TriRNASP: A knowledge-based potential with three-body effects for accurate RNA structure evaluation.

Accurate RNA 3D structure evaluation on candidates is critical for accurate RNA 3D structure predictions. Although some knowledge-based potentials and scoring functions have been developed for RNA 3D structure evaluation, their performance still remains rather limited, especially for the challenging data sets from RNA 3D structure prediction methods. In this work, we developed TriRNASP, an efficient statistical potential with three-body effects, aiming for accurate RNA 3D structure evaluation. TriRNASP integrates coarse-grained three-body correlations with Kullback-Leibler divergence and atom clash penalty. Our benchmarks on extensive test data sets from diverse RNA 3D prediction methods demonstrate that TriRNASP consistently exhibits outstanding performance in identifying native and relatively near-native structures, compared with current state-of-the-art statistical potentials and deep-learning-based scoring functions. Particularly, the superior performance of TriRNASP is more pronounced for the newly released data sets from the CASP15 and CASP16 assessments. Moreover, TriRNASP achieves exceptional computational efficiency, making it ideally suited for large-scale structure evaluation tasks. TriRNASP is freely available at GitHub.