If you don’t have access to the within-sample LD matrix (e.g., using 23andMe sum stats), how can you control for false positives generated by using an external LD matrix.
For example:
European 23andMe GWAS sum stats
LD matrix:
1000 Genomes European HapMap3 LD matrix.
Replying to “If you don’t have access to the within-sample LD m…”:
This is a great question and one that regularly crops up - unfortunately there isn’t a way to guarantee that it will be well matched - there is a related issue that meta-analysis can cause some challenges in the behavior of the relationship with LD [related to Simpson’s paradox] - which is nicely laid out in this paper: Meta-analysis fine-mapping is often miscalibrated at single-variant resolution - PubMed
Replying to “If you don’t have access to the within-sample LD m…”:
You may want to check out the PolyFun paper Functionally informed fine-mapping and polygenic localization of complex trait heritability | Nature Genetics . They have given good benchmarking for how to choose the best matching LD matrices, and how to mitigate false positives. For example setting the number of causal variants to a lower number, like 2-3, can lower false positive rates. There’s always the option of using ABF
Replying to “If you don’t have access to the within-sample LD m…”:
Matching on genetic ancestry as much as possible is important but in general in sample LD is very important