Correction of scaling mismatches in oligonucleotide microarray data

Background: Gene expression microarray data is notoriously subject to high signal variability. Moreover, unavoidable variation in the concentration of transcripts applied to microarrays may result in poor scaling of the summarized data which can hamper analytical interpretations. This is especially relevant in a systems biology context, where systematic biases in the signals of particular genes can have severe effects on subsequent analyses. Conventionally it would be necessary to replace the mismatched arrays, but individual time points cannot be rerun and inserted because of experimental variability. It would therefore be necessary to repeat the whole time series experiment, which is both impractical and expensive. Results: We explain how scaling mismatches occur in data summarized by the popular MAS5 (GCOS; Affymetrix) algorithm, and propose a simple recursive algorithm to correct them. Its principle is to identify a set of constant genes and to use this set to rescale the microarray signals. We study the properties of the algorithm using artificially generated data and apply it to experimental data. We show that the set of constant genes it generates can be used to rescale data from other experiments, provided that the underlying system is similar to the original. We also demonstrate, using a simple example, that the method can successfully correct existing imbalancesin the data. Conclusion: The set of constant genes obtained for a given experiment can be applied to other experiments, provided the systems studied are sufficiently similar. This type of rescaling is especially relevant in systems biology applications using microarray data.