AlzyFinder: A Machine-Learning-Driven Platform for Ligand-Based Virtual Screening and Network Pharmacology

Area of Science:

• Computational chemistry and pharmacology
• Bioinformatics and cheminformatics
• Neuroscience and drug discovery

Background:

• Alzheimer's disease (AD) drug development faces challenges due to its complex nature.
• Identifying effective multitarget ligands is crucial for AD therapeutics.
• Existing tools lack integrated virtual screening and network pharmacology for AD.

Purpose of the Study:

• To introduce AlzyFinder, a free, web-based platform for Alzheimer's disease research.
• To enable parallel ligand-based virtual screening and network pharmacology analysis.
• To accelerate the identification and analysis of potential multitarget ligands for AD.

Main Methods:

• Development of machine learning models (XGBoost, Optuna) for ligand-based virtual screening.
• Validation of models using active, inactive, and decoy molecules with metrics like balanced accuracy, precision, and F1 score.
• Integration of a soft-voting ensemble approach for refined classification and comprehensive analysis of interaction data.

Main Results:

• AlzyFinder successfully screened five known active compounds against key AD targets, predicting them as true positives (>0.70 probability).
• The platform demonstrated high precision in identifying potential therapeutic compounds.
• Network pharmacology analysis provided insights into systemic drug actions and interactions.

Conclusions:

• AlzyFinder is the first open-access platform combining virtual screening and network pharmacology for AD.
• The platform enhances visualization and analysis of drug-protein and protein-protein interactions.
• AlzyFinder serves as a valuable resource for accelerating Alzheimer's disease therapeutic development.