DESeq - estimateDispersion doubt

[alisrpp]

Hi,

I'm using DESeq for the first time and at the step of variance estimation i'm obtaining very different plots from the manual and i have no idea about how to interpretate them.

My data are: 2 different conditions with 2 replicates each

              condition    libType
C4                   SP paired-end
CRL_2APR10           SP paired-end
CRL_1_15JUL11        EB paired-end
CRL_2_15JUL11        EB paired-end

'data.frame':	307048 obs. of  4 variables:
 $ C4           : int  5 7 0 2 3 0 0 0 3 0 ...
 $ CRL_2APR10   : int  0 0 0 0 0 0 0 0 0 0 ...
 $ CRL_1_15JUL11: int  0 0 0 0 0 0 0 0 0 10 ...
 $ CRL_2_15JUL11: int  0 0 0 0 0 0 0 0 0 3 ...

My size factors are:

C4    CRL_2APR10 CRL_1_15JUL11 CRL_2_15JUL11 
    2.6549225     0.8623452     1.3987823     0.3507545

If i run:

> cds <- estimateDispersions( cds, method="per-condition", 
+                             sharingMode="maximum", 
+                             fitType="local")
> plotDispEsts <- function( cds )
+ {
+    plot(
+       rowMeans( counts( cds, normalized=TRUE ) ),
+       fitInfo(cds, name="SP")$perGeneDispEsts,
+       pch = '.', log="xy" )
+    xg <- 10^seq( -.5, 5, length.out=300 )
+    lines( xg, fitInfo(cds, name="SP")$dispFun( xg ), col="red" )
+ }
> plotDispEsts(cds)
Warning messages:
1: In xy.coords(x, y, xlabel, ylabel, log) :
  115300 x values <= 0 omitted from logarithmic plot
2: In xy.coords(x, y, xlabel, ylabel, log) :
  30294 y values <= 0 omitted from logarithmic plot

my dispersion plot looks like:

plotDispEsts_fitInfo(cds,name=SP).jpeg (see attach.)

Doing the same by fitInfo(cds, name="EB"):

plotDispEsts_fitInfo(cds,name=EB).jpeg (see attach.)

And just running:

> cds <- estimateDispersions( cds, 
+                             sharingMode="maximum", 
+                             fitType="local")
> plotDispEsts <- function( cds )
+ {
+    plot(
+       rowMeans( counts( cds, normalized=TRUE ) ),
+       fitInfo(cds)$perGeneDispEsts,
+       pch = '.', log="xy" )
+    xg <- 10^seq( -.5, 5, length.out=300 )
+    lines( xg, fitInfo(cds)$dispFun( xg ), col="red" )
+ }
> plotDispEsts(cds)
Warning messages:
1: In xy.coords(x, y, xlabel, ylabel, log) :
  115300 x values <= 0 omitted from logarithmic plot
2: In xy.coords(x, y, xlabel, ylabel, log) :
  28498 y values <= 0 omitted from logarithmic plot

My dispersion plot is:

plotDispEsts_fitInfo(cds).jpeg (see attach.)

I need some help to interpretate the plots.
Can anyone?

Thanks in advanced!

[muthu545]

Hi,
plotDispEst plots the dispersion vs rowCounts of DESeq object.
Usually as shown in the Vignette, the dispersions will be higher with lower row counts and go on to decrease with increasing row counts.

You have modified by the plotDispEst function, that might be one possible reason that
your plot did not meet the above explanation.

open new R file, load the DESeq package and run plotDispEsts, the output should be

------------------------------------------------
> plotDispEsts
function (cds, name = NULL, ymin, linecol = "#ff000080", xlab = "mean of normalized counts",
ylab = "dispersion", log = "xy", cex = 0.45, ...)
{
px = rowMeans(counts(cds, normalized = TRUE))
sel = (px > 0)
px = px[sel]
py = fitInfo(cds, name = name)$perGeneDispEsts[sel]
if (missing(ymin))
ymin = 10^floor(log10(min(py[py > 0], na.rm = TRUE)) -
0.1)
plot(px, pmax(py, ymin), xlab = xlab, ylab = ylab, log = log,
pch = ifelse(py < ymin, 6, 16), cex = cex, ...)
xg = 10^seq(-0.5, 5, length.out = 100)
lines(xg, fitInfo(cds, name = name)$dispFun(xg), col = linecol,
lwd = 4)
}
<environment: namespace: DESeq>
--------------------------------------

To run DispEst plot for the DESeq object, just run:
plotDispEsts(cds)

To run DispEst plot for any specific condition use the following:
plotDispEsts(cds, name='Cond')

Suggestion:
since you have only 2 replicates, you the following parameters:
cds <- estimateDispersions( cds,
method="pooled",
sharingMode="fit_only",
fitType="parametric")

Thanks
--
Muthu

[alisrpp]

Hi Muthu,

I'm trying to run your suggestion:

Suggestion:
since you have only 2 replicates, you the following parameters:
cds <- estimateDispersions( cds,
method="pooled",
sharingMode="fit_only",
fitType="parametric")

and this is what is happening to me:

> cds <- estimateDispersions( cds,
+                             method="pooled",
+                             sharingMode="fit-only", 
+                             fitType="parametric")
Error in parametricDispersionFit(means, disps) : 
  Parametric dispersion fit failed. Try a local fit and/or a pooled estimation. (See '?estimateDispersions')

Why are you suggesting me a "parametric" fitType instead of "local"? Do you think that i can use local without a problem?

Thanks.

[muthu545]

Hi,

I think you could try local fit, if it does not throw out error.

Are you using DESeq1.X version or DESeq2.X

Thanks
--
Muthu

[alisrpp]

Hi,

I'm using DESeq version 1.10.1

[alisrpp]

I'm still generating the same plot, any ideas about how to interpretate it?

Thanks.