[BioC] scale questions

[Sean Davis]

On Wed, Oct 22, 2008 at 8:16 PM, Hui-Yi Chu <huiyi.chu at gmail.com> wrote:
> Hi Sean,
>
> Yes, they are replicate in ratios. In other words, these ratios are from
> untreatment and treatment of wt1, wt2, mut1,mut2, so total is 8 ratios as
> below.
>
> untreatmen wt1, wt2, mut1, mut2
> treatment   wt1, wt2, mut1, mut2
>
> And I wanna get the second values like:
> r1: untreatment  mut1/wt1
> r2: untreatment  mut2/wt2
> r3: treatment     mut1/wt1
> r4: treatment     mut2/wt2
>
> Now we have 4 ratios. And then I will compare r3 vs r1, r4 vs r2 to get most
> fold changes genes. (I know I need three replicates, but  I cannot convince
> my adviser, therefore, this is the strategy I can use so far. ) So the
> question is should I scale the 8 ratios from wt and mut separately (twice)
> or together(once) before I get the four values?

Hi, Hui-Yi.

You should not do any scaling.  You can use the log fold changes
directly.  You seem to understand that this is not an optimal design,
so I won't belabor that point.

Sean

>
> On Wed, Oct 22, 2008 at 7:51 PM, Sean Davis <sdavis2 at mail.nih.gov> wrote:
>>
>> On Wed, Oct 22, 2008 at 5:39 PM, Hui-Yi Chu <huiyi.chu at gmail.com> wrote:
>> > Dear List,
>> >
>> > I think this may be a simple question for you but I wanna make it sure
>> > for
>> > further steps.
>> > I have already done some of data pre-processing procedures for my
>> > affymetrix
>> > yeast2 arrays. My next step is to get *ratios* from various conditions
>> > in wt
>> > and mutant following by fold-change comparison. So my question is which
>> > step
>> > I should scale my dataframe for comparison?
>> >
>> > Here are parts of my codes (codes with underline are the questions):
>> >
>> > wt.pt.f1 <- exprs(esetsub[, 1])- exprs(esetsub[, 17])
>> > wt.pt.f2 <- exprs(esetsub[, 2])- exprs(esetsub[, 18])
>> > wt.pt.f <- cbind(wt.pt.f1, wt.pt.f2)
>> > wt.pt.f <- new("ExpressionSet", exprs= as.matrix(wt.pt.f))
>> > *wt.pt.f <- scale(exprs(wt.pt.f))      ### not sure
>> >
>> > *mut.pt.f1 <- exprs(esetsub[, 9])- exprs(esetsub[, 21])
>> > mut.pt.f2 <- exprs(esetsub[, 10])- exprs(esetsub[, 22])
>> > mut.pt.f <- cbind(mut.pt.f1, mut.pt.f2)
>> > mut.pt.f <- new("ExpressionSet", exprs= as.matrix(mut.pt.f))
>> > *mut.pt.f <- scale(exprs(mut.pt.f))    **### not sure*
>> >
>> > gg      <- cbind(wt.pt.f, mut.pt.f)
>> > *gg      <- scale(gg)*              *### not sure*
>> > pt.f1   <- gg[,3]-gg[,1]
>> > pt.f2   <- gg[,4]-gg[,2]
>> > gg1     <- cbind(gg, pt.f1, pt.f2)
>> > gg2     <- new("ExpressionSet", exprs=as.matrix(gg))
>> > ....... followed by further steps
>> >
>> >
>> > As seen above, where should I scale the ratios? Scale wt and mut
>> > separately
>> > or together before getting pt.f1 and pt.f2 ratios??
>> > Many many thanks!!!!!
>>
>> I may be misunderstanding, but why do you want to scale the log
>> ratios?  Generally, you would not scale them at all.  And I really
>> cannot tell why you are forming ratios in the first place?  Do you
>> have replicates of any kind?
>>
>> Sean
>
>