Multidimensional scaling (MDS) clustering functionality?

[Michael]

Hello, 

Would you consider adding the ability to cluster using MDS in addition to the already available PCA? MDS is distance based and Plink 2.0 can quickly compute the all-by-all distance  matrix  (make-king-table).  It would be nice to be able to quickly generate the MDS coordinates using Plink 2.0. 

    Best, 

            Michael 

 

     

[Christopher Chang]

That is likely to be added at some point, but to clarify your use case: is this something you're currently able to do by e.g. running --make-king and then loading its output into R, or do you have too many samples for the distance matrix to fit in memory?

[Michael Harris]

I can currently use make-king to calculate the distances.  I have about 6k bacterial genomes in a multi-sample VCF.  The distance calculation in Plink 2.0 works well. I have not tried loading the distance matrix into R yet.  

--
You received this message because you are subscribed to a topic in the Google Groups "plink2-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/plink2-users/PvyQalRzeoI/unsubscribe.
To unsubscribe from this group and all its topics, send an email to plink2-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/plink2-users/c2b5f227-2eaf-4080-9f90-4f9c9c3838c0n%40googlegroups.com.

[Christopher Chang]

Ok.  It should be practical to work with a 6k x 6k matrix in R.  However, I don't think --make-king (without -table) generates the most useful matrix for MDS purposes; you'd probably want to postprocess the output of --make-king-table to compute a more appropriate distance function, instead.  (plink 1.x --cluster --mds-plot was based on (1 - identity-by-state); that might already be good enough for your purposes?)

[Michael Harris]

Thanks very much for the suggestion.  I will check out -mds-plot in Plink 1.9 .  

To view this discussion on the web visit https://groups.google.com/d/msgid/plink2-users/482625e5-4d2e-4c7c-b713-799818f8bd57n%40googlegroups.com.