Hi all.
I have a chromosome segment of roughly 4 Mb that almost does not recombine (barely 1cM overall) but the genetic map was built using another population. The DNA sequence I have from the genome assembly consist of 19 scaffolds (between 3000bp and 1.6 million bp) that are mostly unordered (because of no recombination in the population used for the genetic map).
I have SNPS from RADsequencing of a single population (55 individuals).
I am trying to order somehow the scaffold in that region using a LD approach: I selected the 19 scaffolds and used the algorithm from Zaykin (program MCLD) to calculate r2 values for each SNP within and between scaffold.
My question is: is there a way to summarize the r2 of all the loci within a scaffold in order to get a small number of value per scaffold and use it to do a linear organization of the scaffolds?
For now, I want to suppose that the assembly is good, so I do not want to break the scaffolds. I cannot use any mapping software since it is not a cross, but I still have useful information in there.
I already thought of a Minimum Spanning Tree approach, but the problem resides in the translation of the cluster matrix to one providing a linear organization (i.e. the scaffolds found to be in one group are still not ordered...)
Anyone for an opinion?
Sorry, if you would not mind, SEQ Answers: http://seqanswers.com/ and Biostar: https://www.biostars.org/ are better platforms for these questions and would be answered more promptly as was was the case for me too- as they are highly related forums for your questions ! best wishes.
Interesting! The problem might be that in natural pops LD is rather short, a few kb at most - and usually shorter. You should be able to evaluate the extent of LD in your data by plotting r2~distance within scaffolds, right? So, I would only hope to see elevated r2 between the abutting ends of adjacent scaffolds, and so I would not try to summarize r2 scaffold-wise but rather look for elevated cross-scaffold r2 between markers within 5-10 kb of scaffold ends... does it makes sense?..
I'm not geneticist but I propose you a naive approach :
(1) with your SNPs (genetic markers) obtained from RADseq you can obtain a genetic map;
(2) map (using BLAST...) your markers to their physical location on the scaffolds;
(3) sort the scaffolds according the markers' genetic distance. As said Mikhail, according to the genetic map, it is expected to see "elevated" r2 between the ends of adjacent scaffolds.
Fred, Inra, Montpellier ;-)
Biswapriya, thanks for the link, I will indeed try to ask on those.
Mikhail and Fred, yes, this is exactly what I was thinking to do. I can see that some of my scaffolds present recombination hotspot within them, so using the terminal sections would make more sense indeed, but I was wondering if there was any software automating it.
As for the short LD problems in natural pops, it is true. However, this region is mostly non recombining (and very close to the centromere), so I still see nice patterns when I plot the LD heatmap, and I can see that the ordering of the scaffolds in the region seems wrong (so far in non recombining regions, we followed the alphanumeric order...not so scientific hehe).
So I will give it a try to see if it helps resolving the region.
Thanks for your answers and have a good day! (Fred, apparemment une bonne journée et une bonne semaine, vu qu'enfin il ne pleut plus!!!)
Céline
Hmm, very interesting idea. Could you just use a standard clustering method to order the SNPs based on r^2 (or D') scores (as similarity scores)? That should put close SNPs near to each other, and you can then look at average r^2 values between clusters of SNPs.
Hi David,
Yes I tried. I had r2 values so I used HClust (R) based on the 1-r2 values (it has to be distances so I had to make it such as the biggest values means that the markers are far). I did the hierarchical clustering. This technique is good for finding pairs of scaffolds that are the closest, but then it clusters the remaining scaffolds with no proper linear order. I could only gather a few scaffolds together. I should reorganize my data according to those groups, place them linearly one after another, and run again the LD calculation and compute the new groups pairwise r2 again, and keep building from it. I wanted to have a method that would be as much automated as possible to have a robust and reproducible way of organizing the data (easier for writing the M&M in the publication afterward).
Would you advice any other clustering method?
- Badges
- Science method