This would normally be handled as a QTL (quantitative trait loci) experiment. You will have enough makers to be able to track chromosomal segments from each parent to the progeny, and therefore be able to test whether the grand-parental origins of each region have an effect on the trait. The down-side is that any interesting regions found will not be localized very well, i.e. there will be evidence for a segregating effect, but it is likely to be known to only ~20cM (depends on the size of the family).

Thank you for the great answer Mr. Dodds. However, if I were looking the problem from the software side, would it complain if I perform association mapping? Would these results make any sense? I read some papers defining the diversity between parents as a key momentum for performing AM, which is logical and also needed for classical QTL analysis, but neither one paper says explicitly AM could be performed in one biparental population. Would these results be comparable to regular QTL analysis results? Also, I was wondering which software could handle data matrix with 1M markers (except Tassel) for regular QTL analysis?

you can try using the linkage analysis and association analysis (i heared someone doing the same thing for the biparental population in tomato and got interesting results )  for thebiparental population, and compare the results.   

but i can only find the paper which using the association analysis in the multiple biparental populations, you can read the paper: Xiao et al., Genome-wide dissection of the maize ear genetic architecture using multiple populations. New Phytologist (2015) doi: 10.1111/nph.13814

GBS data requires extensive filtering and I would be very surprised if you have 1 million markers in a biparental population post-filtering by missing data, MAF, etc. The number remaining will likely be few enough that you can use Rqtl for traditional QTL mapping, especially if you bin the redundant markers. If you use association mapping you will probably still identify the major QTL but will have trouble publishing the results.

I agree with Sandra that you will most likely identify the same regions with both analyses. I also agree that the QTL analysis can be done with fewer makers (so removing some with filtering is no great loss); you just need enough markers to track inheritance from each parent (i.e., find the recombination points).

Doing a QTL analysis will force you to think about your data structure more clearly. It will identify in which parent the QTL is segregating. LDLA (linkage + association) will use the association in the parents, but that is only a sample of 2. It would be better to understand the QTL results first, and then look at the regions where both parents are segregating and for them find the region(s?) where the phase is the same.

The population structure, which is normal in biparental populations, will increase the number of false positive associations using GLM or MLM.  I would stick to traditional QTL approache.

Thank you all for the answers. After these suggestions and some reading it seems standard QTL analysis is the best choice for my data. I still have to find out how to perform this analysis with such a large set of markers. In addition, i would like to coment the amount of markers. 950000 SNP markers have been filtered and imputed (partially by the PANZEA.org crew, partially by me) from GBS data. There still is some filtering to do, but I believe the number of markers will not deviate a lot from this number. Also, I am aware of QTL analysis false positives proportional increase with increasing number of markers, but since I have two sets of genotyping data (aforementioned large GBS and SSR analysis with some 2100 loci), I hope differentiation of false positives should not be a problem. Thankful to all for the great and useful advices!

Thanks again for the answers!

I would have another question regarding qtl mapping. If I understood it correctly, rrBLUP package or any blup package does not need any genetic map information to provide the assumptions of marker effects. However, for QTL analysis it seems to be crucial to have a genetic map. So, is it possible to somehow fix the genetic map that looks as the one in attatchement? That is actually somewhat fixed one mapped with "morgan" function of R/QTL using thinned by position SNP markers from GBS.

Is there even some robust (and maybe easy) solution for usage of GBS data in QTL mapping?

Thanks!

Vlatko

Hi Vlatko,

Yes, you will need a map for QTL mapping, and one created from the current dataset should be OK. You should be OK with a thinned map, as all you require are enough markers to help you track recombination events - additional makers within the same recombination bin will not give any extra information for QTL mapping.

I would recommend following Karl Broman's tutorial for map construction with Rqtl: http://www.rqtl.org/tutorials/geneticmaps.pdf

The tutorial includes many tips for shrinking your map size. If you are still concerned with map inflation, the Rqtl google group is a very active forum specializing in map construction and QTL mapping. You could post your map there and request help.

I have posted in the group allready, and am waiting for a response. However, I am aware that greater map density produces a larger map since it is able to fine map recombination events, so I am interested to what extent an expansion can be considered tolerable? Also, i can't tell why some intervals that do not fall into centromere positions, appear so unsaturated with markers?

There is a part of mr. Broman's tutorial by which it is possible to identify (and delete) potentially missplaced markers, so i will try that one, but still, what map size can be considered a good one?