Recommender Systems in Antiviral Drug Discovery

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Recommender systems (RSs) are widely used in e-commerce.
• Their application in drug discovery, particularly for antiviral agents, is an emerging area.
• Predicting compound-virus interactions is crucial for identifying potential antiviral drugs.

Purpose of the Study:

• To investigate the utility of RS methods for predicting antiviral activity of compounds.
• To compare the performance of collaborative filtering and content-based filtering approaches.
• To assess the effectiveness of RS in predicting interactions for known and novel compounds.

Main Methods:

• Applied collaborative filtering (Surprise implementation) and content-based filtering (sparse-group inductive matrix completion - SGIMC).
• Utilized compound and virus interaction data from the ChEMBL database.
• Evaluated prediction accuracy using cross-validation and external validation with receiver operating characteristic (ROC) scores.

Main Results:

• Both RS approaches demonstrated good prediction quality for binary classification of antiviral activity.
• Achieved a receiver operating characteristic (ROC) score greater than 0.9 for individual interactions and compound profiles.
• Successfully predicted antiviral activity for both known and novel compounds.

Conclusions:

• Simple recommender systems can be effective tools in antiviral drug discovery.
• RS methods offer a valuable approach for predicting compound-virus interactions.
• The findings support the integration of RS into early-stage drug discovery pipelines.