I have done phylogenetic analyses for both individual nuclear and mitochondrial markers- I ran the analyses with the correct model of substitution for each marker, appropriate outgroup and my chains converged. I did not think that the resulting tree was problematic as species were monophyletic with significant posterior probabilities. However, it was suggested that I exclude the burnin from my analyses as " including the burnin will severely bias results of this analysis."
I honestly have no idea why this is. In my limited knowledge I have always understood burnin to be a standard practice. Could anyone enlighten me please? Thanks in advance.
Excluding some iterations from the beginning of your runs (burn-in) or using the entire runs should not make any difference in your results as long as you run them long enough. Burn-in is used to remove initial iterations that are far from equilibrium and may end been overrepresented in relatively short runs.
Excluding some iterations from the beginning of your runs (burn-in) or using the entire runs should not make any difference in your results as long as you run them long enough. Burn-in is used to remove initial iterations that are far from equilibrium and may end been overrepresented in relatively short runs.
Hi Kerry!
As explained above, burn-in is common practice in order to remove the initial iterations that are far from (but leading to) the equilibrium. Therefore, burn-in should not change the actual results (represented by the equilibrium) but make them more robust. This is the case if you are applying a "reasonable" burn-in, I would say no more than 25-30%. If your bayesian analysis did run long enough, you should not need a higher burn-in. To answer your question, maybe the reviewer is worried of a high burn-in? If your burn-in is not too high, then it is correct to apply it. To demonstrate this to the reviewer, you could provide the graph showing that you just removed the initial iterations. In order to do so, you can use a software such as Tracer (very simple and immediate to use!), that allow to visualize the log of the iterations and understand how many iterations you should burn.
Hope this was useful!
Yes to both of the above - removing the burn-in generally increases your scores, as less optimal trees are discarded. When you run a nice, long analysis, it is not a problem.
- Badges
- Science topic
- Similar topics
- Biological Science
- Evolutionary Biology
- Phylogenetics