This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Bayesian or maximum likelihood

While construction of trees with bayesian (BEAST PLATFORM) and maximum likelihood (ON MEGA) I am getting a slight different results with similar models. Is there any scope for publishing both the results and and if not what method I should consider and what changes I should do?

beast
• 5,323 views • link
updated by Ram Ram • written by nitishd.dave • nitishd.dave

1 answer

Be careful. You might start a war between Frequentist and Bayesians :-)

Both are useful for phylogenetic reconstruction. Max. Likelihood (ML) can be quite rapid. However, Bayesian frameworks are more readily available for parameter rich models. I tend to use Bayesian more partially for this reason.

Making trees can be dangerous, which is illustrated by your problem. You will always get a tree from an analysis. You need to know which tree to 'trust' and perhaps more importantly you need to know when you cant 'trust' a tree. I often see trees being published that have been made with little to no effort, which undermines the study.

Is your phylogenetic model actually fitting your data? If your getting different trees then at least one of them is not - maybe both are not good!

Phylogenetics is very easily taken for granted, because its easy to get a tree. However, if you really want a robust tree - do a proper phylogenetic analysis. Read this and contact me if you need help:

http://www.mdpi.com/2079-3197/3/2/177

Good luck :-)

R

• 0 views • link
updated by Ram Ram • written by moranr moranr

sir I am using same model gtr +i+ g in beast and in mega (maximum likelihood) I am getting different results in internal clading however results for my hypothesis are similar in both the methods. what should I do??

• 0 views • link
updated by Ram Ram • written by nitishd.dave • nitishd.dave

There could be a number of problems for example:

  1. Your Trees are not converged. -- Probable
  2. This is a promiscuous clade in the sense that your model can't identify its true position. If this clade is not directly relevant to your study , simply remove it , run again and you should have the same answer.
  3. In theory ML can use priors to act similarly to Bayesian methods, when this is the case both answers should be the same. It's just a philosophical approach. Sometimes, there are distinct differences in the frameworks. If this is the case you need to evaluate this and decide which is more appropriate.
• 0 views • link
updated by Ram Ram • written by moranr moranr

Log in to answer this question.

More posts like this