RNA Small-Molecule Binding-Site Prediction by Integrating RNA Language Models and Structural Geometry

RSMBind predicts nucleotide-level RNA small-molecule binding sites, achieving the highest MCC on three of four benchmarks, including 17.1% gains on RB9.

• Why it matters: RNA structural complexity and scarce experimentally characterized RNA-ligand complexes limit computational binding-site prediction, slowing RNA-targeted drug discovery.
• What they did: The graph attention model combines local 3D geometric descriptors with pretrained RNA language model embeddings (RNA-BERT, ERNIE-RNA, RNA-FM), trained on TR60 and HARIBOSS and tested on RB9, TE18, TL12, and JL10.
• The result: It improves over prior best results by 17.1% on RB9, 7.2% on TL12, and 16.0% on JL10, stays competitive on TE18, and still works using only language-model embeddings.

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