Component-wise gradient boosting and false discovery control in survival analysis with high-dimensional covariates

Area of Science:

• Genetics and Bioinformatics
• Statistical Genomics
• Computational Biology

Background:

• High-dimensional data in genetics studies necessitates advanced statistical methods.
• Variable selection for censored outcomes and controlling false discoveries are key challenges.
• Existing methods may struggle with the complexity of modern genetic datasets.

Purpose of the Study:

• To develop a computationally feasible method for high-dimensional variable selection in genetics.
• To control false discoveries using stability selection in the presence of many predictors.
• To improve the utilization of rich genetic information for survival analysis.

Main Methods:

• Developed a novel method combining boosting and stability selection.
• Modified component-wise gradient boosting for enhanced computational feasibility.
• Incorporated random permutation within stability selection to control false discoveries.

Main Results:

• The proposed method demonstrated fewer false discoveries compared to univariate and Lasso approaches.
• Applied to cutaneous melanoma (CM) patients, analyzing 2339 single-nucleotide polymorphisms (SNPs).
• Identified associations between BRCA2 and Fanconi anemia (FA) pathway SNPs with patient survival.

Conclusions:

• The new method offers improved accuracy for variable selection in high-dimensional genetic studies.
• Stability selection effectively controls false discoveries, enhancing reliability.
• The findings provide insights into genetic factors influencing cutaneous melanoma patient survival.