Because large-scale GWAS summary statistics are accessible for an ever-increasing number of traits and tools are now available for using summary statistics to easily identify clusters of genetically correlated traits,20 there will be many sets of traits to which MTAG could be applied. Which potential applications will be most fruitful? Our theoretical results indicate that, relative to the single-trait GWASs, MTAG will improve polygenic prediction quite generally. For identifying individual loci, MTAG will yield the greatest gains in statistical power and little inflation of the FDR for traits with high genetic correlation. We caution, however, that the FDR can become substantial if MTAG is applied to a large number of low-powered GWASs or to GWASs that differ a great deal in power—conditions that do not apply to our empirical application here. In all applications of MTAG, we recommend conducting FDR calculations and, of course, conducting replication analyses if possible.
