venessa's worflow

A workflow for venessa chin

This should look like a hack version of those fancy seurat tutorials that i told you to look at

getting onto the cluster and firing up R

cluster # mounts cluster
d1 # connect to dice
brain # get an interactive compute session
scr # currently goes to /share/ScratchGeneral/briglo/scRNA
cd venchi
R

an R cheatsheet

library(Seurat) #loads the library Seurat, required for basically everything
library(venchi) #loads the functions i wrote for you
setwd("NEW/PATH") #move to a new directory
ls() # gives a list of all objects currently loaded (in case you forgot what name it was)
load("PATH/TO/SEURAT.rdata") #loads the seurat object, takes a while
head(integrated@meta.data) # shows the metadata columns of the seurat bject you can group by
#this will be useful when you're playing in the cluster
pdf('test.pdf',width=36,height=16) #makes a pdf  of set dimensions to plot into, good if hard to see...
dev.off() #stops whatever graphics device R is plotting to, necessary to view pdfs

reading in files

there are two ways really

1. manually genelist=c("GZMA","GZMB","GZMH","GZMK","GZMM","PFN1","FASLG","TNFSF10")
2. from files (dependent on type) read.table("FILENAME", header=T,stringsAsFactors=F) #pretty common, will read a text file saved from an excel doc

simple seurat functions

• the heatmap DoHeatmap(integrated,group.by="orig_final.ident",features=genelist,disp.max=200)
• The violin plot VlnPlot(integrated,split.by="orig.ident",group.by="orig_final.ident",features=genelist,pt.size=.1)
• the dot plot DotPlot(integrated,features=genelist,group.by="orig_final.ident",do.return=T) + theme(axis.text.x = element_text(angle = 90))
• the UMAP plot UMAPPlot(integrated,group.by='orig_final.ident',label=T)
• add a meta score integrated<-AddModuleScore(integrated,features = genelist, name="my_meta_score")
• plot expression of something in UMAP space FeaturePlot(integrated,reduction="umap",features=genelist,split.by=NULL)) or FeaturePlot(integrated,reduction="umap",features='my_meta_score',split_by="orig_final.ident"

some of the "bespoke functions"

• The kinds of comparisons we were talking about counting

#for a simple single gene, single cutoff in all cells
countTab<-countByCut(integrated=integrated,geneName="PFN1",expCut=2,groupBy="orig_final.ident",splitBy="orig.ident")

#for a more complicated mulltigene, multi cutoffs
tab<-data.frame(gene=c("PFN1","GZMA","GZMB"),cuts=c(2,0,0))
#adds new meta.data columns containing info of interest, can be used like below in Reactome Analysis
integrated<-annotateByCuts(seuratObj=integrated,geneTab=tab,groupBy="orig_final.ident",groupID="CD8",splitBy="orig.ident")

outTab #a table of counts for cells matching groupID FYE

• Reactome Analysis e.g.

#for all groupings made by a particular metadata column
markers<-makeReactomePipe(integrated,"SCT_snn_res.0.8")
for (i in 1:length(markers$CP_result)) dotplot(markers$CP_result[[i]]) + ggtitle(names(markers$CP_result)[i])

#for combining two metadata columns and directly comparing two groups(split by something else if you want)
#first extract the relevant markers and universe
enrichment<-makeSpecificMarkers(integrated,groupBy_1="orig_final.ident",groupID_1="CD8 T-cell 1",groupBy_2="orig.ident",groupID_2=c("ACITE","BCITE"),splitBy=NULL,splitID=NULL)
out<-makeReactomeForMarkers(enrichment$markerLists,enrichment$hasEntrez)
dotoplot(out)

bits and bobs

• terminal is your friend
• we need to make sure you are running a "current" version of R R --version
• youll need to make a RStudio project somewhere on your laptop
• you'll need to get the r object pandora/volumes/Cancer/brian\ gloss/190531_integrated.rdata and put it in your project path
• we'll need to install a bunch of R libraries, plus maybe some others that ive forgotten

install.packages("BiocManager")
library(BiocManager)
install(c("devtools","Seurat","dplyr","biomaRt","ggplot2","clusterProfiler","ReactomePA","igraph","cowplot"))
devtools::install_github("https://github.com/briglo/venchi_lung.git")

• CellPhoneDB e.g.

ti<-subsetSeurat2cellPhone(seuratObj=integrated,annoColumn="SCT_snn_res.0.15",no.cells=50,prefix="small")

#running cellphoneDB
module load briglo/miniconda/3
source activate cellphone
qsub -V -cwd -b y -j y -pe smp 8 -N cpdb_1 "cellphonedb method statistical_analysis small_meta.txt small_counts.txt --project-name small --threshold 10 --threads 8"
#EASY!!!

#back in R
cd("PATH/TO/CELLPHONEDB/OUT/small")
cellphoneDB_data<-preprocCellphone(varval=0,pval=.05)
intgraph(cellphoneDB_data$countdat, scoreCut = 0.3, numberCut = 0, numberSplit = 35)

setting up: for later

getting onto the cluster

so first you want to login to the cluster this was originally done by doing this really

this website is where our group put stuff to get on the cluster

getting started

1. once you are done, login
2. make an interactive HPC job qrsh -pe smp 4 -l mem_requested=20G
3. navigate to the results folder you need cd /share/ScratchGeneral/PATH
4. setup your R environment module load briglo/R/3.6.0
5. Fire up R R
6. Set up tools you need

library(scFuncs) 
library(reticulate)
use_python("/share/ClusterShare/software/contrib/briglo/miniconda3/envs/magic/bin/python")

7. load your seurat object load("PATH/TO/SEURAT.rdata")