PLSDA-batch: a multivariate framework to correct for batch effects in microbiome data

Abstract

Microbial communities are highly dynamic and sensitive to changes in the environment. Thus, microbiome data are highly susceptible to batch effects, defined as sources of unwanted variation that are not related to and obscure any factors of interest. Existing batch effect correction methods have been primarily developed for gene expression data. As such, they do not consider the inherent characteristics of microbiome data, including zero inflation, overdispersion and correlation between variables. We introduce new multivariate and non-parametric batch effect correction methods based on Partial Least Squares Discriminant Analysis (PLSDA). PLSDA-batch first estimates treatment and batch variation with latent components, then subtracts batch-associated components from the data. The resulting batch-effect-corrected data can then be input in any downstream statistical analysis. Two variants are proposed to handle unbalanced batch x treatment designs and to avoid overfitting when estimating the components via variable selection. We compare our approaches with popular methods managing batch effects, namely, removeBatchEffect, ComBat and Surrogate Variable Analysis, in simulated and three case studies using various visual and numerical assessments. We show that our three methods lead to competitive performance in removing batch variation while preserving treatment variation, especially for unbalanced batch $\times $ treatment designs. Our downstream analyses show selections of biologically relevant taxa. This work demonstrates that batch effect correction methods can improve microbiome research outputs. Reproducible code and vignettes are available on GitHub.

Best practice for wildlife gut microbiome research: A comprehensive review of methodology for 16S rRNA gene investigations

“…Researchers should try to ensure that factors such as age/sex/genetics of their samples, sampling location/ time, kit type/processing time, etc. (Wang and Lê Cao, 2020) do not overlap to large degrees with the actual biological variation they are testing in their experiments. If, however, such conflation is unavoidable due to the nature of the study system, there are a number of post hoc statistical computational methods that have been developed for dealing with such batch effects, specifically for microbiome data (Gibbons et al, 2018;Ma et al, 2020;Wang and Lê Cao, 2020).…”

Evaluating cancer cell line and patient-derived xenograft recapitulation of tumor and non-diseased tissue gene expression profilesin silico

“…We evaluated the difference in variability across samples by gene set size with principal component analysis (PCA) with the mixOmics R package (version 6.18.1) pca() function to determine principal components, 23 and then used the PLSDAbatch package (version 0.2.1) Scatter_Density() function to generate PCA and density rugplots to investigate the variance between sample origin types (i.e., cell line, PDX, non-disease tissue, tumor tissue; between brain-derived tissues and GBM-derived models in Figure 2B; between all samples of all sources in Supplemental Figure 1A), as well as with respect to other variables that may impact variance (i.e., read length, tissue type, sample source, sex; Supplemental Figure 1B-E). 24 We identified tumor purity-correlated genes based on TCGA tumor gene expression significantly correlated with ABSOLUTE-generated tumor purity. We also defined tissue-specific gene expression from GTEx using the Harminozome (accessed November 2022).…”

Evaluating cancer cell line and patient‐derived xenograft recapitulation of tumor and non‐diseased tissue gene expression profiles in silico

“…We evaluated the difference in variability across samples by gene set size with principal component analysis (PCA) with the mixOmics R package (version 6.18.1) pca() function to determine principal components, 24 and then used the PLSDAbatch package (version 0.2.1) Scatter_Density() function to generate PCA and density rugplots to investigate the variance between sample origin types (i.e., cell line, PDX, non-disease tissue, tumor tissue; between brainderived tissues and GBM-derived models in Figure 2B; between all samples of all sources in Figure S1A), as well as with respect to other variables that may impact variance (i.e., read length, tissue type, sample source, sex; Figure S1B-E). 25 We identified tumor purity-correlated genes based on TCGA tumor gene expression significantly correlated with ABSOLUTEgenerated tumor purity. We also defined tissue-specific gene expression from GTEx using the Harminozome (accessed November 2022).…”