How to set numsites for consensus tree？

[NERV-Asuka]

Hello,

When using treePL, the numsites parameter must be configured. According to the tutorial, this parameter should correspond to the number of sites in the sequence. However, in cases where the phylogenetic tree is derived from multiple genes—with individual trees constructed for each gene and then combined into a consensus tree—we are unsure about the most appropriate way to set the numsites parameter.

Could you please provide guidance on how to set this parameter reasonably in such a scenario? For instance, would it be appropriate to set it as the total length of all genes, the average site length, or are there other recommended practices?

Thank you for your time and assistance.

[josephwb]

I think using the total number of sites is correct. The units of branch lengths is expected number of substitutions per site. All of the sites are contributing to the realized branch length estimates. The purpose of specifying the numsites variable is to convert the branch lengths into an actual number of substitutions along branches.

I could be wrong.

To be honest, I am unaware of anyone looking at the sensitivity of this variable. Presumably, for the same set of branch lengths, a larger number of sites will result in more precise estimates? Since treePL is so fast, you could run some sensitivity analyses to gain some piece of mind.

HTH.

[NERV-Asuka]

Thank you very much for your reply.

I conducted sensitivity analyses as suggested. When using the total number of sites, the optimal smoothing parameter identified by cross‑validation was 0.1, while using the average number of sites resulted in an optimal smoothing parameter of 10. The outputs from the two parameter sets showed some differences, with the former producing overall younger node ages across the tree. Based on the results, I find that the divergence time estimates obtained using the total number of sites appear more reasonable (particularly in terms of the ages of key nodes). I think you are correct.

Thank you once again for your helpful response.

[josephwb]

٩(◕‿◕｡)۶