Biological data are obtained using different instruments, by different researchers, or under other different conditions. As a result non-biological variation between data sets is inevitable. This non-biologically (technical) variance is called a batch effect.
When analyzing a data set with potential batch effects, one must be careful to first remove as much of this variation as possible. The process of removing batch effects is referred to as batch correction. If proper batch correction is not applied then the results of any downstream analyses performed may be useless, since there is no way for these downstream tasks to distinguish between real biological signal and batch-effect-related noise.
On the other hand, it is also possible to over batch correct and inadvertently remove important biologically-relevant variation from the data. The process of merging data sets produced or measured under different conditions is called integration. Striking an appropriate balance with your batch correction algorithm and parameters is a vital step of integration.
Several R packages exist that score batch effects in a given data set. These scores are useful measures of how much batch effect may exist in a given data set. An example of such a score is kBET. kBET measures the mixing of batches by comparing the local distribution of batch labels in a sample of neighborhoods to the global distribution of batch labels.
The objective of IntegrationScore is to allow researchers working with scRNA-seq data to compare the effectiveness of batch correction methods using a different measure. That measure being: in a given batch, are the differentially expressed genes (DEGs) between clusters somewhat invariant under the batch correction method? A measure such as this gives researchers an idea of which genes were significantly changed by the correction algorithm. Additionally, this measure can be used to get a general picture of how the correction algorithm changed each cluster. This R package enables scientists to quickly perform this sanity check and interpret the results using clear visualizations.
IntegrationScore was developed on a Mac using R 4.1.1.
You can install the development version of IntegrationScore from GitHub with:
require("devtools")
#> Loading required package: devtools
#> Loading required package: usethis
devtools::install_github("eliaswilliams/IntegrationScore")
#> Skipping install of 'IntegrationScore' from a github remote, the SHA1 (145b726f) has not changed since last install.
#> Use `force = TRUE` to force installation
library("IntegrationScore")
#> Registered S3 method overwritten by 'spatstat.geom':
#> method from
#> print.boxx cliTo run the shinyApp: run runIntegrationScore().
ls("package:IntegrationScore")
#> [1] "corrDEG" "deg_df" "findDEG"
#> [4] "ifnb_corrected" "ifnb_split" "ifnb_split_corrected"
#> [7] "plotCorrByCutoff" "plotDelta" "plotUMAP"
#> [10] "runIntegrationScore"This package contains 5 functions available to users. It is recommended that users start by calling findDEG which returns a data frame of differentially expressed genes (DEGs) in your Seurat objects. corrDEG is then used to calculate the correlation of the DEGs under batch correction. The remaining three functions are for plotting. plotDelta returns a plot that illustrates change in log2FC (delta) under the correction algorithm. plotCorrByCutoff shows how the correlation of the DEGs changes as you consider more or less DEGs. And finally, plotUMAP plots the UMAP reduction pre and post batch correction.
browseVignettes("IntegrationScore")
#> No vignettes found by browseVignettes("IntegrationScore")The author of this package is Elias Williams. All functions make use of the Seurat library for single-cell analysis. ggplot2 is used for plotting functions. Seurat’s integration vignette is borrowed from in this package’s vignette, and it is attributed there. The vignette also makes use of a sample data set from the SeuratData package.
Büttner, M., Miao, Z., Wolf, F.A. et al. (2019) A test metric for assessing single-cell RNA-seq batch correction. Nat Methods 16, 43–49. https://doi.org/10.1038/s41592-018-0254-1
Hoffman, P. (2021) Satija Lab. Introduction to scRNA-seq integration. https://satijalab.org/seurat/articles/integration_introduction.html
R Core Team (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/
Satija, R. (2021) Satija Lab. Seurat Data R Package. https://github.com/satijalab/seurat-data
Wickham, H. and Bryan, J. (2019). R Packages (2nd edition). Newton, Massachusetts: O’Reilly Media. https://r-pkgs.org/
This package was developed as part of an assessment for 2021 BCB410H: Applied Bioinfor-matics, University of Toronto, Toronto, CANADA.