Consensus holistic virtual screening for drug discovery: a novel machine learning model approach.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Virtual screening is essential for identifying hit compounds in drug discovery.
• Existing methods often have limitations in accurately prioritizing potential drug candidates.

Purpose of the Study:

• To develop and validate a novel virtual screening pipeline integrating multiple established methods.
• To introduce a new metric, "w_new", for ranking machine learning models in virtual screening.
• To enhance the accuracy and efficiency of hit compound identification.

Main Methods:

• A pipeline was developed combining Quantitative Structure-Activity Relationship (QSAR), Pharmacophore, docking, and 2D shape similarity scoring.
• Machine learning models were employed and ranked using a novel "w_new" formula.
• Consensus scoring integrated results from individual methods.
• Enrichment studies and external validation were performed for various protein targets.

Main Results:

• The consensus scoring approach outperformed individual methods for specific targets like PPARG (AUC 0.90) and DPP4 (AUC 0.84).
• The pipeline consistently prioritized compounds with higher experimental PIC50 values.
• External validation demonstrated moderate to high performance with good R² values.

Conclusions:

• The novel consensus scoring workflow significantly improves hit compound identification in drug discovery.
• The integration of diverse screening methods and the "w_new" metric offer a robust approach to virtual screening.
• This holistic strategy enhances the reliability of identifying optimal virtual screening methodologies.