[BioC] A time course study: using edgeR
Gordon K Smyth
smyth at wehi.EDU.AU
Tue Jun 18 00:34:07 CEST 2013
Hi Michael,
Further to Ryan's post, your experimental design seems to fit exactly into
the framework of Section 3.3 of the edgeR User's Guide. The approach that
Ryan recommends is that of Section 3.3.1. Section 3.3.2 describes another
equivalent way to setup the comparisons you want.
Note that you have previously considered additive models like:
~ timecourse + condition
but this is not correct because it assumes that the time course effects
are exactly the same for both conditions. If you want to test for time
effects for each condition separately, then you must use the
all-combination models considered in Section 3.3 of the edgeR User's
Guide.
Best wishes
Gordon
On Mon, 17 Jun 2013, Ryan C. Thompson wrote:
> Hi Michael,
>
> The simplest and most understandable approach I've found is to define a
> single factor that splits your data into all the groups. So there would be
> levels for disease-T0, disease-T1, disease-t3, control-T0, control-T1, and
> control-T3. Then you can use ~0+group as your model formula and define
> whatever contrasts you desire between these six groups.
>
> -Ryan Thompson
>
> On Mon 17 Jun 2013 10:54:59 AM PDT, Michael Breen wrote:
>>
>> Gordon!
>>
>> On second pass, edgeR doesn't seem to address my question.
>>
>> We have disease and control data spread over 3 time points (baseline-T0,
>> time after disease causing event-T1, elapsed time after event-T3) for each
>> of these two conditions. I want to be able to find DEGs accordingly to
>> condition across time-points to decipher which times (either at treatment
>> or elapsed time after treatment) our genes are DE.
>>
>> Any further insight would be fantastic.
>>
>> Yours,
>>
>> Michael
>>
>>
>>
>> -----Original Message-----
>> From: "Michael Breen" <mbreen at vapop.ucsd.edu>
>> To: "Gordon K Smyth" <smyth at wehi.EDU.AU>
>> Cc: Bioconductor mailing list <bioconductor at r-project.org>
>> Date: Mon, 10 Jun 2013 20:47:10 -0700
>> Subject: Re: [BioC] DESEQ ANODEV : A time course study
>>
>> Cheers!
>>
>> I think this is applicable to our needs.
>>
>> Thank you Gentlemen!
>>
>>
>> -----Original Message-----
>>
>> From: Gordon K Smyth <smyth at wehi.EDU.AU>
>> To: Michael Breen <mbreen at vapop.ucsd.edu>
>> Cc: Bioconductor mailing list <bioconductor at r-project.org>
>> Date: Tue, 11 Jun 2013 10:44:34 +1000 (AUS Eastern Standard Time)
>> Subject: DESEQ ANODEV : A time course study
>>
>> Dear Michael,
>>
>> What you want to do is easy and fast using the edgeR package, without any
>> need for ad hoc workarounds like subsetting your data. See McCarthy et
>> all (NAR 2012):
>>
>> http://www.ncbi.nlm.nih.gov/pubmed/22287627
>>
>> Best wishes
>> Gordon
>>
>>>
>>> Date: Mon, 10 Jun 2013 08:41:49 +0200
>>
>>
>>>
>>> From: Simon Anders <anders at embl.de>
>>
>>
>>>
>>> To: bioconductor at r-project.org
>>
>>
>>>
>>> Subject: Re: [BioC] DESEQ ANODEV : A time course study
>>
>>
>>>
>>>
>>
>>
>>>
>>> Hi Michael
>>
>>
>>>
>>>
>>
>>
>>>
>>> On 08/06/13 03:00, Michael Breen wrote:
>>
>>
>>
>>
>>>
>>>>
>>>> What we aim to do is to test for DE of transcripts across all 3 time
>>>
>>
>>
>>>
>>>>
>>>> points for disease and controls seperatly (using DESeq ANODEV) but we
>>>
>>
>>
>>>
>>>>
>>>> want to be able to identify at which time points these transcripts are
>>>
>>
>>
>>>
>>>>
>>>> being DE. In other words, we want to compare DE transcripts with
>>>
>>
>>
>>>
>>>>
>>>> respect to specific time points between cases and controls. Our
>>>
>>
>>
>>>
>>>>
>>>> remaining code looks like this:
>>>
>>
>>
>>>
>>>>
>>>>
>>>
>>
>>
>>>
>>>>
>>>> fit0 <- fitNbinomGLMs (cds, count ~ timecourse)
>>>
>>
>>
>>>
>>>>
>>>> fit1 <- fitNbinomGLMs ( cds, count ~ timecourse + condition )
>>>
>>
>>
>>>
>>>>
>>>> str(fit1)
>>>
>>
>>
>>>
>>>
>>
>>
>>>
>>> One possibility would be to subset your data to only samples from one
>>
>>
>>>
>>> time point and then test cases against control to see the genes that are
>>
>>
>>>
>>> DE at this time point, then go on to the next one. If you consider this
>>
>>
>>>
>>> a post-hoc test and only look at the genes which show overall
>>
>>
>>>
>>> sensitivity, you can probably be more lenient on the significance
>>
>>
>>>
>>> threshold. Maybe other people on the list have input on this point.
>>
>>
>>>
>>>
>>
>>
>>>
>>> Simon
>>
>>
>>
>>
>> ______________________________
>> ________________________________________
>>
>> The information in this email is confidential and intend...{{dropped:11}}
>>
>> _______________________________________________
>> Bioconductor mailing list
>> Bioconductor at r-project.org
>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>> Search the archives:
>> http://news.gmane.org/gmane.science.biology.informatics.conductor
>>
>>
>>
>>
>>
>> On Tue, Jun 11, 2013 at 11:32 AM, Michael Breen
>> <mbreen at vapop.ucsd.edu>wrote:
>>
>>>
>>>
>>>
>>>
>>> -----Original Message-----
>>> From: Gordon K Smyth <smyth at wehi.EDU.AU>
>>> To: Michael Breen <mbreen at vapop.ucsd.edu>
>>> Cc: Bioconductor mailing list <bioconductor at r-project.org>
>>> Date: Tue, 11 Jun 2013 10:44:34 +1000 (AUS Eastern Standard Time)
>>> Subject: DESEQ ANODEV : A time course study
>>>
>>> Dear Michael,
>>>
>>> What you want to do is easy and fast using the edgeR package, without any
>>> need for ad hoc workarounds like subsetting your data. See McCarthy et
>>> all (NAR 2012):
>>>
>>> http://www.ncbi.nlm.nih.gov/pubmed/22287627
>>>
>>> Best wishes
>>> Gordon
>>>
>>>
>>>>
>>>> Date: Mon, 10 Jun 2013 08:41:49 +0200
>>>> From: Simon Anders <anders at embl.de>
>>>> To: bioconductor at r-project.org
>>>> Subject: Re: [BioC] DESEQ ANODEV : A time course study
>>>>
>>>> Hi Michael
>>>>
>>>> On 08/06/13 03:00, Michael Breen wrote:
>>>
>>>
>>>>
>>>>>
>>>>> What we aim to do is to test for DE of transcripts across all 3 time
>>>>> points for disease and controls seperatly (using DESeq ANODEV) but we
>>>>> want to be able to identify at which time points these transcripts are
>>>>> being DE. In other words, we want to compare DE transcripts with
>>>>> respect to specific time points between cases and controls. Our
>>>>> remaining code looks like this:
>>>>>
>>>>> fit0 <- fitNbinomGLMs (cds, count ~ timecourse)
>>>>> fit1 <- fitNbinomGLMs ( cds, count ~ timecourse + condition )
>>>>> str(fit1)
>>>>
>>>>
>>>> One possibility would be to subset your data to only samples from one
>>>> time point and then test cases against control to see the genes that are
>>>> DE at this time point, then go on to the next one. If you consider this
>>>> a post-hoc test and only look at the genes which show overall
>>>> sensitivity, you can probably be more lenient on the significance
>>>> threshold. Maybe other people on the list have input on this point.
>>>>
>>>> Simon
>>>
>>>
>>> ______________________________________________________________________
>>> The information in this email is confidential and intended solely for the
>>> addressee.
>>> You must not disclose, forward, print or use it without the permission of
>>> the sender.
>>> ______________________________________________________________________
>>>
>>>
>>
>>
>> [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> Bioconductor mailing list
>> Bioconductor at r-project.org
>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>> Search the archives:
>> http://news.gmane.org/gmane.science.biology.informatics.conductor
>
______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}
More information about the Bioconductor
mailing list