This dataset supports SMARTBind (Small Molecule Approaches to RNA Targeting Binder Discovery), a structure-agnostic ligand discovery framework that combines an RNA large language model with contrastive learning and a ligand-specific decoy enhancement strategy. Please cite the following publication when using the dataset: Jiang, Shiyu, Amirhossein Taghavi, Tenghui Wang, Samantha M. Meyer, Jessica L. Childs-Disney, Chenglong Li, Mattew D. Disney, and Yanjun Li. "Small Molecule Approach to RNA Targeting Binder Discovery (SMARTBind) Using Deep Learning Without Structural Input." bioRxiv (2025): 2025-09. Overview The dataset contains model checkpoint and training data for SMARTBind including RNAmigos1 10-fold random-split, Hariboss 10-fold random-split, and Hariboss 5-fold sequence-based-split. All data is organized under the archive file SMARTBind_dataset.zip. Contents SMARTBind_weights.zip: Saved checkpoint of 10-fold SMARTBind model. hariboss_merged_5fd.pkl: SMARTBind training data from the HARIBOSS database under 5-fold sequence-based-split cross-validation. hariboss_merged_10fd.pkl: SMARTBind training data from the HARIBOSS database under 10-fold random-split cross-validation. rnamigos_10fd.pkl: SMARTBind training data from the RNAmigos1 under 10-fold random-split cross-validation. Decoy library.smi: a chemical diverse decoy library with 92,626 entries that is curated for the ligand-specific decoy enhancement strategy.