Clustering microarray data to determine normalization method.

Most of the scientific journals require published microarray experiments to meet Minimum Information About a Microarray Experiment (MIAME) standards. This ensures that other researchers have the necessary information to interpret the results or reproduce them. Required MIAME information includes raw experimental data, processed data, and data processing procedures. However, the normalization method is often reported inaccurately or not at all. It may be that the scaling factor is not even known except to experienced users of the normalization software. We propose that using a seeded clustering algorithm, researchers can identify or verify previously unknown or doubtful normalization information. For that, we generate descriptive statistics (mean, variance, quantiles, and moments) for normalized expression data from gene chip experiments available in the ArrayExpress database and cluster chips based on these statistics. To verify that clustering grouped chips by normalization method, we normalize raw data for chips chosen from experiments in ArrayExpress using multiple methods. We then generate the same descriptive statistics for the normalized data and cluster the chips using these statistics. We use this dataset of known pedigree as seeding data to identify normalization methods used in unknown or doubtful situations.