fgvieira
commented
6 days ago Hi @baileyjc,thanks for your interest in ngsLD!
Before going into your questions, I'll just mention that the --max-kb-distoption is meant to speed-up analysis since (as you notice) an LD analyses on a full chromosome can be quite slow. Depending on the objective of the analyses, it might not make sense to use it (more on that later).
Just to make sure I understood your question:
- when running with
--max-kb-dist 0, ngsLD will also output comparisons between different chromosomes. Are you analyzing a single chromosome or several? - you say you have ~80k SNPs, but 200k linked sites with
dist>20Mb and r2> 0.5? Or do you mean 200k interactions? If the latter, how many sites do you have in these 200k interactions?
Going into your questions, and even though it is a bit hard without looking at the data:
What I found from the smaller file was LD decay decreased to a more or less asymptote at 200kb (x-axis) at 0.05 LD (I think that is the y-axis label?). However, when I filter my file from the
--max_kb_dist 0, keeping only linked sites with an R-squared > 0.5 (column 7) the mean value in column 3 of the output file is 6 million. I wanted to see the sites that were in linkage and the distance between them on average. I also get a max distance of ~30 million, the actual value is slightly less than the full chromosome length. When I only include linked sites that have >20 million base pairs between them and have an R-squared > 0.5, I get 200k linked sites. For a chromosome that is only 30 million base pairs, it is difficult to believe they are linked via recombination but instead by some other evolutionary influence.
Two sites can be in LD for several reasons: (e.g.) selection, mutation rate, genetic drift, or gene interactions. In your case, I'd say (but please check) that most of your sites in LD are at relatively short distances (hence the 200kb asymptote), but that you also have some very distant ones. These sites seem to be LRLD and can reflect (e.g.) drift or gene interactions (these could even be between different chromosomes!).
How to define
--max-kb-dist
It depends on the goal of your study. If you want to study short range LD to look for effects of (e.g.) demography or selection, then your threshold of 200kb sounds reasonable. However, if you want to study LRLD, then I'd use --max-kb-dist 0 (prob with all chromosomes at the same time) and then filter the output for long range interactions.
Hope this makes sense! Best,
PS - You can also calculate LD on your entire dataset, but perform LD_decay analyses on a random subset of interactions. You should get an LD_decay plot similar to when you used --max_kb_dist 1000.
baileyjc
Hello, thank you for providing a great package and for being willing to address issues on Git Hub. It has been very helpful to me in refining my code. I am running into an issue that I hesitate to bring up because I keep wondering if I have not looked hard enough for the information so if that is the answer to my question please let me know and my apologies for the open issue.
That being said, my issue is deciding on what number to give
--max_kb_dist. I have not found a similar issue but my apologies if one has already been opened. I ran ngsLD over a single chromosome ~30 million base pairs in length and ~80,000 SNPs present between the 154 individuals in my dataset with--max_kb_distset to 0. I tried to run the LD_decay script with the file but I believe the ngsLD output file is too large (unzipped ~220 Gb). I used a smaller file (~20Gb) setting--max_kb_distto 1000 and I did not run into issues with the LD_decay script until I used the larger file. What I found from the smaller file was LD decay decreased to a more or less asymptote at 200kb (x-axis) at 0.05 LD (I think that is the y-axis label?). However, when I filter my file from the--max_kb_dist 0, keeping only linked sites with an R-squared > 0.5 (column 7) the mean value in column 3 of the output file is 6 million. I wanted to see the sites that were in linkage and the distance between them on average. I also get a max distance of ~30 million, the actual value is slightly less than the full chromosome length. When I only include linked sites that have >20 million base pairs between them and have an R-squared > 0.5, I get 200k linked sites. For a chromosome that is only 30 million base pairs, it is difficult to believe they are linked via recombination but instead by some other evolutionary influence.However, this brings me to my main question, how are people determining
--max_kb_distgiven the challenges of plotting LD_decay for the full chromosome because of its size and my other larger chromosomes with more SNPs won't finish running in ngsLD given my cluster's 3 day limit? I have seen values from 1 to 10 to 50 for--max_kb_dist(but no real evidence for why they chose the value) and with regard to the LD_decay plot people determine the distance once the exponential line passes 0.1 (is that right?). At what distance are SNPs considered linked versus maybe due to selection or another explanation? I have read through the ngsLD github page and your ngsLD manuscript and I have not been able to find an answer to fully explain what is the appropriate value to use with--max_kb_dist. I have also been trying to find common linkage distances for fish but have not found this information so far. Thank you for your time and again for providing this useful package!