Datasets for "Exploring Conformational Transitions of RNA Dimers via Machine Learning Potentials"

Here, you can access the molecular dynamics trajectories for the PMD and TREMD simulations of ApA dimer, as well as the quantum-mechanical datasets generated to train the equivariant ML potentials.

ABSTRACT

RNA is a highly flexible biopolymer that adopts a wide range of conformations while forming well-defined structural motifs essential for its biological function. Elucidating these structure–function relationships therefore requires a thorough exploration of its conformational landscape. In this sense, all-atom molecular dynamics (AAMD) simulations provide a powerful framework for this purpose. However, current classical RNA force fields often suffer from limited transferability and inefficient sampling of transitions between stable states, particularly in moderately large RNA systems. To overcome similar limitations in biomolecular modeling, quantum-informed machine learning (ML) potentials have recently emerged as a promising alternative, offering improved accuracy and transferability compared to classical force fields. In this work, we assess the performance of ML potentials in the exploration of RNA conformational space. As a model system, we focus on the adenine–adenine dinucleoside monophosphate (ApA) dimer, a fundamental RNA building block. We generated an extensive quantum-mechanical (QM) dataset of physicochemical properties for ApA conformations obtained from temperature replica exchange molecular dynamics (TREMD) simulations. Despite its small size, the ApA dimer exhibits a complex energetic landscape with six well-defined conformational clusters in which quantum effects and solvent-mediated interactions play a crucial role. Using this dataset, we parameterized ML potentials built on the equivariant MACE architecture and informed by both ab initio and semi-empirical property data. The resulting potentials successfully reproduce key conformational features of the ApA system, including base stacking, sugar puckering, and backbone flexibility. Moreover, they provide broader coverage of relevant structural transitions compared to the general-purpose SO3LR and MACE-OFF24 models. These findings underscore the importance of comprehensive QM datasets for RNA building blocks and highlight the need for robust and efficient validation metrics for assessing the performance of ML potentials.

PREPRINT

https://www.biorxiv.org/content/10.64898/2026.02.25.707885v1 

FILES:

dft_data.xz: quantum-mechanical data of ApA conformations computed at PBE0+MBD level.

dftb_dat.xz: quantum-mechanical data of ApA conformations computed at DFTB3+MBD level.

plainMD.xz: MD trajectory of solvated ApA dimer at constant temperature.

tremd.xz: TREMD trajectory of solvated ApA dimer.

MLmodels.xz: ML potentials developed in this work.

dihedral_transitions.xz: Scripts for the structural analysis and conformational classification of MD trajectories.