[BioC] Dose response linear correlation using Limma

[James W. MacDonald]

Hi Christian,

If you used a previous version of limma the first time, then this might 
be the case.

Older version of limma:

 > library(limma)
 > dha <- rnorm(10)
 > dat <- matrix(rnorm(10000), ncol=10)
 > design <- model.matrix(~dha)
 > fit <- lmFit(dat, design)
 >
 > fit2 <- eBayes(fit)
 > topTable(fit2)
     X.Intercept.        dha        F      P.Value adj.P.Val
399  -1.16703330  0.9726376 6.951548 0.0009571525 0.5130698
556  -1.24250007  0.2182220 6.881951 0.0010261396 0.5130698
305   0.02123802 -1.1632224 5.965466 0.0025658479 0.7115675
496  -0.97817815 -0.2335099 5.776557 0.0030993670 0.7115675
835  -1.00327277  0.9266638 5.530451 0.0039641990 0.7115675

Current version of limma:

 > library(limma)
 > dha <- rnorm(10)
 > dat <- matrix(rnorm(10000), ncol=10)
 > design <- model.matrix(~dha)
 > fit <- lmFit(dat, design)
 > fit2 <- eBayes(fit)
 > topTable(fit2)
Removing intercept from test coefficients
          logFC     AveExpr         t      P.Value adj.P.Val         B
281  1.0398797 -0.49364355  3.724635 0.0001969434 0.1969434 -1.930158
78   0.9062016 -0.05566960  3.245828 0.0011759163 0.5879581 -2.637635
149  0.8217517 -0.06901555  2.943345 0.0032561888 0.9545421 -3.034482
52   0.7695025  0.21765738  2.756199 0.0058609793 0.9545421 -3.260587
394  0.7458264 -0.01242832  2.671397 0.0075689249 0.9545421 -3.358155


Since the intercept is (when analyzing microarrays) an uninteresting 
coefficient, it is now automatically removed, as the message above 
notes. And the F-statistic that used to be computed when you didn't 
specify a contrast (in the older versions of limma) is not testing 
something useful.

Best,

Jim



On 7/10/2014 10:20 AM, Christian De Santis wrote:
> Dear all,
>
> i tried to extract genes that are linearly correlated to a dose
> response in a dietary experiment using the following script.
>
> dm <- model.matrix(~ DHA, data=targets) fit <- lmFit(MA.list,dm) fit2
> <- eBayes(fit) table <- topTable(fit2,coef=2, adjust.method="none",
> number = 10000, p.value = 0.05)
>
> the first time i did it i had in the output two columns with the
> intercept and the slope but i did it again and now I do not get them
> anymore. I can't figure out why as i dont think i have changed
> anything, but obviously i must have. Just to confirm however, is the
> slope equivalent to the fit2$coefficients values?
>
> Thanks, Christian
>

-- 
James W. MacDonald, M.S.
Biostatistician
University of Washington
Environmental and Occupational Health Sciences
4225 Roosevelt Way NE, # 100
Seattle WA 98105-6099