XGBLC

XGBLC implements an XGBoost-based Lasso-Cox survival prediction model that predicts patient survival from high-dimensional gene expression profiles for cancer prognostic analysis.

Key Features:

• XGBoost integration: Leverages the XGBoost framework to handle large-scale and high-dimensional genomic feature spaces.
• Lasso-Cox enhancement: Implements Lasso-Cox regression incorporating first- and second-order gradient statistics into the loss function for feature selection and weighting.
• High-dimensional genomics: Specifically tailored for survival analysis using gene expression profiles in high-dimensional settings.
• Performance evaluation: Evaluated on 20 cancer datasets from The Cancer Genome Atlas (TCGA) and compared against five state-of-the-art survival methods.
• Statistical metrics: Assessed using concordance index (C-index), Brier score, and area under the curve (AUC).
• Robustness testing: Validated on both simulated and real-world datasets to assess accuracy and consistency across dataset scales.

Scientific Applications:

• Cancer prognostic modeling: Predicts patient survival outcomes from tumor gene expression to inform prognostic assessments.
• Clinical research and biomarker discovery: Supports identification of genomic features associated with survival for translational and clinical studies.
• Personalized medicine: Provides survival risk estimates that can inform personalized treatment decision-making in oncology research contexts.

Methodology:

Combines XGBoost with a Lasso-Cox objective that integrates first- and second-order gradient statistics into the loss function; performance was evaluated on 20 TCGA cancer datasets and on simulated and real-world datasets using C-index, Brier score, and AUC; implemented in R.