Methods that remove batch effects while retaining group differences may lead to exaggerated confidence in downstream analyses

Abstract

Removal of, or adjustment for, batch effects or center differences is generally required when such effects are present in data. In particular, when preparing microarray gene expression data from multiple cohorts, array platforms, or batches for later analyses, batch effects can have confounding effects, inducing spurious differences between study groups. Many methods and tools exist for removing batch effects from data. However, when study groups are not evenly distributed across batches, actual group differences may induce apparent batch differences, in which case batch adjustments may bias, usually deflate, group differences. Some tools therefore have the option of preserving the difference between study groups, e.g. using a two-way ANOVA model to simultaneously estimate both group and batch effects. Unfortunately, this approach may systematically induce incorrect group differences in downstream analyses when groups are distributed between the batches in an unbalanced manner. The scientific community seems to be largely unaware of how this approach may lead to false discoveries.

Letter to the Editor response: Nygaard et al .: Table 1.

“…While we agree with Nygaard and others (2016) premise that the utilization of ANOVA or empirical-Bayes derived approaches to correct for batch effects may in certain cases overestimate significance, the approach they propose is surprisingly sensitive to differences in how data are preprocessed, specifically the handling of technical replicates. We therefore sought to identify alternate methods that address the points raised by Nygaard and others (2016).…”

“…The article by Nygaard and others (2016) proposes that applying batch correction approaches to microarray data from studies with unbalanced designs may inadvertently exaggerate the differences observed. In seeking to illustrate their point, Nygaard and others (2016) utilized a dataset (GSE61901) from a study we published (Towfic and others , 2014) and showed that one analysis pipeline utilizing the traditional approach to batch correction (ComBat) yielded over 1000 differentially expressed probesets, while an alternative approach proposed by Nygaard and others (2016; utilizing batch as a fixed effect and averaging technical replicates).…”

Cell Cycle M-Phase Genes Are Highly Upregulated in Anaplastic Thyroid Carcinoma

“…Combining microarray gene expression data from multiple laboratories or array platforms can have confounding batch effects leading to false discoveries (30). Many methods exist for removing batch effects from data, however, batch adjustments may bias the results and systematically induce incorrect group differences in downstream analyses (31). Rather than follow a more traditional approach (combined analysis), we instead performed separate comparisons in each dataset followed by identification of statistically significant, differentially expressed genes, which repeatedly appeared in each comparison.…”

Systems Genetics Approach Identifies Gene Pathways and Adamts2 as Drivers of Isoproterenol-Induced Cardiac Hypertrophy and Cardiomyopathy in Mice

“…In designing our study, we were cautious and distributed the treated and control conditions evenly across our three batches as well as endeavoring to include a diverse set of genetic backgrounds in each batch. Thus, we do not believe that our data suffer from the potential batch artifacts as reported in Nygaard and Rødland, 2016. To check our results, however, we used wMICA on data from a single batch of our data and recovered a module similar to module 5 (significance of overlap = 5.1E-26) and with similar GO enrichments.…”