Other models

Evaluation of the proposed amyloid transformation model is limited due to the lack of alternative numerical models geared specifically toward predicting amyloid structures. In the case of native proteins, a clear alternative model is the one implemented by AlphaFold2, which utilizes AI and deep learning mechanisms ( [1] , [2] , [3] ) and focuses on optimization of inter-residue distances. Numerous protein structure databases exist, supplying input data for training—in the case of AlphaFold, the model is based on a set of 30,000 non-redundant sequences/structures (in line with Anfinsen's dogma—“one sequence, one structure”). Nevertheless, amyloid forms invalidate this assumption by providing alternative conformational arrangements for identical sequences, thus revealing the polymorphic nature of proteins ( Pauwels et al., 2007 ). Researchers who apply AlphaFold in their studies of protein complexes tend to restrict themselves to globular proteins ( Pinheiro et al., 2021 )—although AlphaFold may also be applied in studies of functional amyloids, under the assumption that these fold in a similar way to other native proteins (primarily globular ones).