A multiple kernel support vector machine scheme for feature selection and rule extraction from gene expression data...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Medicine

Background:

• Gene expression profiling via microarray is vital for cancer diagnosis and treatment.
• High noise and numerous genes in expression data pose challenges for statistical and machine learning methods.
• Support Vector Machines (SVM) are effective for cancer tissue classification but lack transparency.

Purpose of the Study:

• To develop a more explainable SVM for cancer gene expression data analysis.
• To improve the transparency and interpretability of machine learning models in medical diagnostics.

Main Methods:

• A multiple kernel support vector machine (MK-SVM) framework integrating feature selection, rule extraction, and prediction modeling.
• Utilizing a 1-norm based linear programming shrinkage approach for sparse parameter selection and feature identification.
• Implementing a novel rule extraction method based on hyperplane and support vectors for enhanced generalization and comprehensibility.

Main Results:

• MK-SVM demonstrated high classification accuracy (>90%) on public leukemia and colon tumor gene expression datasets.
• The method effectively reduced the number of genes analyzed, focusing on the most informative ones.
• Extracted rules were simple, interpretable with linguistic labels, and possessed high diagnostic power.

Conclusions:

• The proposed MK-SVM approach offers a powerful and interpretable tool for cancer gene expression data analysis.
• This method enhances the diagnostic capabilities of SVM by providing transparent and accurate classification.
• The extracted rules contribute to better understanding and potentially improved clinical decision-making in oncology.