MEvA-X

MEvA-X applies multiobjective evolutionary algorithms combined with extreme gradient boosting (XGBoost) to perform simultaneous hyperparameter optimization and feature selection for biomarker discovery from omics and clinical datasets.

Key Features:

• Hybrid Approach: Combines multiobjective evolutionary algorithms and extreme gradient boosting (XGBoost) to optimize hyperparameters and feature selection simultaneously.
• Multiobjective Optimization: Balances conflicting objectives such as maximizing classification accuracy and minimizing model complexity to select predictive, non-redundant biomarkers.
• Pareto-Optimal Solutions: Identifies multiple Pareto-optimal solutions to provide a set of models that trade off performance metrics.
• Class Imbalance Handling: Designed to address class imbalance in biomarker discovery tasks, improving balanced classification relative to single-objective XGBoost optimization.

Scientific Applications:

• Microarray gene expression and omics analysis: Applied to omics data from microarray gene expression experiments for feature selection and classification.
• Clinical datasets with demographics: Applied to clinical datasets that include demographic information for biomarker identification.
• Precision medicine: Produces low-complexity models applicable to disease prognosis and drug discovery in precision medicine contexts.
• Biomarker discovery for weight loss prediction: Identified blood circulatory markers from gene expression data predictive of weight loss, with further validation required for clinical relevance.

Methodology:

Deploys a multiobjective evolutionary algorithm to optimize XGBoost hyperparameters and feature selection; identifies important biomarkers while maintaining model simplicity; benchmarked against state-of-the-art methods, demonstrating superior balanced classification capabilities.