A missing link on the road to protein fold prediction from first principles

AlphaFold solved the problem of protein folding based on the accumulated body of known protein structures combined with the most advanced computerized search methods. However, can we predict structures of common globular proteins (average size 150 residues) based on first principles alone? The answer here is that we still have long way to go. We hypothesize that an essential element that can lead to solution of the prediction problem is knowledge of few fast formed closed long loops in the protein backbone based on the sequence information alone. identification of key early formed non-local contacts that form closed long loops in the still disordered state of the backbone should reduce the complexity of the problem. Current search methods can solve the task of search for the native structure of the constraint protein backbone. Methods based on empirical knowledge of essential closed long loops were successful in search of the fold of native proteins and in design tasks and thus support this hypothesis. Time resolved FRET experiments combined with ultrafast mixing methods were used for detection of long loops’ closure at the initiation of the folding pathways of model proteins. Statistical mechanical model was developed for prediction of fast formed long loops in the disordered chain and those are tested experimentally by the FRET measurements. The long-range goal of this research is development of a sequence dictionary where loop nodes would be recognized and thus serve as input for the computerized search of native structures and de-novo design of new proteins.