what genetic variations in linear regression are affected by the dominance deviation parameter d
not only Aa, but also AA and aa
>>> both get lower compared to when d = m
what are the 5 assumptions of the classical twin design
- equal environments assumption
- no GE correlation (passive, actieve or evocative)
- no GxE interaction
- no assortative mating - random mating is assumed
- no gene-gene interaction
> non additive epistatic effects
what happens when the following assumptions are violated
- C = 0
- D = 0
- equal environment
- random mating
- Cov (AC) = 0
- AxE = 0
- AxC = 0
violation
- C = 0 > A overestimated, C underestimated
- D = 0 > A overestimated, D underestimated
- equal environment
> C mimc A >> A overestimated, C underestimated
- random mating
> can result in r(A,A)DZ > 0.5 >> C overestimated, A/D underestimated
- Cov(AC) = 0
> mimic C, C overestimated, A underestimated
- AxE = 0
> mimic E, E overestimated, A underestimated
- AxC = 0
> mimic C, C overestimated, A underestimated
how to choose starting model
rDZ < 0.5 rMZ >>> ADE
rDZ = 0.5 rMZ >>> AE
rDZ > 0.5 rMZ >>> ACE
>>> but you never know
explain
AxE/AxC
interaction
1.AxE: effects of E depends on level of A
> genetic control of environmental effects
2.AxE: effects of A depends on level of E
> environmental control of genetic effects
how to test for GxE interaction?
test for heteroskedasticity
> regress phenotype on A/E
> is the variance of the residuals equal over all levels of A/E?
> if so, then homoskedasticity, no interaction
> if not, then the model is heteroskedastic, GxE interaction
what are 2 binary examples of moderating effects
- unmarried women show greater levels of genetic influence on depression
- religious upbringing diminishes A effects on the personality trait of disinhibition
GxE moderation: what to do when M seems to be a measured variable, but really is influenced by genotype?
treat M like a phenotype and include it in your model
>> go full cholesky
what model can be used to analyse moderation in bivariate modeling?
what are the advantages/disadvantages?
Purcell’s model (especially when moderator is continuous)
> path loadings are allowed to vary as function of moderator
> moderator features twice: as phenotype and as moderator
advantage: covariance between M and T can also fluctuate as a function of the moderator itself
disadvanteages:
- lots of parameters need to be estimated
- moderator and trait variables need to have the same measurement level
