A feature selection method based on multiple kernel learning with expression profiles of different types.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput technology generates large, complex expression datasets with numerous features and few samples.
• Effective and robust feature selection is critical for analyzing such high-dimensional biological data.
• Existing feature selection methods often lack comprehensive performance evaluation across diverse expression data types.

Purpose of the Study:

• To propose a hybrid feature selection method integrating Multiple Kernel Learning (MKL) and Support Vector Machine (SVM).
• To evaluate the proposed method's performance on various expression datasets, assessing its effectiveness and robustness.
• To compare the proposed method against existing techniques using multiple performance metrics beyond simple classification accuracy.

Main Methods:

• A hybrid feature selection approach combining Multiple Kernel Learning (MKL) and Support Vector Machine (SVM).
• Utilizing an MKL optimizing function to measure feature-sample relevance, refined through iterative gradient descent.
• Employing an embedded forward selection scheme to derive compact feature subsets from initially selected relevant features.

Main Results:

• The proposed MKL-based method demonstrates strong performance in feature selection across different expression datasets.
• Feature relevance is effectively quantified using MKL optimization, guiding the selection process.
• The embedded forward selection refines feature sets, yielding compact and informative subsets.

Conclusions:

• The developed hybrid method offers a satisfactory capability for feature selection in diverse expression data analysis.
• The approach provides robust performance evaluations, considering stability, similarity, and consistency alongside classification accuracy.
• This method advances the analysis of high-throughput expression data, enabling more reliable biological insights.