Determining best practices for using genetic algorithms in molecular discovery.

Area of Science:

• Computational chemistry
• Cheminformatics
• Bioinformatics

Background:

• Genetic algorithms (GAs) are widely used for inverse molecular design to navigate vast chemical spaces.
• The performance of GAs is sensitive to hyperparameter choices, which require systematic investigation for chemical search applications.

Purpose of the Study:

• To investigate the impact of key genetic algorithm hyperparameters on performance metrics relevant to molecular design.
• To identify an optimal set of hyperparameters for general-purpose molecular design GAs.

Main Methods:

• Systematic examination of hyperparameters including population size, elitism rate, selection method, mutation rate, and convergence criteria.
• Evaluation of GA performance based on metrics such as champion identification, elite target coverage, and relative speedup.
• Utilized a self-termination method based on Spearman's rank correlation coefficient for convergence assessment.

Main Results:

• A population size of 32, 50% elitism rate, three-way tournament selection, and a 40% mutation rate were found to be effective.
• A self-termination criterion requiring a Spearman's rank correlation coefficient of at least 0.8 for 50 consecutive generations demonstrated optimal convergence.
• These settings achieved a favorable balance between identifying superior molecular candidates and maintaining search efficiency.

Conclusions:

• The identified hyperparameter set offers a robust configuration for genetic algorithms in molecular design tasks.
• This research provides practical guidance for optimizing GA performance in chemical space exploration.
• The findings contribute to more efficient and effective computational drug discovery and materials science workflows.