Joint estimation of tree topology and branch rates with only continuous traits

[joelh251]

Hi all.

I'm attempting to build a dated tree using only continuous morphological traits. No molecular trees currently exist for my taxa.

I can construct trees with strict clocks pretty easily, but things become harder when I try to relax the clock. Since I don't have a pre-existing tree, I can't run a relaxed tree such as described in the relaxed Brownian Motion tutorial.

I'd like to estimate branch specific rates of evolution and jointly infer the topology of my tree using an FBD process and a morphological evolution model. I also want to see if there is variation in the rates of evolution between characters.

Has anyone done anything similar and could offer advice?

[wrightaprilm]

We recently published a manuscript doing joint estimation. We did not relax the characters in the way you’d like to, but we did do some relaxed clock estimation and relaxed BM estimation.

something that is really tricky is that these models are very parameter rich relative to the amount of data. It’s very hard to initialize these models. You may need to include a starting tree.

[joelh251]

Thank you for the fast reply. I actually just read your paper about an hour ago, it was very illuminating.

Would you be able to suggest any good ways to get a starting tree? Unfortunately, I can't give too many details to preserve my supervisor's privacy, but I am confident there are no existing trees to draw from. Could I build a rough one using hierarchical clustering or similar as a starting point?

[joelh251]

Or would it be valid to run just the FBD process first, then use the consensus tree from that for my other analyses?

[wrightaprilm]

There are lots of ways to get a starting tree. A parismony or ML analysis would be fine. Or the consensus from prior work is fine.

[davidcerny]

I think the full joint estimation of topology and branch rates would definitely be preferable, but it might be okay to run a strict-clock analysis, summarize its posterior, and use that summary as the starting tree for a subsequent relaxed-clock analysis...