pcangsd --admix VS. fastmixture for assignment analyses?

[KChristie11]

Hi Jonas,

I'm wondering about the similarities/differences and pros/cons of using pcangsd --admix vs. fastmixture for assignment/admixture/STRUCTURE analyses on datasets with called genotypes.

Does fastmixture outperform pcangsd, for example in the case we converted to a PLINK output and how the programs perform admixture analysis on the same input dataset?

Thanks for your thoughts!

[Rosemeis]

Hi!

If you have called genotypes, then I would definitely use fastmixture! The --admix option in pcangsd will be less accurate, and potentially not even faster even though it is based on a least-square approach. We showed in the fastmixture paper that least-square approaches are far less accurate than likelihood-based approaches.

You could try both options and see if they do not converge to the same solution. :-)

Best,
Jonas

[KChristie11]

Hi Jonas,

Thanks for the clarifications! I'll try both was to see how pcangsd compares to fastmixture.

Best,
Kyle