compound.Cox: Univariate feature selection and compound covariate for predicting survival

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Univariate feature selection is crucial for developing multigene survival predictors.
• Existing software lacks tailored solutions for univariate feature selection and predictor construction.

Purpose of the Study:

• To develop the compound.Cox R package for univariate feature selection and multigene predictor construction.
• To provide robust statistical tools for survival data analysis.

Main Methods:

• Implemented univariate significance tests (Wald, score tests) for feature selection.
• Incorporated cross-validation for predictive capability assessment and permutation tests for false discovery rate.
• Developed three multigene predictor construction algorithms: compound covariate, compound shrinkage, and copula-based methods.

Main Results:

• The compound.Cox R package is available on CRAN.
• The package enables determination of optimal significance levels for gene selection.
• Researchers can compute false discovery rates and utilize various prediction algorithms.

Conclusions:

• The compound.Cox package provides a comprehensive solution for univariate feature selection and multigene predictor development.
• It empowers researchers to build accurate survival prediction models using optimal gene subsets.
• The package facilitates robust statistical analysis of survival data, particularly in cancer research.