In debsin/dropClust: Single Cell Transcriptome Analysis

knitr::opts_chunk$set(
  collapse = TRUE,
  eval=TRUE,
  comment = "#>"
)

Loading datasets

Each dataset represents one batch and must be a SingleCellExperiment object. The objects are are merged by passing a list in the next step.

library(dropClust)
load(url("https://raw.githubusercontent.com/LuyiTian/CellBench_data/master/data/sincell_with_class.RData"))

objects = list()

objects[[1]] = sce_sc_10x_qc

objects[[2]] = sce_sc_CELseq2_qc

objects[[3]] = sce_sc_Dropseq_qc

Merge datasets using dropClust

Datasets can be merged in two ways: using a set of DE genes from each batch or, using the union of the sets of highly variable genes from each batch.

all.objects = objects
merged_data<-Merge(all.objects)

Perform correction and dimension reduction

set.seed(1)
dc.corr <-  Correction(merged_data,  method="default", close_th = 0.1, cells_th = 0.1,
                       components = 10, n_neighbors = 30,  min_dist = 1)

Perform Clustering on integrated dimensions

dc.corr = Cluster(dc.corr,method = "hclust")

Visualizing clusters

Compute 2D embeddings for samples followed by post-hoc clustering.

ScatterPlot(dc.corr, title = "Clusters")

Optional Batch correction

Users can use fastmnn method for batch correction. Specific arguments of fastmnn can also be passed through the Correction module.

merged_data.fastmnn<-Merge(all.objects,use.de.genes = FALSE)

set.seed(1)
mnn.corr <-  Correction(merged_data.fastmnn,  method="fastmnn", d = 10)

mnn.corr = Cluster(mnn.corr,method = "kmeans",centers = 3)

ScatterPlot(mnn.corr, title = "Clusters")

Marker discovery from the merged dataset

de<-FindMarkers(dc.corr,q_th = 0.001, lfc_th = 1.2,nDE = 10)
de$genes.df

debsin/dropClust documentation built on Nov. 4, 2019, 10:22 a.m.