Empirical vs ML-optimised frequencies

[Kenta Renard]

Dear All,

In your experience, to what extent does the use of ML-estimated vs empirical frequencies (F vs FO in RAxML-NG) impact phylogenetic analyses? I am aware that in most cases the likelihood is better, but I have often seen that the best trees in terms of likelihood do not always reflect what is the most biologically plausible hypothesis.

I specifically worried about over-fitting of the data due to too many free parameters, but in practice how does using ML-estimated frequencies impact the number of free parameters? Have you seen cases where ML-estimated frequencies impacted the inferred trees in negative ways (e.g. artefacts or topologies that are likely not correct or inaccurate branch lengths)?

Best wishes,

Kenta

[Alexandros Stamatakis]

Dear Kenta,

You can use ModelTest-NG to determine the best-fit model.

My gut feeling is that the base frequencies shouldn't have too much of
an impact on topologies, but there are some studies on this:

https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-016-0985-x

if you od an ML estimate of base freqs on a DNA data partition the
increase in number of free parameters is 3 and 19 on AA data respectively.

Alexis

> --
> You received this message because you are subscribed to the Google
> Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to raxml+un...@googlegroups.com
> <mailto:raxml+un...@googlegroups.com>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/raxml/b3eb0ed9-cb43-448a-8086-356c1426bf21n%40googlegroups.com <https://groups.google.com/d/msgid/raxml/b3eb0ed9-cb43-448a-8086-356c1426bf21n%40googlegroups.com?utm_medium=email&utm_source=footer>.

--
Alexandros (Alexis) Stamatakis

ERA Chair, Institute of Computer Science, Foundation for Research and
Technology - Hellas
Research Group Leader, Heidelberg Institute for Theoretical Studies
Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology

www.biocomp.gr (Crete lab)
www.exelixis-lab.org (Heidelberg lab)

[Oleksiy Kozlov]

Dear Kenta,

IIRC empirical frequencies are also considered free parameters in ModelTest-NG and RAxML-NG.

The reasoning is that in both +FO and +F modes frequencies are estimated from the MSA, just using
different methods (ML vs. simple counting).

Best,
Oleksiy