Limits of deep-learning-based RNA prediction methods.

AlphaFold3 and Boltz-1 deliver the best RNA structure predictions on an independent non-redundant dataset, but accuracy drops sharply for structurally divergent monomeric RNAs.

• Why it matters: Progress in RNA modeling has lagged behind proteins, and it is unclear whether recent deep-learning accuracy improvements generalize beyond familiar RNA folds.
• What they did: The study benchmarks 10 deep learning methods on monomeric RNAs and RNA complexes using an independent, non-redundant dataset, comparing performance across structural similarity and complex types.
• The result: For monomers, reliable predictions concentrate on recurring, well-defined folds (e.g., L-shaped and simple loops), while divergent RNAs are poorly modeled; for complexes, AlphaFold3 and Boltz-1 outperform HelixFold3 and RosettaFoldNA with no significant difference between the top two on most measures.

AI summary of the abstract · may contain errors · not medical advice · read the paper.