregation within amyloid fibrils. This approach could help elucidate how exogenous proteins, including those of microbial origin, may promote the formation of pathological amyloids in humans.

These studies have important potential biomedical implications. As whole-genome sequencing becomes increasingly accessible, our in silico predictors could help translate individual genetic variation into personalized risk profiles for neurodegenerative, age-related, and other amyloid-associated diseases, opening new avenues toward personalized medicine.