Fraction of explained variance in the R function

[Luca Albergante]

Hi all,

I've playing around with the R version of flashpca and it work very well! However I have a small issue: I noticed that the command line version provides information concerning the percentage of explained variance for each component (via the pve.txt file), but I don't seem to be able to get this information directly from the R function. Is there a way to do that?

Best wishes,
Luca

[Gad Abraham]

Hi Luca,

At the moment it's not returned from the R function, but you can do it
yourself by dividing the eigenvalues by the sum of squares of the
matrix entries and divide by n-1, as long as you use the same
standardisation as used for the PCA. For example,

f <- flashpca(X, ndim=10, tol=1e-6)
s <- sum(scale(X, center=f$center, scale=f$scale)^2) / (nrow(X) - 1)
pve <- f$values / s
cumsum(pve)

Gad

> --
> You received this message because you are subscribed to the Google Groups
> "flashpca-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to flashpca-user...@googlegroups.com.
> To post to this group, send email to flashpc...@googlegroups.com.
> To view this discussion on the web, visit
> https://groups.google.com/d/msgid/flashpca-users/f0c8b412-7587-46fc-b180-2c9489646894%40googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Nicole Torosin]

Hello, 

I have been attempting these operations and I keep getting the error: 

Error in scale.default(X, center = f$center, scale = f$scale) : 

  length of 'center' must equal the number of columns of 'x'

I only changed ndim to 6 to accomodate my file f <- flashpca(X, ndim=6, tol=1e-6) 

Any suggestions about what to do to avoid this error?

Thank you, 

Nicole 

On Sunday, September 11, 2016 at 6:37:08 PM UTC-6, Gad Abraham wrote:

Hi Luca,

At the moment it's not returned from the R function, but you can do it
yourself by dividing the eigenvalues by the sum of squares of the
matrix entries and divide by n-1, as long as you use the same
standardisation as used for the PCA. For example,

f <- flashpca(X, ndim=10, tol=1e-6)
s <- sum(scale(X, center=f$center, scale=f$scale)^2) / (nrow(X) - 1)
pve <- f$values / s
cumsum(pve)

Gad


On 9 September 2016 at 20:15, Luca Albergante <luca.al...@gmail.com> wrote:
> Hi all,
>
> I've playing around with the R version of flashpca and it work very well!
> However I have a small issue: I noticed that the command line version
> provides information concerning the percentage of explained variance for
> each component (via the pve.txt file), but I don't seem to be able to get
> this information directly from the R function. Is there a way to do that?
>
> Best wishes,
> Luca
>
> --
> You received this message because you are subscribed to the Google Groups
> "flashpca-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an

> email to flashpc...@googlegroups.com.

[Gad Abraham]

Hi Nicole,

Can you paste all the steps here to reproduce your problem?

Gad

> To unsubscribe from this group and stop receiving emails from it, send an email to flashpca-user...@googlegroups.com.

> To post to this group, send email to flashpc...@googlegroups.com.

> To view this discussion on the web, visit https://groups.google.com/d/msgid/flashpca-users/b2cbe65b-1974-4a25-8197-47ffa957b726%40googlegroups.com.

[Nicole Torosin]

library(flashpcaR)

ACTLRs <- ("AC_TLRs")

 f <- flashpca(ACTLRs, ndim=4, tol=1e-6)

s <- sum(scale(ACTLRs,center=f$center,scale=f$scale)^2)/(nrow(ACTLRs)-1)

And then I receive the following error:

Error in scale.default(X, center = f$center, scale = f$scale) : 

  length of 'center' must equal the number of columns of 'x'

Nicole

[Gad Abraham]

That's because you can't call scale() on a character, it has to be a
matrix (my example code my for a matrix of genotypes X).

If your data is small enough to load into R, then you can just read it
into a matrix, and do as above. Otherwise, it's not currently
supported (although easy to do in principle if I get the time...)

> To unsubscribe from this group and stop receiving emails from it, send an email to flashpca-user...@googlegroups.com.

> To post to this group, send email to flashpc...@googlegroups.com.

> To view this discussion on the web, visit https://groups.google.com/d/msgid/flashpca-users/85991e5b-ade6-4c6e-b4eb-bfb542dae11d%40googlegroups.com.

[Nicole Torosin]

Hi Gad, 

Thank you so much for clarifying. How would I convert my PLINK data into a numeric matrix? Or the original vcf file?

Nicole  

[Gad Abraham]

Forgot another alternative: if you don't mind running the commandline
flashpca, you can get the proportion of variance explained in the file
pve.txt. It's computed internally, but simply not returned in the
current version of flashpcaR...

> To unsubscribe from this group and stop receiving emails from it, send an email to flashpca-user...@googlegroups.com.

> To post to this group, send email to flashpc...@googlegroups.com.

> To view this discussion on the web, visit https://groups.google.com/d/msgid/flashpca-users/445dd263-5604-4e64-84aa-1195ffb2f589%40googlegroups.com.

[Nicole Torosin]

Unfortunately, I have tried multiple times and I cannot get flashpca installed to run from the command line. 

Nicole

[Gad Abraham]

Have you tried the precompiled binaries?

> To unsubscribe from this group and stop receiving emails from it, send an email to flashpca-user...@googlegroups.com.

> To post to this group, send email to flashpc...@googlegroups.com.

> To view this discussion on the web, visit https://groups.google.com/d/msgid/flashpca-users/34fa652a-d51c-4bb9-b9b1-b86432eb4090%40googlegroups.com.

[Nicole Torosin]

Hi Gad, 

I found these and have gotten that to work, thank you!

Nicole