4
$\begingroup$

I am starting to use microarrays and maybe this is a dumb question:

Using Illumina microarrays, linear gene expression can be negative? Or maybe some artefacts have been introduced?

And, in this case, how to correct them? With scaling (adding the absolute value of the minimum negative value) or flooring (converting negative values to zeros)?

Thanks

$\endgroup$

1 Answer 1

6
$\begingroup$

I'm not personally familiar with Illumina arrays, but I can give some notes here. This link is a paper which describes the array quality controls specifically. This presentation describes the calculation of the intensities in bioconductor.

The answer is yes: you will find negative numbers sometimes. They should be rare. The Intensities numbers from a scanner which essentially takes an image of the bead fluorescence on the glass slide and tries to subtract a background signal.

The background is the typical level of signal you see on a bead with no sample DNA bound to it. You will get some pixels lighting up a bit even there.

In expression microarrays, this is not a perfect system as each bead has a different nucleotide sequence on it. Each bead has a specific DNA sequence has non-specific binding which is pretty much unique on the slide. That is to say, the oligomer on the bead might bind strongly to DNA from your sample which is not its reverse complement to varying degrees.

I think what happened is that there will be a handful of cases you can find a bead which has less fluorescence than the background controls on the slide. It's possible that the probe has a design flaw but I would generally assume that the negative number means that there is no detectable target cDNA in the sample for that oligo.

I probably wouldn't convert the number to a zero, but you could probably justify it to yourself in some cases. Most difference experiments are the logarithm of the ratio so zero is not a great number in those cases.

@Luke 's comments are well said.

I think that negative values can still represent signal because of the variance of sequence-dependent effects for the purposes of scaling, and difference experiments. But I'd tend to think of them as zero signal in any case. There are probably genes which express at or below detection threshold all the time, which is not zero information, so the number has some value.

$\endgroup$
3
  • 2
    $\begingroup$ This is pretty spot on. The "intensity" values you end up getting are relative values to the background levels. These tend to be normalized to an arbitrary value - "100" could be the arbitrary threshold for background, above which you see a reliable signal. A probe with very low intensity could therefore be below 0. $\endgroup$
    – Luke
    Oct 9, 2013 at 17:23
  • $\begingroup$ do you have any suggestions about what to do with a negative intensity @Luke ? I tend to ignore things with such low values, but it can make a difference in comparing 2 small amplitude measurements i guess. (i'd still tend to ignore them in that case!) $\endgroup$
    – shigeta
    Oct 9, 2013 at 18:01
  • 1
    $\begingroup$ Anything below the background threshold is by definition an unreliable signal (i.e. detection P-value > 0.05). The 2 options for dealing with this is either to set values below the background threshold to "missing", or to leave them in but be aware that below the background threshold the values are no longer meaningful (you might do this if, for instance, 60% of your sample were below background for this probe, but the rest were above - this is still meaningful because it means that 60% of your sample express less than the 40%, so you do not want to set them as missing, as this may be relevant $\endgroup$
    – Luke
    Oct 11, 2013 at 14:33

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .