Comparison of commercially available genetic algorithms: gas as variable selection tool.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Variable selection is crucial for building accurate Quantitative Structure-Activity Relationship (QSAR) models.
• Genetic algorithms (GAs) are popular evolutionary computation techniques for identifying relevant variables.
• Numerous GA software tools exist, but their comparative performance in QSAR is not well-established.

Purpose of the Study:

• To compare the performance of different commercially available genetic algorithm tools for variable selection in QSAR modeling.
• To determine the most effective GA method for identifying significant molecular descriptors.
• To assess the utility of GA-selected variables in developing robust QSAR models.

Main Methods:

• Comparison of multiple genetic algorithm tools (GFA, QuaSAR-Evolution, Partek's GA).
• Application of genetic algorithms for variable selection on proprietary and Selwood datasets.
• Generation of stepwise multiple linear regression models using selected variables.

Main Results:

• Genetic algorithms successfully identified relevant variables for QSAR model development.
• The selected variables enabled the generation of predictive QSAR models for both proprietary and public datasets.
• Demonstrated the practical utility of different GA tools in cheminformatics workflows.

Conclusions:

• Genetic algorithms are effective tools for variable selection in QSAR.
• The comparative study provides insights into choosing appropriate GA software for cheminformatics tasks.
• Successful QSAR model generation confirms the value of GA-based variable selection.