RNA Structure Prediction Takes a Leap Forward with RNAbpFlow
In a blind test against a community benchmark, RNAbpFlow produced correct overall structures for 12 of 14 RNA targets, compared to 8 of 14 for AlphaFold 3.
Blind Test Results
Researchers at Virginia Tech have developed RNAbpFlow, a new method for predicting RNA three-dimensional structures. In a head-to-head blind test against a community benchmark, RNAbpFlow significantly outperformed AlphaFold 3, producing correct overall structures for 12 of 14 RNA targets versus AlphaFold 3's 8 of 14.
A Key Advantage
Crucially, RNAbpFlow does not require large evolutionary sequence databases, which are often unavailable or incomplete for RNA molecules. This makes it especially useful for studying RNAs with few known relatives, including a conserved element from SARS-CoV-2 and a laboratory ribozyme.
How It Works
RNAbpFlow uses flow matching, a generative AI technique, to create complete all-atom 3D structures from sequence and base pair information. The model begins from random noise and folds into structures guided by base pairs. This approach allows it to generate multiple structures, capturing the natural conformational flexibility of RNA molecules.
Real-World Implications
Accurate RNA structure prediction could accelerate drug discovery. Notably, risdiplam, a drug targeting RNA shape for spinal muscular atrophy, demonstrates the potential impact of this technology.
Current Limitations
While powerful, RNAbpFlow has limitations. On larger, complex RNAs, established servers that rely on evolutionary data still perform better.
Looking Ahead
The research team is already developing an improved version for the upcoming CASP competition. In a move that will accelerate the field, the implementation, training data, and code have been publicly released.
Study Details
- Publication: Nature Methods, June 30
- Lead Author: Sumit Tarafder, doctoral student
- Senior Author: Debswapna Bhattacharya, associate professor
- Funding: National Institutes of Health and National Science Foundation