R help - Nov 2010 - One question on heatmap

Dear R-helper:

Suppose we have a matrix:

Gene		sample1	sample2

Gcnt1            12.0000    52.80000
Max               8.8000    39.10000
Tmem176b         67.9000   304.70000
Shmt2             8.6000    42.40000
Rtn4             11.5000    57.70000
Il17re            7.6000    38.80000
Bclp2             6.2000    32.10000
Mobkl3            4.4000    32.20000
Akr1b10           3.4000    30.10000
Atp6ap2           5.4000    48.20000
Snx2              5.7000    63.10000
Tmem176a          7.6000    91.40000
Klhl9             1.7000    30.30000
Fbxo27            1.0000    28.90000
Scd1             34.6000     0.70000
Tspan9           35.8000     4.20000
2210016L21Rik    39.1000     4.90000
Ctnnb1          212.1000    33.10000
Apoe            397.2000    74.20000
H2-DMb1          72.3000    14.10000
Ryk              31.7000     6.40000
Dapk2            85.4000    17.30000
Gzmm            179.4000    36.80000
Actb          12993.4000  2678.10000
Faim3           758.0000   157.60000
Aktip           209.4000    46.00000
Tbrg1            93.3000    21.30000

When I try to make heatmap based on this gene expression value table, I found
that, when I set 'scale' to 'column', the heatmap will be always
be red. I think this is because, there's very large values in the matrix
(gene Actb), while the most are just very small. Thus, the color will be very
ugly. I just wonder, how to set the color to make the heatmap look better?  I
have tried log-tranformation on the matrix and it's better now. But I do
want to know if you have better ways to set the color span manually and make the
heatmap look better without any log-transformation?

Thanks in advance!

Best, Hua

Before plotting a heatmap we usually standardize all genes to mean
zero and variance 1. That way the green/red represent under/over
expression with respect to the mean expression, which is roughly what
the original 2-color arrays (that literally produced such heatmaps)
were measuring. Of course, standardization assumes you have more than
2 samples.

If your expression matrix is stored in the variable expr, with columns
corresponding to samples and rows to genes, you can obtain a
standardized expression matrix as

stdExpr = t(scale(t(expr)))


Peter


On Tue, Nov 2, 2010 at 8:50 AM, Hua <kaixinsjtu at gmail.com>
wrote:
> Dear R-helper:
>
> Suppose we have a matrix:
>
> Gene ? ? ? ? ? ?sample1 sample2
>
> Gcnt1 ? ? ? ? ? ?12.0000 ? ?52.80000
> Max ? ? ? ? ? ? ? 8.8000 ? ?39.10000
> Tmem176b ? ? ? ? 67.9000 ? 304.70000
> Shmt2 ? ? ? ? ? ? 8.6000 ? ?42.40000
> Rtn4 ? ? ? ? ? ? 11.5000 ? ?57.70000
> Il17re ? ? ? ? ? ?7.6000 ? ?38.80000
> Bclp2 ? ? ? ? ? ? 6.2000 ? ?32.10000
> Mobkl3 ? ? ? ? ? ?4.4000 ? ?32.20000
> Akr1b10 ? ? ? ? ? 3.4000 ? ?30.10000
> Atp6ap2 ? ? ? ? ? 5.4000 ? ?48.20000
> Snx2 ? ? ? ? ? ? ?5.7000 ? ?63.10000
> Tmem176a ? ? ? ? ?7.6000 ? ?91.40000
> Klhl9 ? ? ? ? ? ? 1.7000 ? ?30.30000
> Fbxo27 ? ? ? ? ? ?1.0000 ? ?28.90000
> Scd1 ? ? ? ? ? ? 34.6000 ? ? 0.70000
> Tspan9 ? ? ? ? ? 35.8000 ? ? 4.20000
> 2210016L21Rik ? ?39.1000 ? ? 4.90000
> Ctnnb1 ? ? ? ? ?212.1000 ? ?33.10000
> Apoe ? ? ? ? ? ?397.2000 ? ?74.20000
> H2-DMb1 ? ? ? ? ?72.3000 ? ?14.10000
> Ryk ? ? ? ? ? ? ?31.7000 ? ? 6.40000
> Dapk2 ? ? ? ? ? ?85.4000 ? ?17.30000
> Gzmm ? ? ? ? ? ?179.4000 ? ?36.80000
> Actb ? ? ? ? ?12993.4000 ?2678.10000
> Faim3 ? ? ? ? ? 758.0000 ? 157.60000
> Aktip ? ? ? ? ? 209.4000 ? ?46.00000
> Tbrg1 ? ? ? ? ? ?93.3000 ? ?21.30000
>
> When I try to make heatmap based on this gene expression value table, I
found that, when I set 'scale' to 'column', the heatmap will be
always be red. I think this is because, there's very large values in the
matrix (gene Actb), while the most are just very small. Thus, the color will be
very ugly. I just wonder, how to set the color to make the heatmap look better?
?I have tried log-tranformation on the matrix and it's better now. But I do
want to know if you have better ways to set the color span manually and make the
heatmap look better without any log-transformation?
>
> Thanks in advance!
>
> Best, Hua
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

On 11/03/2010 02:50 AM, Hua wrote:
> ...
> When I try to make heatmap based on this gene expression value table, I
found that, when I set 'scale' to 'column', the heatmap will be
always be red. I think this is because, there's very large values in the
matrix (gene Actb), while the most are just very small. Thus, the color will be
very ugly. I just wonder, how to set the color to make the heatmap look better? 
I have tried log-tranformation on the matrix and it's better now. But I do
want to know if you have better ways to set the color span manually and make the
heatmap look better without any log-transformation?
>
Hi Hua,

It is not all that easy, but can be done. I read in your data as 
"genexp". Notice how I split up the data into three ranges, adding the
range extremes in color.scale (plotrix) and then removing the extremes.

expcol[genexp$sample2<100]<-
  color.scale(c(0,100,genexp$sample2[genexp$sample2<100]),
  c(1,0.5),c(0,0.5),0)[-(1:2)]
expcol[genexp$sample2>=100&genexp$sample2<1000]<-
 
color.scale(c(100,1000,genexp$sample2[genexp$sample2>=100&genexp$sample2<1000]),
  c(0.5,0.1),c(0.5,0.9),0)[-(1:2)]
expcol[genexp$sample2>1000]<-
  color.scale(c(1000,10000,genexp$sample2[genexp$sample2>=1000]),
  c(0.1,0),c(0.9,1),0)[-(1:2)]
barplot(rep(1,27),col=expcol)
color.legend(0,-0.1,15,-0.05,c(0,100,1000,10000),
 
rect.col=c("red","#808000","#1ae600","green"))

Jim