Feedback 2020-06-30

[richelbilderbeek]

There are three remaining questions which I hope you can clarify:

1. The main text now says nothing about how the twin tree is generated. Is that on purpose?

Yes, due to the words count. It is in the Supplementary Materials.

2. The supplementary material is quite clear on that this was done with a BD model (because it is the default). However, as I stated earlier, even though BD is used, it may be then in practice extinction was zero, so it was a Yule model after all. If this is so, then I think this should be added to the text in the supplementary material. I am willing to do this, once you tell me whether this is indeed the case. What were the estimates for lambda and mu? Must be easy to replicate.

These are the parameters used to generate the twin tree:

• lambda_bd: 0.199967443181565
• mu_bd: 0.000157467752061131

Look quite Yule to me as well :+1:

Details at https://github.com/richelbilderbeek/pirouette_article/issues/118

As I stated previously, to show the baseline inference error, the inference models MUST be identical to the generating models for the twin tree. Thus, this point is critical. No re-analysis needs to be redone if the extinction rate used to generate the BD tree was (close to) 0. But I do think that the pirouette package should by definition use the models used to generate the twin tree for inference for both the true and twin tree (at least in the generative experiment), or equivalently, use the models used for inference of the true tree for generating and inferring the twin tree. Hence the default for the twin tree should not be a particular model, but should depend on what models were chosen to infer the true tree. I might even argue more strongly that this should not even be a default, but the only possible setting. However, I do see merit in leaving some flexibility to the user, so a default is fine with me.

3. The exemplary tree consists of 5 taxa instead of 6. That's fine of course as it's just an example. But this makes me wonder about two things:

a) Is this reduction in number of taxa the reason why the ESSes are now a lot higher?

Yup, that is indeed the case.

b) Is this reduction also done for all the other experiments in the supplementary material (looking at the distribution of trees, looking at DNA sequence length)? It's not necessary of course, but I just wonder about this now.

For those experiments, a distribution of trees was used, that were generated in the same fashion. The main example simply picks another one of those trees.

[richelbilderbeek]

Sent email. Thanks @rsetienne!