A Machine Learning-Guided Approach for Identifying Potential HCAR1 Antagonists in Lactate-Driven Cancers.

Area of Science:

• Pharmacology
• Computational Chemistry
• Machine Learning

Background:

• GPR81 (HCAR1), a lactate-sensing receptor, is implicated in cancer progression and therapeutic resistance.
• Selective antagonists for HCAR1 are currently lacking, representing a significant unmet need in oncology.

Purpose of the Study:

• To develop a statistically validated Support Vector Machine (SVM) model for classifying HCAR1 ligands as agonists or antagonists.
• To identify key molecular substructures driving HCAR1 antagonism and receptor selectivity.
• To screen compound libraries for potential HCAR1 antagonists.

Main Methods:

• Training an SVM model on 144 known HCAR1 ligands using physicochemical descriptors and ECFP4 fingerprints.
• Employing molecular docking against active and inactive receptor conformations to calculate ΔAffinity scores.
• Utilizing SHAP analysis for feature interpretation and model explainability.
• Screening 3,377 compounds from diverse libraries.

Main Results:

• The SVM model achieved 79.3% accuracy on the test set, with an AUC of 0.94, indicating robust predictive performance.
• SHAP analysis identified polar, rigid, and aromatic substructures as critical for HCAR1 antagonism and selectivity.
• Screening identified Ketanserin, Cryptopyranmoscatone A1 diacetate, and Cefuroxime as potential HCAR1 antagonists, with two being FDA-approved drugs.

Conclusions:

• The study presents a validated computational framework for discovering selective HCAR1 antagonists.
• The identified structural determinants provide valuable insights for fragment-based and bioisosteric design of novel cancer therapeutics.
• This approach can accelerate the development of drugs targeting lactate-driven cancers.