Canek

Canek is an R package to correct batch effects from single-cell RNA-seq biological replicates.

Motivation to develop Canek

As single-cell genomics technologies become mainstream, more laboratories will perform experiments under different conditions with biological replicates obtained using a common technology. In this scenario, integration of datasets with minimal impact on cell phenotype is essential.

The workflow

Canek leverages information from mutual nearest neighbors to combine local linear corrections with cell-specific non-linear corrections within a fuzzy logic framework.

A. Canek starts with a reference batch and query batch, assuming a predominantly linear batch effect.

B. Cell clusters are defined on the query batch and MNN pairs (arrows) are used to define batch effect observations.

C. The MNN pairs from each cluster are used to estimate cluster specific correction vectors. These vectors can be used to correct the batch effect or, (D) a non-linear correction can be applied by calculating cell-specific correction vectors using fuzzy logic.

Results

Canek was the highest scored method in tests specifically designed to assess over-correction, where Canek corrected batch effects without distortion to the structures of cells as compared with a gold standard.

For more information about Canek check out our manuscript in NAR Genomics and Bioinformatics.

What’s new in 0.3.1

• Latest updates: RunCanek() on Seurat objects now defaults to correctEmbeddings = TRUE — correction happens in PCA-embedding space instead of on gene expression directly, and the result is a new "canek" reduction rather than a "Canek" assay. Pass correctEmbeddings = FALSE to keep the previous behavior.
• pcaDim is inferred automatically from an existing "pca" reduction when present, and reused directly instead of being recomputed.
• Added SCTransform support, including automatic reconciliation checks for batches normalized separately.
• Added a repeat-correction loop (maxLoop/loopTol) that refines the correction across multiple passes.
• Sped up MNN pair finding, and added an opt-in ncores parameter to parallelize it.

See the full changelog for details.

Usage

You can use Canek directly with normalized-count matrices, Seurat objects or SingleCellExperiment objects. As of Canek 0.3.1, correcting Seurat objects defaults to PCA-embedding space (correctEmbeddings = TRUE) rather than gene expression directly, and both log-normalized and SCTransform-normalized data are supported. For more details, check out our GitHub page and vignettes:

• Canek website
• Run Canek on a toy example vignette
• Correct log-normalized Seurat data
• Correct SCTransform-normalized Seurat data
• Run Canek on SingleCellExperiment objects vignette
• Speed up batch correction with parallel processing

For vignettes related to the previous v0.2.x version, see Previous versions.

Installation

You can install the release version of Canek from CRAN with:

install.packages("Canek")

You can install the development version from GitHub with:

# install.packages("remotes")
remotes::install_github("MartinLoza/Canek")

Citation

If you use Canek in your research please cite our work using:

Loza M, Teraguchi S, Standley D, Diez D (2022). “Unbiased integration of single cell transcriptome replicates.” NAR Genomics and Bioinformatics, 4(1), lqac022. doi:10.1093/nargab/lqac022 https://doi.org/10.1093/nargab/lqac022, https://martinloza.github.io/Canek/.