BiGAMi: Bi-Objective Genetic Algorithm Fitness Function for Feature Selection on Microbiome Datasets.

Area of Science:

• Microbiome research
• Bioinformatics
• Machine learning applications in biology

Background:

• The host-microbiome relationship is critical for health and disease.
• High-dimensional microbiome data presents significant analytical challenges for machine learning (ML) and deep learning (DL) models.
• Effective feature selection is essential for building accurate phenotype classifiers from complex microbial datasets.

Purpose of the Study:

• To introduce BiGAMi, a bi-objective genetic algorithm fitness function for feature selection in microbial datasets.
• To develop high-performing phenotype classifiers using significantly reduced feature sets.
• To demonstrate the efficacy of BiGAMi in identifying biologically relevant microbial features.

Main Methods:

• Development of BiGAMi, a bi-objective genetic algorithm fitness function.
• Application of BiGAMi for feature selection on microbial datasets.
• Training and evaluation of phenotype classifiers using selected features.
• Comparison of BiGAMi with existing feature selection methods (sequential forward selection, SelectKBest, GARS).

Main Results:

• Classifiers built with BiGAMi significantly outperformed baseline performance, utilizing only 0.04% to 2.32% of original features.
• BiGAMi selected 6-93% fewer features compared to other evaluated methods across 35 out of 42 comparisons.
• The selected microbial subsets were experimentally validated for their contribution to known diseases and host states.

Conclusions:

• BiGAMi is a highly effective feature selection tool for microbiome data analysis.
• This approach enables the identification of minimal, yet crucial, microbial signatures for disease classification.
• BiGAMi facilitates rapid discovery of relevant microbes for novel disease conditions.