Numerical Deconvolution of cDNA Microarray Signal: Simulation Study

Authors

  • SIMON ROSENFELD,

    Corresponding author
    1. Biometry Research Group, Division of Cancer Prevention, National Cancer Institute, National Institutes of Health, Department of Health and Human Services, Rockville, Maryland 20892, USA
    Search for more papers by this author
  • THOMAS WANG,

    1. Phytonutrients Laboratory, USDA, Beltsville, Maryland 20705, USA
    Search for more papers by this author
  • YOUNG KIM,

    1. Nutritional Sciences Research Group, Division of Cancer Prevention, National Cancer Institute, National Institutes of Health, Department of Health and Human Services, Rockville, Maryland 20892, USA
    Search for more papers by this author
  • JOHN MILNER

    1. Nutritional Sciences Research Group, Division of Cancer Prevention, National Cancer Institute, National Institutes of Health, Department of Health and Human Services, Rockville, Maryland 20892, USA
    Search for more papers by this author

Address for correspondence: Simon Rosenfeld, Biometry Research Group, Division of Cancer Prevention, National Cancer Institute, National Institutes of Health, Department of Health and Human Services, EPN 3136, 6130 Executive Boulevard, Rockville, MD 20892. Voice: 301-496-7748; fax: 301-402-0816. sr212a@nih.gov

Abstract

Abstract: A computational model for simulation of the cDNA microarray experiments has been created. The simulation allows one to foresee the statistical properties of replicated experiments without actually performing them. We introduce a new concept, the so-called bio-weight, which allows for reconciliation between conflicting meanings of biological and statistical significance in microarray experiments. It is shown that, for a small sample size, the bio-weight is a more powerful criterion of the presence of a signal in microarray data as compared to the standard approach based on t test. Joint simulation of microarray and quantitative PCR data shows that the genes recovered by using the bio-weight have better chances to be confirmed by PCR than those obtained by the t test technique. We also employ extreme value considerations to derive plausible cutoff levels for hypothesis testing.

Ancillary