[BioC] Detection calls and LIMMA

[Gordon K Smyth]

Deaer Avhena,

I agree with Wolfgang that filtering is useful.  In my lab, the standard 
practice is to filter probes that fail to show some modest evidence for 
expression on at least n arrays, where n is the minimum group size.  For 
example, if we compare wt (with 2 replicate arrays) to a mutant (with 3 
replicate arrays), we filter probes that are Present on fewer than 2 
arrays.

This is because we want to keep any probe that is expressed in at least 
one of the experimental conditions.  If a probe is expressed in one of the 
conditions, then it should appear consistently across the replicates for 
that condition.

Best wishes
Gordon

> Date: Mon, 28 Feb 2011 11:35:42 -0500
> From: avehna <avhena at gmail.com>
> To: whuber at embl.de
> Cc: bioconductor at r-project.org
> Subject: Re: [BioC] Detection calls and LIMMA
>
> Dear Wolfgang,
>
> Thank you for your response, I agree with you. I will read the paper now...
>
> Best Regards,
> Avhena
>
> On Mon, Feb 28, 2011 at 4:38 AM, Wolfgang Huber <whuber at embl.de> wrote:
>
>> Hi Avhena
>>
>> it is not required, but properly applied filtering can increase detection
>> power in your experiment while still controlling type-I error (false
>> positives). The example you mention seems to be one that you want to keep
>> though, since it is a good candidate for being up-regulated in the Treatment
>> condition. One possibly reasonable criterion would be, e.g., to filter out
>> all probesets that are called 'Absent' on all arrays. Some further
>> discussion on the topic is also here:
>>
>> [1] Bourgon, Gentleman and Huber. Independent filtering increases detection
>> power for high-throughput experiments. PNAS, 107(21):9546-9551,
>>
>>        Best wishes
>>        Wolfgang
>>
>>
>> Il Feb/28/11 6:51 AM, avehna ha scritto:
>>
>>> Hi All,
>>>
>>> I have a basic question. Is it required to filter the microarray data
>>> based
>>> on the detection calls (A/M/P) before analyzing it with LIMMA?
>>>
>>> What if I have the following scenario (for example):
>>>
>>>                            Control Control Control         Treatment
>>> Treatment Treatment
>>> 1367813_at           A            A             P
>>> P                   P                 P
>>>
>>> Please note that this gene is just "present/detected"  once in the
>>> Control,
>>> but it is present in all the replicates of the treatment. In this case:
>>> what
>>> would be the right thing to do? To eliminate it from the analysis or keep
>>> it
>>> and consider it up or down depending on the signal of the treatment?
>>>
>>> Thank a lot!
>>> Avhena

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}