Gene shaving using a sensitivity analysis of kernel based machine learning approach, with applications to cancer data

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Gene shaving (GS) is crucial for biomedical research but limited by large gene numbers and complex networks.
• Existing GS methods struggle with non-linear and multi-view data.
• A robust positive definite kernel-based GS method for biomedical data was lacking.

Purpose of the Study:

• To introduce a novel kernel-based gene shaving (KBGS) method.
• To evaluate KBGS performance against state-of-the-art methods using simulated and real data.
• To demonstrate KBGS's utility in identifying disease-associated gene subsets.

Main Methods:

• Developed a novel KBGS method utilizing the influence function of kernel canonical correlation analysis.
• Compared the proposed method with existing state-of-the-art gene shaving techniques.
• Assessed performance using metrics like true positive rate, false discovery rate, and AUC on simulated and colon cancer microarray data.

Main Results:

• The proposed KBGS method demonstrated superior performance compared to existing methods.
• Identified a significant subset of 210 genes from 2000 in colon cancer data.
• Revealed more significant gene interactions, suggesting concerted gene function in colon cancer.

Conclusions:

• The novel KBGS method effectively addresses the need for advanced gene selection in complex biological data.
• KBGS offers improved performance and identifies more significant gene interactions than current methods.
• Kernel-based methods, particularly positive definite ones, are valuable for non-linear biomedical data analysis and disease prediction.