(1) For time varying factors like the BMI and smoking example, can we really study G by E interaction with cross-sectional analysis when we don’t know what comes first.
(2) For studying disease risk similarly I don’t know how to interpret any G by E finding for time varying exposure unless incidence disease outcome is being used. I insist this point as I m seeing reports of PRS by context interaction in studies based on EHR where complete incidence outcomes cannot often be clearly defined. In UKB, where incidence disease outcomes can be clearly defined, there is a very little evidence of non-multiplicative effects of PRS and E.
(3)what is the impact of population stratification that can create G-E correlation and also confounding through other mechanism for G-E interaction study.
For (1) this is very likely having some confounding effects. For example in the GxSES interaction on Education observed in Mostafavi, Harpak et al. , there are almost certainly causal paths going from Education -> SES (and vice versa). Ultimately I think we will need quasi-experimental analyses (e.g. raising school leaving age) or (more speculatively) using non-transmitted effects in parents where the exposure is likely to precede the outcome.
Point (2) is really interesting and I agree we need much better tracking of incident events than we currently have (would also allow for much more interesting longitudinal modeling). Do you have a cite for the null PRSxE effects when looking at incidence by any chance?
I’m not seeing how proportional amplification would give a perfect genetic correlation. Wouldn’t I see a component not explained by genetics. I..e a model that had no environmental term would appear to have a heteroskedastic error term that depended on the genetics( that is the presence of a Y chromosome).
The genetic correlation here is computed only over the autosomes, so if the environmental term is independent of genetic variation on the autosomes then it will only impact heritability (i.e. amplify/dampen) without influencing the genetic correlation.
It would be great if you could invite other experts in the field of behavioral genetics such as Turkheimer, Kendler, Visscher, Plomin, etc., but also Steve Steward-Williams as an evolutionary psychologist or Razib Khan to the discussion, because at this level of discussion it is hard for informed laypeople to evaluate the arguments you make and the implications they have.
I need time to absorb this.
Great summary. A few thoughts
(1) For time varying factors like the BMI and smoking example, can we really study G by E interaction with cross-sectional analysis when we don’t know what comes first.
(2) For studying disease risk similarly I don’t know how to interpret any G by E finding for time varying exposure unless incidence disease outcome is being used. I insist this point as I m seeing reports of PRS by context interaction in studies based on EHR where complete incidence outcomes cannot often be clearly defined. In UKB, where incidence disease outcomes can be clearly defined, there is a very little evidence of non-multiplicative effects of PRS and E.
(3)what is the impact of population stratification that can create G-E correlation and also confounding through other mechanism for G-E interaction study.
Nilanjan
Thanks Nilanjan, these are great points!
For (1) this is very likely having some confounding effects. For example in the GxSES interaction on Education observed in Mostafavi, Harpak et al. , there are almost certainly causal paths going from Education -> SES (and vice versa). Ultimately I think we will need quasi-experimental analyses (e.g. raising school leaving age) or (more speculatively) using non-transmitted effects in parents where the exposure is likely to precede the outcome.
Point (2) is really interesting and I agree we need much better tracking of incident events than we currently have (would also allow for much more interesting longitudinal modeling). Do you have a cite for the null PRSxE effects when looking at incidence by any chance?
Point (3) is definitely a big deal though I'm more optimistic that within-family MR can resolve it (some initial promising results in Howe et al. https://academic.oup.com/ije/article/52/5/1579/7193346).
I’m not seeing how proportional amplification would give a perfect genetic correlation. Wouldn’t I see a component not explained by genetics. I..e a model that had no environmental term would appear to have a heteroskedastic error term that depended on the genetics( that is the presence of a Y chromosome).
The genetic correlation here is computed only over the autosomes, so if the environmental term is independent of genetic variation on the autosomes then it will only impact heritability (i.e. amplify/dampen) without influencing the genetic correlation.
It would be great if you could invite other experts in the field of behavioral genetics such as Turkheimer, Kendler, Visscher, Plomin, etc., but also Steve Steward-Williams as an evolutionary psychologist or Razib Khan to the discussion, because at this level of discussion it is hard for informed laypeople to evaluate the arguments you make and the implications they have.
Would love to do that. Though I don't think I'm making any particularly controversial or disputed claims this post.