Gene Selection using a High-Dimensional Regression Model with Microarrays in Cancer Prognostic Studies

Area of Science:

• Bioinformatics
• Genomics
• Biostatistics

Background:

• Identifying genes linked to patient survival from gene expression data is crucial for cancer prognosis.
• High-dimensional microarray data presents challenges in gene selection and accurate parameter estimation.
• The least absolute shrinkage and selection operator (lasso) is commonly used but can yield false positives.

Purpose of the Study:

• To propose a novel method for estimating the false positive rate (FPR) in high-dimensional Cox models utilizing lasso estimates.
• To address the issue of false positive genes identified through cross-validation in lasso regression for cancer prognostics.

Main Methods:

• Development of a statistical method to estimate the FPR specifically for lasso-selected genes in a Cox proportional hazards model.
• Conducting simulation studies to rigorously evaluate the precision and reliability of the proposed FPR estimation technique.
• Application of the developed method to real-world gene expression datasets from cancer patients.

Main Results:

• The proposed method demonstrated precision in estimating the false positive rate (FPR) for lasso-based gene selection.
• Simulation studies confirmed the accuracy of the FPR estimation in high-dimensional settings.
• The method successfully identified false positive genes when applied to actual patient data, aiding in refining prognostic markers.

Conclusions:

• The novel FPR estimation method offers a valuable tool for improving the reliability of gene selection in cancer prognostic studies.
• Accurate identification of false positives enhances the precision of prognostic models derived from microarray data.
• This approach contributes to more dependable cancer outcome predictions by filtering out spurious gene associations.