MEvA-X: a hybrid multiobjective evolutionary tool using an XGBoost classifier for biomarkers discovery on biomedical...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Biomarker discovery is vital for precision medicine, disease prognosis, and drug discovery.
• Challenges include a low sample-to-feature ratio and class imbalance in datasets.
• Existing methods like XGBoost struggle with multi-objective optimization and class imbalance.

Purpose of the Study:

• Introduce MEvA-X, a hybrid ensemble for feature selection and classification.
• Address limitations of current methods in handling class imbalance and multiple objectives.
• Optimize XGBoost hyperparameters and perform feature selection simultaneously.

Main Methods:

• Developed MEvA-X, integrating a niche-based multiobjective evolutionary algorithm (EA) with XGBoost.
• Utilized multiobjective EA for hyperparameter optimization and feature selection.
• Identified Pareto-optimal solutions balancing classification and model simplicity.

Main Results:

• MEvA-X outperformed state-of-the-art methods on omics and clinical datasets.
• Achieved balanced class categorization and generated low-complexity models.
• Identified non-redundant biomarkers, including potential blood circulatory markers for precision nutrition.

Conclusions:

• MEvA-X offers a robust approach for biomarker discovery in bioinformatics.
• The tool effectively addresses class imbalance and multi-objective optimization challenges.
• Demonstrated utility in identifying relevant biomarkers for applications like precision nutrition.