AlphaFold 3 Prediction Accuracy and Reliability across RNA Structures and Protein-RNA Interfaces

AF3 preserves protein fold accuracy on structurally novel RNAs, but RNA-fold and interface accuracy drop sharply with little change in predicted RNA confidence score across 1,300+ protein-RNA complexes and ~270 RNA-only systems.

• Why it matters: It is unclear when AF3 confidence metrics reliably indicate correct RNA structure and protein-RNA interface placement, especially for novel RNAs and structurally similar but incorrect candidates.
• What they did: The study benchmarked AF3 on 1,300+ protein-RNA complexes and ~270 RNA-only systems, using post-cutoff RNAs versus pre-cutoff candidates selected by sequence and fold similarity, plus input ablations and motif checks against CLIP-seq, SELEX, and RNAcompete.
• The result: Protein alignments/templates from RNA-bound proteins drive complex assembly, RNA alignments matter most for single-chain and long RNAs, and a 2026 released-complex ranking score combining AF3 confidence with RNA familiarity and template use ranked accurate folds and correctly placed interfaces better than AF3’s ranking score.

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