[BioC] maSigPro and "vars" argument

Mª José Nueda mj.nueda at ua.es
Tue Sep 13 13:33:34 CEST 2011


Hi Andrea,

In the regression model of each gene you will have included the 
"statistically significant"  variables that are the variables that make 
change the response (gene-expression). For each gene you can have several 
significant variables. Examples:
- A gene with flat profile for group1 and linear trend for group2, in the 
model: time and time^2 will not be significant and time_group2 will be 
significant.
- A gene with linear profile for group1 and quadratic profile for group2, 
will have time, time2_group2 significant and time2 no significant.
Anyway this is difficult to interpret.
I think that if you are looking for genes that change in whatever way you 
have to choose vars="all" and you will have only one group with significant 
genes. If you need an specific question, then you have to decide how to use 
the 6 groups of genes you have when vars="each".

Good luck,

María.

--------------------------------------------------
From: <andrea.grilli at ior.it>
Sent: Monday, September 12, 2011 2:59 PM
To: "Mª JoséNueda" <mj.nueda at ua.es>
Cc: <bioconductor at r-project.org>
Subject: Re: [BioC] maSigPro and "vars" argument

> Hi María,
> thank you for your detailed explanation.
>
> Only some doubt remain regarding point 3: looking at my data as  example, 
> I get 6
> different variables (and than 6 different results), but it's difficult  to 
> me to
> understand to what correspond each variable, in particular how to  manage 
> the 6 different
> groups of clustering I get as results.
> Maybe my problem is to understand how variable itself is created: I  know 
> it comes from
> regression model, but nothing more.
>
> Thanks in advance,
> Andrea
>
>
>
> Citando Mª José Nueda <mj.nueda at ua.es>:
>
>> Dear Andrea,
>>
>> 1) Your experimental design is correct.
>> 2) Your explanation about the 2 groups you have when vars="groups" is
>> also correct. Normally the first group is a reference (the control
>> group) and maSigPro looks for genes that have differences between other
>> treatments and the control. If you want to find genes with changes in
>> time for the second group you can make 2 things: -Selecting group2 as
>> the reference (first group) or, as you say, spliting the data in 2
>> groups. But this last option doesn't give you genes with differences
>> between groups.
>> 3) Using vars="each" you get a many lists as variables you have in the
>> model. The meaning "biologically speaking" depends on the study. This
>> is an option that allows look for specific questions (differents to
>> "all" or "groups") that a user can be interested in.  For instance,  if
>> you are looking for all the genes with linear changes but not quadratic
>> changes or whatever. You can manage these lists of genes to get the
>> question you desire.
>>
>> If you don't understand my answer, please contact me again. Thank you
>> for using maSigPro.
>>
>> María J. Nueda.
>>
>> --------------------------------------------------
>> From: <andrea.grilli at ior.it>
>> Sent: Thursday, September 08, 2011 5:41 PM
>> To: <bioconductor at r-project.org>
>> Subject: [BioC] maSigPro and "vars" argument
>>
>>> Hi to all,
>>> I'm analyzing time series experiment with maSigPro package as first 
>>> time, and I get problems to understand if experimental design is 
>>> correct or not, in particular I'm doubtful with "vars" argument.
>>>
>>> Data comes from Affymetrix gene chip from 2 different cell lines, 4 
>>> time points, 2 replicates at each time. I normalized with RMA,  and 
>>> filtered out low expressed/low changing genes, getting from   initial 
>>> 54k probes about 12k probes.
>>>
>>> I'm interested in genes varying (i)in either cell lines between the 
>>> different time points (ii) between the two cell lines across time.
>>>
>>> I did the analysis with vars argument as "groups", getting these 
>>> comparisons:
>>>> (ts.analysis$sig.genes$)
>>> ts.analysis$sig.genes$Group1        ts.analysis$sig.genes$Group2vsGroup1
>>>
>>> So, If I well understood, I have 2 gene sets of significant genes,   the 
>>> first with those changing across time in Group1 cells, the   second with 
>>> those changing in Group2 vs Group1 cells across time.
>>>
>>> My questions: how can I also get significant genes for Group2??   Should 
>>> I split the experiment in two parts and performing separately?
>>> Last question: using vars = "each", what I exactly get? I mean 
>>> biologically speaking...
>>>
>>>
>>>
>>> This is my design matrix:
>>>          Time Replicates Group1 Group2
>>> wt22_g21    21          1            1    0
>>> wt22_g7      7          2            1    0
>>> wt36_g21    21          1            1    0
>>> wt36_g7      7          2            1    0
>>> Saos1_g21   21          5            0    1
>>> Saos2_g21   21          5            0    1
>>> Saos1_g7     7          6            0    1
>>> Saos2_g7     7          6            0    1
>>> wt22_g0      0          3            1    0
>>> wt22_g14    14          4            1    0
>>> wt36_g0      0          3            1    0
>>> wt36_g14    14          4            1    0
>>> Saos1_g0     0          7            0    1
>>> Saos2_g0     0          7            0    1
>>> Saos1_g14   14          8            0    1
>>> Saos2_g14   14          8            0    1
>>>
>>> This is the command line:
>>>> ts.analysis <- maSigPro (Data, parameters2, min.obs=4, rsq=0.7, 
>>>> step.method="backward", pdf = TRUE, main = "./results.pdf", alfa = 
>>>> 0.05, degree = 2, k = 9, vars = "groups")
>>>
>>> I checked in Bioconductor documentation, but things remain confused to 
>>> me.
>>> Any clarification is really appreciated,
>>> Thanks,
>>> Andrea
>>>
>



More information about the Bioconductor mailing list