What does the weight file mean?

[melop]

What does the weight file mean?

[FerRacimo]

Hi Ray,

Using the outgroup map would be best, as recombination rate changes would be equally biased to be wrong in either of the two daughter populations. That said, if you only have a recombination map for one of the daughter populations and it's not very distantly related from the other two, then it should be fine.

The weight file should be a 3-column file with a header (like the input file header). The first column is the chr, the second column is the position and the third column is the LD weight. The number of positions should be equal to the number of positions in your input file. I would first try running things without the weight file. If the selection signals are strong enough, it won't make much difference to use it or not.

Hope that helps!
Fernando

[melop]

Dear Fernando,

Thank you for the detailed reply.

What if I use the average rec map for the two ingroups? I ran ldhat separately on the two ingroups, and simply averaged the normalized maps. The outgroup rec map is quite different from both ingroups, while the two ingroups have more similar maps.

Perhaps it would help if you can post an example of the weight file in the README file? For now I will try running without.

After I obtained the normalized CLR scores by dividing the raw scores with the chromosome average and standard deviation, what would be a reasonable cutoff to use as the significant threshold?

Thanks a lot!
Best
Ray

[FerRacimo]

Re: recombination maps - yeah, that sounds reasonable if the outgroup map is very different.

Re: cutoffs - the most appropriate (but cumbersome) thing to do would be to generate a set of replicate simulated windows under whatever neutral demographic model you fit your data to, compute a 99% or 99.9% quantile from the distribution of simulations, and then use that quantile as a cutoff for the real data; a quicker and more practical (but less statistically sound) route would be to use an empirical quantile cutoff on the data itself, which in general gives very similar results to the first method. As to the exact cutoff number, that's up to you. People use 99% of 99.9% as a rule of thumb. You could also just rank your results, and look at the top X regions, without specifying a particular quantile.