[BioC] Differential expression in time series

[James W. MacDonald]

Hi Sanjat,

Sanjat Kanjilal wrote:
> James W. MacDonald <jmacdon at ...> writes:
> 
>>>   I have two strains (B and W) and two different time points (2 and 1 
> hours). I am trying to find genes which
>> respond differently:
>>>   1. in B vs W, when comparing time points 1 and 2 (i.e. I compare BvsW in 
> time point 1 to BvsW in time point 2)
>>>   2. in time in different strains (i.e. I compare B1vsB2 versus W1vsW2) 
>>>    
>>>   I am getting the same results, or they should be the same?
>>>   Could you comment on whether this is a right way to analyse time series?
>> It appears you want to test for the interaction between time and strain. 
>>   Both of your points (1 and 2 above) are essentially identical. You 
>> want to know what genes react differently over time in the two strains 
>> (and worded differently, but meaning the same thing - what genes react 
>> differently between the strains at different times). These two things 
>> are verbo-algebraically the same (yup, new word. My next move is to 
>> start a new Wikipedia entry describing exactly what it means ;-D)
>>
>>>    
>>>   My code is below.
>>>    
>>>    
>>>   Thank you,
>>>   Lev.
>>>    
>>>   > temp<-rma(data)
>>>   > targets <- readTargets("Targets.txt")
>>>
>>>> lev <- c("W.1","B.1","W.2","B.2")
>>>> f <- factor(targets$Target, levels=lev)
>>>> design <- model.matrix(~0+f)
>>>> colnames(design) <- lev
>>>> fit <- lmFit(temp, design)
>>>> cont.dif <- makeContrasts(Diff.Time=(B.2-W.2)-(B.1-W.1), Diff.Strain=(B.2-
> B.1)-(W.2-W.1), levels=design)
>> As the statements above are verbo-algebraically the same, these two 
>> terms are algebraically the same.
>>
>> (B.2-W.2)-(B.1-W.1) = (B.2-B.1)-(W.2-W.1)
>>
>> Hence you should get the same results from each contrast. And yes, this 
>> contrast does give you the interaction.
>>
>> Best,
>>
>> Jim
>>
> Thanks for your helpful explanation Jim.
> 
> As a follow-up question:
> 
> Can the (B.2-W.2) - (B.1-W.1) contrast be thought of as comparing B @ time 2 
> to B @ time 1 while 'normalizing' to the effects of the wild type (ie I'd like 
> to subtract out the effects of the control group)?  If not, is there a term 
> that would do that?

Yes, you can think of it that way. That's the same idea behind a paired 
t-test. Each individual might have an inherently different baseline (the 
wt in this case), and you are just interested in seeing if the treatment 
has the same relative effect in both samples. Since the difference in 
baseline between samples is probably not interesting, you subtract it out.

Best,

Jim


> 
> Thanks,
> Sanjat
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor

-- 
James W. MacDonald, M.S.
Biostatistician
Affymetrix and cDNA Microarray Core
University of Michigan Cancer Center
1500 E. Medical Center Drive
7410 CCGC
Ann Arbor MI 48109
734-647-5623