Automated pharmacophore query optimization with genetic algorithms - a case study using the MC4R system.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Virtual screening, particularly pharmacophore-based virtual screening (PBVS), is crucial for identifying drug candidates.
• The accuracy of PBVS heavily depends on the quality of the pharmacophore query.
• Existing methods for deriving pharmacophore queries from single structures can lack specificity, leading to suboptimal performance.

Purpose of the Study:

• To develop and evaluate a genetic algorithm (GA)-powered optimization approach for enhancing pharmacophore query accuracy and specificity.
• To improve the identification of biologically relevant molecules in drug discovery campaigns.

Main Methods:

• Development of a GA-based method to optimize pharmacophore queries.
• Application of the method to the human melanocortin type 4 receptor (hMC4R) using a known rigid cyclic peptide agonist structure.
• Validation using training and testing datasets of hMC4R agonists and nonagonists.

Main Results:

• The optimized hMC4R pharmacophore query identified 37 true positive agonists with zero false positives from a training set (55 agonists, 51 nonagonists), a significant improvement over the initial query (37/32 hit rate).
• In a testing set (55 agonists, 50 nonagonists), the optimized query achieved a 33/8 hit rate, outperforming the initial query's 40/31 hit rate.
• The study analyzed the impact of various GA parameters (mutation rate, crossover rate, population size, etc.) on optimization performance.

Conclusions:

• Genetic algorithm optimization substantially enhances the accuracy and specificity of pharmacophore queries for virtual screening.
• This approach offers a powerful strategy for improving hit rates and reducing false positives in drug discovery, exemplified by the hMC4R case study.
• The findings underscore the utility of GA-driven optimization for refining molecular recognition models.