[Paper Review] gRNAde: Geometric Deep Learning for 3D RNA inverse design

Chaitanya K. Joshi, Arian R. Jamasb+6 more|arXiv (Cornell University)|May 24, 2023

RNA and protein synthesis mechanisms7 citations

TL;DR

gRNAde uses a geometric multi-graph neural network to design RNA sequences conditioned on multiple 3D backbone conformations, improving native sequence recovery for multi-state RNA designs.

ABSTRACT

Computational RNA design tasks are often posed as inverse problems, where sequences are designed based on adopting a single desired secondary structure without considering 3D conformational diversity. We introduce gRNAde, a geometric RNA design pipeline operating on 3D RNA backbones to design sequences that explicitly account for structure and dynamics. gRNAde uses a multi-state Graph Neural Network and autoregressive decoding to generates candidate RNA sequences conditioned on one or more 3D backbone structures where the identities of the bases are unknown. On a single-state fixed backbone re-design benchmark of 14 RNA structures from the PDB identified by Das et al. (2010), gRNAde obtains higher native sequence recovery rates (56% on average) compared to Rosetta (45% on average), taking under a second to produce designs compared to the reported hours for Rosetta. We further demonstrate the utility of gRNAde on a new benchmark of multi-state design for structurally flexible RNAs, as well as zero-shot ranking of mutational fitness landscapes in a retrospective analysis of a recent ribozyme. Experimental wet lab validation on 10 different structured RNA backbones finds that gRNAde has a success rate of 50% at designing pseudoknotted RNA structures, a significant advance over 35% for Rosetta. Open source code and tutorials are available at: https://github.com/chaitjo/geometric-rna-design

Motivation & Objective

• Motivate RNA design to account for conformational flexibility beyond single-state designs.
• Introduce a geometric multi-graph neural network pipeline (gRNAde) for multi-state RNA inverse design.
• Demonstrate improved native sequence recovery on a large-scale 3D RNA design dataset, especially for structurally diverse RNAs.
• Provide a dataset and benchmarks to enable multi-state RNA design research.

Proposed method

• Represent RNA backbones as coarse-grained 3-bead graphs per nucleotide with P, C4’, and N1/N9 coordinates.
• Construct independent geometric graphs for each RNA backbone conformation and merge them into a geometric multi-graph.
• Apply a multi-Graph Neural Network encoder with conformer-wise message passing that is equivariant to conformer order.
• Pool across conformations using conformer order-invariant Deep Set pooling to obtain a single per-nucleotide representation.
• Use autoregressive GVP-GNN decoders to predict the nucleotide at each position conditioned on multi-state representations.

Experimental results

Research questions

• RQ1 Does incorporating multiple backbone conformations improve native sequence recovery compared to single-state designs?
• RQ2 How does performance vary with the number of conformations (k) used during design?
• RQ3 Can the model handle variable numbers of conformations and biologically diverse RNA ensembles?
• RQ4 Is multi-state design more reliable for case studies with structurally diverse RNAs (e.g., aptamers)?

Key findings

• Multi-state designs using gRNAde consistently improve native sequence recovery compared with single-state baselines across multiple data splits.
• Using multiple conformations as input boosts sequence recovery particularly for multi-state and structurally diverse RNAs.
• A case study on theophylline aptamer backbones shows that single-state models can be unreliable due to conformer choice, while gRNAde yields more consistent sequences by accounting for multiple states.
• The multi-state architecture enables the model to capture conformational diversity without increasing parameter count, and the approach is validated on a large-scale 3D RNA design dataset.