Likelihood Score Question

[Maya Gupta]

Hi,

 

I have a few questions about the scores in the log file in BAli-Phy’s output folder.

 

In BAli-Phy’s output C1.log file there is a likelihood score (second column in the log file) assigned for every iteration run. I want to understand how this likelihood score is computed. Specifically, what is the sequence evolution model? How does this relate to maximum average likelihood or most parsimonious likelihood (the terminology from POY5)?

[Benjamin Redelings]

Hi Maya,

In the terminology of POY5, bali-phy is using the "maximum average likelihood", just like RAxML, BEAST, MrBayes, etc.  I think you can safely assume that no Bayesian software uses the "most parsimonious likelihood".  I would also say that the "most parsimonious likelihood" does not qualify as a likelihood.

I am not quite sure what you mean by "what is the sequence evolution model?".  Is that a different question, or is it the same question I (tried to) answer above?  I am slightly confused, because BAli-Phy does not have a single fixed model -- you can select different substitution models using the --smodel argument.

Does that make sense?

-BenRI

P.S. If you want more detail on the probability expression for BAli-Phy, you could take a look at Redelings and Suchard (2005): https://academic.oup.com/sysbio/article/54/3/401/1727336 . In that paper we explain how the model works by factoring the probability expression into the likelihood, the alignment prior, and the priors on other parameters.

--
You received this message because you are subscribed to the Google Groups "bali-phy-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bali-phy-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bali-phy-users/0d71a9fe-3817-40c3-a8ee-c3edbb648885o%40googlegroups.com.

[Maya Gupta]

Hey Ben,

Thank you for your response! To clarify, I want to understand how the model BAli-Phy uses takes indels into account when computing the likelihood score. For example, when specifying the GTR model, does the likelihood score account for indels as missing data, characters, etc.?

Thanks!

Maya

To unsubscribe from this group and stop receiving emails from it, send an email to bali-ph...@googlegroups.com.

[Benjamin Redelings]

Hi Maya,

Oh, I see.  I think I should refer you to the "Probabilistic Model" section of the 2005 paper, because it was written basically in order to answer that question.  I could retype it here, but I think any explanation I type by e-mail won't be as complete as the paper.  If you have any questions about that section, feel free to ask!

The short answer is that the likelihood term depends only on the substitution process, and gaps are treated as missing data.

-BenRI

To unsubscribe from this group and stop receiving emails from it, send an email to bali-phy-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bali-phy-users/080b0d18-a8cf-4ff5-a053-c85268a8a8e9o%40googlegroups.com.