A Machine Learning Method for RNA–Small Molecule\nBinding Preference Prediction

The\ninteraction between RNA and small molecules is crucial in various\nbiological functions. Identifying molecules targeting RNA is essential\nfor the inhibitor design and RNA-related studies. However, traditional\nmethods focus on learning RNA sequence and secondary structure features\nand neglect small molecule characteristics, and resulting in poor\nperformance on unknown small molecule testing. To overcome this limitation,\nwe developed a double-layer stacking-based machine learning model\ncalled ZHMol-RLinter. This approach more effectively predicts RNA–small\nmolecule binding preferences by learning RNA and small molecule features\nto capture their interaction information. ZHMol-RLinter also combines\nsequence and secondary structural features with structural geometric\nand physicochemical environment information to capture the specificity\nof RNA spatial conformations in recognizing small molecules. Our results\ndemonstrate that ZHMol-RLinter has a success rate of 90.8% on the\npublished RL98 testing set, representing a significant improvement\nover existing methods. Additionally, ZHMol-RLinter achieved a success\nrate of 77.1% on the unknown small molecule UNK96 testing set, showing\nsubstantial improvement over the existing methods. The evaluation\nof predicted structures confirms that ZHMol-RLinter is reliable and\naccurate for predicting RNA–small molecule binding preferences,\neven for challenging unknown small molecule testing. Predicting RNA–small\nmolecule binding preferences can help in the understanding of RNA–small\nmolecule interactions and promote the design of RNA-related drugs\nfor biological and medical applications.