Feature selection and tumor classification for microarray data using relaxed Lasso and generalized multi-class...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression data analysis is crucial for molecular-level tumor diagnosis.
• High-dimensional, small-sample gene expression data presents challenges in feature selection and classification.
• Accurate classification is needed to differentiate tumor subtypes and normal from patient samples.

Purpose of the Study:

• To develop a novel method for tumor classification using gene expression data.
• To address the challenges of feature selection in high-dimensional and small-sample datasets.
• To improve the accuracy of distinguishing between tumor subtypes and normal versus patient samples.

Main Methods:

• Data preprocessing involved z-score normalization of tumor datasets.
• Feature gene selection was performed using the relaxed Lasso (least absolute shrinkage and selection operator) method on the training set.
• Classification was achieved using the generalized multi-class support vector machine (GenSVM) algorithm.

Main Results:

• The proposed relaxed Lasso and generalized multi-class support vector machine (rL-GenSVM) method demonstrated superior performance compared to other classifiers.
• Experiments on four two-class and four multi-class datasets showed higher classification accuracy.
• The method effectively selected a smaller set of feature genes while maintaining high accuracy, validated by 10-fold cross-validation.

Conclusions:

• The rL-GenSVM method offers a robust approach for tumor classification with high-dimensional and small-sample gene expression data.
• The use of regularization parameters in rL-GenSVM helps prevent overfitting.
• This method holds potential for wide application in cancer research and clinical diagnostics.