Skip to content

Basic: setting resolution and size of an R graphic

5 messages · Dr. med. Peter Robinson, Marc Schwartz (via MN), Marc Schwartz +1 more

Original
Dr. med. Peter Robinson
# 
Dear List,

I am sorry if this perhaps a too basic question, but I have not found an
answer in Google or in the R help system. I am trying to use R to do a
very simple analysis of some data (RT-PCR and Western analysis) with a
T-test and
to plot the results as a histogram with error bars. (I have pasted an
example script at the bottom of this mail).
In order to use this for publication, I would like to adjust the
resolution and size of the final image. However, even using file types
such as postscript or pdf that are vector based, I get rather bad-looking
results with
using either pdf or postscript or jpg devices.


Therefore I would like to ask the list, how to best produce a graphic from
the script below that would fit into one column of a published article and
have a high resolution (as eps, or failing that tiff or png)?
Thanks in advance for any advice,

Peter




## Western.R
## A script to display the results of quantitative Western blotting with 6
repeats each at three dosages.
## This particular script has data from stimulation of fibroblasts with M-wt.


# ----------   CONVENIENCE FUNCTIONS --------------------- #
## Define a simple function to draw the error bars.
makeBars <- function(x,mean,se){
	segments(x,mean - se/2,x,mean+se/2)
	segments(x-0.1,mean - se/2,x+0.1,mean - se/2)
	segments(x-0.1,mean + se/2,x+0.1,mean + se/2)
	}
##Define a simple function to write p values
writeP <- function(x,mean,se,pval) {
	if (pval >= 0.01) {
		# text(x, mean + se/2 + 0.25, sprintf('p=%.2f',pval),cex=1.5)
		text(x +0.05, mean + se/2 + 0.4, sprintf('*'),cex=1)
	} else {
		text(x +0.05, mean + se/2 + 0.4, sprintf('**'),cex=1)
		}
	}


## define function to draw entire group
## A,B,C refer to the x positions of the x,y,z observations
drawBarsAndPValueForGroup <-
function(A,B,C,x.mean,x.se,y.mean,y.se,z.mean,z.se,xy.pval,xz.pval) {
	makeBars(A,x.mean,x.se)
	makeBars(B,y.mean,y.se)
	makeBars(C,z.mean,z.se)
	writeP(B - 0.05, y.mean,y.se,xy.pval)
	writeP(C -0.05,z.mean,z.se,xz.pval)
}



## We will make a two part graphic
par(mfrow=c(1,2))

## X: 0
## y: 0.2 ??M
## z: 0.4 ??M

yTop <- 12  ## Limit for Y axis

##  ---  RT-PCR MMP1   --- ##

x <- c(0.8839034,0.42011158,0.65318013 , 0.88494528,1.900606, 1.2572536 )
x.mean <- mean(x)
x.se <-	sd(x) / sqrt(length(x))
y <- c(5.067579666,2.630677502,1.881902881,1.61994864,3.356066695 )
y.mean <- mean(y)
y.se <- sd(y)/ sqrt(length(y))
z <-
c(13.38923048,3.677270765,3.559984278,10.83628903,12.20110874,12.8957108)
z.mean <- mean(z)
z.se <- sd(z)/ sqrt(length(z))

## -- Do t test and calculate the p values -- ##
xy.t <- t.test(y,x, alternative=c("greater"),var.equal=TRUE)
xy.pval <- xy.t$p.value

xz.t <- t.test(z, x, alternative=c("greater"),var.equal=TRUE)
xz.pval <- xz.t$p.value


arr <- c(x.mean,y.mean,z.mean)
mat <- matrix(arr,nrow=3,byrow=F)


barplot(mat,  ## The data
	beside=TRUE, ## juxtapose values in each column rather than stacking them
	ylim=c(0,yTop), ## limits for y axis
	xlim=c(0,3),
	width=1,
	space=c(0,0.1),##space: the amount of space (as a fraction of the average
bar width)
                            ##  left before each bar. beside=TRUE, can be
given
			    ## by 2 numbers, the space between bars within a group
			    ## and space betweens bars of different groups
	names=c('0.0 ??M','0.2 ??M','0.4 ??M'),
	ylab='Relative Change',
      col=c('white'),
      cex.names=1,
	cex.axis=1,##cex.axis: expansion factor for numeric axis labels.
	cex.lab=1)

## First for the RT-PCR group
start <- 0.1
first <- start + 0.5
second <- start + 1.5
third <- start + 2.5

drawBarsAndPValueForGroup(first,second,third,x.mean,x.se,y.mean,y.se,z.mean,z.se,xy.pval,xz.pval)


##  --- Western MMP1 --- ##

x2 <- c(1.117373856,0.690266558,1.192359586 )
y2 <- c(3.53806369,3.895634049,6.653024511 )
z2 <- c(8.609814741,3.858564979,8.492977115)

x2.mean <- mean(x2)
x2.se <- sd(x2) / sqrt(length(x2))
y2.mean <- mean(y2)
y2.se <- sd(y2)/ sqrt(length(y2))
z2.mean <- mean(z2)
z2.se <- sd(z2)/ sqrt(length(2))

## -- Do t test and calculate the p values -- ##
xy2.t <- t.test(y2,x2, alternative=c("greater"),var.equal=TRUE)
xy2.pval <- xy2.t$p.value

xz2.t <- t.test(z2,x2,alternative=c("greater"),var.equal=TRUE)
xz2.pval <- xz2.t$p.value



arr <- c(x2.mean, y2.mean,z2.mean)
mat <- matrix(arr,nrow=3,byrow=F)

## mat now has the values of each type of experiment in individual columns




barplot(mat,  ## The date
	beside=TRUE, ## juxtapose values in each column rather than stacking them
	ylim=c(0,yTop), ## limits for y axis
	xlim=c(0,4),
	width=1,
	space=c(0,0.1),##space: the amount of space (as a fraction of the average
bar width)
                            ##  left before each bar. beside=TRUE, can be
given
			    ## by 2 numbers, the space between bars within a group
			    ## and space betweens bars of different groups
	names=c('0.0 ??M','0.2 ??M','0.4 ??M'),
	ylab='Relative Change',
      col=c('white'),
      cex.names=0.8,
	cex.axis=1, ##cex.axis: expansion factor for numeric axis labels.
	cex.lab=1)


first <- 0.6
second <- first + 1
third <- first + 2
drawBarsAndPValueForGroup(first,second,third,x2.mean,x2.se,y2.mean,y2.se,z2.mean,z2.se,xy2.pval,xz2.pval)
Marc Schwartz (via MN)
#
On Mon, 2005-10-24 at 22:32 +0200, Dr. med. Peter Robinson wrote:
<Snip of code>

What OS are you on?

Running your example on FC4, I get the attached output for a pdf().

I suspect that on your OS, the height and width arguments are not
appropriate by default.

Thus, you may need to adjust your pdf (or postscript) function call to
explicitly specify larger height and width arguments.

Also note that to generate an EPS file, pay attention to the details
section of ?postscript, taking note of the 'onefile', 'horizontal' and
'paper' arguments and settings.

Also, check with your journal to see if they specify dimensions for such
graphics so that you can abide by their specs if provided. If they are
using LaTeX, there are means of specifying and/or adjusting the height
and/or width specs in the code based upon proportions of various
measures (ie. \includegraphics[width=0.9\textwidth]{GraphicsFile.eps} ).

HTH,

Marc Schwartz

-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.pdf
Type: application/pdf
Size: 5321 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-help/attachments/20051024/fbf70820/test.pdf
Dr. med. Peter Robinson
# 
Thanks Marc and Jim for the tips. The PDF file that I create with R looks
about the same as the one you created. However, I need to get the graphic
to be a certain size (300 pixels wide). I have been using the ImageMagick
program to do so for other graphics:

convert test.pdf -resize 300x300 out.pdf

then out.pdf looks rather poor (pixelly). The original image is too big.
ANy ideas?
Thanks a lot,Peter


Am Mo, 24.10.2005, 22:53, schrieb Marc Schwartz (via MN):
Marc Schwartz
# 
If you specifically need the plot to have a dimension measured in
pixels, then you need to use a bitmapped format such as png and specify
the output to be the size you require:

  png("test.png", width = 300, height = 300, ...)
  DoYourPlotHere()
  dev.off()

Do this directly using the png() device, rather than trying to convert
the image format, which almost always introduces "noise".

Since you are using a bitmapped format, you will experience the tradeoff
with respect to the image quality (ie. pixelated) as compared to a
vector based format such as PDF or PS.

I would re-verify the requirements for the journal to which you are
submitting the article relative to what they need for image specs. It
seems unusual for a journal to request an image in this fashion, unless
it is a photograph where a jpg format may be preferred or perhaps for
online publication on a web page.

HTH,

Marc Schwartz
On Tue, 2005-10-25 at 08:57 +0200, Dr. med. Peter Robinson wrote:
Romain Francois
# 
Le 25.10.2005 14:59, Marc Schwartz a ??crit :
Hi folks,

I just would like to highlight the argument 'pointsize' of the png 
function.
For graphics in RGG i use pointsize=4 and i'm happy with it.

Romain