Genetic algorithms and self-organizing maps: a powerful combination for modeling complex QSAR and QSPR problems

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Modeling complex descriptor-target relationships is crucial for designing biologically active molecules.
• Large numbers of descriptors in quantitative structure-activity relationship (QSAR) and quantitative structure-property relationship (QSPR) analyses can lead to spurious correlations and interpretation challenges.

Purpose of the Study:

• To develop a novel method for modeling non-linear descriptor-target relationships with reduced descriptor complexity.
• To enhance the interpretability and efficiency of QSAR/QSPR models in molecular design.

Main Methods:

• Coupling the supervised self-organizing map (SOM) with a genetic algorithm (GA) for descriptor selection.
• Utilizing GA for efficient reduction of descriptors to a manageable set.
• Performing feasibility studies on six diverse biological datasets with external validation.

Main Results:

• The combined GA-SOM method achieved comparable accuracy to supervised SOM alone but with significantly fewer descriptors.
• Models generated by the GA-SOM approach demonstrated consistently better performance than partial least squares (PLS) models.
• Validation confirmed the predictive quality of the models, addressing variability from dataset partitioning and method stochasticity.

Conclusions:

• The integration of genetic algorithms with supervised self-organizing maps offers a powerful tool for QSAR/QSPR modeling.
• This combined approach enhances model efficiency and interpretability in the design of biologically active molecules.
• The method shows significant potential for advancing quantitative structure-activity/property relationship studies.