Ensemble feature selection for stable biomarker identification and cancer classification from microarray expression...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray technology allows simultaneous measurement of tens of thousands of gene expressions, crucial for molecular-level cancer and tumor studies.
• Microarray data often presents challenges with small sample sizes and high dimensionality, making accurate and stable feature selection vital for diagnostics and understanding disease mechanisms.

Purpose of the Study:

• To develop an ensemble feature selection framework to enhance the discrimination and stability of selected features from high-dimensional microarray data.
• To improve the diagnostic accuracy and understanding of disease mechanisms by refining feature selection processes.

Main Methods:

• Utilized sampling techniques to generate multiple subsets from the original microarray dataset.
• Employed base feature selectors on each sampled dataset to identify feature subsets.
• Developed two aggregation strategies to combine multiple feature subsets into a single, robust set.

Main Results:

• The proposed ensemble framework demonstrated superior stability scores compared to existing methods.
• Achieved classification performance comparable to, and in some cases better than, competing methods across various microarray datasets.
• Validated on four publicly available microarray datasets, including both binary and multi-class scenarios.

Conclusions:

• The ensemble feature selection framework offers improved stability and robust classification performance for high-dimensional microarray data.
• This approach is effective for both binary and multi-class classification tasks in cancer and tumor molecular studies.
• The findings contribute to more accurate diagnostics and a deeper understanding of complex disease mechanisms through advanced bioinformatics techniques.