A comprehensive benchmarking and validation study of cross-trait association methods

Cross-trait analyses are a powerful approach for refining our understanding of the genetic architectures underlying complex traits. Although many cross-trait association methods exist, a systematic evaluation of their performance is lacking. We compare true and false positive rates of several other methods using numerical (up to 300 traits) and genotype (up to 4 traits) simulations and introduce a new cross-trait association method, SumRank. For two traits, ConjFDR, SumRank, and GPA showed high true positive rates while maintaining false positive rates below 5%. SumRank and ASSET outperformed other methods for more than two traits. Most other methods had false positive rates well above 5%, with the false positive rates rising with the number of traits. Application of SumRank to eight psychiatric disorders yielded 658 novel loci, tripling the number of known loci. We discuss the differences in performance in cross-trait analyses as well as the risk of inflated false positives.