Question

Filter Out Low Read Counts

0

Entering edit mode

pthom010 • 0

@pthom010-23694

Last seen 2.3 years ago

United States

I ran a time course with a resistant and a susceptible genotype over 22 days (0 dpi, 1 dpi, 7 dpi, 14 dpi, 22 dpi). I have a dds object from DESeq2 made from RSEM for all thirty time points. I would like to filter out lowly expressed genes by rowMeans, but want to do so in two batches (filter out the low row means for all of the susceptible samples, then the resistant ones). I should note that the two groups of 15 are separated by a condition (Genotype) included in the DESeqDataSetFromTximport function. Would I be able to just run a subset function on the two genotypes?

deseq2 r • 2.3k views

ADD COMMENT • link updated 4.5 years ago by Michael Love 43k • written 4.5 years ago by pthom010 • 0

score 0 · Answer 1 · 2020-06-16

0

Entering edit mode

Michael Love 43k

@mikelove

Last seen 5 hours ago

United States

I wouldn't use a filter that takes into account the conditions at all. If you want to pre-filter, i would recommend just filtering such that each row has x or more samples with a count of 10 or more for example.

keep <- rowSums(counts(dds) >= 10) >= x
dds <- dds[keep,]
# then run DESeq()

ADD COMMENT • link 4.5 years ago Michael Love 43k

0

Entering edit mode

Thanks, Michael. After filtering, I ended up with this error:

>dds = DESeq(dds)
Error in estimateSizeFactorsForMatrix(counts/normMatrix, locfunc = locfunc,  :
  every gene contains at least one zero, cannot compute log geometric means

Would you have any insight on how I could resolve said error? Thank you.

ADD REPLY • link 4.5 years ago pthom010 • 0

0

Entering edit mode

This reflects probably some aspects that make DESeq2 model not appropriate. Why are there no genes that are expressed for all samples?

ADD REPLY • link 4.5 years ago Michael Love 43k

0

Entering edit mode

The RSEM values were assembled through Trinity. I created my DDS object using transcript values. Might that explain it?

ADD REPLY • link 4.5 years ago pthom010 • 0

0

Entering edit mode

Eyeball your gene counts. Usually that error means you have a few samples that have zero reads for nearly every gene.

ADD REPLY • link 4.5 years ago swbarnes2 ★ 1.4k

0

Entering edit mode

Thanks. Not entirely sure how I should progress. Do you think I should filter those transcripts with gene counts lower than 1 in any column?

ADD REPLY • link 4.5 years ago pthom010 • 0

0

Entering edit mode

Swbarnes2 said "samples".

Take a look at the column sum of the counts. colSums(counts(dds))

ADD REPLY • link 4.5 years ago Michael Love 43k