Could heatmap default to scale="none" ?

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Could heatmap default to scale="none" ?

Kevin Wright
(I have searched the email archives for discussions on this topic but
have found nothing.)

The help page for heatmap says:
By default (scale = "row") the rows are scaled to have mean zero and
standard deviation one. There is some empirical evidence from genomic
plotting that this is useful.

I offer two comments on that.

1. Since the heatmap function is general-purpose (that is, not in a
genomics package), wouldn't it be better to just use no scaling?

2. Why assume that the rows should be scaled instead of the columns?
There appears to be an assumption on the structure of the data.  Is
this assumption really warranted for a general-purpose function?

For the data I'm analyzing, scaling has hidden a great deal of
structure and I've just had to re-do many hours of work because I
didn't thoroughly read the help page and was not expecting the
scaling.  Further, given the structure of my data I did try (manually)
centering (not scaling) the COLUMNS, not the rows.

The heatmap.2 function in the gplots package defaults to scale="none"
and I argue that it would be better for the heatmap function to do the
same.

Kevin Wright

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
Reply | Threaded
Open this post in threaded view
|

Re: Could heatmap default to scale="none" ?

Earl F. Glynn
"Kevin Wright" <[hidden email]> wrote in message
news:[hidden email]...
> (I have searched the email archives for discussions on this topic but
> have found nothing.)
> 1. Since the heatmap function is general-purpose (that is, not in a
> genomics package), wouldn't it be better to just use no scaling?
>
> 2. Why assume that the rows should be scaled instead of the columns?
> There appears to be an assumption on the structure of the data.  Is
> this assumption really warranted for a general-purpose function?
. . .
> The heatmap.2 function in the gplots package defaults to scale="none"
> and I argue that it would be better for the heatmap function to do the
> same.

I second this recommendation for exactly the same reasons.
Last week I spent an hour or so trying to figure out why the heatmap colors
for the same number were different across several related heatmaps. Turning
the scaling off removed the surprise color differences.  I will always want
scale="none" as my default and now must remember to always specify it.

If the default is not changed, I think a WARNING should be included in the
documentation about this.

efg

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel