Does AlphaFold predict the spatial structure of a protein from physics or recognize it (its main parts and their association) using databases?

Abstract

The great success of the AlphaFold programs poses the following questions: (i) What is the main reason for this success? (ii) What exactly do AlphaFolds do: prediction of the 3D protein structure based on its amino acid sequence and knowledge of the protein physics or recognition of this 3D structure, based on the similarity between some parts of its amino acid sequence and parts of sequences with already known 3D structures? The answers given in this paper are: The main reason for the tremendous success of the AlphaFold is (i) the usage of huge protein databases, which already cover all or almost all of the protein superfamilies existing in nature; (ii) using these databases and the resulting multiple sequence alignments and coevolutionary information (like correlations in pairs and especially in triplets of amino acid residues in the contacting chain regions), AlphaFold recognizes a 3D structure of the examined amino acid sequence by a similarity of this sequence (or its parts) to related sequences with already known 3D structures. Concluding, I have to emphasize that this paper does not diminish the merit and utility of AlphaFold; it only explains the basis of its success.