Sparsity regularization enhances gene selection and leukemia subtype classification via logistic regression.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional gene expression data presents challenges in classifying leukemia subtypes.
• Overfitting and dimensionality issues can hinder accurate cancer subtyping models.

Purpose of the Study:

• To investigate sparsity regularization methods for improving leukemia subtype classification.
• To compare the effectiveness of Ridge, Lasso, and Elastic Net regularization techniques.

Main Methods:

• Utilized multinomial logistic regression with Ridge, Lasso, and Elastic Net regularization.
• Applied methods to a leukemia gene expression dataset (CuMiDa) with 16,383 genes and 281 samples.
• Evaluated models using Accuracy, Kappa statistics, AUC, and F1-score.

Main Results:

• Elastic Net regularization outperformed Ridge and Lasso in overall classification performance and achieved the highest accuracy and Kappa values.
• Lasso and Elastic Net demonstrated superior feature selection capabilities, creating sparse models.
• Sparsity methods effectively reduced dimensionality and enhanced model interpretability.

Conclusions:

• Sparsity regularization significantly enhances accuracy and knowledge in leukemia subclass classification.
• Elastic Net is a promising technique for precise leukemia subtyping and personalized treatment strategies.
• Effective gene selection via sparsity methods aids in discriminating leukemia subtypes.