DRABAL: novel method to mine large high-throughput screening assays using Bayesian active learning.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• High-throughput screening (HTS) assays are crucial for drug discovery and repositioning.
• Extracting actionable insights from HTS data presents significant challenges.
• Existing virtual screening methods have limitations in fully addressing these challenges.

Purpose of the Study:

• To develop a novel multi-label classification (MLC) technique for modeling correlations between HTS assays.
• To enhance the accuracy of predicting compound interactions in assays where they haven't been tested.
• To improve decision-making in drug discovery and repositioning.

Main Methods:

• Implemented a multi-label classification (MLC) approach using a Bayesian network structure learning.
• Developed DRABAL, a novel MLC solution incorporating Bayesian active learning.
• Processed over 1.4 million interactions from 400,000 compounds across five large HTS assays.

Main Results:

• DRABAL significantly improves the F1 Score by approximately 22% compared to other MLC methods.
• Identified potential drug repositioning candidates, including Thiabendazole for Niemann-Pick type C disease.
• Demonstrated utility in screening FDA-approved drugs for multi-target interactions.

Conclusions:

• A novel MLC solution, DRABAL, was developed using a Bayesian active learning framework.
• DRABAL effectively models dependencies between HTS assays, overcoming limitations of incomplete training data.
• The method enhances prediction performance and facilitates drug multi-target repositioning.