gboosting

gboosting performs high-dimensional variable selection for genome-wide association studies (GWAS) with censored survival outcomes by applying modified component-wise gradient boosting together with stability selection to control false discoveries.

Key Features:

• Boosting Methodology: Employs a modified component-wise gradient boosting approach adapted for computational feasibility with large-scale genetic data.
• Stability Selection with Random Permutation: Integrates stability selection combined with random permutation techniques to control the false discovery rate (FDR) and reduce inclusion of irrelevant variables.
• High-Dimensional Censored Outcome Selection: Tailored for variable selection in high-dimensional settings involving censored survival data typical of genetic studies.
• Comparison to Other Methods: Incorporates stability selection within boosting to provide an alternative to univariate and Lasso approaches for survival GWAS variable selection.

Scientific Applications:

• Cutaneous Melanoma SNP–Survival Associations: Applied to identify associations between single-nucleotide polymorphisms (SNPs) and overall survival in cutaneous melanoma (CM) patients.
• BRCA2 and Fanconi Anemia Pathway Findings: Confirmed associations with BRCA2 pathway SNPs and identified potential modulators within the Fanconi anemia (FA) pathway.
• General Survival GWAS: Applicable to survival-related genetic association studies across diseases involving censored outcomes and high-dimensional predictors.

Methodology:

Uses component-wise gradient boosting adapted for computational efficiency and applies stability selection with random permutations to control false discoveries.