Experimental process
Data
- genepix file have .gpr extension
- 8 data points recorded for each spot (mean & median, foreground & background, red & green)
- usually use median as less affected by outliers
Goal of experiment is to discover genes undergoing differential expression, ie. different intensities between treatments.
Background correction
Foreground intensity consists of both signal and noise. Background correction attempts to remove the noise by subtracting the background intensity.
Cons:
- sometimes background intensities higher than foreground (“black hole effect”)
- often very little background noise spread uniformly
- binding characteristics of background (ie. blank slide) are quite quite to those of foreground (ie. DNA)
Pros:
- can help to remove spatial effects
Fold changes
Regardless of whether background correction is performed, we have two data points for each spot: a red intensity (Cy5) and a green intensity (Cy3). Using these we wanted to work out if there is a difference in expression between the two treatments. Differential expression is often reported as fold change = red intensity / green intensity, or log(fold change). Initially biologists used a cutoff of > 2 and < 0.5.
There are a number of things that might cause different intensity to be measured (differential expression, hybridisation variability, dye effects). A statistical approach needs to take this into account, using replication to estimate variability and experimental design to balance unwanted effects.
The effect of hybridisation variability can be reduced by increasing replication of genes on the array, but this reduces the number of genes we can test. Dye effects can reduced be balancing across treatments, but we can’t double label each sample on a single array. Solution: use more arrays! This introduces another issue: variability between arrays.
Dye Swap
- Split each sample into two subsamples
- Label one green and one red.
- On each array use one colour from each sample
Equivalent to a Latin-square design of size 2. The array, treatment, dye and gene effects are balanced, allowing their effects to be separated and estimated.