HCAR1 antagonist screening based on boundary-selected negative sampling strategy and multi-level graph neural network

Area of Science:

• Oncology
• Pharmacology
• Artificial Intelligence in Drug Discovery

Background:

• Hydroxycarboxylic acid receptor 1 (HCAR1), or lactate receptor, is implicated in cancer progression and is a target for cancer treatment.
• Traditional drug screening for HCAR1 antagonists is hindered by imbalanced data and incomplete molecular representations, limiting therapeutic options.

Purpose of the Study:

• To develop an accurate prediction model for HCAR1 antagonists to aid tumor immunotherapy.
• To overcome limitations of traditional drug screening methods for identifying potential HCAR1-targeted cancer therapeutics.

Main Methods:

• Constructed a balanced HCAR1 target activity dataset using negative sampling.
• Developed a Multi-GNN model integrating fingerprints, molecular graphs, and fragment features for activity prediction.
• Screened millions of compounds using the Multi-GNN model, physicochemical filtering, and molecular docking.

Main Results:

• The Multi-GNN model outperformed eight state-of-the-art methods in predicting HCAR1 target activity.
• Identified five candidate compounds from screening approximately ten million molecules.
• Discovered a promising HCAR1 inhibitor with an IC50 of 22.39 μM through in vitro assays.

Conclusions:

• Introduced a novel AI-based framework for HCAR1-targeted drug discovery.
• Highlighted potential lead compounds for further development in cancer therapy.
• Demonstrated the efficacy of the Multi-GNN approach in identifying novel drug candidates.