Metrology for Gene Expression: Measurement Batch Effects, Probe Sensitivity, Gene List Reproducibility

As for other measurements, metrology for gene expression involves issues such as sources of measurement variation, measurement calibration, and inference on comparisons. These issues are addressed here in a highly multiplexed case of gene expression measurement, the EMERALD dataset of CAMDA08, which contains replicate measurements on liver, kidney, and mixtures of these two RNAs in six animals (Rattus norvegicus), made with Affymetrix, Agilent, and Illumina platforms.

We have obtained insight into the relative size of measurement batch effects and biological variation as represented by the animal-to-animal differences. These differences provide a practical benchmark because the animals were all subject to the same control-group treatment.

Although calibration curves for individual probes are unknown, a platform-to-platform correspondence identifies probes that measure the same transcripts and allows us to examine the relative sensitivity of probes from different platforms.

For biologists, gene expression microarrays provide an approach to identifying genes with particular properties such as change in expression with experimental treatment. The genes thus identified populate a gene list. Because we have measurements on six animals, we can obtain insight into such gene lists.