[BioC] statistical test for time course data

[Richard Friedman]

Dear Gordon,

	Thank you very much for the clarification. Now that I think of it, the one-against all
is straightforward. However,   If there are any worked examples
you could point me towards for polynomial and spline modeling of the time 
series I would greatly appreciate it. I am especially interested in testing the 
hypothesis that the temporal behavior of 2 treatments are different.

Best wishes,
Rich

On Feb 2, 2013, at 6:50 PM, Gordon K Smyth wrote:

> Dear Rich,
> 
>> Date: Fri, 1 Feb 2013 10:30:16 -0500
>> From: Richard Friedman <friedman at cancercenter.columbia.edu>
>> To: chris Jhon <cjhon217 at gmail.com>
>> Cc: Bioconductor mailing list <bioconductor at r-project.org>
>> Subject: Re: [BioC] statistical test for time course data
>> 
>> Dear Chris,
>> 
>> The F-test in Limma will tell you if at least one point is different from the others. It won't tell you which one.
> 
> The time course example in the limma User's Guide does use an F-test, but limma isn't limited to that approach.
> 
>> The program EDGE (not EdgeR which is a different program entirely) will do the same based on an model which explicitly takes temporal variation into account (this is why I mention it).
> 
> limma can easily do explicit temporal modelling by fitting time course trends, for example as polynomials or as spline curves.  It's just a matter of defining the design matrix.
> 
>> There is also a t-test that deals with whether a single measurement is a member of the same normal distribution as other measurements. I am not sure off hand, how to do this in R. If you implement this in R and do it for all the rows, it may be the test that you want. If you do this, you will be forgoing the empirical Bayesian increase in accuracy in Limma or the explicit temporal modeling in EDGE, but that is up to you. If you do this, you should current for false discoveries.
> 
> limma can test for a single observation as an outlier by creating a design matrix column specific for that observation.  This produces an empirical Bayes version of the classical t-test for an outlier.
> 
> Best wishes
> Gordon
> 
>> With hopes that this help,
>> 
>> Best wishes,
>> Rich
>> ------------------------------------------------------------
>> Richard A. Friedman, PhD
>> Associate Research Scientist,
>> Biomedical Informatics Shared Resource
>> Herbert Irving Comprehensive Cancer Center (HICCC)
>> Lecturer,
>> Department of Biomedical Informatics (DBMI)
>> Educational Coordinator,
>> Center for Computational Biology and Bioinformatics (C2B2)/
>> National Center for Multiscale Analysis of Genomic Networks (MAGNet)
>> Room 824
>> Irving Cancer Research Center
>> Columbia University
>> 1130 St. Nicholas Ave
>> New York, NY 10032
>> (212)851-4765 (voice)
>> friedman at cancercenter.columbia.edu
>> http://cancercenter.columbia.edu/~friedman/
>> 
>> In Memoriam, Hymie Simon
>> On Jan 31, 2013, at 10:19 PM, chris Jhon wrote:
>> 
>>> Hi Richard,
>>> 
>>> Thank you for help.
>>> In my data ,i have one point which i think it is different from other points and i would like to test statistical significance of the difference of this point.
>>> Your suggestion means that there is no direct function in R that i can use,i have to use a package which implement an algorithm.
>>> If so, i think edgeR can do the same analysis too,Am i right?
>>> 
>>> Best Reagards,
>>> Chris
>>> 
>>> On Thu, Jan 31, 2013 at 11:53 PM, Richard Friedman <friedman at cancercenter.columbia.edu> wrote:
>>> Dear Chris,
>>> 
>>> 	Limma can be used to test between time points
>>> treating each time point as a categorical variable.
>>> The program "EDGE" from the Storey lab, can test whether
>>> there is significant change over a whole time course.
>>> 
>>> http://www.ncbi.nlm.nih.gov/pubmed/16357033
>>> 
>>> with hopes that the above helps,
>>> Rich
>>> Richard A. Friedman, PhD
>>> Associate Research Scientist,
>>> Biomedical Informatics Shared Resource
>>> Herbert Irving Comprehensive Cancer Center (HICCC)
>>> Lecturer,
>>> Department of Biomedical Informatics (DBMI)
>>> Educational Coordinator,
>>> Center for Computational Biology and Bioinformatics (C2B2)/
>>> National Center for Multiscale Analysis of Genomic Networks (MAGNet)/
>>> Columbia Initiative in Systems Biology
>>> Room 824
>>> Irving Cancer Research Center
>>> Columbia University
>>> 1130 St. Nicholas Ave
>>> New York, NY 10032
>>> (212)851-4765 (voice)
>>> friedman at cancercenter.columbia.edu
>>> http://cancercenter.columbia.edu/~friedman/
>>> 
>>> "Complex numbers! Ha! Ha! There is nothing weirder
>>> than imaginary numbers. Architects don't need to know
>>> complex numbers. Whenever I get a  negative root for
>>> an area, I throw it out. And don't talk to me about
>>> quaternions. I am not going into computer animation."
>>> -Rose Friedman, age 16
>>> 
>>> 
>>> On Jan 30, 2013, at 11:43 PM, chris Jhon wrote:
>>> 
>>>> Hi All,
>>>> 
>>>> I have data at different time points for time course experiment.
>>>> I have a response for each time point and i would like to test whether the
>>>> difference between response of two time points is statistically significant
>>>> or not.
>>>> my data is linear plot where response on y axis and time on x axis.
>>>> 
>>>> what statistical test shall i use?
>>>> 
>>>> 
>>>> I appreciate any help.
>>>> 
>>>> Best Regards,
>>>> Chris
> 
> ______________________________________________________________________
> The information in this email is confidential and inte...{{dropped:6}}