PVE formulas

[Corin Stansfield]

Hi all,  I have a question about PVE calculations.

I'm currently performing MQM using the textbook method (fitqtl, refineqtl etc.), and while in many cases my models are enhanced by the use of cofactors, this is no always the case.

If the model has cofactors hen I take the PVE value from fitqtl. If there are no cofactors, then I would use the formula from the textbook: 1 - 10^(-2/N*Lod). However, as has been pointed out elsewhere on the forums, the formula used in fitqtl is rather more complex. Thus there are instances where two QTL will have similar lod scores, but the different formulas will mean that PVE values vary several-fold.

My question is: how should we deal with and compare PVE values that have been calculated in different ways? Is it better to apply the textbook formula to all, even if this is less sophisticated than the formula from fitqtl?

Thanks,

Corin

[Karl Broman]

The results from fitqtl are not more sophisticated than the formula you cite. In fact, we just use that formula, applied to the LOD scores from the "drop one QTL" analysis.

The PVE results from fitqtl are particularly problematic in the case of linked QTL, and I would not trust the values provided, in such cases. And in general, I don't think I would try to compare PVE values calculated in different ways, as you can't tell whether the differences are due to the methods or to the QTL effects themselves.

I unfortunately don't have much useful advice on this issue; sorry. I've never much focused on PVE, particularly as the values can be so strongly biased due to the selection process in identifying QTL.

karl