gRNAde: A Geometric Deep Learning Pipeline for 3D RNA Inverse Design.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• RNA Design

Background:

• RNA's biological functions depend on its 3D structure and flexibility, allowing single sequences to adopt multiple states.
• Current RNA design methods often focus on secondary structure, neglecting 3D geometry and conformational diversity.

Purpose of the Study:

• To introduce gRNAde, a novel geometric RNA design pipeline.
• To enable sequence design that explicitly considers RNA 3D structure and dynamics.

Main Methods:

• gRNAde utilizes a graph neural network with an SE(3) equivariant encoder-decoder framework.
• It generates RNA sequences conditioned on unknown base identities within given 3D backbone structures.

Main Results:

• gRNAde successfully re-designs existing RNA structures (riboswitches, aptamers, ribozymes) from the Protein Data Bank (PDB).
• The pipeline demonstrates higher native sequence recovery accuracy compared to physics-based tools.
• gRNAde offers significantly faster computation times than existing 3D RNA inverse design methods like Rosetta.

Conclusions:

• gRNAde provides an effective computational approach for 3D RNA inverse design.
• The method advances RNA sequence design by integrating structural and dynamic considerations.