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

By Andrew M. Watkins, R. Rangan, Rhiju Das. 2020

223 citations recorded by the source corpus.

Summary Predicting RNA 3D structures from sequence could accelerate understanding of the growing number of RNA molecules being discovered across biology. Rosetta’s Fragment Assembly of RNA with Full-Atom Refinement (FARFAR) has shown promise in community-wide blind RNA-Puzzle trials, but lack of a systematic and automated benchmark has left unclear what limits FARFAR performance. Here, we benchmark FARFAR2, an algorithm integrating RNA-Puzzle-inspired innovations with updated fragment libraries and helix modeling. In 16 of 21 RNA-Puzzles revisited without experimental data or expert intervention, FARFAR2 recovers native-like structures more accurate than models submitted during the RNA-Puzzles trials. Remaining bottlenecks include conformational sampling for >80-nucleotide problems and energy function limitations more generally. Supporting these conclusions, pre-registered blind models for adenovirus VA-I RNA and five riboswitch complexes predicted native-like folds with 3–14 Å RMSD accuracies. We present a FARFAR2 webserver and three large model archives (FARFAR2-Classics, FARFAR2-Motifs, and FARFAR2-Puzzles) to guide future applications and advances.

Subfields: Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Molecular Biology

Citations in the corpus

Recent advances in RNA structurome (2022, methodology): “Several approaches have been developed to perform all-atom assembling using known fragments, such as FARNA/FARFAR/FARFAR2 (Das and Baker, 2007; Das et al., 2010; Watkins et al., 2020), RNAComposer (Popenda et al., 2012), and 3dRNA (Zhang et al., 2020c; Zhao et al., 2012).”
Accurate prediction of nucleic acid and protein-nucleic acid complexes using RoseTTAFoldNA (2022, background): “Over the 78 RNA structures with no homologs in our training data, FARFAR2’s top-ranked models have an average lDDT of 0.47 [4], while DeepFoldRNA has an average lDDT of 0.61 [15], compared to 0.67 for RoseTTAFoldNA and 0.79 for RoseTTAFoldNA high-confidence predictions (Suppl.”
Accurate prediction of protein–nucleic acid complexes using RoseTTAFoldNA (2023, methodology): “For predicting protein components, machine learning-guided approaches like RoseTTAFold 4 and Alpha-Fold 5 are highly accurate, while RNA structure prediction has used a combination of Monte Carlo sampling approaches 6–9 as well as deep learning methods 10,11 .”
Interpretable RNA Foundation Model from Unannotated Data for Highly Accurate RNA Structure and Function Predictions (2022, background): “Then, their 3D structures can be thereby optimized with RNA 3D modeling tools, including 3dRNA and FARFAR2 [53].”
Geometric Prediction: Moving Beyond Scalars (2020, methodology): “We generate data for our structure refinement task by using the structural candidates for each of 21 RNA sequences provided in the FARFAR2-Puzzles set [27].”
Integrating end-to-end learning with deep geometrical potentials for ab initio RNA structure prediction (2022, unknown): “For example, the fraction of the test targets for which DRfold achieved a lower RMSD than the control methods was 80.0% (to 3dRNA), 82.5% (FARFAR2), 72.5% (RNAComposer), 75.0% (RNA-BRiQ), and 80.0% (SimRNA), respectively, as shown in Fig.”
RNA-Puzzles toolkit: a computational resource of RNA 3D structure benchmark datasets, structure manipulation, and evaluation tools (2019, background): “(41); (ii) a decoy dataset to develop effective structure scoring function, e.”
Pairing a high-resolution statistical potential with a nucleobase-centric sampling algorithm for improving RNA model refinement (2021, methodology): “Comparison to FARFAR2 .”