Improving Omics-Based Classification: The Role of Feature Selection and Synthetic Data Generation.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Omics datasets present challenges in classification due to high dimensionality and limited samples.
• Current models often lack interpretability, hindering trust and reproducibility in clinical applications.

Purpose of the Study:

• To develop a machine learning framework integrating feature selection and data augmentation for improved omics-based classification.
• To enhance model transparency and reliability in disease classification using omics data.

Main Methods:

• A machine learning pipeline combining feature selection with data augmentation techniques.
• Bootstrap analysis on the E-MTAB-8026 dataset across six binary classification scenarios.
• Evaluation of cross-validated performance and generalization to larger test sets.

Main Results:

• The proposed framework achieves high classification accuracy and improved interpretability.
• Cross-validated performance on small datasets was maintained when applied to larger test sets.
• Synthetic data augmentation positively impacted model generalization, especially with limited sample availability.

Conclusions:

• The framework offers a balance between accuracy and feature selection for reliable omics-based classification.
• Data augmentation is crucial for enhancing generalization in omics studies with scarce data.
• This approach supports the development of explainable, reproducible diagnostic tools for clinical decision-making.