Hello,

I have a fairly simple question that I know has been addressed many times: I want to import RSEM data into DESeq2 for modeling and DE. For reproducibility, the workflow is:

Unfortunately, it is costing me an inordinate amount of time, and I cannot find a perfectly analogous example either in any vignette (for DESeq2, tximport, tximeta, tximportData, etc) or in other posts.

Load libraries

libsNeeded<-c( "cBioPortalData", "DESeq2", "Tximport", "SummarizedExperiment")
BiocManager::install(libsNeeded, update = TRUE, ask = FALSE)
lapply(libsNeeded, library, character.only = TRUE)

Get a MultiAssayExperiment using cBioPortal

cbio <- cBioPortal()
getStudies(api = cbio)
studies <- getStudies(cbio, buildReport = TRUE)
CptacMae<-cBioDataPack("brain_cptac_2020", ask=FALSE)
CptacRnaSeqSE<-experiments(BrainCptac2020Mae)$mrna_seq_v2_rsem

Look at resulting MAE object

CptacMae # Returns, a MultiAssayExperiment with 6 listed experiments
CptacRnaSeqSE # returns a Summarized Experiment, and:
assays(CptacMae)$mrna_seq_v2_rsem # returns a DataFrame of 18,000 rows and 188 cols. 
row.names(CptacRnaSeqSE) # returns HUGO gene symbols - so this is **not** transcript level info
all(row.names(colData(CptacRnaSeqSE) ) == colnames(assays(CptacRnaSeqSE)[[1]])) # TRUE

Determine prior processing of Rna-Seq data by reading the metadata from cBioPortal:

cancer_study_identifier: brain_cptac_2020

stable_id: rna_seq_v2_mrna

profile_name: RNA expression

profile_description: Log2 of FPKM + 1 expression values. STAR v2.6.1d aligned + RSEM v1.3.1 quantification.

genetic_alteration_type: MRNA_EXPRESSION

datatype: CONTINUOUS

show_profile_in_analysis_tab: false

data_filename: data_mrna_seq_v2_rsem.txt

Of course, I cannot load these data directly with DESeqDataSetFromMatrix() because the values are decimal and it wouldn't be appropriate.

Using RSEM with tximport and DESeq2 provides a workflow using:

library(tximeta)
se <- tximeta(ColData, type="rsem", txIn=FALSE, txOut=FALSE, skipMeta=TRUE)
assays(se)$length[ assays(se)$length == 0] <- NA # set these as missing

but trying this on my data returns:

tximeta(CptacRnaSeqSE, type = "rsem", txIn = FALSE, txOut = FALSE, skipMeta = TRUE)

all(c("files", "names") %in% names(coldata)) is not TRUE

I do not have files, I am working from a SE directly. This would seem like it should be the easiest scenario on paper - it's already in the native language in an appropriate data object - but its proving to be more difficult than just downloading flat files would have been ...

What am I missing here? Thank you!

profile_description: Log2 of FPKM + 1 expression values. STAR v2.6.1d aligned + RSEM v1.3.1 quantification.

If the data is indeed log2(fpkm) then it is not suited for DESeq2 anyway. Note that the code does not run as such, there is no package called cBioPortal, tximport must be lowercase "t" and CptacMae<-cBioDataPack("brain_cptac_2020", ask=FALSE) throws an error for me that this studyId does not exist. Anyway, tximport requires the rsem output, you cannot run this with the data you have, but if it was raw gene level estimates you could round to nearest integer, this was suggested here on the support site on multiple occasions.

A DESeqDataSet is a SummarizedExperiment with a few extra slots. You should be able to do something like

se <- as(se, "DESeqDataSet")
design(se) <- <put correct design here>

Maybe you need to do other things, but your code is non-functional and I can't find the data you are using, so cannot tell you any more.