You would have to clarify how you ran PLINK (command line arguments) and why you think it is taking away too many SNPs. Have you adjusted the window, step and VIF?

Thank you Dr Tromp

Maybe I made that mistake. I used command --indep. What would be the optimum adjustment for window, step and VIF.

Hard to tell a priori, you could run it for a chromosome with a few settings. The smaller the window, the more SNPs retained. The larger the VIF the more SNPs retained. Default appears to be 50 5 2. So, you might want to keep 50 and 5 and then use 2.5 or 3. VIF of 2.5 corresponds to multiple R^2 of about 0.8. 

Another alternative is to use the pair-wise flag --indep-pairwise, which has similar parameters although the third is the simple r^2 (pairwise SNP).

Thank you.

Muhammad

One would need an order of magnitude fewer SNPs than 1 million, while maintaining complete linkage information. If I remember correctly 10k to 20k was used in the dying days of linkage, before GWAS became feasible. Eg. Affy's 10k array, which achieved ~60% linkage information (probably sample size dependent).

If you fine-tune your VIF to keep 50-100k SNPs, you should be absolutely fine. But the other issue is to ensure even coverage across the full genome. I don't imagine that LD-pruning would leave big gaps, but you should consider checking the range of inter-SNP intervals doesn't rise above some number of the order of 100kbp (I just pulled that off a back-of-the-envelope calculation for 50k SNPs).

Thanks Ricardo. It is very helpful.

I have seen a publication where they used 250 K for initial linkage and then in couple of areas where there were linkage signals they reduced the markers to address the issue of false positive results. 

Muhammad

You may check linkage disequilibrium publications in potato by Dave Douches, he works with 8333 highly informative SNPs http://www.biomedcentral.com/1471-2164/12/302. Now he is probably going to 12000 specially informative in potatoes. From his work linkage disequilibrium can be measured appropriately.

Please take a look to  http://solcap.msu.edu/pdf%20files/6PAAextension.pdf

Dr Marcelo Atilio Huarte

Thank you. This is very helpful.

You really don't need many SNPs for linkage analysis; the initial linkage arrays used ~6,000 SNPs, if I'm not mistaken. Mind you, linkage disequilibrium is only a problem for multipoint linkage analysis, and when founders are missing. 

A few answers above mention 'highly informative SNPs'. 'Informative' in linkage analysis means polymorphic. For SNPs, that means the closer the minor allele frequency is to 0.5, the more informative it is. 

Thank you Dr. Robert Wojciechowski. I'll keep that in mind.

Muhammad