Predicting small molecule-RNA interactions without RNA tertiary structures.

Area of Science:

• Computational biology
• Drug discovery
• Molecular biology

Background:

• Small molecules regulating RNA (RNA-binding small molecules) offer therapeutic potential.
• Current prediction tools for small molecule-RNA interactions (SRIs) require RNA tertiary structures, limiting their applicability.

Purpose of the Study:

• To develop a deep learning method (SMRTnet) for predicting SRIs based on RNA secondary structures.
• To overcome the limitations of existing SRI prediction tools by eliminating the need for RNA tertiary structures.

Main Methods:

• SMRTnet integrates multimodal data using two large language models, convolutional neural networks, and graph attention networks.
• The method leverages RNA secondary structure information for SRI prediction.

Main Results:

• SMRTnet demonstrated high performance across multiple experimental benchmarks, outperforming existing tools.
• Predictions for disease-associated RNA targets identified 40 potential RNA-targeting small molecules.
• In silico screening of the MYC internal ribosome entry site identified molecules with binding scores correlating with validation rates.
• One validated small molecule downregulated MYC expression, inhibited cancer cell proliferation, and promoted apoptosis.

Conclusions:

• SMRTnet effectively predicts small molecule-RNA interactions using RNA secondary structures, expanding the range of druggable RNA targets.
• The method accelerates the discovery of novel RNA-targeting therapeutics by removing the requirement for complex RNA tertiary structure data.