We discuss several complementary computational approaches towards understanding, controlling, and predicting protein self-assembly and aggregation, with a specific focus on the protein tau that is implicated in a range of neurodegenerative disorders. At small scales, we use molecular simulations of multimers and fibril fragments to characterize hydrophobic driving forces and identify hydrophobic “hot spot” regions, through the response of hydration water structure, to identify novel binding targets for manipulating aggregation pathways. At intermediate scales, we use relative entropy coarse-graining to create simplified simulation models capable of mapping out the phase behavior of tau and its response to co-factors. At larger scales, we use recent AI protein embedding platforms to create predictive models of protein aggregation and liquid-liquid phase separation, and in turn, screen the entire human intrinsically disordered proteome for protein classes with high propensities for either or both. These efforts demonstrate the potential to combine simulations and modern ML tools to understand and predict self-assembly properties of many kinds of intrinsically disordered proteins.
