In amsszlh/scMC: Integrating and comparing multiple single cell genomic datasets

---

title: 'Demo of scMC Seurat Wrapper using mouse skin dermis scRNA-seq data ' author: "Lihua Zhang" date: "r format(Sys.time(), '%d/%m/%y')" output: html_document: default mainfont: Arial vignette: | %\VignetteIndexEntry{Integrating and comparing multiple single cell genomic datasets using scMC} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8}

---

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  root.dir = './'
)

To make it easy to run scMC in most common scRNA-seq data analysis pipelines, the implementent of scMC is seamlessly compatible with the workflow of Seurat package. In sum, scMC pipeline consists of the following three major parts:

• Set up a list of Seurat objects (one per dataset) and preprocess each dataset seperately

• Perform an integrated analysis using scMC Seurat Wrapper by taking an input a list of Seurat objects

• Run the standard workflow for downstream analyses such as clustering and visualization

This demo closely follows the Seurat vignette: https://satijalab.org/seurat/v3.0/immune_alignment.html

Load the required libraries

library(scMC)
library(Seurat)
library(patchwork)
library(dplyr)
library(ggplot2)

Part I: Load a list of preprocessed Seurat objects, one per dataset

Here we load preprocessed Seurat object for each dataset. These preprocessed Seurat objects are obtained using Seurat's standard workflow. Please see the tutorial "demo_scMC_drmis" for details.

load("/Users/suoqinjin/Documents/scMC-master/tutorial/SeuratObject.list_dermis.RData")

We also perform normalization, feature selection and scaling of the data in each dataset.

Normalizing, scaling the data and feature selection

object.list <- lapply(X = object.list, FUN = function(x) {
  x <- NormalizeData(x, verbose = FALSE)
  x <- FindVariableFeatures(x, verbose = FALSE)
  # perform scaling on the previously identified variable features
  x <- ScaleData(x, verbose = FALSE)
})

Part II: Perform an integrated analysis using scMC

We have wrritten a Seurat Wrapper function RunscMC to simply run the integrated analysis via scMC.

future::plan("multiprocess", workers = 4)
options(future.rng.onMisuse="ignore")
combined <- RunscMC(object.list) # return an updated Seurat object with a new reduced dimensional space named "scMC"

scMC outputs a shared reduced dimensional embedding of cells that retains the biological variation while removing the technical variation. This shared embedding can be used for a variety of single cell analysis tasks, such as low-dimensional visualization, cell clustering and pseudotemporal trajectory inference.

Part III: Run the standard Seurat workflow for downstream analyses such as clustering and visualization

We can now run standard Seurat workflow for downstream analyses. Users only need to make one change to your code, i.e., using the scMC embeddings instead of PCA.

Run the standard workflow for clustering

nPC = 40
combined <- FindNeighbors(combined, reduction = "scMC", dims = 1:nPC)
# scMC uses Leiden algorithm for clustering by default. However, users can also use other algorithms inplemented in Seurat
combined <- FindClusters(combined, algorithm = 4, resolution = 0.05)
combined <- BuildClusterTree(combined, reorder = T, reorder.numeric = T, verbose = F)
combined <- RunUMAP(combined, reduction='scMC', dims = 1:nPC)

Quick visualization of cells onto the low-dimensional space

DimPlot(combined, reduction = "umap", group.by = c("sample.name","ident"))

Visualization and annotation

Feature plot of known marker genes

features = c('Lox','Ptch1','Cd68','Pecam1','Myh11','Plp1')
FeaturePlot(combined, features = features, ncol = 3)

Annotation

new.cluster.ids <-c("Immune", "Hh-inactive Fib", "Hh-active Fib",  "Endotheial", "Schwann","Muscle")
names(new.cluster.ids) <- levels(combined)
combined <- RenameIdents(combined, new.cluster.ids)
new.order <- c("Hh-inactive Fib", "Hh-active Fib","Immune","Endotheial", "Muscle", "Schwann")
[e-mail] <- factor([e-mail], levels = new.order)

Visualize cells onto the low-dimensional space

DimPlot(combined, reduction = "umap", group.by = c("sample.name", "ident"))
# Split the plot into each dataset
DimPlot(combined, reduction = "umap", split.by = "sample.name")

Heatmap of marker genes

combined <- ScaleData(combined, feature = rownames(combined), verbose = FALSE)
markers <- FindAllMarkers(combined, only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0.25)
top10 <- markers %>% group_by(cluster) %>% top_n(n = 10, wt = avg_log2FC)
DoHeatmap(combined, features=top10$gene)

Violin plot

Stacked violin plot

features = c('Lox','Ptch1','Cd68','Pecam1','Myh11','Plp1')
gg <- StackedVlnPlot(combined, features = features, colors.ggplot = T)
gg

Splitted violin plot

features = c('Lox','Ptch1','Cd68','Pecam1','Myh11','Plp1')
gg <- StackedVlnPlot(combined, features = features, split.by = "sample.name", colors.ggplot = T)
gg

Part IV: Downstream analysis

combined$clusters.final <- Idents(combined)

Compute the cellular compositions

gg1 <- computeProportion(combined, x = "clusters.final", fill = "sample.name")
gg2 <- computeProportion(combined, x = "sample.name", fill = "clusters.final")
gg1 + gg2

Save data

combined <- ScaleData(combined) # only store the highly variable genes
save(combined, file = "demo_scMC_dermis.RData")

amsszlh/scMC documentation built on Jan. 2, 2021, 1:51 p.m.