META-BOA

META-BOA balances metabolomic and lipidomic datasets by applying over-sampling algorithms to correct class imbalance for machine learning analyses.

Key Features:

• Targeted Data Types: Operates on metabolomic and lipidomic datasets to address domain-specific class imbalance issues.
• Over-sampling Algorithms: Implements Synthetic Minority Over-sampling Technique (SMOTE), Borderline-SMOTE (BSMOTE), Adaptive Synthetic (ADASYN), and Random Over-Sampling Examples (ROSE) to generate additional minority-class samples.
• Visualization: Provides principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) visualizations both pre- and post-over-sampling to show distributional changes.
• Classification Evaluation: Uses random forest classification to compare classification performance on original and balanced datasets.
• Model Performance Assessment: Enables comparison of how different over-sampling methods affect downstream machine learning model performance.

Scientific Applications:

• Class Imbalance Correction: Corrects unequal class representation in metabolomics and lipidomics datasets to reduce bias in analyses.
• Supervised Learning Improvement: Supports improvement and validation of supervised classification models via balanced training data.
• Method Comparison: Facilitates comparative evaluation of SMOTE, BSMOTE, ADASYN, and ROSE on biological datasets.
• Exploratory Data Analysis: Visualizes high-dimensional metabolomic and lipidomic sample distributions before and after balancing.

Methodology:

Applies SMOTE, Borderline-SMOTE (BSMOTE), ADASYN, and ROSE for over-sampling; generates PCA and t-SNE visualizations pre- and post-over-sampling; and performs random forest classification to evaluate performance on original versus balanced datasets.