The basic metrics of RNA 3D structure comparison include:

• Root Mean Square Deviation (RMSD)
• P-value
• Deformation Index (DI)
• Interaction Network Fidelity (INF)
  this metric can be measured by different types of contacts, including:
  • all contacts (INF all)
  • Watson-Crick interactions (INF wc)
  • non-Watson-Crick interactions (INF nwc)
  • stacking interactions (INF stacking)

All of these metrics have been implemented in Basic Assessment Metrics (author: Chichau).

Deformation profile is a metric first described in the paper by Parisien et al. (2009) “New metrics for comparing and assessing discrepancies between RNA 3D structures and models”. RNA 15(10):1875-1885. doi: 10.1261/rna.1700409)

A full implementation can be found at github, maintained by Chichau.

MCQ (Mean of Circular Quantities) has been first presented in the paper by Zok et al. (2014) “MCQ4Structures to compute similarity of molecule structures”. Central European Journal of Operations Research 22(3):457-474. doi: 10.1007/s10100-013-0296-5).

LCS-TA (Longest Continuous Segments in Torsion Angle space) applies a measure first described by Wiedemann et al. (2017) “LCS-TA to identify similar fragments in RNA 3D structures”. BMC Bioinformatics 18(1):456. doi: 10.1186/s12859-017-1867-6).

Their most important features:

• MCQ
  • The method applies to a pair of 3D structures and is size independent.
  • It can be used for a set of structures in all-against-all or all-against-target mode.
  • It translates typical algebraic representation of a 3D structure into the trigonometric one (a set of torsion angles).
  • It computes the distance between structures in torsion angle space.
  • The distance is measured as mean of local distances between the corresponding angles, and provided in degrees.
  • The measure is sequence independent and does not need sequence/structure alignment.
• LCS-TA
  • The method applies to a pair of 3D structures and is size independent.
  • It uses MCQ-based measure for structure comparison.
  • Within the compared structures, it finds the longest continuous segments which display similarity in torsion angle space.
  • Two segments are considered similar if their MCQ is below the predefined threshold.
  • The method provides segment length and its position in the structure.
  • The length of the longest continuous segment is a measure of similarity of two structures.
  • The method can be run in sequence-dependent or sequence-independent mode.

A full implementation of both methods can be found on github, MCQ maintained by Zok, LCS-TA by Wiedemann.

We provide all the native and predicted structures from RNA-Puzzles here as a dataset resource.

Currently, there are few gold standard datasets available in benchmarking RNA 3D structure prediction. RNA-Puzzles is a well-established blind assessment of RNA 3D structure prediction. Therefore, the collection of its structure data provide a good benchmark set for RNA 3D structure prediction and assessment. Besides, models submitted to RNA-Puzzles are generated by various prediction methods that generally cover all the RNA 3D structure prediction landscape, and thus, can be considered as perfect decoy sets in deriving energy functions for RNA structure folding.

We provide raw predictions, evaluation-ready data, and structures standardized using RNA-PDB-tools.

After the first RNA-Puzzles meeting (2016 Oct, Strasbourg), the RNA-Puzzles community have agreed to unify all predictions to the same PDB format.

To check if the submitted structures are in good format, a simple script based tool is provided:

RNA-Puzzles format – a tool used to check if a predicted structure is in good format.

To format/normalize RNA structures, please use structure normalization tools.

RNAQUA (RNA QUality Assessment) is a web service-based wrapper of basic RNA comparison metrics. It is a RESTful web service client developed in Java. The tool provides a set of web services initially designed for RNAssess (Lukasiak et al. (2105) Nucleic Acids Research 43(W1):W502-W506. doi:10.1093/nar/gkv557) to support the quality assessment of RNA 3D structures. RNAQUA requires a stable release of JRE 7 (or later) installed on a user workstation with an internet access. Most important features of RNAQUA:

• At the input, a user provides RNA 3D structure(s) in PDB format.
• Output data are returned in XML format.
• RNAQUA allows to compute the following measures:
  • ClashScore,
  • Root-mean-square deviation (RMSD),
  • Interaction network fidelity (INF),
  • Deformation index (DI),
  • P-value (either for entire RNA 3D structure(s) or for a set of discontinuous 3D substructures).
• Two processing modes are available:
  • An analysis of single RNA 3D structure (PDB validation, ClashScore, sequence-based analysis, structure unification),
  • An analysis of RNA 3D model(s) with respect to the reference structure (RMSD, INF, DI, P-value, Deformation profile, sequence-based differences between structures, multiple models over the reference structure superposition).
• Optionally, the user can define alignment between the reference structure and all the corresponding RNA 3D model(s), which is helpful if there are differences in sequence, distribution of chains or residue numbering.

Full implementation of the tool is stored on github and maintained by Antczak.

RNA structures in RNA-Puzzles are deposited in the PDB format. However, we need to consider the variations when participants generate their predictions, e.g., the nomenclature of the residue/atom names, and the variations in the format of deposited native structure. Sometimes, the sequence used in prediction may slightly deviate from the solved crystal structure: in such a case, we need to compare only the common part of the structure.

Considering these aspects, we provide tools to normalize the PDB format of an RNA 3D structure:

PDB normalizer – this tool is the PDB normalization tool used by RNA-Puzzles, and is part of the basic assessment metrics tool (maintained by Chichau)

RNA PDB tools - alternatively, you may use a more functional tool RNA-PDB-tools (mained by Magnus).