Enhanced Rosetta-based Protein Structure Prediction For non-Beta Sheet Dominated Targets

Protein structure prediction has been one of the most challenging tasks undertaken in bioinformatics. Fragment assembly methodologies have emerged as the most accurate approaches to predict protein conformations without the need of homologues. Rosetta - a fragment-based tool - has consistently been at the forefront for two decades. Rosetta assembles candidate conformations using fragments of length 9 and 3 extracted from a pool of high-resolution proteins. Herein, an extensive study has been conducted highlighting the importance of the size 3 fragments - 3-mers - the role of which is both refining and correcting. Reduction of the number of those fragments from 200 to 100 revealed that Rosetta was able to produce first models of improved accuracy (+8%) for alpha and alpha-beta targets by 8%. Accordingly, an amended pipeline was proposed: it involves adjusting the number of 3-mers according to sequence-based structural class prediction of the protein target.