Different result among versions, same dataset

[Ittybittyweeny]

Hi!

I was running first version 0.97, and my PCA looked like this:

Now I updated to version 1.11 and ran the exact same code and dataset, but my PCA changed a lot:

I can not find information about how PCAngsd changed so to explain how I am getting different results. What do you think is the issue here? Thank you!

[Rosemeis]

Hi,

Can you also share the commands you used for both runs. :-) Did the number of SNPs differ between the runs?

Best, Jonas

[Ittybittyweeny]

Hi, Thank you for your help. It was super simple since I was just interested in the covariance matrices at the moment. I ran: v0.97 python2 pcangsd097_path/pcangsd.py -threads 2 -beagle "my_beagle.gz" -o out

v1.11 (installed so no need for path) module load python/3.7 pcangsd/1.10 pcangsd -t 2 -b "my_beagle.gz"-o out

In my humble opinion,it looks like the new version is more conservative or stringent. The number of SNPs and samples is exactly the same, these are both runs using the same beagle file as input. Thank you again!

Best!

[Rosemeis]

Okay that is good to know. What about the number of eigenvectors inferred to use in the iterative process? Is there a difference? You can try to add the following parameter in both cases, "-e 2", and see if that is the case.

Best, Jonas

[Ittybittyweeny]

Hi Jonas!

When specifying -e 2 in version 1.11 I get a similar structure than when not specifying it in the older one. Explained variance in both PCs is still less than the calculated with the version 0.97 but I think that is much more similar now and will be using such parameter. Thank you for your help!!!