Graph Neural Networks Model Based on Atomic Hybridization for Predicting Drug Targets

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Quantitative structure-activity relationship (QSAR) models are crucial for drug discovery but often struggle to integrate local structural patterns and global physicochemical properties.
• Existing QSAR models have limitations in capturing the complex interplay of molecular features influencing bioactivity.

Purpose of the Study:

• To develop a hybrid deep learning framework that improves the accuracy and interpretability of predicting half-maximal inhibitory concentration (IC50) values.
• To address the limitations of traditional QSAR models by integrating graph neural networks with explicit molecular descriptors.

Main Methods:

• Developed a hybrid deep learning framework combining graph neural networks (GNNs) with explicit molecular descriptors.
• The model processes molecular graphs with atomic and bond features, alongside interpretable physicochemical properties and structural fingerprints.
• Trained and validated the model on a dataset of 14,316 compounds across nine diverse biological targets (kinases, nuclear receptors, proteases).

Main Results:

• Achieved an overall test R-squared (R²) of 0.87, demonstrating high predictive accuracy.
• Outperformed previously reported methods by 6-42% across various biological targets.
• Showcased robust generalization with comparable training and test performance, indicating reliability.

Conclusions:

• The hybrid framework synergistically combines data-driven learning with domain knowledge for superior structure-activity modeling.
• Offers enhanced accuracy and interpretability, facilitating efficient compound prioritization and optimization in early-stage drug discovery.
• Represents a significant advancement in computational approaches for accelerating the drug discovery pipeline.