Naïve Bayes classification using 2D pharmacophore feature triplet vectors

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Predicting molecular activity is crucial for efficient drug discovery.
• Existing methods may lack accuracy or scalability.
• Feature representation significantly impacts predictive model performance.

Purpose of the Study:

• To develop and validate a novel computational approach for predicting molecular activity against biological targets.
• To assess the efficacy of 2D pharmacophore feature triplet vectors in describing molecular properties for classification.
• To evaluate the performance of a naïve Bayes classifier built upon these feature vectors.

Main Methods:

• Utilized 2D pharmacophore feature triplet vectors to represent molecular descriptors.
• Developed a naïve Bayes classifier based on these feature vectors.
• Performed retrospective validation using diverse chemical libraries and known active compounds.
• Evaluated classifier performance using enrichment curves, enrichment factors, and BEDROC metrics.

Main Results:

• The naïve Bayes classifier, using feature triplet vectors, successfully predicted molecular activity.
• Significant enrichments were observed across various test sets in retrospective validation experiments.
• The method proved effective in identifying potential active molecules from large chemical libraries.

Conclusions:

• 2D pharmacophore feature triplet vectors provide a robust molecular representation for predictive modeling.
• Naïve Bayes classifiers built on this representation are effective for virtual screening and hit identification.
• This approach offers a valuable tool for accelerating early-stage drug discovery efforts.