Question

DESEQ2 blocking design doesn't match expectation !

0

Entering edit mode

akp • 0

@akp-8846

Last seen 9.1 years ago

I would like to estimate DE between a gene in a case and control, however, there are two underlying teams (A and B) (the experiment was run in two batches) so I would like to consider them as well.

the data format is sth like below and I have made sure, "condition" and "team" are factor

gene_in_sample	condition	team
4.1	control	team A
3.1	control	team B
4.8	case	team A
3.8	case	team B

but in reality there are 12 controls and 12 cases, which each are also split into half team A and team B

diagdds = phyloseq_to_deseq2(mydata, ~ condition + team)
diagdds = DESeq(diagdds, test="Wald", fitType="parametric")
res = results(diagdds, cooksCutoff = FALSE, contrast = c("condition", "A", "B"))
sigtab.deseq = res[which(res$padj < 0.05), ]

I wonder if the design should be sth like design = ~ condition + team

I am not sure, if it is right, since I can see a clear distinction in my PCA between these case/control so I expected to see some significant taxa - but by adding "+ condition", there is nothing significant.

I appreciate your comments,

deseq2 block pca • 2.1k views

ADD COMMENT • link 9.4 years ago akp • 0

0

Entering edit mode

Can you edit your question to say how many samples there are in each group/team, and also add column headings to your table?

ADD REPLY • link 9.4 years ago Ryan C. Thompson ★ 7.9k

0

Entering edit mode

Does it look better ?

ADD REPLY • link 9.4 years ago akp • 0

score 1 · Answer 1 · 2016-07-02

1

Entering edit mode

manasishah86 ▴ 30

@manasishah86-11034

Last seen 9.1 years ago

Hi akp,

Please make sure both the Treatment as well as Condition variables are factor variables (using as.factor)

dds$Treatment <- as.factor(dds$Treatment)

dds$Condition <- as.factor(dds$Condition)

That makes a difference with the results! Also make sure you are calling results correctrly:

deseq2res_comparison = results(DESeq2result, contrast = c("Condition", "case", "control"))

Besides that as Ryan said, reproducible code and proper titles will help.

Best,

Manasi

ADD COMMENT • link 9.4 years ago manasishah86 ▴ 30

0

Entering edit mode

Thanks, in fact it looks more reasonable - but are you sure, the contract should be like that but not the other way around ? ((I have updated my code with the contrast thing))

ADD REPLY • link 9.4 years ago akp • 0

0

Entering edit mode

Swapping "case" and "control" in that contrast will not change the p-values. It will simply switch the sign on the log fold change values.

ADD REPLY • link 9.4 years ago Ryan C. Thompson ★ 7.9k

0

Entering edit mode

Hi akp,

As Ryan said it will change the fold change signs.

if you do ?results in DESeq2 and look at the contrast section , the syntax is "name of the factor in the design formula, the name of the numerator level for the fold change, and the name of the denominator level for the fold change (simplest case)". So hence I am assuming case would be your numerator?

Anyhow, if my answer above was helpful, please do mark it as accepted. Would like to build me some point karma. ;)

Thanks,

Manasi

ADD REPLY • link 9.4 years ago manasishah86 ▴ 30