AlzyFinder: A Machine-Learning-Driven\nPlatform for\nLigand-Based Virtual Screening and Network Pharmacology

Alzheimer’s disease (AD), a prevalent neurodegenerative\ndisorder, presents significant challenges in drug development due\nto its multifactorial nature and complex pathophysiology. The AlzyFinder\nPlatform, introduced in this study, addresses these challenges by\nproviding a comprehensive, free web-based tool for parallel ligand-based\nvirtual screening and network pharmacology, specifically targeting\nover 85 key proteins implicated in AD. This innovative approach is\ndesigned to enhance the identification and analysis of potential multitarget\nligands, thereby accelerating the development of effective therapeutic\nstrategies against AD. AlzyFinder Platform incorporates machine learning\nmodels to facilitate the ligand-based virtual screening process. These\nmodels, built with the XGBoost algorithm and optimized through Optuna,\nwere meticulously trained and validated using robust methodologies\nto ensure high predictive accuracy. Validation included extensive\ntesting with active, inactive, and decoy molecules, demonstrating\nthe platform’s efficacy in distinguishing active compounds.\nThe models are evaluated based on balanced accuracy, precision, and\nF1 score metrics. A unique soft-voting ensemble approach is utilized\nto refine the classification process, integrating the strengths of\nindividual models. This methodological framework enables a comprehensive\nanalysis of interaction data, which is presented in multiple formats\nsuch as tables, heat maps, and interactive Ligand–Protein Interaction\nnetworks, thus enhancing the visualization and analysis of drug–protein\ninteractions. AlzyFinder was applied to screen five molecules recently\nreported (and not used to train or validate the ML models) as active\ncompounds against five key AD targets. The platform demonstrated its\nefficacy by accurately predicting all five molecules as true positives\nwith a probability greater than 0.70. This result underscores the\nplatform’s capability in identifying potential therapeutic\ncompounds with high precision. In conclusion, AlzyFinder’s\ninnovative approach extends beyond traditional virtual screening by\nincorporating network pharmacology analysis, thus providing insights\ninto the systemic actions of drug candidates. This feature allows\nfor the exploration of ligand–protein and protein–protein\ninteractions and their extensions, offering a comprehensive view of\npotential therapeutic impacts. As the first open-access platform of\nits kind, AlzyFinder stands as a valuable resource for the AD research\ncommunity, available at http://www.alzyfinder-platform.udec.cl with supporting data and scripts accessible via GitHub https://github.com/ramirezlab/AlzyFinder.