Terry Speed, Ph.D.
Department of Statistics and Program in Biostatistics
Low level analysis of microarray expression data
With all of the microarray platforms, there are common features in the conversion of raw data into biologically meaningful measurements or answers to questions. Measurements could be "absolute" (up to a scale factor) such as the concentration of a given mRNA, or they could be relative, such as the fold-change of a transcript between two samples. Questions might concern the presence or absence of a given mRNA species in the sample, or the differential expression of a gene between two samples. How we carry out the low level processing of the raw microarray data, the image, has an impact on our measurements or answers. I will illustrate this with examples drawn principally from the Affymetrix short oligo platform, but the message is general. The issues are familiar: background adjustment, normalization, dealing with outliers. The problems are also familiar: models and bias-variance trade-off.