In qingnanl/SCAVG: Single-Cell Averaging Tool

knitr::opts_chunk$set(echo = TRUE)
library(knitr)
opts_chunk$set(tidy.opts = list(width.cutoff = 60), tidy = T)

Introduction

Here is a demonstration of how to use the package SRAVG to create a Seurat object with cells grouped and averaged.

Load libraries

library(Seurat)
library(SRAVG)
library(dplyr)
library(Matrix)
library(SeuratData)

Use the pbmc3k dataset from the SeuratData

data("pbmc3k")
#pbmc3k

The pbmc3k is a Seurat object while it is not processed. To run the SRAVG, we require the preprocessing of the object. Two things are needed: first, a dimension reduction (must be one of names(object@reductions)), for grouping nearest neighbors; second, a predefined classification of cells (must be one of colnames([e-mail])), because we don't want to group and average cells from different cluster/type/sample/donor etc.

Regular Seurat workflow:

pbmc3k <- pbmc3k %>% 
  NormalizeData() %>% 
  FindVariableFeatures() %>% 
  ScaleData() %>% 
  RunPCA(verbose = FALSE) %>% 
  FindNeighbors(dims = 1:10) %>%
  FindClusters(resolution = 0.5) %>%
  RunUMAP(dims = 1:10, verbose = FALSE)

Visualize the clustering:

DimPlot(pbmc3k, group.by = 'seurat_clusters', label = TRUE) + NoLegend()

Run SRAVG

Here we use 'pca' as the dimension reduction to evaluate the distances between cells, and top 10 PCs are used (dr_dims). We try to form meta-cells by averaging 10 cells to 1. The group will be within each individual 'seurat_clusters'. We also average two other columns in the meta.data and transfer to the output seurat, which are 'nCount_RNA', and 'nFeature_RNA'. At this time we only support numerical columns for 'extra_meta'.

start_time <- Sys.time()

pbmc_avg <- sravg(object = pbmc3k, dr_key = 'pca', dr_dims = 1:10, group_size = 10,
                             group_within = 'seurat_clusters',
                             extra_meta = c('nCount_RNA', 'nFeature_RNA'))
end_time <- Sys.time()
end_time - start_time

The output is a Seurat object

pbmc_avg

We can still run UMAP on the pbmc_avg object. The 'pca' in this averaged object is calculated with averaging the original 'pca' coordinates (for each meta-cell). Basically, the dimension reduction key used in the original object will be succeeded by the averaged object (if dr_key = 'umap', then the pbmc_avg will have a 'umap' in 'reductions'). Also, the 'group_within' column will be retained in the meta.data of the averaged object.

pbmc_avg <- RunUMAP(pbmc_avg, dims = 1:10, verbose = FALSE)
DimPlot(pbmc_avg, group.by = 'seurat_clusters', label = T)

Compute the sparsity of matrix (proportion of zeros) before and after averaging

sparsity <- function(matrix){
  sparsity <- sum(matrix == 0)/(dim(matrix)[1] * dim(matrix)[2])
  return(sparsity)
}

print(sparsity(pbmc3k@assays$RNA@counts))

print(sparsity(pbmc_avg@assays$RNA@counts))

qingnanl/SCAVG documentation built on June 25, 2022, 6:26 a.m.