Integrating Coarse-Grained Simulations and Deep Learning to Unveil Entropy-Driven dsRNA Unwinding by DDX3X

Abstract

DEAD-box RNA helicases (DDXs) are essential RNA metabolism regulators that typically unwind dsRNA in an ATP-dependent manner. However, recent studies show some DDXs can also unwind dsRNA without ATP, a phenomenon that remains poorly understood. Here, we developed HelixTriad coarse-grained RNA model, incorporating Watson-Crick base pairing, base stacking, and electrostatics within a three-bead-per-nucleotide scheme to accurately reproduce experimental RNA melting curves. Molecular dynamics simulations showed that weak, specific DDX3X-dsRNA interactions drive stochastic strand separation without ATP. Free energy analysis revealed that successful unwinding via high-entropy, stand-displacing intermediates. Furthermore, we introduced Entropy-Unet, a deep learning framework for entropy prediction, which corroborated theoretical estimates and uncovered a hierarchical pattern of entropy contributions. Together, our findings suggest that ATP-independent dsRNA unwinding by DDXs is predominantly entropy-driven, offering new mechanistic insights into RNA helicases versatility.