$k$-Shape Clustering Enhances Group Lasso for Gene Selection and Sample Classification

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput biological data requires efficient knowledge discovery tools.
• Group Lasso for logistic regression is effective for sample classification and gene selection but depends on robust clustering.
• Traditional k-means clustering variants can be unstable.

Purpose of the Study:

• To enhance the stability and performance of group Lasso for logistic regression.
• To introduce k-shape clustering as an alternative to k-means variants within the group Lasso framework.
• To evaluate the impact of k-shape clustering on gene selection and sample classification.

Main Methods:

• Integration of k-shape clustering into the group Lasso for logistic regression framework, termed GLKSH.
• Comparative analysis of GLKSH against traditional k-means variants using simulated and real-world biological datasets.
• Evaluation of classification accuracy, robustness, and gene identification capabilities.

Main Results:

• GLKSH demonstrated superior accuracy and robustness compared to k-means variants in both simulated and real-world datasets.
• GLKSH effectively identified informative genes relevant to sample classification.
• The proposed method achieved superior sample classification performance.

Conclusions:

• K-shape clustering significantly improves the performance of group Lasso for logistic regression.
• GLKSH offers a robust and accurate approach for gene selection and sample classification in high-throughput biological data.
• This work highlights the critical role of clustering in enhancing group Lasso methodologies.