Filtering on absolute gene expression level in microarrays

Question

[mi_kappa]

Hello all,

I have a normalized microarray dataset (affy_hg_u133_plus_2) of 23100 genes and 582 individuals. I have been in a discussion about filtering my dataset on absolute gene expression levels by looking into whether Y chromosome genes are expressed in the female individuals in my dataset. Y chromosome genes should not be expressed in women thus expression values can be reflective of background noise. The idea is to take the mean Y chromosome gene expression in women and below that it will be background noise and above that we will have the final expressed dataset. I have three questions:

1. Is this step necessary to my analysis?
2. If it is would that be the way to go or there are other analysis that I could use?
3. if we filter as I describe above does it make sense to use the mean? I feel the median would be more appropriate.

Answer 1

[Asaf]

1. Yes, this type of QC is necessary, I've never analyzed an experiment with zero errors, with that many individuals you'll find some surprises.
2. You can plot the expression on the Y axis and the Y chromosome genes on the X axis, color according to sex and figure out what the best statistic is, I assume it will be pretty straightforward. If you want to be meticulous you can train a simple naive-Bayes machine that will give a probability for each sample being male or female but that's unnecessary.

[mi_kappa]

Thank you for your help!