[BioC] Filtering gene list prior to statistical testing

[Mark Cowley]

Hi Johan,
You've raised some good points, and i hope that your post attracts a  
number of replies.

For me, i routinely filter out genes that are Absent on all arrays, as  
detected by mas5calls on older affy data, or some threshold on DABG  
calls for newer affy data; This removes ~30% of the genes on the array  
that are detected too close to background in ALL SAMPLES.
These are the data points that are least likely to be measured  
correctly, and least likely to be validated by other approaches. Note  
I'm not saying that there aren't interesting changes down here, is  
just that you have to understand/work within the limits of the  
technology itself.

my 2 cents

Mark
-----------------------------------------------------
Mark Cowley, BSc (Bioinformatics)(Hons)

Peter Wills Bioinformatics Centre
Garvan Institute of Medical Research
-----------------------------------------------------
On 25/06/2008, at 5:36 PM, Johan van Heerden wrote:

> Dear Jim,
>
> Thanks for the feedback. Your point about the implications for  
> Normalization is noted, but my question relates to filtering post  
> Normalization. I.e. the scenario is as follows: Data is normalized  
> (using all "good" data points - "good" being based on Genepix  
> quality criteria) and effeciency of Normalization is assessed (using  
> visualizations and stats) based on the assumptions that you stated.  
> This data is then pre-processed (missing values are imputed,  
> replicates are averaged etc. Once a "clean" data set is obtained  
> (where we are fairly certain that most technical noise has been  
> removed), we filter out all genes that don't show at least a 1.2fold  
> change between classes (i.e. not very stringent, the aim is to  
> remove within class "flat" patterns), which reduces our sample space  
> by about 1/3 (highlighting again that most genes show no or very  
> little change). Furthur analysis is then done on this data set (i.e.  
> Functional enrichment of gene groups as well as
> "straight" differential testing).
>
> What I would like to know is if there is any good reason why we  
> should keep the genes that show "no" class-based behaviour (in terms  
> of fold-changes) - as these contain very little or no information.  
> Including them does not change the types of functional classes found  
> to be enriched, but it pushes all adjusted p-values (doing  
> differential testing) into very non-significant ranges.  Eliminating  
> them improves the adjusted p-values greatly.  The "top" x genes  
> using the "filtered" data set is exactly the same as the unfiltered  
> one, except they have significant adjusted p-values. I am aware that  
> a significant p-value is not the be-all-and-end-all of microarray  
> research, but it is an unfortunate reality that most people do  
> attach great importance to this and "significant" values are  
> required to make any substatiated assertions (prior to downstream  
> validation of course!).
>
> Can any serious criticisms be raised against this type of "post- 
> normalization" sample space reduction?
>
> Thanks,
> Johan van Heerden
>
>
>
>
> --- On Tue, 6/24/08, James W. MacDonald <jmacdon at med.umich.edu> wrote:
>
>> From: James W. MacDonald <jmacdon at med.umich.edu>
>> Subject: Re: [BioC] Filtering gene list prior to statistical testing
>> To: jvhn1 at yahoo.com
>> Cc: bioconductor at stat.math.ethz.ch
>> Date: Tuesday, June 24, 2008, 5:30 PM
>> Hi Johan,
>>
>> Johan van Heerden wrote:
>>> Dear All,
>>>
>>> I have scoured the BioC mailing list in search of a
>> clear answer
>>> regarding the filtering of a data sets prior to
>> differential testing,
>>> in an attempt to circumvent the multiple testing
>> problem. Although
>>> several opinions have been expressed over the last
>> couple of years I
>>> have not yet found a convincing argument for or
>> against this
>>> practice.  I would like to make a comment and would
>> appreciate any
>>> constructive feedback, as I am not a Statistician but
>> a Biologists.
>>>
>>> As far as I can see the problem has been divided into
>> 2 categories:
>>> (1) "Supervised" and (2)
>> "Unsupervised" filtering, where (1) is based
>>> on some knowledge regarding the functional classes
>> present in the
>>> data, as opposed to (2) which does not consider any
>> such information.
>>> Several criticism have been raised against the
>> "Supervised" approach,
>>> with many people calling it flawed logic. My first
>> comments are
>>> regarding the logic of "Supervised"
>> filtering.
>>>
>>> As an example: A data set consisting of two classes
>> (Treatment 1 and
>>> Treatment 2) has been generated.  A fold-change is
>> then used to
>>> enrich the data set for genes that show within class
>> activity (i.e.
>>> select only genes that show a mean x-fold change
>> between classes).
>>> This filtered data set is then used for differential
>> testing.
>>>
>>> My first question is: How is this different
>> (especially when working
>>> with "whole-genome" arrays) from having
>> custom arrays constructed
>>> from genes known show a response to some treatment.
>> I.e. Arrays will
>>> then be selectively printed with genes that are known
>> to or expected
>>> to show a response. This is a type of
>> "filtering" step that will
>>> yield arrays with highly reduced gene sets. This
>> scenario can result
>>> from known knowledge about pathways or can arrise from
>> a discovery
>>> based microarray experiment, where a researcher
>> produces whole genome
>>> arrays and from there select "responsive"
>> genes for the creation of
>>> targeted (or custom arrays). Surely this step-wise
>> sample space
>>> reduction should be subject to the same criticism?
>>
>> It is. When normalizing microarray data, the assumption
>> being made is
>> that many (most) of the genes being measured are not
>> actually changing
>> expression. What most normalization schemes do is line up
>> the bulk of
>> the data so on average the log fold change is zero. If we
>> can't make
>> this assumption (e.g., it is possible that _all_ genes are
>> up-regulated
>> in one sample), then without having some housekeeping genes
>> to use for
>> the normalization, there is no way to normalize the data
>> without making
>> some strong and possibly unwarranted assumptions.
>>
>> So the main argument as I see it against doing supervised
>> sample space
>> reduction is that you may be removing the main assumption
>> of most
>> normalization schemes. The normalization is really the
>> important thing
>> here, as you are trying to remove unwanted technical
>> variation that will
>> have a much larger effect on your statistics than the
>> multiple testing
>> issue.
>>
>>>
>>> Secondly, the supervised fold-change filter should not
>> affect the
>>> statistic of each individual gene, but will have
>> profound effects on
>>> the adjusted p-values. I have checked this only for
>> t-tests and am
>>> not sure what the effect on more complex statistical
>> differential
>>> testing methods would be. If the only effect of the
>> "supervised"
>>> filtering step is the enrichment of class-specific
>> responsive gene
>>> and a reduction in the severity of the p-value
>> ADJUSTMENT (without
>>> affecting the actual statistic), this could surely be
>> a very useful
>>> way of filtering data?
>>>
>>> Wrt the "unsupervised" approaches: These
>> approaches define some
>>> overall variability threshold which can be used to
>> filter out genes
>>> that don't show a minimum degree of variability
>> regardless of class.
>>> As far as I can tell there are several issues wrt this
>> approach. (1)
>>> Some genes will be naturally "noisy", i.e.
>> will show high levels of
>>> fluctuation regardless of class. These genes are
>> likely to be
>>> included in a filter based on degree of varilablity.
>> (2) Some genes
>>> might show low levels of variability (with small
>> changes between
>>> classes) and could be important, but will be excluded
>> if a filter is
>>> based on degree of variability.
>>
>> This is all true, but again I think the normalization issue
>> is much more
>> important, and that is where we really want to make sure we
>> are doing a
>> good job.
>>
>> These days people are getting much less interested in a
>> list of
>> differentially expressed genes, as these are often too
>> large to be
>> useful anyway. The real underlying goal of most experiments
>> IMO, is to
>> find pathways that are perturbed by some
>> treatment/condition/whatever.
>> In this case one really doesn't care about multiple
>> testing, and instead
>> is just using the t-stats (or whatever) in a GSEA type
>> statistic to
>> measure the difference in sets of genes.
>>
>> Best,
>>
>> Jim
>>
>>
>>
>>>
>>> I would greatly appreciate some feedback on these
>> comments,
>>> specifically some statistical substantiation as to why
>> a "supervised"
>>> approach is "flawed", given the similar
>> experimental strategies
>>> included in the paragraph on this approach.
>>>
>>> Many Thanks!! Johan van Heerden
>>>
>>> _______________________________________________
>> Bioconductor mailing
>>> list Bioconductor at stat.math.ethz.ch
>>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>> Search the
>>> archives:
>>>
>> http://news.gmane.org/gmane.science.biology.informatics.conductor
>>
>> -- 
>> James W. MacDonald, M.S.
>> Biostatistician
>> Affymetrix and cDNA Microarray Core
>> University of Michigan Cancer Center
>> 1500 E. Medical Center Drive
>> 7410 CCGC
>> Ann Arbor MI 48109
>> 734-647-5623
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor