DynaRNA: accurate dynamic RNA conformation ensemble generation with diffusion model.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Non-coding RNAs (ncRNAs) are crucial for diverse biological functions, with their dynamic structures mediating these roles.
• Traditional methods for characterizing RNA conformational dynamics, including experimental techniques and molecular dynamics (MD) simulations, face limitations due to cost and methodology.
• Understanding RNA dynamics is essential for various biological processes and therapeutic development.

Purpose of the Study:

• To present DynaRNA, a novel diffusion-based generative model for predicting RNA conformation ensembles.
• To enable rapid and accurate exploration of RNA conformational space.
• To provide a complementary computational tool for RNA structural biology.

Main Methods:

• DynaRNA utilizes a denoising diffusion probabilistic model (DDPM) integrated with an equivariant graph neural network (EGNN).
• The model directly learns RNA 3D coordinates, enabling end-to-end generation of conformation ensembles.
• It does not require Multiple Sequence Alignments (MSA) information for generating experimental geometries.

Main Results:

• DynaRNA accurately generates RNA conformation ensembles, including a tetranucleotide ensemble with a lower intercalation rate than MD simulations.
• The model successfully captures rare excited states, such as the HIV-1 Trans-Activation Response (TAR) element.
• DynaRNA can recapitulate de novo folding of tetraloops.

Conclusions:

• DynaRNA is an efficient and versatile platform for modeling RNA structural dynamics.
• It serves as a valuable complementary tool to existing methods like MD simulations.
• The model has broad implications for RNA structural biology, synthetic biology, and therapeutic development.