How to interpret significant negative genetic correlation?

[Qiang Xu]

Hi ldsc developer:

                         I have found many statistically significant negative genetic correlation(GC) between two traits in your article"An atlas of genetic correlations across human diseases and traits".For example,HDL and cadiovascular disease,anorexia nervosa and obesity,college attendence and AD.However,you did not give us detailed disscussion on these relationship.In my opnion,there are one explanation:Each pair of traits were influenced by common genetic factors,but they have different effect on phenotypes or traits due to different environmental exposures.And the genetic factors making individuals more susceptible to one disease or trait have opposite effect on another disease or trait.

                        Am I right? Can you provide your plausible explanation on negative GC? Thanks a lot!

                          

[Raymond Walters]

Hello,

In the simplest sense, the interpretation is the same as a positive genetic correlation with one of the traits reversed. I.e. a positive genetic correlation between HDL and not having heart disease, between anorexia and not being obese, etc.

More strictly, ldsc is designed to estimate the genetic correlation defined as the correlation between the best linear predictor from SNPs of the first phenotype (i.e. sum of beta*x with true population betas) and the best linear predictor from SNPs of the second phenotype. As with any correlation, this can be positive or negative. Non-zero values indicate some degree of association, but permit a broad range of possible interpretations. Described in terms of negative correlations, these include:

1) The same causal SNPs affect both phenotypes directly, and the SNPs that increase Phenotype 1 reduce Phenotype 2 and vice versa

2) The same causal SNPs affect both phenotypes indirectly, affecting some mediator that has a positive effect on  Phenotype 1 and a negative effect on Phenotype 2

3) The same causal SNPs affect both phenotypes indirectly, affecting some mediator that has a positive effect on  Phenotype 2 and a negative effect on Phenotype 1

4) Phenotype 1 reduces/protects against Phenotype 2

5) Phenotype 2 reduces/protects against Phenotype 1

6) The causal SNPs for Phenotype 1 are in LD with the causal SNPs for phenotype 2, with the SNPs that increase Phenotype 1 being in LD with SNPs that reduce Phenotype 2 and vice versa

7) Any combination of the above occurring at different loci

LD score regression does not distinguish between these scenarios.

LD score regression does not model the possibility of environmental exposures changing the genetic effects (i.e. gene x environment interaction). It’s only focused on correlation of the genetic risk, which is what helps distinguish it from observing correlations between the phenotypes directly. If GxE interactions are present, LD score estimates should roughly correspond to the main effect of SNPs in an average environment.

Cheers,

Raymond

--
You received this message because you are subscribed to the Google Groups "ldsc_users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ldsc_users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ldsc_users/712dc5a4-4115-421b-812f-58b48dee8736%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.