We envisage LD Hub as a useful hypothesis generating tool, providing an easy method of screening hundreds/thousands of traits for interesting genetic correlations that could subsequently be followed up in further detail by other approaches such as pathway analysis (Segrè et al., 2010) or Mendelian randomization (Davey-Smith and Ebrahim, 2003). For example, under most models, a causal relationship between two heritable traits should induce a genetic correlation between the two phenotypes (assuming individual differences in the causal trait are influenced by genetic variation). LD Hub could be used to screen a large number of putatively causally related phenotypes quickly and easily for evidence of genetic correlation, and the most promising candidate pairs could then be followed up by selecting appropriate genetic instruments and performing formal instrumental variables analysis (Evans Davey-Smith, 2015) which can be implemented via the online platform MR-Base (www.mrbase.org/beta). This framework could be particularly useful in the dissection of high dimensional molecular networks where the number of possible pair-wise relationships may be extremely large.
