TRANSCRIPTOME 2002: From Functional Genomics to Systems
Application of Statistical Modeling to Microarray Gene Expression Data in Human Lung Cancer
Najma Khalid, Lue Ping Zhao, Fred Hutchinson Cancer Research Center Public Health Sciences, Seattle, WA
Microarray technology allows investigators to genotype thousands of transcripts in experimental studies. Cluster analysis techniques, typically used to classify expression profiles, cannot utilize clinical information. We applied a more robust statistical approach to the analysis of expression data collected in a study to subclassify lung adenocarcinomas. (Bhattacharjee et al, PNAS 2001). Expression data on 12,600 genes were available for 17 normal lung samples and 139 adenocarcinoma tumors (125 samples associated with clinical data). Using an estimating equation technique that adjusts for heterogeneity across microarray chips, and for multiple comparisons, we compared normal samples with tumor samples and identified nearly 700 genes that were differentially expressed between the two groups (p< 0.001). We had similar results by including clinical variables (tumor stage, size, vital status recurrence, survival, age at resection, smoking history and sex), as covariates in the model. Finally for adenocarcinomas only, we compared groups by vital status and by recurrence. No differentially expressed genes were identified with either outcome. Further adjustment for age at resection and survival also failed to identify any differentially expressed genes. We conclude that microarray data of samples taken at resection yield diagnostic markers but do not predict prognosis.
Return to Table of Contents * Speaker Abstracts * Poster Abstracts * View the Photos
Return to Meetings Home Page
This site produced by the Human Genome Management Information System of Oak Ridge National Laboratory.