I’ve posted a brief comment on PubPeer to address questions on my preprint. The main points are about population history under neutrality, and how explaining the data is training on the test set.

In the London Review of Books. Malaparte creeps me out, and I don't trust him, but he works something like an ostinato here that I admire. https://t.co/0sdwEx6m0i

The Traitor Curzio Malaparte, translated by Walter Murch

It should have been an opinion piece.

The main points are that I treat intra-host divergence the same as Pekar et al., and that assuming a condition is satisfied by an unspecified process is the equivalent to not applying that condition.

It’s up :

Argh! It's Hensel, not Hensell. Sorry, Zach.

Zach Hensell has a reply on PubPeer - some ideas about intra-host divergence, and reiteration of the argument that separate introductions of lineages A and B were in the hypothesis posed by Pekar et al. My response might clear moderation tomorrow. https://t.co/IRShZ8kINb

bending the algorithm to my will by providing more detailed information about my interests

I think there’s a reasonable biological case for exploring the effects of a different clock, as suggested by @dr_handler. I think neutrality should be kept by default.

For a Bayes factor to be meaningful, its models must express prior expectations. Their performance can then reflect the plausibility of explanations. Were a chosen to explain the data, its performance would merely reflect the post-hoc fit, making the Bayes factor meaningless.

To obtain a higher Bayes factor, a biological reason is needed to relinquish neutrality or relax the clock. However, many comments eschew biology and ask instead for a model that explains the data, which is not relevant to the Bayes factor.

Were I to model reservoir population history, it would induce a distribution over (tx, ty). Marginalising over that distribution would produce a weighted average of the surface, which cannot exceed the maximum. Thus, reservoir population history cannot increase the upper bound.

Under neutrality, upstream population history affects evolution through the branch lengths, i.e. the times from the MRCA to each introduction. I denote these times tx and ty. I explore the Bayes factor across (tx, ty) and report the maximum as a best‑case upper bound (~0.25).

To summarise, I model the upstream ancestry of two lineages from their most recent common ancestor (MRCA) to their introductions. I assume neutral evolution with a strict molecular clock, which are standard assumptions for these data, e.g. when calculating likelihoods in BEAST.

I am looking for a postdoc to develop high-performance algorithms in computational genomics. Email or DM me if interested. For more information, see https://t.co/6JJn0i5k3F. RTs appreciated!

@nizzaneela @norvid_studies @pleometric @voooooogel Mine is always like "Your tone and rhetorical mannerisms conflict with the serious content and desired rigor in your writing. This might be off-putting to some readers."

Michael Weissman has posted an eLetter to Science addressing Zach Hensel's arguments and approach. I think it clarifies the main issues well.

My first international flight was on a C-130J. They gave us a brown paper bag with sandwiches and an apple. The sandwiches were pretty good. The apple was great.