[BioC] besides batch effects, whether adjust for specific quality metric variables (pos_control_mean, neg_control_mean, etc) in the analysis models

[shirley zhang]

Thanks Kasper and Jeff for your quick response.

I did try to use PC and SVA method as Jeff suggested. But my local
statistician really suggest me to adjust those Affy specific quality
metric variables suck as pos_control_mean, neg_control_mean, etc
(totally 19 this kind of variables as I listed in my previous email).
He also said since our sample size is pretty big (>1000), it won't be
a big problem in terms of degree of freedom. But my concern is I did
not see such kind of adjusting in the literature or in this list
besides SVA, PC, limma for batch effects, combat, linear mixed model
for random batch effects, etc.

Thanks again for your reply.

Shirley

On Tue, Dec 20, 2011 at 9:48 AM, Kasper Daniel Hansen
<kasperdanielhansen at gmail.com> wrote:
> On Tue, Dec 20, 2011 at 9:27 AM, shirley zhang <shirley0818 at gmail.com> wrote:
>> I have about 1000 samples run on Affy's exon gene expression arrays.
>> For differential analysis, I was suggested by one of our statistician
>> that in my model, besides batch effects, I have to adjust for the
>> following technical variables.
>>
>> RNA concentration
>> RNA Quality (RIN number)
>> cell counts
>>
>> all_probeset_mean
>> pos_control_mean
>> pos_control_stdev
>> neg_control_mean
>> neg_control_stdev
>> pos_control_mad_residual_mean
>> all_probeset_stdev
>> all_probeset_rle_stdev
>> all_probeset_rle_mean
>> spike-in control variables
>>
>> I have adjusted the first 3 variables in my previous analysis besides
>> treating batch effect as a random effect, but for other variables, I
>> think they are quality metrics to check whether the quality of the
>> Affy array data is high or low during processing the chip
>> (hybridization, scanning, etc.).
>>
>> I've also tried to use principle components and SVA method to deal
>> with hidden variables, but I have not heard about to adjust these
>> specific  quality metric variables (pos_control_mean,
>> neg_control_mean, etc) in the analysis models.
>>
>> Could anybody give me more comments and suggestions on this?
>
> I don't think your local statistician really intended for you to
> control for these variables.  But really, why don't you ask him/her
> again?  That is going to be much more profitable than for us to guess
> at what the intention is.
>
> Kasper



-- 
Xiaoling (Shirley) Zhang

M.D., Ph.D. (Bioinformatics)
Boston University, Boston, MA
Tel: (857) 233-9862
Email: zhangxl at bu.edu