Multi-objective genetic algorithm with maximum-entropy classification and domain-specific evolutionary operators for microorganism selection on microbiomes

The field of microbiome research has experienced exponential growth in recent years, driven by advancements in next-generation sequencing technologies. A particular topic of interest in this field lies in studying the influence of the microbiome on the host organism’s state, with the aim of boosting the potential use of microorganisms as indicators of phenotypic conditions. However, microorganism selection is a hard-to-tackle problem involving multiple optimality criteria. This work introduces MADMISM, a Multi-objective Approach based on Dominance for MIcroorganism Selection on Microbiomes. Based on the Non-dominated Sorting Genetic Algorithm III, the proposed approach includes maximum-entropy classification with stochastic gradient descent learning and domain-specific operators through all evolutionary stages. Hence, MADMISM combines the search power of evolutionary algorithms with the enhanced guidance provided by the domain-specific operators. The evaluation of results on eight real microbiome datasets showcases MADMISM’s superiority in providing solutions with a minimal set of Operational Taxonomic Units (OTUs) while achieving high classification performance, with statistically significant improvements over state-of-the-art methods in most scenarios. Specifically, MADMISM accomplishes an average OTU reduction of 46.35 % and a classification improvement of 12.05 % compared to state-of-the-art methods, while also showcasing remarkable scalability. Furthermore, MADMISM exhibited superior multi-objective performance with an average hypervolume of 90.06 %, and presented statistically significant improvements over well-established multi-objective algorithms. Therefore, MADMISM represents an effective approach for identifying microbial markers. This is evidenced by the quality of its results and the biological relevance of the selected microorganisms. Materials for this work can be found at https://github.com/frodbej/madmism . • Novel multi-objective approach for microorganism selection on microbiomes. • Definition of domain-specific operators for accurate microorganism selection. • In-depth experimental evaluation on eight real microbiome datasets. • Improvements over the state of the art in terms of AUC and number of OTUs. • Analysis of the biological relevance of the identified microorganism sets.