are branch lengths necessary?

[kmko...@gmail.com]

Hello,

I have a network that I coded up by hand based on previous work, and I want to use PCM on this network.

Is it possible to use just the network topology, or do I have to optimise the edge lengths first?

Cheers,

Chris

[Cécile Ané]

Hi Chris,

Yes, you need branch lengths, and inheritance values also at hybrid nodes. You could set them all to some arbitrary value (say 1 for each edge length, 0.5 for each inheritance). This was done on trees back in the days when researchers didn't have good tools to estimate branch lengths. But it's best if you could estimate branch lengths from data.

We have tools to do this in PhyloNetworks, using pairwise genetic distances between taxa. The advantage to using pairwise distances is that we do not have to assume that all genes has the same trees (as those may be different due to reticulation and/or ILS), and we do not need to know which genes were affected by which reticulation. This approach was used in Karimi et al. 2020 for example and their code for network calibration is on github here. Perhaps this approach might be applicable for the data you have. Or tweakable.

Cecile.

[kmko...@gmail.com]

Hi,

Thanks a lot, I will generate an estimate of some sort.

Best,

KK