QTL Multi-model

[Othmane Lamoumni]

Hello everyone,

I have been conducting multi-qtl analysis using the following procedure:

1- Sim.geno()

2- Scanone: Whole Genome + some chrs

3- Selection of significant markers

4- Definition of the object "qtl" with the selected markers

5- addqtl() then adding significant markers (if there is) to the "qtl" object

6- Stepwiseqtl() using the most recent model

7- Fitqtl() to fit the model defined by stepwiseqtl()

I would like if you have any comments or remarks on the following pipeline, but my main question is linked with the results obtained by stepwiseqtl(), apparently, when I re-run the same analysis of the same data, with the same defined model, I get largely different results. If you have any idea about the possibly involved factors, I would be very thankful

P.S: I've tried the MQM process, apparently, all I get is:

Error in mqmscan() : Singular matrix

Othmane

[Thierry Hodehou]

Hi, I'm experiencing the same thing using the CIM method, getting different results after a re-run. Any guidance will help.

Thierry

--
You received this message because you are subscribed to the Google Groups "R/qtl discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rqtl-disc+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl-disc/4a2c3b84-b535-4d96-8d41-99d747c151a0n%40googlegroups.com.

[Karl Broman]

Procedures that involve any kind of randomness may give different results with each run.

In cim() as implemented in R/qtl, missing marker data is first imputed, by default using a single random imputation.

This is very likely the source for differences in results between runs.

karl

[Othmane Lamoumni]

Hello Dr.Broman

Thank you for your response 

I would really appreciate your comment as, you see, one can choose to define a qtl model and use addqtl + addint to fit a multi qtl model.

Yet, when wanting to define qtl intervals, either bayesint() or lodint() accept scanone or qtl objects only, leading to the use of refineqtl, which is also a problem as qtl positions are switched every re-run. 

The stepwiseqtl object though is used to define the intervals with either if the mentioned functions, yet what confuses me about this function is the max.qtl argument, is there a recommended value? As it seems that higher the value more qtls are added, changing the whole model properties which is more confusing?

You received this message because you are subscribed to a topic in the Google Groups "R/qtl discussion" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/rqtl-disc/hhCATy6FB-Q/unsubscribe.
To unsubscribe from this group and all its topics, send an email to rqtl-disc+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl-disc/40566905-dc7e-4388-8e85-37446cb1ca02n%40googlegroups.com.

[Karl Broman]

I'm not able to provide general advice or to say that a particular method or approach is correct.

But I do try to explain use of the R/qtl software, and to resolve issues that arise in its use.

refineqtl() is seeking to optimize the positions of QTL in the context of a multiple-QTL model, and lodint() and bayesint() use the LOD profiles produced.

stepwiseqtl() performs forward selection up to some fixed pre-specified number of QTL (defined by the max.qtl argument) followed by backward elimination.

The "best" value for max.qtl is not known and likely varies across datasets.

But if stepwiseqtl() is used to try to identify the model that maximizes the penalized LOD score, you could then compare the penalized LOD score for the selected model from different runs, and pick the model with maximum value.

karl

[Othmane Lamoumni]

Thank you Dr.Broman

I've already thought of and looped the stepwiseqtl function to maximize the model pLOD, while also looking into conserved qtls throughout different models, I guess that'll help.

Thank you again!

To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl-disc/bb76beec-9fe4-4563-a18d-7058e99da07an%40googlegroups.com.