Researchers have developed AI tool to rival the behemoth of Google’s Alphafold 3 in predicting 3D shapes in RNA using less data.
RNAbpFLow could shed new light on the RNA conformational dynamics that underpin diverse cellular processes.
It could also lead to novel RNA-based treatments, such as the messenger RNA vaccines used to prevent COVID-19.
The invention, by two computer scientists at Virginia Tech, generates all-atom RNA conformational ensembles for single-chain RNA monomers.
Unlike several existing deep-learning methods, it can do this without using evolutionary information or homologous structural templates.
The approach is outlined in Nature Methods and the researchers have made the training data, and code freely available.
Study first author Sumit Tarafder, a PhD student, flagged the importance of knowing the shape of an RNA so that it could be targeted.
“In the shape, there are pockets where a drug can attach,” he explained. “If you can’t predict the shape, your pockets are wrong—and the drug won’t work.”
Growing interest in RNA-based therapeutics has driven efforts to determine the 3D structures of RNA.
However, the intrinsic conformational flexibility of RNA presents major challenges when using methods such as X-ray crystallography, nuclear magnetic resonance spectroscopy, and cryo-electron microscopy.
Computer-based methods have emerged as an attractive alternative, but several of these approaches are constrained by the scarcity of RNA structural data in the Protein Data Bank.
While a growing number of methods based on Deep Learning have emerged, most are highly dependent on explicit evolutionary sequence information derived from multiple sequence alignments (MSA) or implicitly make use of homologous information learned by biological language models.
Tarafder and associate professor Debswapna Bhattacharya therefore developed RNAbpFlow, a sequence- and base pair-conditioned all-atom RNA 3D structure generation method based on SE(3)-equivariant flow matching model.
RNAbpFlow incorporates conditions on the nucleotide sequence and base-pairing information from three complementary base pair annotation methods to comprehensively capture canonical and noncanonical interactions.
By incorporating a nucleobase center representation that enables the optimization of angles of all rotatable bonds of nucleobases, it directly outputs all-atom RNA structures in an end-to-end fashion.
This bypasses the need for a post-hoc geometry optimization module, which is impractical in the context of large-scale sample generation.
Base pair-centric auxiliary-loss functions maximize the realization of canonical and noncanonical base-pairing interactions. This enables efficient generation of all-atom RNA conformational ensembles while explicitly modeling nucleobase orientation and flexibility.
Experimental results demonstrated that the introduction of base-pairing conditioning led to improved performance and accuracy connected to the quality of the base pairs.
In blind testing, RNAbpFlow produced a correct overall structure for 12 of 14 RNA targets, compared with eight out of 14 for AlphaFold 3, from Google DeepMind.
“We wanted to keep it simple and predict the structure from scratch, using just the sequence and the base pairs,” Tarafder said.
“The model starts from complete noise and, guided by those base pairs, folds into the right 3D shape.
“That’s the beauty of flow matching, and we can generate as many structures as you want, which lets us capture how the molecule actually moves.”
The post AI Tool Outperforms Google Rival at 3D RNA Shape Prediction appeared first on Inside Precision Medicine.