A new deep-learning framework called SMRTnet predicts small-molecule–RNA interactions using RNA secondary structure and multimodal AI models. The method fuses large-language models with convolutional and graph-attention networks to score interactions without prior tertiary structures, and authors report high performance across experimental benchmarks. SMRTnet identified nanomolar-to-micromolar hits against ten disease-linked RNAs and validated a compound that downregulated MYC and impaired cancer cell growth. The tool promises to broaden RNA-targeting small-molecule discovery where 3D RNA structures are unavailable.