HERCULES: a deep-learning framework for RNA-binding region localization and mutation-effect prediction

HERCULES, combining a fine-tuned ProteinBERT with residue physicochemical descriptors, discriminates RNA-binding proteins from non-RBPs with an AUROC of 0.86 on an independent benchmark.

• Why it matters: Identifying RNA-interacting regions and prioritizing sequence variants that alter binding remains difficult from protein sequence alone, limiting variant interpretation and target discovery.
• What they did: The framework localizes interaction sites and estimates amino-acid substitution effects by integrating domain-scale sequence context from ProteinBERT with local residue-change sensitivity from physicochemical descriptors.
• The result: It enriches high-scoring positions within annotated interaction regions and prioritizes deleterious substitutions and non-catalogued candidate RBPs, with protein microarray support using CGG, CAG, and interrupted CAACAG repeats.