ROMET-RNA benchmarks and selects RNA 3D structures using multiple predictors and alignment strategies

Accurate prediction of RNA tertiary structure is essential for understanding RNA function, yet systematic, controlled benchmarking and integration of RNA three-dimensional structure prediction methods, particularly with respect to multiple sequence alignment strategies used by deep learning models, remain limited. Here, we present ROMET-RNA, an automated benchmarking framework and meta-selection toolkit that benchmarks eleven state-of-the-art RNA structure predictors within a standardized workflow and evaluates three complementary MSA pipelines, rMSA, RNAcmap3 MARS, and Infernal RNAcentral. Model selection is performed using a consensus scoring strategy combining TM-scoreRNA based structural similarity with the deep learning derived statistical potential lociPARSE. Benchmarking across four non-redundant test sets comprising 205 RNAs, including CASP15, RNA Puzzles, and recent PDB structures, demonstrates that ROMET-RNA consistently selects more accurate models than individual methods would on its own and we confirm that this is not an artefact of training-test overlap. Further analysis reveals that coevolutionary information substantially improves accuracy, that optimal MSA strategies are method specific, and that MSA depth exhibits a nonlinear relationship with performance. ROMET-RNA provides an open-source standardized framework that enhances RNA 3D structure model selection, supporting advances in RNA structural biology and genomics.