Statistical issues and methods for meta-analysis of microarray data: a case study in prostate cancer

Funct Integr Genomics. 2003 Dec;3(4):180-8. doi: 10.1007/s10142-003-0087-5. Epub 2003 Jul 22.

Abstract

With the proliferation of related microarray studies by independent groups, a natural step in the analysis of these gene expression data is to combine the results across these studies. However, this raises a variety of issues in the analysis of such data. In this article, we discuss the statistical issues of combining data from multiple gene expression studies. This leads to more complications than those in standard meta-analyses, including different experimental platforms, duplicate spots and complex data structures. We illustrate these ideas using data from four prostate cancer profiling studies. In addition, we develop a simple approach for assessing differential expression using the LASSO method. A combination of the results and the pathway databases are then used to generate candidate biological pathways for cancer.

Publication types

  • Meta-Analysis
  • Research Support, Non-U.S. Gov't

MeSH terms

  • Algorithms
  • Computational Biology
  • Databases, Factual
  • Expressed Sequence Tags
  • False Positive Reactions
  • Gene Expression Profiling
  • Gene Expression Regulation, Neoplastic
  • Humans
  • Male
  • Models, Statistical
  • Oligonucleotide Array Sequence Analysis*
  • Prostatic Neoplasms / genetics*
  • Prostatic Neoplasms / pathology
  • Statistics as Topic / methods*