[BioC] Confusing colours using geneplotter functions and heatmap()

Sean Davis sdavis2 at mail.nih.gov
Wed Aug 25 13:03:09 CEST 2004


There are times when using row-based scaling makes sense.  For example, 
if one is comparing two samples via a common reference (2-color 
design), then absolute ratios are not meaningful, only the ratios 
relative to each other.  In this case, and probably others, having 
row-based scaling (or normalization) makes contrasts easier to see and 
more intuitive.

In your case, it seems you feel that scaling is not appropriate and 
probably isn't.  I believe the default for heatmap is to have row 
scaling.  From heatmap documentation (see ?heatmap):

    scale: character indicating if the values should be centered and
           scaled in either the row direction or the column direction,
           or none.  The default is '"row"' if 'symm' false, and
           '"none"' otherwise.

So, you probably want to do heatmap(...,scale="none")


> Clearly this is not how it should be, UNLESS heatmap() only assigns
> colours relative to each row individually, and does not do any
> comparison between rows.  Is this the case?  Won't this lead to
> misleading results?  Surely the colour assigned to a spot in a heatmap
> should reflect that spot's "position" in the whole dataset, not just 
> the
> data for that row of the matrix?  If the latter is true, then values of
> 5000 and 0.5 could be assigned the same colour, simply because they are
> in the same position relative to their respective genes - and that 
> would
> render making comparisons vertically in a heatmap() useless...?

More information about the Bioconductor mailing list