FoldARE, an RNA secondary structure analysis and prediction tool via generative pseudo-SHAPE modeling

RNA secondary structure prediction remains limited by RNA conformational heterogeneity and scarcity of experimental data: many RNAs populate ensembles of near-isoenergetic folds, and structure-probing data such as SHAPE are often unavailable. Here we introduce FoldARE (Folding and Analysis of RNA Ensembles), a two-step framework that derives pseudo-SHAPE constraints from in silico structural ensembles and uses them to guide a downstream SHAPE-aware predictor. In the first ("ensembler") step, an ensemble is generated and parsed position-by-position to estimate single-strandedness frequencies, which are converted into a pseudo-SHAPE reactivity profile via a weight-and-threshold scheme. In the second ("predictor") step, this profile is supplied as a constraint to a SHAPE-compatible folding algorithm to produce an improved secondary structure. We systematically evaluated all combinations of four ensemble-capable predictors: ViennaRNA, RNAstructure, LinearFold, and EternaFold. We optimized parameters on a manually curated, structurally diverse 25-RNA training set and validated robustness using multiple scoring schemes, including identity-based measures and an exact base-pair-matching metric. The optimal configuration uses EternaFold as the ensembler and RNAstructure as the predictor, yielding consistent gains over all standalone methods and over other ensembler/predictor pairings. On external benchmark RNA structure collections (RNAstrand, ArchiveII, and bpRNA; total n = 1964 after filtering) and on the experimentally derived eFold dataset spanning human mRNAs, pre-miRNAs, and lncRNAs (n = 1024), FoldARE achieved the highest accuracy across datasets with highly significant improvements in paired comparisons. Beyond prediction, FoldARE provides modules for ensemble-level comparative analysis, including pairwise and multi-tool consensus assessment, per-nucleotide variability metrics, and interactive visualizations. It also supports the evaluation of m6A modification effects on folding ensembles using modification-aware engines. Together, our results show that mining ensemble statistics to generate pseudo-probing constraints is an effective, accessible strategy to improve RNA secondary structure prediction and to support ensemble-focused structural analysis. FoldARE is freely available on GitHub (https://github.com/TebaldiLab/FoldARE).