Large differences in "--cluster --mds-plot" results using plink1.07 or 1.90

[Ezequiel Luis Nicolazzi]

Dear Chris,

I'm running a relatively small (160 inds, 49k SNPs) dataset. One of the individuals was included from a completely different population on purpose (its a dataset for a course). However, when I run the same exact command ( plink --cow --file <inputfile> --cluster --mds-plot 2 --out <outputfile> ) with PLINK 1.07 and 1.90 I get completely different results. Plink 1.9 does not seem to identify the deviant individual, whereas plink 1.07 does. 

I work with multiple livestock populations, and I heavily rely on this analysis! I'm attaching you the 2 mds plots. If you need extra files or the dataset itself, let me know and I can share the link (in private) with you.

Best wishes,

ezequiel

[Christopher Chang]

Hi,

Can you send me the .log files and the dataset?  Thanks!

[Christopher Chang]

This was due to the switch from singular-value decomposition to eigen-decomposition in --mds-plot, since my reading of the multidimensional scaling literature suggested that the latter was more standard.

But apparently I was wrong about that.  The R cmdscale() function uses SVD, so I will switch PLINK 1.9 back to SVD later today.

On Friday, June 12, 2015 at 12:34:53 AM UTC-7, Ezequiel Luis Nicolazzi wrote:

[Christopher Chang]

The June 13 builds should match PLINK 1.07's (and R cmdscale's) output, up to sign flips and rounding errors.

On Friday, June 12, 2015 at 12:34:53 AM UTC-7, Ezequiel Luis Nicolazzi wrote:

[Ezequiel Luis Nicolazzi]

Thank you very much for this and all your efforts in this tool. I'm usong plink1.9 as the underlying engine of a tool for that combines multiple livestock science analysis. MDS is one of our routine analyses (lots of pop structure in livestock).

Thanks a lot. 

Ezequiel

Inviato da iPhone

--
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/mbAHeXikyf4/unsubscribe.
To unsubscribe from this group and all its topics, send an email to plink2-users...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Till Andlauer]

Dear Chris,

I have just tested your new implementation of --mds-plot. I have
compared PLINK v1.90b3r to PLINK v1.90b3q and could hardly find any
differences in the results, yet the runtime increased significantly: For
12710 people and about 500000 SNPs it takes me about 20 minutes with b3q
and 2 hours and 20 minutes with b3r. I would therefore very much
appreciate if you introduced a flag with which one can choose the
algorithm for --mds-plot.

Thanks,

Till

> --
> You received this message because you are subscribed to the Google
> Groups "plink2-users" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to plink2-users...@googlegroups.com
> <mailto:plink2-users...@googlegroups.com>.

[Christopher Chang]

Okay, the eigendecomposition algorithm can be requested with the 'eigendecomp' modifier (e.g. "--mds-plot 3 eigendecomp" instead of "--mds-plot 3") in the June 15 development build.

> an email to plink2-users+unsubscribe@googlegroups.com
> <mailto:plink2-users+unsub...@googlegroups.com>.