CRANBERRY: An RNA Dynamics Model with Sugar Puckering and Noncanonical Base Pairing

We introduce a new coarse-grained model "CRANBERRY" that incorporates sugar puckering and non-canonical base pairing, two factors central to RNA structure and dynamics, yet rarely included in most coarse-grained models. Our model is parameterized through a contrastive divergence approach, combined with fine-tuning strategies to improve accuracy in generating disordered states, a feature that is critical for the accurate description of thermodynamics. This two-stage training procedure greatly enhances cooperative folding behavior. Due to these advances, the models predictive performance is comparable to that of all-atom force fields for native-state structural fluctuations. Furthermore, CRANBERRY exhibits better agreement with experimental data on stacking free energies and disordered structures measured by Small Angle X-ray Scattering. In addition, CRANBERRY can reversibly fold tetraloops with an RMSDmin of 1.4 [A] de novo, which continues to be challenging for all-atom models. It predicts melting temperatures in agreement with experimental values, and with a greater cooperativity than all-atom predictions.