FARFAR2: Improved de novo Rosetta prediction of complex global RNA folds

Abstract

Methods to predict RNA 3D structures from sequence are needed to understand the exploding number of RNA molecules being discovered across biology. As assessed during community-wide RNA-Puzzles trials, Rosettas Fragment Assembly of RNA with Full-Atom Refinement (FARFAR) enables accurate prediction of complex folds, but it remains unclear how much human intervention and experimental guidance is needed to achieve this performance. Here, we present FARFAR2, a protocol integrating recent innovations with updated RNA fragment libraries and helix modeling. In 16 of 21 RNA-Puzzles revisited without experimental data or expert intervention, FARFAR2 recovers structures that are more accurate than the original models submitted by our group and other participants during the RNA-Puzzles trials. In five prospective tests, pre-registered FARFAR2 models for riboswitches and adenovirus VA-I achieved 3-8 [A] RMSD accuracies. Finally, we present a server and three large model archives (FARFAR2-Classics, FARFAR2-Motifs, and FARFAR2-Puzzles) to guide future applications and advances.