Bioinformatics with Python Cookbook - Second Edition

There's more...

Unfortunately, you cannot run a single instance of admixture to get a result. The best practice is to actually run 100 instances and get the one with the best log likelihood (which is reported in the admixture output). Obviously, I cannot ask you to run 100 instances for each of the 7 different Ks for this recipe (we are talking about two weeks of computation), but you will probably have to perform this if you want to have publishable results. A cluster (or at least a very good machine) is required to run this. You can use Python to go through outputs and select the best log likelihood. After selecting the result with the best log likelihood for each K, you can easily apply this recipe to plot the output.