Supervised evolution: research concerning the number of evolutions that occur under certain constraints

Area of Science:

• Computational chemistry
• Cheminformatics
• Evolutionary computation

Background:

• Evolutionary algorithms are used to model complex chemical phenomena.
• Understanding evolutionary dynamics is crucial for optimizing predictive models.
• Polychlorinated biphenyls (PCBs) present complex partitioning behavior relevant to environmental and toxicological studies.

Purpose of the Study:

• To analyze the number of evolutions in a genetic algorithm experiment.
• To investigate the distribution of evolutions across nine different evolution strategies.
• To determine the suitability of statistical distributions for modeling evolutionary processes in cheminformatics.

Main Methods:

• A genetic algorithm was employed to study the octanol/water partition coefficient of PCBs.
• The number of evolutions (objective function score improvements) was recorded over 20,000 generations from 46 independent runs.
• Distribution analysis was performed for each of the nine implemented evolution strategies.

Main Results:

• The Weibull distribution was found to provide a good fit for the number of evolutions across all nine strategies at a 5% significance level.
• The Weibull distribution model remained robust even when analyzing merged samples from different runs and strategies.
• This indicates a consistent pattern in evolutionary progress regardless of the specific strategy employed.

Conclusions:

• The Weibull distribution is a suitable statistical model for describing the number of evolutions in genetic algorithm-based cheminformatics studies.
• This finding aids in predicting and understanding the evolutionary trajectory of predictive models for chemical properties.
• Consistent distributional patterns suggest underlying principles governing evolutionary optimization in this domain.