Question

Complicated paired multi factor design

0

Entering edit mode

Vladimir Krasikov ▴ 90

@vladimir-krasikov-5097

Last seen 5.1 years ago

Hello Experts!

I encountered somewhat difficult experimental design.

Briefly, blood cells from 2 individuals where treated with 2 different compounds (this compounds supposedly have to force cells switch their type from Phen1 to Phen2). Afterwards cells of different phenotypes were physically separated and sequenced. So this groups are paired, as well as each Treatment paired as applied to the same individual. So design is kind of paired in two-dimensions.

exp.design file:

Sample	ind	phenotype	treatment
sample1	1	Phen1	Comp1
sample2	1	Phen1	Comp2
sample3	2	Phen1	Comp1
sample4	2	Phen1	Comp2
sample5	1	Phen2	Comp1
sample6	1	Phen2	Comp2
sample7	2	Phen2	Comp1
sample8	2	Phen2	Comp2

I found nothing better than to create combined Phenotype and Treatment into one Group (as suggested in edgeR):

Ind <- factor(exp.design$ind)

Phenotype <- factor(exp.design$phenotype, levels = c('Phen1', 'Phen2'))

Treatment <- factor(exp.design$treatment, levels = c('Comp1', 'Comp2'))

Group <- factor(paste(exp.design$phenotype, exp.design$treatment, '.',

                       levels = c('Phen1.Comp1', 'Phen1.Comp2', 'Phen2.Comp1', 'Phen2.Comp2'))

Exp.Design <- data.frame(Group, Ind)

and finally

model.matrix( ~ Ind + Group, Exp.Design) to account for paired samples.

I am interested in all types of comparisons:

Phen1.Comp1 vs Phen1.Comp2 - difference in treatments in Phen1

Phen2.Comp1 vs Phen2.Comp2 - difference in treatments in Phen2

Phen1.Comp1 vs Phen2.Comp1 - difference in Phenotypes caused by Comp1

Phen1.Comp2 vs Phen2.Comp2 - difference in Phenotypes caused by Comp2

Is this approach valid?

Any suggestions and remarks are welcome

deseq2 multiple factor design • 1.6k views

ADD COMMENT • link 7.5 years ago Vladimir Krasikov ▴ 90

0

Entering edit mode

Now if I want to add to the experiment another compound (Comp3), how can I extract genes responded differently to the Comp2 and Comp3 related to the Comp1 in first Phenotype?

ADD REPLY • link 7.5 years ago Vladimir Krasikov ▴ 90

0

Entering edit mode

results(dds, contrast=c("group","Phen1.Comp3","Phen1.Comp2"))

This has been asked a few times on the support site, but it would be hard to search for. The Comp1 as baseline for the fold change drops out when you compare Comp3 and Comp2:

(Comp3/Comp1) / (Comp2/Comp1) = Comp3/Comp2

ADD REPLY • link 7.5 years ago Michael Love 42k

score 1 · Answer 1 · 2017-05-17

1

Entering edit mode

Michael Love 42k

@mikelove

Last seen 1 day ago

United States

hi,

You never mention what kind of comparisons you are interested in. That will help inform the easiest way to set up design and extract results.

ADD COMMENT • link 7.5 years ago Michael Love 42k

0

Entering edit mode

I updated initial post to reflect what comparisons I am interested in

ADD REPLY • link 7.5 years ago Vladimir Krasikov ▴ 90

1

Entering edit mode

In that case your design is the right one, you can just use:

design(dds) <-  ~ Ind + Group
dds <- DESeq(dds)
res <- results(dds, contrast=c("group","Phen1.Comp2","Phen1.Comp1"))

Note that I put Comp 2 first, then Comp 1. This will give you a comparison of 2 / 1, which is more typical than 1 / 2. Usually the first condition goes in the denominator, e.g. control which is taken as a "reference".

ADD REPLY • link 7.5 years ago Michael Love 42k

0

Entering edit mode

Thanks a lot,

this is what I expected

ADD REPLY • link 7.5 years ago Vladimir Krasikov ▴ 90