The Practical Effect of Batch on Genomic Prediction

• Activity Number: 188
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Hilary Parker*+ and Jeff Leek and Rafael Irizarry
• Companies: The Johns Hopkins University and The Johns Hopkins University and The Johns Hopkins University
• Keywords: Genomics ; Batch Effects ; Microarrays ; High-throughput ; Prediction ; Bioinformatics

Abstract

Measurements from microarray and other high-throughput technologies are susceptible to a number of non-biological variables such as temperature and reagent lots. It has been shown that these artefacts, collectively called batch effects, can severely alter the outcome of any differential expression analysis, resulting in misleading biological conclusions. Here we examine the impact of batch effects on predictors built from genomic technologies - specifically gene expression microarrays. We compare single microarray (fRMA) and multiple microarray (RMA, MASS5) preprocessing methods and both rank-based (top-scoring pairs) and continuous (PAM) predictors. We show that in general, prediction is made more difficult by batch effects. We also show that when there is perfect confounding of batch and the outcome being predicted, then accuracy is substantially reduced. In an effort to mitigate this effect, we determine which probes from commonly used Affymetrix arrays are most susceptible to batch and investigate their properties. Down-weighting these "batch-affected" probes may lead to increased predictive accuracy when building gene expression based predictors.