A Comparison of High Dimensional Variable Selection Methods with Missing Covariates in a Prostate Cancer Study

Area of Science:

• Oncology
• Biostatistics
• Genetics

Background:

• Prostate cancer is a prevalent malignancy in American men.
• Numerous genes are implicated in prostate cancer development and progression.
• Accurate gene identification is crucial for understanding benign versus non-benign cases.

Purpose of the Study:

• To apply penalized logistic regression models for gene selection in prostate cancer.
• To compare gene selection results using complete versus imputed data, addressing missing values.
• To validate the proposed methodology through a simulation study.

Main Methods:

• Utilized penalized logistic regression with various penalty functions for gene selection.
• Employed cross-validation to determine the optimal tuning parameter.
• Implemented multiple imputation to handle missing gene expression data.

Main Results:

• Identified specific genes associated with benign and non-benign prostate cancer.
• Demonstrated the impact of multiple imputation on gene selection outcomes.
• Simulation study confirmed the robustness and effectiveness of the proposed method.

Conclusions:

• Penalized logistic regression effectively selects genes relevant to prostate cancer subtypes.
• Multiple imputation enhances the analysis of prostate cancer data with missing values.
• The validated method provides a robust approach for identifying cancer-associated genes.