Comparative analysis of RNA 3D structure prediction methods: towards enhanced modeling of RNA-Ligand interactions

This dataset accompanies the publication "Comparative Analysis of RNA 3D Structure Prediction Methods: Towards Enhanced Modeling of RNA-Ligand Interactions." Our study's primary objective was to evaluate the accuracy of various methods in modeling RNA structures, with a particular focus on RNA-small molecule complexes and ligand-binding sites. We scrutinized the performance of six RNA 3D structure prediction programs—DeepFoldRNA, RhoFold, BRiQ, FARFAR2, SimRNA, and Vfold2—using RNA sequences as a standard input across all methods. Methods like FARFAR2, SimRNA, and Vfold2 were examined both with and without the inclusion of secondary structure information. Notably, BRiQ requires secondary structure restraints for its operation and was, therefore, only run under these conditions. The dataset is meticulously organized into sub-directories named according to each method. For SimRNA, FARFAR2, and Vfold2, directories without secondary structure input maintain the method's name, whereas runs that included secondary structure information are denoted with an '_ss' suffix (e.g., SimRNA_ss). For instances where secondary structures were utilized, we employed ideal secondary structures derived from the reference structure, extracted using the x3dna-dssr program v1.9.10. All secondary structures were subject to manual inspection and refinement to address any anomalies introduced by x3dna-dssr, ensuring the highest fidelity in our modeling efforts. During the final stages of preparing this publication, AlphaFold 3 was released. To benchmark the performance of all ML-based methods (AlphaFold 3, DeepFoldRNA, and RhoFold), we developed two new datasets: Blind set 1 (B1) and Blind set 2 (B2) (see Supplementary Table S2).