Ultrafast and accurate clustering through imputation and dimensionality reduction for single-cell RNA-seq data.

Most existing dimensionality reduction and clustering packages for single-cell RNA-Seq (scRNA-Seq) data deal with dropouts by heavy modelling and computational machinery. Here we introduce CIDR (Clustering through Imputation and Dimensionality Reduction), an ultrafast algorithm which uses a novel yet very simple ‘implicit imputation’ approach to alleviate the impact of dropouts in scRNA-Seq data in a principled manner.

For more details about CIDR, refer to the paper: Peijie Lin, Michael Troup, Joshua W.K. Ho, CIDR: Ultrafast and accurate clustering through imputation for single-cell RNA-seq data. Genome Biology 2017 Mar 28;18(1):59.

CIDR is maintained by Dr Joshua Ho [email protected].

• Make sure your version of R is at least 3.1.0.
• CIDR has been tested primarily on the Linux and Mac platforms. CIDR has also been tested on the Windows platform - however this requires the use of an external software package Rtools.
• If you are on the Windows platorm, ensure that Rtools is installed. Rtools is software (installed external to R) that assists in building R packages, and R itself. Note that the downlaod for Rtools is in the order of 100M.
• Install the CRAN package devtools package which will be used to install CIDR and its dependencies:

## this is an R command
install.packages("devtools")

• Install the CIDR package directly from the Github repository (including any dependencies):

## this is an R command
devtools::install_github("VCCRI/CIDR")
## Note that for some Windows platforms, you may be asked to re-install RTools
## - even though it may already have been installed.  Say yes if prompted.
## Your windows platform may require the specific version of RTools being suggested.
##
## For Mac platforms, ensure that the software "Xcode" and "Command Line Tools" are
## installed, by issuing the following command from a terminal prompt:
##  /usr/bin/clang --version
##

Test the newly installed CIDR package:

library(cidr)
example("cidr")
#> 
#> cidr> par(ask=FALSE)
#> 
#> cidr> ## Generate simulated single-cell RNA-Seq tags.
#> cidr> N=3 ## 3 cell types
#> 
#> cidr> k=50 ## 50 cells per cell type
#> 
#> cidr> sData <- scSimulator(N=N, k=k)
#> 
#> cidr> ## tags - the tag matrix
#> cidr> tags <- as.matrix(sData$tags)
#> 
#> cidr> cols <- c(rep("RED",k), rep("BLUE",k), rep("GREEN",k))
#> 
#> cidr> ## Standard principal component analysis.
#> cidr> ltpm <- log2(t(t(tags)/colSums(tags))*1000000+1)
#> 
#> cidr> pca <- prcomp(t(ltpm))
#> 
#> cidr> plot(pca$x[,c(1,2)],col=cols,pch=1,xlab="PC1",ylab="PC2",main="prcomp")

#> 
#> cidr> ## Use cidr to analyse the simulated dataset.
#> cidr> ## The input for cidr should be a tag matrix.
#> cidr> sData <- scDataConstructor(tags)
#> 
#> cidr> sData <- determineDropoutCandidates(sData)
#> 
#> cidr> sData <- wThreshold(sData)
#> 
#> cidr> sData <- scDissim(sData)
#> 
#> cidr> sData <- scPCA(sData)

#> 
#> cidr> sData <- nPC(sData)
#> 
#> cidr> nCluster(sData)

#> 
#> cidr> sData <- scCluster(sData)
#> 
#> cidr> ## Two dimensional visualization: different colors denote different cell types,
#> cidr> ## while different plotting symbols denote the clusters output by cidr.
#> cidr> plot(sData@PC[,c(1,2)], col=cols,
#> cidr+      pch=sData@clusters, main="CIDR", xlab="PC1", ylab="PC2")

#> 
#> cidr> ## Use Adjusted Rand Index to measure the accuracy of the clustering output by cidr.
#> cidr> adjustedRandIndex(sData@clusters,cols)
#> [1] 0.9203693
#> 
#> cidr> ## 0.92
#> cidr> 
#> cidr> 
#> cidr>

Examples of applying CIDR to real biological datasets can be found at this Github repository. The name of the repository is CIDR-examples.

Clicking on the Clone or Download button in the Github repository for CIDR-examples will enable the user to download a zip file containing the raw biological data and the R files for the examples. The user can then extract the files and run the provided R examples.

CIDR-examples contains a human brain single-cell RNA-Seq dataset, located in the Brain folder. In this dataset there are 420 cells in 8 cell types after we exclude hybrid cells.

Reference for the human brain dataset:

Darmanis, S. et al. A survey of human brain transcriptome diversity at the single cell level. Proceedings of the National Academy of Sciences 112, 7285–7290 (2015).

CIDR-examples contains a human pancreatic islet single-cell RNA-Seq dataset, located in the PancreaticIslet folder. In this dataset there are 60 cells in 6 cell types after we exclude undefined cells and bulk RNA-Seq samples.

Reference for the human pancreatic islet dataset:

Li, J. et al. Single-cell transcriptomes reveal characteristic features of human pancreatic islet cell types. EMBO Reports 17, 178–187 (2016).

CIDR utilises the hclust function from the base stats package. Loading CIDR masks hclust in other packages automatically. However, if any package with an hclust function (e.g., flashClust) is loaded after CIDR, the name clashing can possibly cause a problem. In this case unloading that package should resolve the issue.

In some cases when installing a new version of CIDR on top of an existing version may result in the following error message:

Error in fetch(key) : lazy-load database '/Library/Frameworks/R.framework/Versions/3.3/Resources/library/cidr/help/cidr.rdb' is corrupt

In this case, one way to resolve this issue is to reinstall the devtools package:

install.packages("devtools")
## Click “Yes” in “Updating Loaded Packages”
devtools::install_github("VCCRI/CIDR",force=TRUE)

Some users might have installed an older version of RcppEigen. CIDR requires RcppEigen version >=0.3.2.9.0. Please re-install the latest version of this package if necessary.