Question about running and merging together multiple BayesAss3 runs to increase total iterations and model convergence likelihood

[lexiebsturm]

Hello,

I am working with a RADseq dataset consisting of ~5,200 SNPs, ~750 samples, and 14 populations. I am running BayesAss on my university's HPC cluster which unfortunately only allows jobs to run for a maximum time of 1 week before cancelling them. In that amount of time I can only run about 6 M iterations but some of my pairwise migration models are not reaching convergence within that time. I read in some papers that said they ran multiple BayesAss runs and then "merged" them together to increase the number of total iterations.

My question is can I launch BayesAss to run beginning at the state that a previous run left off on? Or any other recommendations of how to increase the number of iterations so that my models hopefully reach convergence?

Any recommendations are greatly appreciated!

[stevemussmann]

Hi Alexis,

Unfortunately I don't have a way of resuming a run at this moment, but that would be a good feature to implement. I will try to work on this in the future, but I am not sure when I will get to it (I am currently in the process of moving and starting a new job).

The only other recommendation I have is partitioning your dataset by population, or otherwise reducing the number of populations you are running (if possible). For example, if you have some sets of populations among which migration is impossible, try running those sets of populations in different runs of the program. Partly I make this suggestion because BayesAss is known to struggle to correctly assign individuals when you have large numbers of populations (see Faubet et al. 2007: https://onlinelibrary.wiley.com/doi/epdf/10.1111/j.1365-294X.2007.03218.x).

Let me know if you have any other questions,

-Steve