Help with smartpca?

[mnievesc]

Hi everyone,

This is my first post so I apologize if its not great, any advice on posting etiquette is valued highly!

I am attempting to use smartpca from Eigensoft to perform a PCA with a subset of genotyped individuals and then project another individual onto the plot. I followed the instructions in the README file and made a list of population names and added the poplistname parameter to my par file:

genotypename: PI.merged.ped
snpname: PI4.merged.map
indivname: PI.merged.ped
evecoutname: PI.merged.evec
evaloutname: PI4.merged.eval
numoutlieriter: 0
poplistname: PI.poplist.txt

However, I keep getting this error message everytime I try to run the program:

"fatalx:
XTX has zero trace (perhaps no data)
Aborted (Core dumped)"

I tried using different formats in the par file but nothing fixed the issue.
Does anyone have any advice regarding what I may be doing wrong?

Many thanks!

[Skiaphrene]

Hi mnievesc,



I'm not familiar with Eigensoft's smartpca so I can't help you with this issue specifically. The error is telling you that somehow, with the input data you're giving it, the empirical covariance matrix (X transposed X) has a zero trace, i.e. the sum of it's diagonal elements is 0. I might be wrong but my intuition is that this means that all of your input variables are constants...



If I'm wrong about this I could propose something else. If you're not afraid of using R to perform your PCA then you could use the package FactoMineR (available on CRAN: http://cran.r-project.org/web/packag...neR/index.html) - their PCA function supports the inclusion of "illustrative" variables and individuals that do not participate in the PC construction but that are projected onto the plots afterwards. You can read about this on this page on their website: http://factominer.free.fr/classical-...-analysis.html.



Alternatively if Eigensoft's smartpca returns the "raw' loadings (eigen vector) matrix W, then you can calculate the coordinates of new individuals in the PC space T by multiplying their original (centred, scaled - as with the input) coordinates X by the loadings: T=XW. Check out the Wikipedia page for PCA for the math: http://en.wikipedia.org/wiki/Princip...onent_analysis.



Hope this helps,

-- Alex

[mnievesc]

This is great advice thank you very much. I do use R regularly so I wll give this idea a try!