scBatch

Correct scRNA-seq count matrix subject to batch effects by sample distance matrix adjustment

scBatch utilizes previous correction on sample distance matrices, such as QuantNorm, to further correct the count matrix. We implemented the method with RcppArmadillo for higher efficiency. The manuscript associated with this tool has been published on Bioinformatics.

Installation

The package requires R version 3.3.0 with prerequisite packages Rcpp, stats and utils. The package can be installed using the following code. The installation will typically complete within a minute.

#install.packages(devtools)
devtools::install_github('tengfei-emory/scBatch')
library(scBatch)

Example with simulated data

The following script utilizes scBatch to conduct batch effect correction on a simulated scRNA-seq data by splatter. It will take around ten minutes to simulate and correct a data with 100 subjects and 10000 genes. It can be observed from PCA plots that the batch effects are mitigated as the biological patterns are restored after correction.

#Loading packages
library(splatter)
library(scater)
library(scBatch)
library(rgl)

Set sample size n = 100, number of genes p = 10000 and the probability of being differentially expressed de = 0.1.

n=100
p=10000
de=0.1

Simulate a scRNA-seq data with four batches and four biological groups by splatter. The location and scale parameter of batch effects are fixed as 0.1 for quicker convergence of algorithm. More complicated batch effect mechanism will take longer to reach satisfied results.

sim.groups <- splatSimulate(nGenes=p, batchCells = c(n/4,n/4,n/4,n/4), batch.facLoc=c(0.1,0.1,0.1,0.1),
                            batch.facScale=c(0.1,0.1,0.1,0.1),seed = 1234, group.prob = c(0.4,0.3,0.2,0.1),
                            de.prob = c(de,de,de,de), method = "groups", verbose = F)
sim.groups <- normalise(sim.groups)

#Extract batch, biological information
batch = sim.groups@colData$Batch
cell.type = sim.groups@colData$Group

#normalized count matrix
exp <- exprs(sim.groups)

#Plot the uncorrected sample pattern
plot3d(princomp(cor(exp))$scores[,1:3],col=as.numeric(as.factor(cell.type)))

Now we conduct batch effect correction. The first step is distance matrix correction by QuantNorm:

correctedD <- QuantNorm(exp,as.numeric(as.factor(batch)),logdat=F,method='row/column',cor_method='pearson',max=5)

#Plot the corrected sample pattern from distance matrix D
plot3d(princomp(correctedD)$scores[,1:3],col=as.numeric(as.factor(cell.type)))

The second step is to use scBatch algorithm to further correct count matrix.

correctedmatrix <-scBatchCpp(c=exp,d=correctedD,w=diag(n),m=5,max=1000,tol=1e-10,step=0.0001,derif=scBatch::derif,verbose=T)

#Plot the corrected sample pattern from corrected count matrix
plot3d(princomp(cor(correctedmatrix))$scores[,1:3],col=as.numeric(as.factor(cell.type)))

For data set with large sample size, please consider utilizing high performance computing (HPC) platforms. The typical running time for data sets with sample size less than 1,000 is between 2 to 3 hours on HPC devices.

Reproducibility

For the results generated for the manuscript, the relevant scripts are available at this repository.

References

Fei, Teng, and Tianwei Yu (2020). "scBatch: batch-effect correction of RNA-seq data through sample distance matrix adjustment." Bioinformatics 36(10): 3115-3123.

Fei, Teng, et al (2018). "Mitigating the adverse impact of batch effects in sample pattern detection", Bioinformatics 34(15):2634–2641.

Zappia, Luke, Belinda Phipson, and Alicia Oshlack (2017). "Splatter: simulation of single-cell RNA sequencing data." Genome biology 18.1: 174.

Eddelbuettel, Dirk, and Conrad Sanderson (2014). "RcppArmadillo: Accelerating R with high-performance C++ linear algebra." Computational Statistics & Data Analysis 71: 1054-1063.