gRNAde: Geometric Deep Learning for 3D RNA inverse design (2305.14749v6)

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 geometry and 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. Under the hood, gRNAde is a multi-state Graph Neural Network that 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. Open source code: https://github.com/chaitjo/geometric-rna-design

• The paper introduces gRNAde, a novel pipeline that leverages a multi-graph neural network architecture to capture RNA’s conformational diversity.
• It utilizes a large-scale 3D RNA dataset with over 11,000 structures, improving upon traditional single-state design methods.
• The study demonstrates that gRNAde achieves higher native sequence recovery, especially in structurally diverse RNA molecules.

Multi-State RNA Design with Geometric Multi-Graph Neural Networks

The paper "Multi-State RNA Design with Geometric Multi-Graph Neural Networks" presents a novel approach to computational RNA design, addressing the intrinsic conformational flexibility of RNA molecules. This flexibility allows a single RNA sequence to adopt multiple distinct three-dimensional (3D) structural states, which is fundamental to its diverse biological functions.

Summary of the Study

The authors propose gRNAde, a geometric RNA design pipeline, which improves upon traditional methods by considering multiple 3D RNA backbone structures simultaneously. This approach contrasts with the conventional single-state design methodologies that focus on a singular desired RNA conformation. By leveraging geometric deep learning and specifically utilizing graph neural networks (GNNs), the pipeline seeks to enhance the recovery of native RNA sequences that are capable of existing in multiple conformations.

Key Contributions

1. Multi-Graph Neural Network Architecture: The core innovation of gRNAde is its use of a multi-Graph Neural Network architecture. This architecture independently processes multiple RNA conformations via message passing and subsequent conformer order-invariant pooling. This design captures the conformational diversity inherent in RNA molecules, which is essential for accurate sequence prediction.
2. Large-scale 3D RNA Dataset: A significant contribution of this paper is the creation of a new dataset for 3D RNA design, derived from the RNASolo repository. With over 11,000 structures, this dataset provides more substantial and diverse training data than previously available collections.
3. Performance Evaluation: The paper illustrates the utility of gRNAde by comparing its sequence recovery performance against single-state approaches. The gRNAde model demonstrates enhanced accuracy in recovering native sequences, particularly for multi-state and structurally diverse RNAs, across various data splits designed to test structural diversity, variable structure quantity, and biological dissimilarity.

Numerical Results

The analysis reveals that gRNAde consistently outperforms single-state models, especially in scenarios with high structural diversity. For example, when tested on datasets with varying average pairwise RMSD among structures, gRNAde achieves higher sequence recovery rates compared to traditional methods, indicating its ability to handle complex RNA structures.

Implications and Future Directions

The implementation of gRNAde marks a significant evolution in RNA design methodologies, emphasizing the necessity to account for RNA's conformational variability. This represents a critical step towards more reliable and functional RNA-based nanotechnology and therapeutic solutions. The methodological advances brought forward by gRNAde, particularly its multi-graph processing capabilities, open new pathways for the expansion of geometric deep learning in other biomolecular design tasks, including proteins and other nucleic acids.

The paper also anticipates that with improvements in RNA structure determination techniques (e.g., cryo-EM), datasets containing 3D structures of RNA will become more abundant, further facilitating the development of sophisticated computational models for RNA design.

Overall, this work highlights a pivotal shift towards integrating multi-state RNA representations in design models, setting a foundation for future developments that may incorporate more intricate aspects of biomolecular design, potentially fostering advances in synthetic biology and therapeutic design.