Joint analysis of two microarray gene-expression data sets to select lung adenocarcinoma marker genes

Area of Science:

• Bioinformatics
• Genomics
• Cancer Research

Background:

• Microarray experiments face challenges with high costs, low reproducibility, and limited sample sizes.
• Integrating multiple microarray datasets increases statistical power and aids in identifying common marker genes for diseases like lung cancer.

Purpose of the Study:

• To develop and validate a method for integrating disparate microarray datasets.
• To identify novel marker genes for distinguishing lung cancer from normal samples.
• To identify genes significantly correlated with patient survival in lung cancer.

Main Methods:

• Combined two lung cancer GeneChip microarray datasets using a distribution transformation method.
• Applied gene shaving (GS) methods based on Random Forests (RF) and Fisher's Linear Discrimination (FLD) for marker gene selection.
• Utilized a two-step survival test, including univariate Cox proportional hazard regression and principal component regression, to identify survival-related genes.

Main Results:

• Identified 13 (RF) and 10 (FLD) marker genes differentiating lung cancer from normal samples, with 5 common genes.
• Classifiers built from one dataset predicted the other with >98% accuracy.
• 36 genes showed significant correlation with patient survival, with 26 previously reported, including tumor suppressors and oncogenes. Principal component regression reduced this to 16 genes.

Conclusions:

• Developed a robust method for integrating microarray data from different sources.
• Demonstrated the effectiveness of gene shaving and survival analysis for identifying minimal marker gene sets for cancer diagnosis and prognosis.
• High prediction accuracy achieved when applying classification models developed from one integrated dataset to another.