Hello,

I am using the estimateCellCounts function that implements the Houseman regression to estimate blood cell types in Infinium 450k data derived from blood.

As far as I understand this is a quadratic programming method that constraints the estimates to sum up to 1 per sample, but in my dataset the sums are almost 1.3. I understand that the estimates can sum to a bit more than 1 because these are predictions from a model based on Houseman's training data.

How do I interpret sums of estimates > 1? Does it mean I'm outside the predicable range (the data is outside the training data range)?

Also another question regarding the estimateCellCounts function: Does it use sex as a covariate? I get warnings (I've only males in my dataset):

An inconsistency was encountered while determining sex. One possibility is that only one sex is present. We recommend further checks, for example with the plotSex function.

   regards,

   Arne

I am running into the same error regarding the "inconsistency" encountered while determining sex.

As for the cell proportion estimates, the original Houseman paper imposed only a constraint that the estimates be non-negative.  That group's software has now been modified by adding in a constraint that the estimates sum to a value less than one ("lessThanOne" argument is set to TRUE). This has not been implemented in the minfi version to my knowledge (as of Feb. 2, 2016).

http://people.oregonstate.edu/~housemae/software/