Leveraging 3D Molecular Spatial Visual Information and Multi-Perspective Representations for Drug Discovery

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Drug discovery is expensive and slow, with current computational methods often ignoring 3D molecular structures.
• Accurate identification of drug associations is crucial for developing new therapeutics.

Purpose of the Study:

• To develop a deep learning framework that leverages 3D molecular spatial information for enhanced drug discovery.
• To improve the prediction of drug-target interactions by integrating spatial and traditional molecular features.

Main Methods:

• A deep learning framework was developed to learn directly from 3D molecular spatial visual data.
• Geometric, topological, and stereochemical features were captured from spatial renderings.
• Unified multi-perspective molecular representations were created by combining spatial information with traditional descriptors.

Main Results:

• The model consistently outperformed conventional fingerprint-based methods in predicting drug-microRNA, drug-drug, and drug-protein interactions.
• Interpretability analyses revealed the model's focus on biologically relevant substructures.
• The value of 3D spatial information in molecular recognition was highlighted.

Conclusions:

• Spatially informed deep learning enhances predictive performance in computational drug discovery.
• This approach offers potential for improved therapeutic development and mechanistic insights.
• 3D molecular representations are critical for understanding molecular recognition and function.