A deep‑learning model named SMRTnet was introduced to predict small‑molecule–RNA interactions using RNA secondary structure and multimodal training data, sidestepping the need for tertiary structures. The authors report that SMRTnet outperforms existing tools across benchmarks, identifies nanomolar‑to‑micromolar binders for disease‑associated RNAs, and produced validated hits that downregulated MYC and impaired cancer cell growth. SMRTnet expands feasible RNA targets for small‑molecule discovery by leveraging language models and graph attention networks to infer binding propensities from sequence and secondary structure features. The approach accelerates early hit identification for RNA‑targeting therapeutics and reduces dependence on high‑resolution structures that are scarce for many RNAs. Clarification: RNA secondary structure is the pattern of local base pairing in a transcript; SMRTnet predicts binding using these features rather than full 3D fold information.