Drug Development.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Neuroscience and neurodegenerative diseases

Background:

• Beta-secretase 1 (BACE1) is a key enzyme in Alzheimer's disease (AD) pathogenesis, producing neurotoxic amyloid-beta peptides.
• Developing effective BACE1 inhibitors is critical for AD intervention strategies.
• This study presents a novel machine learning approach to identify BACE1 inhibitors.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting BACE1-ligand binding affinity (pIC50).
• To integrate BACE1-ligand interaction features with quantitative structure-activity relationship (QSAR) descriptors for enhanced prediction.
• To facilitate efficient virtual screening for potent BACE1 inhibitors.

Main Methods:

• Utilized 284 BACE1-ligand complexes from the Protein Data Bank, extracting binding interaction features via docking simulations.
• Extracted 881 QSAR features per ligand using the PubChem molecular fingerprint calculator.
• Developed 43 machine learning models using three input sets: binding features, top 12 QSAR features, and combined features, evaluating performance with R².

Main Results:

• Models integrating both binding and QSAR features showed superior prediction accuracy compared to individual feature sets.
• The NuSVR model achieved the highest prediction performance with R² = 0.78 using combined features.
• A fuzzy Gaussian regression model also demonstrated acceptable performance (R² = 0.642) using binding features and molecular weight.

Conclusions:

• Machine learning models combining BACE1-ligand interaction and QSAR features effectively predict binding affinity, enabling efficient virtual screening.
• This approach accelerates the identification of potent BACE1 inhibitors for Alzheimer's disease drug discovery.
• Future work will involve expanding these models to larger chemical databases to further advance AD intervention strategies.