Source-linked AI summary
Correcting pervasive errors in RNA crystallography through enumerative structure prediction
Fang-Chieh Chou, Parin Sripakdeevong, Sergey M. Dibrov, Thomas Hermann, Rhiju Das
TL;DR
RNA crystallographic models contain bond-geometry errors, sugar-pucker and backbone-conformer problems, and steric clashes. ERRASER-PHENIX evaluates and refines these structures, improving base-pair geometry, base orientation, and planarity across the benchmark while independent base-pair validation remains unavailable.
Problem
RNA crystallographic models exhibit pervasive bond-geometry errors, anomalous sugar puckers, and backbone conformer problems.
Method
ERRASER-PHENIX assesses RNA models using backbone-rotamer and sugar-pucker geometric criteria, with paired-sample comparisons across 24 benchmark datasets.
Results
ERRASER-PHENIX improves base-pairing geometry, base orientation, and base-pair planarity, with automatically assigned base-pairs improved in 21 of 24 cases.
Takeaways & Limitations
Refinement improves structural features used to represent RNA base-pairing, orientation, and planarity, while some refined base-orientation changes agree with available models.
Takeaways & Limitations
Independent base-pair validation tools are not currently available, limiting unbiased assessment of improvement.
Abstract
from arXiv · showhide
Three-dimensional RNA models fitted into crystallographic density maps exhibit pervasive conformational ambiguities, geometric errors and steric clashes. To address these problems, we present enumerative real-space refinement assisted by electron density under Rosetta (ERRASER), coupled to Python-based hierarchical environment for integrated 'xtallography' (PHENIX) diffraction-based refinement. On 24 data sets, ERRASER automatically corrects the majority of MolProbity-assessed errors, improves the average Rfree factor, resolves functionally important discrepancies in noncanonical structure and refines low-resolution models to better match higher-resolution models.
conformational ambiguities, geometric errors, and steric clashes. To address these
RNA crystallographic models contain frequent geometric and conformational errors, especially at medium-to-low resolution. ERRASER-PHENIX enumeratively rebuilds RNA while retaining diffraction agreement and substantially improves model quality across 24 structures.
- Problem: The pipeline addresses steric clashes, bond-length and angle outliers, backbone conformer outliers, and sugar-pucker errors identified in deposited RNA models.These features occur more often in 2.5–3.5 Å models than in models below 2.0 Å, suggesting inaccurate fits.
- Geometric correction: ERRASER-PHENIX eliminated all benchmark bond-length and angle outliers after PHENIX alone had already reduced their frequencies.Starting mean frequencies were 0.53% for bond lengths and 1.18% for angles.
- Steric correction: The average MolProbity clashscore fell from 18.0 to 7.0, while a low-resolution test case showed an 80% reduction in clashes.Less stringent or absent steric criteria produced higher average clashscores.
- Conformational correction: Backbone outlier rates fell from 19% to 8% in 22 of 24 cases, despite the Rosetta modeling not using the 54-rotamer classification.A functionally relevant kink-turn suite changed from an outlier to a rotamer later recovered in an independently released structure.
- Conformational correction: Mean sugar-pucker error rates fell from 5% to 0.2%, with zero errors in 19 cases, including improved agreement at a ribozyme active site.The remodeled active-site pucker and hydrogen-bonding network agreed with independent double-mutant analyses.
- Diffraction agreement: Across the benchmark, ERRASER-PHENIX lowered average R from 0.210 to 0.199 and average Rfree from 0.255 to 0.243.Rfree decreased in 22 of 24 cases, and low-resolution models showed improved agreement with higher-resolution references.
Methods
ERRASER-PHENIX combines PHENIX preparation and diffraction refinement with Rosetta-based exhaustive nucleotide rebuilding and minimization. Electron-density scoring helps select conformations that remain compatible with experimental data.
- Pipeline: The pipeline performs initial minimization, single-nucleotide rebuilding, final minimization, and subsequent PHENIX refinement against diffraction data.The rebuilding-minimization cycle was iterated three times to produce the final ERRASER-PHENIX model.
- PHENIX integration: PHENIX-generated models are prepared with added hydrogens and constraints, then refined before and after ERRASER using diffraction-based procedures.Rfree reflections were excluded during map generation to avoid directly fitting the validation set.
- Conformational search: Single-nucleotide rebuilding exhaustively samples torsions and the common 2′-endo and 3′-endo sugar puckers, then uses kinematic loop closure.Candidate conformations are evaluated with Rosetta’s all-atom energy function supplemented by electron-density correlation.
- Density restraint: The electron-density score is computed from pre-calculated atomwise correlations and was an order of magnitude faster than the previous Rosetta implementation.The faster scoring term substantially reduced total computational time.
- Target selection: Problematic nucleotides are identified from bond, angle, pucker, backbone-rotamer, and nucleotide-wise RMSD criteria before rebuilding.Nucleotides with large movement after minimization are also selected because their starting conformations may be energetically unfavorable.
2 OH i
The implementation includes automated model preparation, density-map generation, Rosetta rebuilding, and final PHENIX model selection. Practical constraints include segmentation for large RNAs and special handling of fixed or modified components.
- Finalization: After ERRASER, PHENIX refines the rebuilt coordinates and the best-scored model is selected as the final structure.The final refinement reinstates relevant coordination constraints and follows the established multi-step PHENIX procedure.
- Automation: The automated script accepts a starting PDB file, CCP4 density map, map resolution, and nucleotides to hold fixed during refinement.The example command specifies both the map-resolution parameter and fixed residues.
- Map generation: PHENIX generates electron-density maps from experimental data and refined models, excluding Rfree reflections during map calculation.Missing or excluded observations are filled with calculated values, and kicked maps reduce noise and model bias.
Supplementary Information
The supplementary analyses document the 24-model benchmark, refinement workflow, validation metrics, and structural improvements produced by ERRASER-PHENIX. Across geometric, steric, base-pairing, base-orientation, and low-resolution comparisons, the refined models generally improved over starting PDB models and alternative protocols.
- Benchmark and workflow: The benchmark contains 24 RNA structural models, with analyses organized by resolution and including high- and low-resolution comparisons.Supplementary tables report benchmark composition, resolution ordering, R factors, geometric errors, clashes, base-pair counts, base orientations, and model similarity.
- Benchmark and workflow: ERRASER-PHENIX combines enumerative Rosetta RNA modeling with electron-density-guided real-space refinement and PHENIX diffraction-based refinement.The Rosetta stage explicitly samples selected torsions by enumeration while automatic loop closure determines others.
- Geometric validation: ERRASER-PHENIX eliminated all benchmark outlier bond lengths and angles, exceeding the improvement obtained with PHENIX alone.Outliers are defined as deviations greater than 4 σ from PHENIX ideal geometry.
- Geometric validation: ERRASER-PHENIX reduced the mean clashscore from 40.8 to 7.9, below the mean value of 9.3 for the comparison high-resolution models.Clashscore counts serious atom-pair overlaps of at least 0.4 Å per 1,000 atoms.
- Base-pairing and orientation: ERRASER-PHENIX increased automatically assigned base-pairs in 21 of 24 cases and improved base-pair planarity, co-planarity, and hydrogen-bonding geometry.For 3P49, the base-pair count increased from 44 in the PDB model to 60, compared with 46 and 44 for RNABC-PHENIX and RCrane-PHENIX.
- Base-pairing and orientation: All 12 assessable base-orientation changes agreed well with reference models, while the lowest-resolution ribosome case retained alternative conformations compatible with density.Across the benchmark, refined models produced improved or alternatively possible base orientations; most changes were syn-to-anti flips, but anti-to-syn flips also occurred.
- Model comparison: ERRASER-PHENIX improved torsional and sugar-pucker similarity metrics in nearly all cases and generally outperformed RNABC-PHENIX and RCrane-PHENIX.Low-resolution models were refined to better match higher-resolution models, and improved models had similar or better fits to set-aside diffraction data.