probe_level

Evalutate this code:

## this section is also current broken...

## # Fit a probe-level model to the soy.ab probe-level data
## library('affyPLM')
## Pset1 <- fitPLM(soy.ab)

## # Display probe-level quality diagnostics for array # 1
## par(mfrow = c(2,2))
## par(oma = c(3,1,3,1))
## image(soy.ab[,1], col = palette.gray)
## image(Pset1, type = 'weights', which = 1)
## image(Pset1, type = 'resids', which = 1)
## image(Pset1, type = 'sign.resids', which = 1)
## mtext('Probe Level Models - QC Checks', side = 3, outer = T)
## par(mfrow = c(1,1))
## # Repeat above commands with .. soy.ab[,i], i = 2-6; which = i

## # Construct Relative Log Expression (RLE) Plot
## library(affyPLM)
## Mbox(Pset1, col = brewer.cols, names = NULL, main = 'Relative Log Expression Plot - H/R vs. T/S')

## # Add ylim = c(-.5, .5) after observing previous plot
## Mbox(Pset1, col = brewer.cols, ylim = c(-.5,.5),  main = 'Relative Log Expression Plot - H/R vs. T/S')


## # Construct Normalized Unscaled Standard Error (NUSE) Plot
## boxplot(Pset1, col = brewer.cols, main = 'NUSE Plot', ylab = 'NUSE - Normalized Unscaled Standard Error')

## # Add ylim = c(.9, 1.1) after observing previous plot
## boxplot(Pset1, ylim = c(.9, 1.1), col = brewer.cols, main = 'NUSE Plot', ylab = 'NUSE - Normalized Unscaled Standard Error', las = 2)

(Complete File)(Rout)

The first of these is for background correction. Essentially, this provides yet another of way of looking for spatial artifacts. The upper right picture, here, is the same as you produced earlier. The upper left is much the same, over normalised data (we will discuss normalisation next). The bottom left is, again, looking for variation from the mean — in this case the value that we have actually measured for the average. The bottom right is the same, but signed rather than absolute.

The second of these is, again, a boxplot. We are asking here about the expression levels of each gene (not intensity of each probe) with respect to the median.

Finally, the NUSE plot; here, we look for the variation between different probes for each gene. This gives us an standard error for each gene. We then normalise, because different genes differ in their expression. And finally, we plot the lot on a box plot. Again, this is based on the idea that genes should show about similar variation between replicate probes.

If you are tired by this stage, you can move onto normalisation. Alternatively, we will take a small detour to quality checks on some lymphoblastic datasets.

CSC8309 -- Gene Expression and Proteomics

Finally (nearly)....