How do you calculate root-to-tip dN/dS using free ratio model in codeml?

[DNAngel]

I've read that calculating root-to-tip dN/dS ratios is somehow different from codeml sites, clade, and branch-site models calculates dN/dS and I am wondering what is the difference? Root-to-tip uses the free ratio model but what values are being used and why is it different from obtaining an overall dN/dS value provided by the other models?

Free ratio model produces dN/dS, dN, and dS values per branch so do I add the individual dN/dS ratios for each branch from an ancestor to a species to obtain the "root-to-tip" dN/dS value? Is this better because it is not averaging the dN/dS from the whole phylogeny but taking each branch into account?

Thank you for any help and insight!

[Ziheng]

i never heard of this concept of root-to-tip dN/dS ratios.  it does not sound to make too much sense.

i suppose if you use model M0, assuming the same dN/dS ratio for the whole tree, the dN/dS ratio you get will apply to the whole tree, and thus it is root-to-tip dN/dS as well? 

if the dN/dS ratio varies among evolutionary lineages or branches, then the idea of root-to-tip dN/dS ratios does not make sense, because the value should depend on which tip you use.

ziheng

[DNAngel]

I am not sure myself, I was reading this paper:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5381557/

They said they tested evolutionary rate of their genes by two models. (1) using branch-site models for molecular evolution between partitions and (2) root-to-tip dN/dS which they phrase as the average rate of evolution during the lineage leading to EACH species (not sure what part of the tree here is being labeled, the branches just before each species...just the tips?)

[DNAngel]

Oh I think it is literally from the root of the tree along all the branches leading to each tip.

So if a tree has 5 species, they average the dN/dS ratio on each branch from the root that leads directly to each tip, so there would be three averaged dN/dS ratios being compared. 

I am not sure though HOW they actually ran the PAML analyses to do so since they used branch models for both of their methods.

I have only ran branch-site models so perhaps the branch models allows one to obtain dN/dS ratios along each branch excluding site heterogeneity, and they averaged it that way. I am still not sure what the purpose of root-to-tip dN/dS is or why it would be different from just running branch-site models.

[Ziheng]

I only glanced at the paper, and did not study the paper carefully.  The main focus of that analysis seems to be to correlate dN/dS with brain size.  I suppose for that purpose they needed to reconstruct the ancestral states (or average over the uncertain ancestral states) of brain size and then correlate with the estimated dN/dS ratio.  I suppose one way of doing it might be to estimate branch-specific dN/dS ratios and then use the reconstructed brain sizes for the branches or ancestral species to do a correlation analysis.  A big concern is then the uncertainties and sampling errors in the dN/dS estimates (which should be large as the free-ratios model is parameter-rich) and the uncertainties and sampling errors in the reconstructed brain size (which should also be very large).  

ziheng