Abstract
Aggregation is a pervasive phenomenon affecting both disordered and folded proteins, modulated by the balance between hydrophobic attraction and electrostatics, and generally explored by means of molecular dynamics simulations, often performed with low-resolution models. On the other hand, the AI based structure prediction AlphaFold software, extremely accurate for single chains, has been recently provided with the capability of predicting the structure of protein complexes. In this work, we compare coarse-grained molecular dynamics simulations to assess the relative performance of different force fields and those of AlphaFold3 (AF3) in predicting the aggregation of four variants of intrinsically fluorescent proteins. Our results are useful for guiding the choice or design of new potentials for systems containing both disordered and folded proteins and outline a possible scheme to integrate AlphaFold with molecular dynamics simulations.