SGL-SVM

SGL-SVM applies Sparse Group Lasso and Support Vector Machine methods to select informative genes from high-dimensional gene expression data and classify tumor samples.

Key Features:

• Sparse Gene Selection: Initial feature genes are selected from normalized tumor datasets using the Kruskal-Wallis rank sum test.
• Sparse Group Lasso Refinement: Sparse Group Lasso is applied to refine feature selection by combining group and individual penalties on regression coefficients to enforce group- and within-group sparsity.
• Support Vector Machine Classification: A Support Vector Machine constructs an optimal hyperplane to separate tumor classes based on the refined gene set.
• High-dimensional, small-sample suitability: The approach is tailored for high-dimensional and small-sample datasets common in clinical genomics.
• Cross-validation evaluation: Model performance is assessed using cross-validation schemes described in the input (10-fold and 5-fold).

Scientific Applications:

• Microarray and NGS datasets: The method was tested on both microarray and Next Generation Sequencing (NGS) datasets and evaluated with 10-fold cross-validation on three two-class and five multi-class microarray datasets, demonstrating superior performance compared to other classifiers.
• Specific tumor types: Application to Breast Cancer (BRCA) and Glioblastoma Multiforme (GBM) used 5-fold cross-validation and achieved high classification accuracy while selecting a minimal effective set of feature genes.

Methodology:

Initial feature gene selection on normalized datasets using the Kruskal-Wallis rank sum test; refinement of selected features with Sparse Group Lasso combining group and individual penalties; final classification using a Support Vector Machine.