Ensemble Refinement of Cryo-Electron Microscopy-Derived RNA Structures Using Molecular Dynamics Simulations

RNA molecules play central roles in biology as catalysts, regulators, and structural scaffolds, with their function tightly linked to their ability to adopt complex and dynamic confor- mations. Despite advances in structural biology, accurately characterizing RNA structures remains difficult because of their intrinsic flexibility and the limitations of experimental meth- ods. Cryo-electron microscopy (cryo-EM) has enabled major progress, but RNA structures often contain unresolved regions or mismodeled secondary structures. This thesis explores the integration of molecular dynamics (MD) simulations with cryo- EM data in order to refine RNA structural models and better capture conformational vari- ability. A central focus is the application of ensemble refinement methods, particularly the Bayesian framework of metainference which explicitly accounts for structural heterogeneity and experimental uncertainty. Specifically, ensemble refinement was applied to two ribozymes: a Group II intron, re- solved in complex with its maturase protein, and a Group I intron, which is an RNA-only structure. In the first case, secondary-structure restraints were used to stabilize peripheral helices, while in the second case water and ions were included in the refinement, allowing the identification of numerous ordered solvent molecules and ions. In both systems, the refined ensembles showed improved agreement with experimental data while remaining physically realistic. In addition, we participated in the CASP16 challenge (task R1260), which required prediction of the solvation shell around the Group I intron. We also developed a validation pipeline for the submitted predictions of solvation shells, which demonstrated the superior accuracy of MD-based predictions in the R1260 task, compared to alternative approaches. Together, these studies demonstrate the potential of ensemble-based refinement methods to improve RNA structural models derived from cryo-EM data. The results highlight both the progress and the limitations of current approaches, and point toward future developments integrating experimental data, advanced ion models, and ensemble computational methods to better capture RNA structural heterogeneity.