library("SCP.replication")
library("tidyverse")
library("data.table")
report <- fread("~/PhD/.localdata/SCP/brunner2022/20210919_DIANN_SingleCellOutput.tsv")
## Build the annotation table from the run names
annot <- DataFrame(File.Name = unique(report$File.Name))
annot$Run <- gsub("^D.*[\\]|[.]d$", "", annot$File.Name)
otherVars <- strsplit(annot$Run, "_")
otherVars <- lapply(otherVars, function(x) {
if (length(x) == 12) x <- x[-4]
x
})
otherVars <- do.call(rbind, otherVars)
colnames(otherVars) <- c("Date", "MsInstrument", "Purification", "User",
"SampleAnnotation", "SampleType", "CellCycleStage",
"..undetermined..", "PlatePosition", "CellNumber",
"RunID")
annot <- cbind(annot, otherVars)
## Format to a QFeatures object
brunner2022 <- readSCPfromDIANN(annot, report)
## The protein data
pgTable <- read.delim("~/PhD/.localdata/SCP/brunner2022/20210919_DIANN_SingleCellOutput.pg_matrix.tsv",
check.names = FALSE)
prots <- readSingleCellExperiment(pgTable, ecol = unique(report$File.Name),
fnames = "Protein.Names")
prots <- prots[rownames(prots) != ""]
rownames(prots) <- make.unique(rownames(prots))
from <- names(brunner2022)
brunner2022 <- addAssay(brunner2022, prots, "proteins")
brunner2022 <- addAssayLink(brunner2022,
from = from,
to = "proteins",
varFrom = rep("Protein.Group", length(from)),
varTo = "Protein.Group")
# Save data as Rda file
save(brunner2022,
file = "~/PhD/.localdata/scpdata/brunner2022.Rda",
compress = "xz",
compression_level = 9)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.