FuzDrop: sequence-based prediction of the propensity of proteins for liquid-liquid phase separation and aggregation

Proteins exhibit complex phase behavior as they convert between the native state, the liquid condensate (or droplet) state and the solid condensate (or amyloid) state. To facilitate the study of these processes, we describe the FuzDrop method of predicting the condensation propensity of proteins to undergo liquid-liquid phase separation and to subsequently form amyloid aggregates. The method is based on the principle that liquid condensations reflect a balance between enthalpic and entropic contributions; FuzPred is an algorithm that provides sequence-based estimates for these contributions in stoichiometric complexes ( https://fuzpred.bio.unipd.it/predictor ). FuzDrop extends this algorithm to protein condensates, and enables prediction of the propensity for amyloid formation within liquid condensates, known as the condensation pathway to protein aggregation ( https://fuzdrop.bio.unipd.it/predictor ). This prediction is based on the principle that the sequence regions that promote aggregation within liquid condensates have a multiplicity of binding modes, because they have a strong propensity for both entropic-driven interactions to stabilize the droplet state and enthalpic-driven interactions to stabilize the amyloid state. The time required for FuzDrop predictions on the web server scales linearly with protein length and is typically ~30 s for a protein of 500 residues. By enabling predictions of protein phase behavior, FuzDrop may facilitate experimental studies directed at the development of therapies for protein condensation diseases.