How to estimate the gene flow betweem sister species using gIMble?

[kuangzhuoran]

Hello, thank you very much for developing this tool.

I saw that the authors used gIMble to estimate gene flow across genome-wide windows in the paper “Chromosome Fissions and Fusions Act as Barriers to Gene Flow between Brenthis Fritillary Butterflies.”

However, I didn’t find information related to gene flow analysis on the GitHub page. Should I expect gene flow estimates to be included in the output files after sequentially running parse, blocks, ......, gridsearch, and simulate?

[GertjanBisschop]

Converting to a discussion, this question might be useful to others.

[GertjanBisschop]

If you fit a model with gene flow, then the gene flow estimates will indeed be included as part of the parameter combinations that are explored during gridsearch. By default processing results get stored in the data store that contains all the data of a single project. You'll have to run query to extract this information from the store.

[kuangzhuoran]

Hi, I have a pair of sister species, A and B. They were originally separated by geography, but that barrier has now disappeared. Species A has undergone several chromosomal rearrangements, and I want to know whether gene flow between the two sister species is reduced across those rearranged chromosomes.

I’ve noticed that gIMble can only estimate gene flow in one direction at a time, so my plan is to calculate gene flow separately for each direction and then take the average.

[GertjanBisschop]

Indeed, gIMble can only fit unidirectional gene flow models.
Obviously, in reality gene flow might indeed go in both directions. The most important question to answer here thus is, what does the gene flow estimate from a model with unidirectional gene flow mean in case we suspect that in reality gene flow goes both ways? One way to interpret this parameter is to see it as the net gene flow. The excess of gene flow in one direction relative to the other. It is definitely not possible to estimate the 'true' gene flow in both directions by seperately fitting two unidirectional models. I would therefore not use these estimate to simply take an average. This will have very little meaning.

[KLohse]

How one should interpret unidirectional gene flow estimates is an interesting and tricky question and answering it would require checks on simulated data.
There are two obvious possible interpretations of a unidirectional m estimate in this:
i) the net difference of m in both directions (Gertjan's suggestion)
ii) the total m in both directions

Given that many of the signals in the data that inform us about m (e.g. topology info in the form of shared heterozygous sites) are not sensitive to the direction of gene flow (all else being equal), I would argue that ii) is more likely. So I think your idea of interpreting the average of unidirectional m estimates for each direction as our best estimate for the the rate of bidirectional gene flow does make sense. However, I would suggest dividing this average by 2! One thing worth trying may be to fit a model of bidirectional gene flow (using SFS based methods) as a sanity check.