CSC8309 -- Gene Expression and Proteomics

Boxplots

Boxplots show the mean, interquartile ranges, max and minimum values.

Evaluate this code in R ## Construct color boxplots boxplot(soy.ab, col = brewer.cols, ylab = 'Unprocessed log (base 2) scale Probe Intensities', xlab = 'Array Names') title('Box Plots of soy Probe Level Data') ## and density plots hist(soy.ab, col = brewer.cols, lty = 1, xlab = 'Log (base 2) Intensities', lwd = 3) samp.leg.names <- new.sampleNames legend(10, .4, legend = samp.leg.names, lty = 1, col = brewer.cols, lwd = 3) title('Density plots - probe level data for six soy arrays', cex = .6) # Construct MAplots par(oma = c(1,1,3,1)) par(mfrow = c(2,3)) MAplot(soy.ab, cex = .5) mtext('M', side = 2, outer = TRUE) mtext('A', side = 1, outer = TRUE) mtext('Six MA-Plots vs. the Median Mock Array', side = 3, outer = TRUE) par(mfrow = c(1,1)) # Construct single (first) MAplot to reproduce figure in paper par(oma = c(1,1,1,1)) MAplot(soy.ab, which=1) (Complete File)(Rout) You might want to do this slowly as this snippet of code produces all the images for the following section (not just boxplots). So evaluate till you get a graphic, then read the information.

Essentially, this gives another visual measure of the intensities of the spots. The max and min statistics are very sensitive to outliers (obviously), but the mean of the intensities should be reasonably equivalent in all the arrays. Look at the image. In this case, all the arrays look okay.

What is an IQR? You can also do boxplots for blocks on the array. Why would you do this? These arrays produce nice box plots. Can you find some examples of poor boxplots on the web?

Density Plots

Next, we try density plots. In this case, we are plotting log intensities against the number of spots with that intensity. In short, it's a histogram, which should have "1" under the curve.

Looking at this density image, it looks fairly good. The lines are all similar, and there are no major bimodalities. The paper, again, suggests that what is bad is "context dependent".

For external examples, try these two:

Good

Bad

(from http://envgen.nox.ac.uk)

You notice that even the good one does have biomodality.

What does the density plot tell us that the boxplot does not?

MAPlots

Finally, we look at "MAPlots" ¹. Here you are comparing the variance of the intensities for each probe on the chip in question against an artificial chip which reflects the average. This gives us a big scatter plot, through the middle of which we add a loess regression line (the average at every point).

Here we want a straight or not too curved regression line and all of the plots to look roughly similar.

I find six on one picture a bit small. Can you work out how to get one on a graphics pane instead?

1. These plots have a name which can also be read "map-lots" which gives you lots of spurious hits on google.