DESeq result interpretation

[huma Asif]

Dear ALL,
this is my first time i use DESeq for two RNA-seq samples control and affected
for one gene using exon data
id baseMean baseMeanA baseMeanB foldChange log2FoldChange pval padj
1 122.3801454 86.64640789 158.113883 1.824817518 0.867752202 0.227324022 1
2 82.8516747 73.99729725 91.70605214 1.239316239 0.30954437 0.701941312 1
3 52.65192304 46.80170937 58.50213671 1.25 0.321928095 0.736690545 1
4 106.726871 98.03060747 115.4231346 1.177419355 0.235628248 0.754103212 1
5 34.94316814 49.3315315 20.55480479 0.416666667 -1.263034406 0.271418019 1
6 17.7087549 25.93067681 9.486832981 0.365853659 -1.450661409 0.375252652 1
7 28.46049894 28.46049894 28.46049894 1 3.20E-16 1 1
8 24.82387963 22.76839915 26.87936011 1.180555556 0.239465935 0.873379607 1
9 13.43968006 18.97366596 7.90569415 0.416666667 -1.263034406 0.521556208 1
10 48.3828482 49.3315315 47.4341649 0.961538462 -0.056583528 0.999156423 1
11 436.2362032 532.527558 339.9448485 0.638361045 -0.647555479 0.274798212 1


i need to know what type of info i can get from this result
i will be obliged if somebody will give a feedback. it will encourage me to explore my results for whole transcriptome data

[dpryan]

It's unclear what you mean by "exon data" in this context, though I'm assuming that you mean "I used counts of the number of reads mapping to the exons as the counts for each gene".

The most important things to look at in such results are the adjusted p-values (the last column) and fold changes (the 5th or 6th column, depending on whether you prefer them log2 transformed). In your case, the snippet you posted contains no differentially expressed genes, at least when one uses a standard threshold for significance.

BTW, since you're starting out, you should use DESeq2, which has a number of improved features.

[velt]

I advise you to sort your files in ascending order on the "adjusted p-value" column. Thus, the most significant genes are in the beginning of file. Generally a p-value is significant if it is less than 0.05. Then, as stated by dpryan, refer you to the "log2 fold-change" column. More the value is high (or low, less than 0), more the change of expression is important.

[huma Asif]

thanks

yes you are right.
can you guide me where i can read more about interpretation.
i am not with statistical background so having problem in understanding in manual and paper they didnt write much.

[huma Asif]

in my case padjusted is all one and p value for any exon is not equal to 0.05
let me clear i am checking just one gene with 11exons

[dpryan]

If you want to look at differential exon usage then you should use DEXseq rather than DESeq/DESeq2. If you want to look at differential expression, then each gene should only have one count associated with it (i.e., you need to sum the exons (depending on how you did the counting)).

[velt]

If you want to learn more about DESeq or DESeq2, the documentation is very well done.

DESeq:

http://bioconductor.org/packages/release/bioc/vignettes/DESeq/inst/doc/DESeq.pdf

DESeq2:

https://bioconductor.org/packages/devel/bioc/vignettes/DESeq2/inst/doc/DESeq2.pdf

It is difficult with the little information that you give to explain the way to go. But the documentation is well, with the "pasilla dataset" that you can use in R to understand the different steps.

[huma Asif]

ok so you mean DESeq work with gene and DEXseq with exon
if yes then i will install DEXseq and try again
i will also try DESeq by summing the counts for every exon into one gene

[dpryan]

Yes, that's exactly what I mean.

[huma Asif]

Error

hi i tried to used DESeq with gene count and got this error after
countDataset<-estimateDispersions(countDataset,method='blind',sharingMode="fit-only")

tried this also but error persist
countDataset<-estimateDispersions(countDataset,method='blind')

Error in glm.fit(x = numeric(0), y = numeric(0), weights = NULL, start = c(0.1, :
object 'fit' not found
In addition: Warning messages:
1: In glm.fit(x = numeric(0), y = numeric(0), weights = NULL, start = c(0.1, :
no observations informative at iteration 1
2: glm.fit: algorithm did not converge

[dpryan]

Is there a reason you're using blind dispersion estimation? Aside from that, it looks like a row with 0 counts didn't get removed prior to testing.

BTW, as I mentioned before, I strongly recommend using DESeq2 instead.

[huma Asif]

for DEXSeq i downloaded Homo_sapiens.gtf from UCSC but it terminate everytime on chr15
from ensembl it is downloaded successfully but now prob is chr in bam and 1,2,... chromosome notation
can you guide how to make it compatible

[dpryan]

Regarding the Ensembl annotation, you could just use sed or awk to change the chromosome names (note that the mitochondria are chrM in UCSC and MT in Ensembl, so just appending "chr" won't fix everything).

Regarding the DEXSeq error, it would be helpful to see the command issued and the exact error message.

[huma Asif]

thanks i downloaded gtf from ucsc

well i am following instruction in
Inferring differential exon usage in RNA-Seq data with the DEXSeq package
done with
dexseq_prepare_annotation.py script and
python /path/to/library/DEXSeq/python_scripts/dexseq_count.py
Dmel_flattenend.gff untreated1.sam untreated1.counts

done with > sampleTable
countFile condition libType
A A.counts control single-end
B B.counts patient single-end


need help here
ecs <- read.HTSeqCounts(
+ sampleTable$countFile,
+ sampleTable,
+ "gene.gff" )

getting this error
Error: unexpected string constant in:
"sampleTable
"gene.gff""

[dpryan]

What version of DEXseq are you using? That entire method has been deprecated. Anyway, try:

flattenedfile="gene.gff"

instead of just "gene.gff".