About
SimRNAweb offers a web interface to support most of the features of the program SimRNA and offers some unique setup features of its own that can be used by advanced users. There is no need to download programs, run programs with various command line options or find the support and resources (80 cores, computational disk space, etc.) of a high performance computing facility.
The program SimRNA can carry out de novo folding of RNA sequences using only the sequence information or can accept a PDB file containing an RNA structure in any state of folding (obtained either from experiment or from a simulation). When starting with a PDB file with a RNA structure in some folding trajectory, the user can either carry out coarse-grained refinements or search the general trajectory of folding from such a starting structure. In benchmark tests, SimRNA was able to recapitulate the secondary structure and 3D structure with fairly high accuracy, including correct prediction of pseudoknots. SimRNA can also fold an RNA sequence under a variety of restraints including secondary structure restraints, distance restraints and, when a PDB file is supplied, parts of the structure in the PDB file can be frozen while other parts are allowed to be flexible.
All of these features and more are provided with the SimRNAweb server and, in many cases, this does not require the reformating of PDB files. SimRNAweb accepts any form or combination of restraints including secondary structure dot bracket notation, distance restraints, and, when one has a PDB file, the position of specific residues can be frozen (while leaving other residues free) by simply indicating the RNA chain and residue index. In addtion, a unique function on the server is the ability to add missing residues to a PDB structure. This allows the possibility of building a complete structure from a PDB file of a crystal structure determination where there are numerous gaps in the sequence.
Since restraints can be derived from experimental or computational analyses (including information about secondary structure and/or long-range contacts) it is possible to use this information in de novo folding or refinement of RNA molecules. With some advanced effort, molecules with folds, pockets, exposed or buried regions of structure can be explored. Moreover, since the trajectory files can also be obtained, SimRNAweb can be used to analyze conformational landscapes and identify potential alternative structures.
To learn more about SimRNA, please read SimRNA: a coarse-grained method for RNA folding simulations and 3D structure prediction
To use the standalone version of SimRNA, please download simrna_package.zip or visit the SimRNA software page.
The SimRNA-package.zip archive was previously distributed as part of the book Protein Structure Prediction (4th Edition) to support the chapter workflow on SimRNA modeling.
WARNING!
A single SimRNA run is not a "structure prediction". Please read this before you download and run SimRNA, and especially before you compare its results to other computational tools:
1. What SimRNA is for:
- Exploring RNA conformational space and generating plausible 3D folds, including pseudoknots and alternative conformations.
- Predicting RNA 3D structures by extensive sampling, preferably guided by reliable secondary-structure, contact, or experimental restraints.
- Studying folding intermediates and conformational landscapes, and remodeling or refining uncertain or missing regions.
2. What SimRNA is NOT for:
- It is not a one-shot, "sequence in → structure out" predictor. A single SimRNA run is not a structure prediction!
- It is not primarily a secondary-structure prediction program. Although SimRNA can recover secondary structure, for routine 3D modeling we recommend providing reliable secondary structure restraints from dedicated methods, particularly for RNAs > 50 nt.
- The best-score frame is not automatically the best structural prediction. Proper prediction requires extensive sampling followed by analysis and clustering of many trajectories. Clustering generally performs better than score alone.
- It is not an all-atom, high-resolution method. SimRNA models are coarse-grained and should be subjected to appropriate all-atom reconstruction/refinement when atomic detail is required. We recommend our method QRNAS for this.
- SimRNA does not generate the predicted 3D structures as PDB or mmCIF files by default. Its primary output is a trajectory containing sampled conformations. To obtain actual structural models, the trajectories must be analyzed: typically combined, filtered by score, clustered, and selected frames converted to the PDB or mmCIF formats. Do not mistake a starting/initial PDB file for "the SimRNA prediction".
3. Don'ts:
- Don't run SimRNA without first reading the SimRNA paper, Supplementary Information, and usage recommendations. Don't modify the recommended parameters unless you understand exactly why you are doing so.
- Don't expect reliable predictions from one short simulation, too few independent runs, too few replicas, or insufficient use of available restraints.
- Don't take the lowest-score frame from one trajectory and report it as "the SimRNA prediction". For most applications this is a fundamentally incorrect use of the method. SimRNA score is neither physical free energy nor a direct measure of structural accuracy.
4. Do's:
- Use all reliable prior information available: at minimum confidently predicted secondary structure, and whenever possible tertiary contacts, experimental restraints, or a good starting model.
- For standard de novo predictions, run multiple independent REMC simulations with different random seeds. In the 2016 benchmark, predictions used 8 independent runs, each with 10 replicas, each for 16 million Monte Carlo steps, followed by pooling and clustering of the trajectories.
- In general, sample as extensively as computing resources allow. Combine trajectories, retain a low-score subset (for example, the best-scoring 1%), cluster it using an appropriate RMSD threshold, and inspect several largest clusters.
- SimRNA produces a conformational landscape, not "the structure". A single model may be selected from this landscape, but it should not be interpreted without first examining the landscape and the various clusters of alternative low-score conformations.