Benchmarking RNA all-atom force fields using hairpin motifs

Ribonucleic acid (RNA) is a multipotent polymer of great biological and engineering interest. Computer simulations that investigate the structure and dynamics of these molecules are powerful tools for understanding and manipulating their physical properties. Despite the progress made towards predictive modeling, many physicochemical aspects remain elusive. Modeling the structural dynamics of RNA is particularly challenging, owing mostly to the flexible, polyanionic backbone and the many polarizable moieties. Several empirical all-atom RNA force fields (FFs) have been developed over the last two decades to address this challenge, albeit with limited reported success. In addition, the sampling profile of such FFs has rarely been explored in detail for RNAs larger than tetranucleotides. This study addresses this issue by benchmarking four state-of-the-art FFs against four full-sequence RNA hairpins, of varying conformational stability, that incorporate a wide range of structural characteristics. Mutli-μs enhanced sampling simulations, starting from distinct and independent conformations, generated well-converged ensembles of loop conformations, enabling an accurate assessment of FF performance. Beyond the robust description of loop conformations, backbone dihedrals, sugar puckers, and contact profiles were thoroughly analyzed and compared to experimental data. Overall, none of the studied FFs was found to quantitatively agree with the available NMR ensembles. Despite this discrepancy, we propose that some FFs might be more appropriate when non-canonical base-pairing or base-backbone interactions are involved. Lastly, we note that different FFs sample significantly different conformational spaces for most cases, making the choice of an appropriate RNA FF non-trivial.