Feature screening for survival trait with application to TCGA high-dimensional genomic data

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Cancer Research
• Biostatistics and Survival Analysis

Background:

• Identifying cancer-related genes in high-dimensional survival genomic data is crucial but challenging.
• Numerous feature screening methods exist, but systematic comparisons are lacking.
• Existing methods often overlook gene-gene dependency information.

Purpose of the Study:

• To systematically compare existing feature screening methods through simulation studies.
• To develop a more accurate patient survival prediction model using The Cancer Genome Atlas (TCGA) datasets.
• To evaluate a novel network-adjusted feature screening approach.

Main Methods:

• Conducted simulation studies to compare feature screening methods.
• Applied network-adjusted feature screening to TCGA survival genomic data.
• Utilized gene-gene dependency information in the proposed method.

Main Results:

• Network-adjusted feature screening demonstrated superior performance compared to univariate independent feature screening methods.
• The proposed approach achieved more accurate survival prediction by incorporating gene-gene dependency.
• Identified cancer-related genes and biomarkers for esophageal, pancreatic, head and neck, lung, and breast cancers.

Conclusions:

• The network-adjusted feature selection method offers significant advantages over methods that ignore gene-gene dependencies.
• The network-based screening method is reliable and credible for identifying cancer-related genes.
• This approach enhances the accuracy of survival prediction models in cancer genomics.