Integrating coarse-grained simulations and machine learning to uncover protein aging mechanisms

Protein condensate aging is a widespread phenomenon, yet quantitative insights into its molecular mechanisms remain rare. Here, we employ the sticker-spacer model and integrate coarse-grained lattice simulations with machine learning to investigate the liquid-to-solid transition in protein condensates. The transition arises from the intricate interplay of heterotypic interactions among stickers within proteins. By quantitatively distinguishing inhomogeneous solid-like and liquid-like structures, we confirm the formation of a solid-like shell at the condensate interface. Proteins at the interface, characterized by extended conformations and orientations, promote the formation of solid-like structures via the aggregation of aging stickers. Notably, variations in the number, location, and intensity of aging stickers significantly influence the structural and viscoelastic properties of the condensates, mediated by the balance between competition and cooperation in heterotypic interactions. Our findings provide a quantitative framework for understanding condensate aging and suggest potential strategies to regulate aging process.