Sue Geller, Ph.D.
Professor of Mathematics
Professor of Veterinary Anatomy and Public Health
Variance stabilization, normalization, and power calculations of microarray data with application to autism
It has long been known that microarray data do not have constant variance across levels nor does taking the log of the data completely solve the problem since the variance of log transformed data is not constant at low or moderate expression levels. Furthermore, there are unresolved issues with normalization and power calculations. I will present a method of doing experiments and simultaneously transforming and normalizing the data so that the transformed data have constant variance and the errors of the transformed data are symmetric and long-tailed. Thus, power calculations can be performed on the transformed data using standard software. This method can be thought of as a machine calibration in that it requires a few biologically identical replicates, called technical replicates by some, of one sample to establish the constant of the transformation and normalize the entire data set.
The efficacy of the above transformation is examined on gene expression data from a cell line derived from an autistic patient. This cell line was grown up four times, separately processed, and hybridized to four separate Affymetrix HG\_U95Av2 oligonucleotide GeneChip arrays, containing 12,625 genes. Despite autism being highly heritable (over 90%), no specific chromosomal loci have yet been located. An alternative approach to the identification of genetic components of autism is the identification of patterns of altered gene expression through microarray analysis of blood and tissue samples from autistic patients and unaffected family members with the objective of identifying patterns of gene expression that associate with the autism spectrum phenotype.
The above is joint work with Jeff
P. Gregg, Paul Hagerman, and David M. Rocke, all at
the