Jie Zheng, Tom Richardson, Louise Millard, Gibran Hemani, Chris Raistrick, Bjarni Vilhjalmsson, Philip Haycock, Tom Gaunt
Review posted on 29th August 2017
Identifying phenotypic correlations between human traits or diseases is a very important problem. The authors developed PhenoSpD to estimate phenotypic correlation structure by utilizing existing summary statistics based methods (metaCCA or LD score regression) and reduce the multiple testing burden for traits by performing a spectral decomposition (SpD) on the estimated correlation matrix. The authors performed simulation studies to assess the performance of PhenoSpD in diverse simulation scenarios. A comparison between metaCCA and LD score regression was also discussed. Using a real data set, the authors well illustrated the performance and possible applications of PhenSpD. However, I have several concerns that I would like to see discussed.
1. My major concern is that PhenSpD is just using existing methods to infer phenotypic correlation structure (metaCCA and LD score regression) and determine the number of independent tests (SpD). What is the novelty or advantages of this method/software compared to a script executing all three existing tools in the order described in Fig1?
2. I am also concerned about the suggested simulation design. While authors simulated diverse variables, they did not simulate important factors in GWASs such as linkage disequilibrium, ethnical diversity/population stratification or number of causal variants. It would be interesting to see the influence of these factors on the accuracy of phenotypic correlation structure estimation.
3. It would be helpful for users if the authors can suggest possible downstream analyses using the output (phenotypic correlation matrix and the number of independent traits) of the software.
4. In page 3 line 40, "Table S2" is the estimated correlation matrix nor 123
metabolites GWAS list used in the study.
5. In page 6 line 43, please correct "correlation structure of the correlation structure of ".