Assessment of a rule-based virtual screening technology (INDDEx) on a benchmark data set.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Drug discovery relies on identifying active compounds and guiding development.
• Virtual screening is crucial for efficient identification of potential drug candidates.
• Machine learning offers novel approaches to analyze molecular data.

Purpose of the Study:

• To introduce the Investigational Novel Drug Discovery by Example (INDDEx) package.
• To demonstrate INDDEx's capability in linking compound activity to chemical substructures.
• To showcase INDDEx's utility in guiding drug development and virtual screening.

Main Methods:

• INDDEx employs a machine-learning technique to create logical rules from active molecule substructures.
• These rules are weighted to form a quantitative model for database screening.
• The method was tested for its learning capacity from small datasets and scaffold-hopping potential.

Main Results:

• INDDEx achieved high retrieval rates in virtual screening, even when learning from few compounds.
• Average enrichment factors were significantly high, particularly at lower data percentages (e.g., 492 at 0.1% with 2 ligands).
• Performance was competitive when compared to other established methods like eHiTS LASSO, PharmaGist, and DOCK.

Conclusions:

• INDDEx is an effective tool for identifying active compounds and supporting drug discovery pipelines.
• Its machine-learning approach provides high performance in virtual screening and scaffold-hopping.
• INDDEx demonstrates utility in drug development by learning from limited active compound data.