[BioC] Linear models?

[Khadeeja Ismail]

Gordon K Smyth <smyth at ...> writes:

> 
> Dear Hajja,
> 
> Your targets frame (in limma) needs to contain three columns: Pair (taking 
> values A, A, B, B, C, C etc), Treatment, before or after (taking values 
> 1,2,1,2 etc) and Treatment length.


Thanks Gordon,

I was actually working on methylation data from 11 pairs.

I followed your instructions and created the targets file and design matrix but
then it gives me a coefficient for every pair in my data.

#####My targets file:######
Pair	Tr	TLength
"T1	1	27.6
"T1	2	21.9
"T2	2	22.2
"T2	1	26.6
...
"T11	2	24.5
"T11	1	28.4



Pair <- factor(Targets$Pair)
Treatment <- factor(Targets$Tr)
design <- model.matrix(~Pair + Treatment + Targets$TLength)


##OUTPUT##################################################################

         ID           X.Intercept.   PairT1          PairT2        ...
43676  cg04570604    -4.944408     -0.09987055       0.2324025     ...
49660  cg05227111     3.895301     -0.17053184      -0.2422336     ...
...
      
           PairT11      TreatmentL  Targets.TLength
43676     -0.23353534   -0.1040837  -0.03335663 
49660      0.16514871    0.3153623   0.07148090
...
              F      P.Value    adj.P.Val
43676  1232.021 1.639072e-33 3.387964e-29
49660  1175.197 3.069305e-33 3.387964e-29
...
############################################################################

Initially I did a one sample t-test on the differences between the pairs (done
by hand and submitted to limma), ignoring the differences in treatment length.
I'm wondering if I can do the same but taking into account the treatment length.
My current objective is to find out the probes that show significant differences
in methylation between pairs. 

Thanks once again for your reply. 

Best regards,
Hajja






> 
>   Pair <- factor(Pair)
>   Treatment <- factor(Treatment)
>   design <- model.matrix(~Pair+Treatment+Treatment:Length)
> 
> The last coefficient tests when the differences in expression are 
> correlated with treatment length.
> 
> Best wishes
> Gordon
> 
> > Date: Tue, 12 Jul 2011 03:57:12 -0700 (PDT)
> > From: khadeeja ismail <hajjja at ...>
> > To: bioconductor at ...
> > Subject: [BioC] Linear models?
> > Message-ID:
> >
> > Dear All,
> > �
> > I have a small problem that is silimar to the case below, and would really
appreciate if someone could give
> me some ideas.
> > �
> > I am doing a�pairwise analysis for some samples (say 8 pairs, A1, A2, B1,
B2, C1, C2,...H1, H2) using linear
> models, to find out gene�expression difference between�before treatment and
after treatment. But if
> the length of the treatment is different for every pair, how can I include�the
treatment length�in my analysis?
> > �
> > If one objective of my study is to see if the difference in gene expression
correlates with the treatment
> length, will adjusting the expression differences relative to the highest
treatment length introduce a bias?
> > �
> > Thanks in advance,
> > Hajja
> 
> 
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at ...
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives:
http://news.gmane.org/gmane.science.biology.informatics.conductor