A Machine Learning Method for RNA-Small Molecule Binding Preference Prediction.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• RNA-small molecule interactions are vital for biological processes.
• Current methods for identifying RNA-targeting molecules often overlook small molecule characteristics, limiting predictive accuracy.
• Developing effective RNA-targeting inhibitors requires understanding these complex interactions.

Purpose of the Study:

• To develop a novel machine learning model for predicting RNA-small molecule binding preferences.
• To improve the accuracy of identifying potential RNA-targeting molecules, especially for novel compounds.
• To enhance the design of RNA-related drugs and facilitate RNA-based studies.

Main Methods:

• Developed ZHMol-RLinter, a double-layer stacking machine learning model.
• Integrated RNA sequence, secondary structure, and spatial conformation features.
• Incorporated small molecule structural geometric and physicochemical environment information.
• Trained and validated the model on established and novel small molecule datasets.

Main Results:

• ZHMol-RLinter achieved a 90.8% success rate on the RL98 testing set, outperforming existing methods.
• Demonstrated a 77.1% success rate on the UNK96 unknown small molecule testing set, indicating strong generalization.
• Structural evaluations confirmed the model's reliability and accuracy in predicting RNA-small molecule binding.

Conclusions:

• ZHMol-RLinter offers a significant advancement in predicting RNA-small molecule interactions.
• The model's ability to integrate diverse features improves prediction accuracy for both known and unknown molecules.
• This approach holds promise for accelerating the development of RNA-targeting drugs and understanding biological mechanisms.