I have been using DESeq2 for a while. works great. I am running into trouble with a custom reference.

I am using DESeq2 to normalize transcript counts. I noticed that about half of the transcripts are getting dropped. Using print statements and the debugger I think there maybe be a problem with my tx2Gene data frame. I think it matches the salmon quant.sf files. using debug(), browse(), trace() I am able to step into tximport(). I am not able to step into sumarizeToGene() to figure out where the match goes wrong.

I tried playing around with the ignoreAfterBar and txOut arguments to tximport() v 1.17.3

When I run salmon I am using a custom reference. The counts have basically to formats

first format example

ENST00000450305.2|ENSG00000223972.5|OTTHUMG00000000961.2|OTTHUMT00000002844.2|DDX11L1-201|DDX11L1|632|transcribed_unprocessed_pseudogene|   632 386.000 0.000000    0.000

second format example. This is the custom part of our reference.

hg38_rmsk_ALR/Alpha_range=chr22_KI270739v1_random:50071-73985_5'pad=0_3'pad=0_strand=-_repeatMasking=none   23915   23653.000   0.000000    0.000

my tx2Gene file has entries that match both formats. It may not be easy to see but this file has 2 columns, txid, gene. The txid is really long. for ENST transcript we use the entire salmon quant.sf 'name' column. I suspect that summarizeToGene make some sort of assumption about the txid that I am violating.

grep ENST00000456328 gencode.v35.ucsc.rmsk.tx.to.gene.csv 
ENST00000456328.2|ENSG00000223972.5|OTTHUMG00000000961.2|OTTHUMT00000362751.1|DDX11L1-202|DDX11L1|1657|processed_transcript|,DDX11L1

$ grep  hg38_rmsk_ALR/Alpha_range=chr22_KI270739v1_random:50071-73985_5\'pad=0_3\'pad=0_strand=-_repeatMasking=none gencode.v35.ucsc.rmsk.tx.to.gene.csv 
hg38_rmsk_ALR/Alpha_range=chr22_KI270739v1_random:50071-73985_5'pad=0_3'pad=0_strand=-_repeatMasking=none,ALR/Alpha

for unknown reasons all the transcripts that start with ENST are dropped.

attributes(txi$abundance)[2]$dimnames[[1]]

values are the gene name column, gene names like DDX11L1 are missing. That is to say, transcripts that started with ENST did not match

I have run tximport() on the same salmon quant.sf file using a tx2gene file that only supported the ENST like entries. If I set ignoreAfterBar=TRUE, it works as expected.

Any idea how I can debug this?

Kind regards

Andy

txi <- tximport(allSalmonQuantFiles, 
                  type="salmon", 
                  tx2gene=transcriptMapperDF, 
                  ignoreAfterBar=ignoreAfterBar,
                  txOut=FALSE)

With tx2gene, you are summarizing transcripts to gene level, so there should not be ENST in the output. I'm confused by what you mean that transcripts are missing, because they should be missing (txOut=FALSE).