3
$\begingroup$

I'm currently studying a behavioral genetic course, but still, I feel that I'm lacking many basic concepts.
A particular topic I don't understand is missing heritability.
Here's an example regarding the genetics of schizophrenia, this is what is stated in my course material: "GWAS studies have identified 240 genes associated with this disorder, but these only explain less than 10% of heritability, the missing heritability for schizophrenia is around 90%".

What does it mean that 90% of heritability is missing?
I'd like to understand the math behind this concept but I don't care if this statement is actually true, I don't know where this data came from.

PS I know how heritability is generally defined in a narrow and broad sense.

$\endgroup$
3
  • $\begingroup$ Can you please clarify what you are asking: it seems clear from what you wrote that missing = unexplained (by known genetic factors). Does that help or is there some more subtle question that I'm missing? $\endgroup$
    – tyersome
    Commented Dec 26, 2021 at 0:47
  • $\begingroup$ Still unclear. Let's say you have a $ h^2 = 0.5 $. What exactly does it mean 90% is unexplained? How do you obtain 0.45 as a value for this missing heritability? $\endgroup$
    – Mirko
    Commented Dec 26, 2021 at 8:04
  • $\begingroup$ @Mirko It means that if we knew the effects of every SNP, we could make a predictor explaining 50% of the variance. But right now, we only can explain 0.10*50% = 5% of the variance with our current predictors. $\endgroup$
    – BigMistake
    Commented Feb 29 at 3:52

1 Answer 1

2
$\begingroup$

Missing heritability refers to the fact that while we make heritability estimates using e.g. midparent regression, we can't actually find gene variants whose effects we can add up to the amount of heritability we think we should find.

This is in spite of the fact that we have sufficient statistical power that we should have found such gene variants given how many people we have genotypes and phenotypes for. (Under the normal assumption that they have effects at least somewhat independent of the environment or of other genes, and that we are actually observing the relevant genetic variation.)

As an allegory, let's say we have a bucket of apples. There are 20 apples in the bucket. I know that I picked one apple, and my friend Jane picked one apple, but I don't know where the other 18 apples came from, even though I have been watching the bucket and I should have seen anyone else put apples in there.

People are getting very clever with machine learning models that try to relax various assumptions and incorporate extra data to detect weaker effects.

At an extreme, people have resorted to arguments that literally everything in the genome contributes to every trait, the so-called omnigenic model. In this case they leverage gene expression data as a covariate to add information to models, since you will probably never have enough power to find all of those individual associations/interactions. I am unsure if this has gained much traction in the years since.

Perhaps more promising (IMO, I am not sure that anyone agrees with me) is that people are starting to incorporate more different kinds of variants (not just relatively common single nucleotide variants). There is perhaps some evidence that these sometimes rarer but more dramatic variants might explain more variance.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .