My view of this study is that it is generally underpowered and that, apart from alleles with large effect sizes, it will be not reliable. The Manhattan plot shows that only six SNPs exceeded the generally accepted level of significance of 10-7
(Figure S5), and none of these had the comforting “climbing tower” of less well-defined SNPs to indicate that they are part of a haplotype block showing association (as one of many of the examples of this, see the Simón-Sánchez et al., 2009
SNCA “climbing towers” in their Nature Genetics study of Parkinson’s disease). This does not mean the associations are wrong, but it does mean they may not be reliable. The replication study is an order of magnitude too small.
A general rule is, the more complex the statistical analysis, the less reliable the outcome, because one does not know how many other complex statistical algorithms were tried. The way to do GWAS is to do simple chi squareds across the SNPs, apply a conservative Bonnferroni (10-7) and believe...