Supervised feature ranking using a genetic algorithm optimized artificial neural network.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Feature selection is crucial for developing predictive models in drug discovery.
• Existing methods like support vector machines (SVM) may not always provide clear feature rankings.

Purpose of the Study:

• To design and evaluate a genetic algorithm optimized artificial neural network (GNW) for feature ranking.
• To compare the performance of GNW against SVM for feature selection in multivariate datasets.

Main Methods:

• A genetic algorithm was used to optimize an artificial neural network (GNW) by encoding weights as chromosomes.
• Features were ranked using a primary index based on self-depleted weights and weight adjustments.
• GNW performance was compared with SVM using F-scores on artificial and real-world datasets.

Main Results:

• The GNW demonstrated superior performance over SVM in feature ranking for artificial and matrix metalloproteinase-1 inhibitor datasets.
• GNW provided a clear separation between relevant and irrelevant features, unlike SVM.
• The proposed feature ranking index effectively distinguished between good and bad features.

Conclusions:

• The GNW is a powerful tool for feature selection in cheminformatics and drug discovery.
• GNW offers improved feature ranking capabilities compared to traditional SVM methods.
• This approach enhances the interpretability of complex datasets for identifying key molecular descriptors.