Include estimates of dominance effects

[kumine]

Mr zhou,

GEMMA only estimates the additive effect. for heterozygous population, it's very important to estimate dominant effect. Maybe I can record the genotype (AA,Aa,aa) as (1,1,0), (0,1,1) and (0,1,0) respectively to estimate dominant effect. But it is reasonable to estimate dominant effect base on additive effect as NULL model . So I hope dominant estimation can be added into GEMMA. thanks.

WangMin

[pcarbo]

Hi Wang Min,

In practice, we often find that the additive genetic variance is much more important than dominant/recessive effects. This is typically the case for complex phenotypes in human populations, and is often the case in other organisms/populations. Incorporating dominance effects into the LMM is challenging (at least if we want the computation to be fast), and is generally not used much in human genetics, therefore we have not put much effort into this direction, but we will keep your request open.

For an LMM method that does estimate for dominance effects, I recommend QTLRel, although note that this package is about 10x slower than GEMMA.

Thank you for your interest in GEMMA, Peter

[pjotrp]

Closed due to inactivity. Please reopen if someone wants to work on it.

[jasongallant]

I'm faced with a similar challenge. Would it be possible to calculate kinship as normally, and then recode as @kumine describes above? Or does this break something downstream? Probably a naive question!

[robwwilliams]

Hmm, Peter is of course correct, but once you have a locus or SNP of interest, it should be trivial to compute the dominance deviation on a per-SNP basis WITHOUT any correction for kinship. This estimate will often be "good enough" for most users. One problem with dominance deviations is that they depend greatly on scale. Log2 expression data will generally show a dominance deviation that is simply due to scaling.