Inferring human demographic history from extant genomes is an important goal of population genetics. To date, the sensitivity of coalescence-based methods in detecting population bottlenecks has not been well characterized. In this study, we find that brief bottlenecks, of just a few generations, are undetectable by current methods. A new approach to population inference, Lineage Time Inference (LiTI), uses data-derived windows to demarcate the limits of the genetic data. We find that a sharp population bottleneck at the time of the Youngest Toba Eruption, and also at more ancient timepoints in the human lineage, would be outside the genetic streetlight.
I’m pleased to let you know about a pre-print we just released. It is under consideration at a journal now, and hopefully will be published soon. However, it is not actually peer reviewed yet.
Let Jack and I know your thoughts on our findings.
Thanks for this! This seems a useful study
Unfortunately I really don’t know the methods or literature here so don’t have much substantive to say, but here are some minor comments to start with
the first sentence mentions ancient genomes, but the methods don’t actually take into account data from ancient genomes. This may be misleading, in implicitly setting up the paper in a direction it doesn’t go. I guess conceivably ancestral population data could actually answer some of the questions raised (we just don’t have that much data).
The second to last sentence in the abstract appears to be incomplete.
I find it odd to have a log scale for the time axis in this context, but maybe it is normal for these simulations.
At the end of p. 3 it says “short and brief”, but as these are synonyms it probably should just be brief - if, though, a distinction is intended between short in an absolute sense and short relative to some other timeframe, then this can be clarified
It might also be appropriate to cite Richard Buggs’ comments on this issue, e.g. Adam and Eve: lessons learned | Nature Portfolio Ecology & Evolution Community
I realise of course that this is not a published paper, but it seems worthwhile acknowledging it somehow given that it is scholarly and his input on these matters I think contributed to this line of exploration (?)
Similarly, the work by Steve Schaffner.
I have not added an acknowledgment section but I do plan to do so. I am not aware of any contributions to this by @glipsnort. I’ll run the acknowledgement section by Richard eventually to be sure he is credited correctly. I’m not sure about how to cite the non journal publications here (e.g. TMR4A).
Back in 2018, I individually invited Buggs, Schaffner, and Venema to collaborate with me in writing a paper on this. They all declined. Then this happened. I can disclose now that a BioLogos staff member intervened to disrupt dissemination of my work. I am grateful that the ASA apologized for their part in this.
This interference derailed things for a few years, in part because I didn’t want to throw an undergrad (Jack at the time) into the line of fire.
Buggs did cite this. But this work on SFS never faced scrutiny. If you look back at the exchange, you’ll see this wasn’t enough to convince Buggs when it was first put forward by @glipsnort. I was very skeptical at the time that it was a valid line of evidence. Several valid statements objections were made to which @glipsnort never responded.
Since then, as I expected, several studies have been published that contradict @glipsnort ’s results, showing how a more recent couple is compatible with observed SFS. This does not surprise me. The approach that @glipsnort takes does not appear to be scientifically valid. I don’t think this is a valid line of evidence for the claims he is making.
@glipsnort is aware of these papers, of course, as I’ve brought them to his attention on this forum over the last couple years. What is surprising to me is that he still claims it’s a valid line of evidence.
It’s worth pointing out some key things:
SFS is a simulation based approach that is entirely distinct from LiTI, which is a measurement based approach. We use simulations to validate (as everyone does), but that’s distinct from using simulations to construct virtual test data, as @glipsnort.
SFS doesn’t estimate an Nmin over time, but LiTI does.
I tried replicating @glipsnort’s results with standard software and couldn’t. So there is something off here for sure, and until he allows his work to be peer reviewed, we won’t be able to sort it out. (Once again, @glipsnort is aware of this).
#1 is a critical weakness because at best he can claim to rule out the specific scenario he simulated. But no one thinks that is a plausible model of human history (not YECs, not WLC, not RTB, not even @glipsnort ). There is very good reason to think that making the simulation more realistic will alter the conclusions. So his work relies critically on a particular type of strawmanning.
SFS doesn’t demonstrate why other approaches don’t work.
#1 and #4 are two reasons (of several) that the SFs reasoning is invalid. This is an exceedingly weak case, so vulnerable to criticism I have always declined to even recognize it as a valid line of evidence.
More to the point, LiTI does not rely at all upon SFS. It is entirely independent. It would be a mistake to think the problems with SFS apply also to LiTI/TMR4A.
SFS = site frequency spectrum. I’m sure SFS has authored no papers.
Perhaps. But I’m unaware of any time a bottleneck of a single generation has been considered in the literature, or any method that has been demonstrated capable of detecting one.
Ayala’s work did consider a tight bottleneck, but all his simulations had the bottleneck last for approx 30 generations (if I recall correctly). Why? Likely because a strawman model was what he needed for his preferred conclusion.
Of note, SFS can’t actually “detect” bottlenecks, even if @glipsnort’s argument is correct. It would only be able to rule out specific hypotheses of demographic history, and the hypotheses @glipsnort considered are not held by anyone. Everyone thinks they are implausible, which makes his argument nearly irrelevant.
I should also add that the simulation technology required to simulate the required hypotheses was only very recently was made available. The validation studies in the preprint were not even possible a couple years ago.
Basically, anything prior to 2019 or so is guaranteed to have some severe methodological deficiencies, at least in how they were validated. We’ve been in communication with the SLIM and tskit teams too, and their code base required some improvements to model our cases.
That alone is one reason I’m pretty sure I haven’t missed anything critical. But by all means let me know if you find a relevant paper.
A single generation would be all it takes to change genetic variation. You would lose a lot of rare variants, and amplify other rare variants. However, I may have this all backwards so I will see if I can’t find some papers that will help me wrap my mind around it.