A benchmarking study of feature screening approaches across type 1 diabetes omics studies classification settings

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• High-dimensional omics analyses are increasingly used to study complex biological systems.
• Machine learning (ML) is often employed to identify key biomolecules predictive of biological outcomes.
• Feature selection is critical in ML for omics data due to high dimensionality and limited, unbalanced sample sizes.

Purpose of the Study:

• To evaluate model-free, filter-based feature selection methods using the sure screening principle for omics data.
• To compare the performance and computational efficiency of various sure screening approaches.
• To identify the most effective sure screening method for ML classification in omics applications.

Main Methods:

• The study focused on filter-based feature selection methods grounded in the sure screening principle.
• A suite of model-free sure screening approaches was applied to several omics biomedical datasets.
• Performance was evaluated in a machine learning classification context, comparing runtime and effectiveness.

Main Results:

• BcorSIS was identified as the most effective and computationally efficient sure screening method across diverse omics datasets.
• BcorSIS consistently outperformed other methods, including CSIS and DCSIS, in terms of runtime.
• The study provides a comprehensive evaluation of sure screening for omics data analysis.

Conclusions:

• Sure screening methods offer analytical guarantees for feature set retention in omics data analysis.
• BcorSIS emerges as a highly recommended tool for feature selection in high-dimensional omics data due to its efficiency and effectiveness.
• This work contextualizes feature screening within broader feature selection strategies for omics research.