Tackling the widespread and critical impact of batch effects in high-throughput data

By Jeffrey T. Leek, R. Scharpf, Héctor Corrada Bravo, David M. Simcha, Benjamin Langmead, W. Johnson, Donald Geman, Keith A. Baggerly, Rafael A. Irizarry. 2010

2,064 citations recorded by the source corpus.

High-throughput technologies are widely used, for example to assay genetic variants, gene and protein expression, and epigenetic modifications. One often overlooked complication with such studies is batch effects, which occur because measurements are affected by laboratory conditions, reagent lots and personnel differences. This becomes a major problem when batch effects are correlated with an outcome of interest and lead to incorrect conclusions. Using both published studies and our own analyses, we argue that batch effects (as well as other technical and biological artefacts) are widespread and critical to address. We review experimental and computational approaches for doing so.

Subfields: Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Molecular Biology

Citations in the corpus

Methods for interpreting and understanding deep neural networks (2017, background): “In practice, it is also essential to verify for a given task, that the high measured accuracy results from the use of a proper problem representation, and not from the exploitation of artifacts in the data [29, 46, 27].”
openTSNE: a modular Python library for t-SNE dimensionality reduction and embedding (2019, background): “This procedure can therefore be used to handle batch effects (10), a key obstacle in molecular biology when dealing with the data from different sources (11).”
WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, neutral): “These batch effects make it difficult to draw conclusions from data across experimental batches (Leek et al., 2010; Parker & Leek, 2012; Soneson et al., 2014; Nygaard et al., 2016; Caicedo et al., 2017).”
Bayesian group factor analysis with structured sparsity (2016, background): “…component mixture is that, in real applications such as the analysis of gene expression data, it has been shown that much of the variation in the observation is due to technical (e.g., batch) or biological effects (e.g., sex, ethnicity), which impact a large number of features (Leek et al., 2010).”
Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization (2019, background): “, batch effects in biology (Leek et al., 2010), or image artifacts (Oakden-Rayner et al.”
Batch Effect Confounding Leads to Strong Bias in Performance Estimates Obtained by Cross-Validation (2014, background): “We distinguished four levels of confounding: none ( 50% g 1 samples in b 1 , 50% in b 2 ; same for g 2 samples), intermediate ( 75% 1 samples in b 1 , 25% in b 2 ; vice versa for g 2 samples), strong ( 95% 1 samples in b 1 , 5% in b 2 ; vice versa for g 2 samples) and full ( 100% g 1 samples in b 1…”
Covariate powered cross‐weighted multiple testing (2017, background): “So-called batch effects are common in large-scale data (Leek et al., 2010) and cause dataset-wide, and generally undesired, dependences.”
Data-analysis strategies for image-based cell profiling (2017, methodology): “However, there is no single rule that applies in all cases, and the best practice is to collect convincing evidence supporting these decisions, especially with the use of quality measures and replicate analysis (described in ‘Downstream analysis’). not unique to imaging and have been widely discusse”
Batch effect removal methods for microarray gene expression data integration: a survey (2013, methodology): “This has been the main motivation for the development of a wide range of methods for integrating (or combining, or merging) MAGE data originating from different studies over the past years [10–14].”
Predicting effective microRNA target sites in mammalian mRNAs (2015, background): “Pairing to the 3 ′ end of the miRNA can supplement canonical sites, although evidence for the use of this 3 ′ -supplementary pairing is observed for no more than 5% of the seed-matched sites ( Brennecke et al., 2005 ; Lewis et al., 2005 ; Grimson et al., 2007 ; Friedman et al., 2009 ).”
Count-based differential expression analysis of RNA sequencing data using R and Bioconductor (2013, methodology): “Here R and Bioconductor are adopted to tie together the workflow and provide data structures, and their unique strengths in workflow components are leveraged, including statistical algorithms, visualization”
Dimension reduction techniques for the integrative analysis of multi-omics data (2016, background): “It summarizes the main characteristics of data and may identify potential issues such as batch effects [2] and outliers.”
CMScaller: an R package for consensus molecular subtyping of colorectal cancer pre-clinical models (2017, methodology): “RNA-sequencing data was first log 2-transformed and batch adjusted according to sequencing platform (HiSeq/GAIIx) using the ComBat method implemented in the R package sva 57 .”
Discovery and Validation of a Prostate Cancer Genomic Classifier that Predicts Early Metastasis Following Radical Prostatectomy (2013, result): “As suggested by a previous study [36], the first 10 principal Table 1.”