AGAPE (Computational G‑Quadruplex Stabilization Prediction): The First Machine Learning Workflow for G‑Quadruplex...

Area of Science:

• Medicinal Chemistry
• Computational Biology
• Cheminformatics

Background:

• G-quadruplexes (G4s) are crucial nucleic acid structures found in telomeres and oncogene promoters.
• G4s represent promising therapeutic targets for various diseases, particularly cancer.
• Designing selective small molecule binders for G4s remains a significant challenge in drug discovery.

Purpose of the Study:

• To develop a novel machine learning (ML)-based tool, AGAPE, for predicting the G4 stabilizing potential of small molecules.
• To integrate diverse molecular descriptors, including quantum chemical features, for enhanced prediction accuracy.
• To provide researchers with an accessible and interpretable platform for identifying potential G4-targeting compounds.

Main Methods:

• Utilized a curated dataset of 1217 compounds with FRET melting assay data for training.
• Integrated 5666 molecular descriptors, encompassing classical and quantum chemical features.
• Employed machine learning algorithms, with XGBoost selected as the best-performing model based on feature selection and performance metrics.

Main Results:

• The AGAPE tool, utilizing XGBoost with 489 selected features, achieved a prediction accuracy of nearly 91%.
• SHAP analysis identified key molecular descriptors, including those related to topology, polarizability, and electrostatics, that drive G4 stabilization prediction.
• The developed tool successfully integrates quantum chemical information within a machine learning framework.

Conclusions:

• AGAPE offers a robust and interpretable computational solution for predicting G4 stabilization by small molecules.
• The tool facilitates accelerated discovery of novel G4-binding ligands, including organic molecules and metal complexes.
• AGAPE is accessible via a user-friendly web interface, supporting batch predictions and secure data handling.