We (with @gatag.bsky.social) are happy to share our discussion, where we explore three cases of matrilocality and genetic matriline connections identified with ancient DNA.
You can find the open access paper published in evolutionary human sciences: www.cambridge.org/core/journal...
Really excited to share our new paper on the Early Iron Age mass grave at Gomolava (9th c. BCE), a study I had the pleasure to work on during my PhD with an amazing group of co-authors across multiple disciplines. Open access link: www.nature.com/articles/s41... 🧵👇
The Association of Greek Archaeologists has published an open letter withdrawing its participation from the EAA Athens advisory committee 🍉
archaeol.gr/anoichti-epi...
To sum up, the effect of the sampling is what we're testing here, and I am confident in our results and their presentation. I am very open to discussing it directly with you on Zoom, if you are interested. I feel like having an actual conversation would be more useful to both of us. Let me know!
It is true that some methods fare better with strong selection than others here, but I think the results from the first part with "ideal" data, and the way they are discussed in the discussion, offer enough context so as not to suggest that some methods are just always bad with strong selection
In the second part, we look at "ancient-like" data by focusing on a sampling scheme based on a real-life scenario. This one also produces streaks of fixed alleles for higher coefficients, albeit to a lesser extent than in the 1,000 gen. version of the "ideal" dataset (Fig. S2).
In that case, while weak selection is now more easily detectable, the sampling does not allow to estimate strong selection. The fact that this is an effect of the sampling is clearly, explicitly described in the figures, results, and discussion.
In this case, we do show that stronger selection is easier to estimate while weaker selection remains undetectable. On the opposite, the dataset covering a longer timespan (1,000 gen.) does produce a long streak of fixed alleles (Fig. S1). This corresponds to not removing the red part on your plot.
I think that last statement oversimplifies what the results show. We assess several sampling strategies. In the first part with "ideal" data, the dataset covering a "short" timespan (100 gen.) corresponds in practice to what you suggest to do on your plot for large selection coeffs. (see Fig. S1).
When sampling over a long period of time, the trajectory can reach fixation too quickly to be effectively covered, and the subsequent streak of time points with a frequency of 1 in itself is uninformative. All methods, not just BMWS, struggle to estimate stronger selection coefficients in that case
We actually discussed a lot about that and decided to keep all simulations. We thought that applying any type of conditioning on loss/fixation would also introduce some kind of bias, as we wanted to also test cases with “bad” trajectories. What you mention here is precisely what our results show.
@anchieri.bsky.social @cegamorim.bsky.social et al. benchmark the inference of selection with aDNA-like time series datasets, showing that ApproxWF can estimate selection with datasets of ∼100 individuals when selection is strong.
🔗 doi.org/10.1093/gbe/evaf234
#genome #evolution #compbio
🧬 Now published in Bioinformatics Advances: "pygenstrat: A Python package for EIGENSTRAT data processing" by @dilekopter.bsky.social
Full article available: https://doi.org/10.1093/bioadv/vbag022
Interested in using aDNA time-series datasets to estimate selection?
Our study "Assessing Ancient DNA Sampling Strategies for Natural Selection Inference in Humans Using Allele Frequency Time Series Data" is now out in GBE! doi.org/10.1093/gbe/... @genomebiolevol.bsky.social @cegamorim.bsky.social
Thanks for putting this together!! I would be happy to be part of the list as well
With all the new people migrating into BlueSky, we have created a small starter pack for aDNA (and other molecules) researchers. go.bsky.app/F4EPLJh 🧪🏺 #evosky #PaleoSky