Posted at 14:35h in Uncategorized by Eric Turkheimer 0 Comments
Charles Murray and I were going back and forth yesterday about the future of polygenic scores, especially as they might be applied in the schools. Murray, as always, looks to a future (now moved back to 2030) in which all behavior is highly predictable from the genome, so schools that don’t use polygenic scores get sued by parents and go bankrupt.
I know Murray wasn’t thinking specifically about the disorders he listed, but his inclusion of “aphasia” as something schools might be able to predict struck me:
1. Per my essay, if you want to predict these things, why not just use parental phenotype? You could do it tomorrow.
2. Aphasia would be a neat trick.
— Eric Turkheimer (@ent3c) August 22, 2019
The more I think about it, the more the genetics of aphasia seems like an important example. Why would it be a neat trick to be able to predict it? The first problem is developmental heterogeneity. Apahasia is caused by brain injuries to the left hemisphere, mostly either stroke or some kind of traumatic brain injury. So the first problem is that any genetics of aphasia– has someone done the GWAS yet?– would be a complex stew of risk factors for stroke (smoking, blood pressure) and risk factors for head injury (sensation seeking, IDK what else) which don’t in fact have anything to do with each other.
Think of depression in these terms. Why do people get depressed? In terms of the developmental processes that cause it, it’s like aphasia, except much much worse. People get depressed because their marriage falls apart, or someone close to them dies, or because they make some serious error that is exposed in public. All of these things interact with a million pre-existing dispositions, leading some people to wind up depressed. The problem is particularly clear for aphasia because it has such well-defined high-level causes, but that fact is that most complex behavioral traits work that way.
Then there is the stochastic problem. Even if you had a genetic score that was informed by risk factors for stroke and head injury, you wouldn’t be able to do a very good job predicting aphasia. The risk factors themselves, never mind their distal genetic predictors, don’t do a very good job because stroke and head injury both have a huge random element. How could you predict if you are ever going to replace the left front quarter panel on your current car? You will replace it if you get in a particular kind of accident. People get aphasia if they have a particular kind of accident.
A couple of posts ago, I wrote about how thinking about genetic causation remains tied to an old-fashioned kind of medical genetics in which developmental pathways work their way up, slowly but inexorably, from genes to outcomes. That is the way Huntington’s disease works– the Huntington gene has low level biological effects, which in turn affect brain development, which in turn affect cognition and motor function. The causal chain is complex, but unitary– that is why we can talk about there being a “Huntington gene”.
But most complex behavioral phenomena don’t work that way. It isn’t as though there is a developmental process working its way up from the genes that one day manifests itself as aphasia. Aphasia is caused by an event at a higher level of analysis than the level of the genes. Genetic variation can be correlated with this high level cause, but they cannot be this cause. No amount of GWAS is ever going to change the fact that aphasia is caused by strokes, and there is no genetic entity that corresponds in a straightforward way to stroke. That is just the way the world is put together.
The consequences of this kind of thinking are extremely important. It means, contra Murray, that there is never going to be a particularly useful genetics of aphasia, nor an ability to predict it from a PGS. Not in 2030, not when we have 100,000 cases and controls. not ever, not in principle. This is much more generally true about the genetics of complex entities than is currently realized. More theoretically, notice that I am not arguing that aphasia is somehow “environmental” as opposed to genetic. That isn’t the point. The point is that aphasia, as a thing, an entity we can recognize in the world, exists at a neuroanatomical and/or linguistic level of analysis, and that level of analysis does not map onto a genetic level of analysis in a useful way.
Once you start thinking this way, you realize that most of the language that we use for the genetics of complex entities is borrowed from old-fashioned medical genetics, and doesn’t really apply. For example, the idea of “causal variants,” as though there were some variants with direct causal relationships to complex outcomes, and others that are only correlated with the causal variants via linkage disequilibrium. There are no “causal variants” for aphasia; there are only a range of variants that wind up associated with aphasia via a vast matrix of tiny correlations. Or pleiotropy: if there are no causal variants, the idea that some variants have multiple “effects” is meaningless. Or, for that matter, “polygenic”, which implies that there is some countable set of variants that have a discernible effect on an outcome. Or, on the bottom line, “heritability.” Aphasia is, beyond any doubt, heritable, but how heritable it is doesn’t make any difference for our understanding of aphasia.
I have an in press chapter on these topics that I can’t share publicly, but if you are interested email or dm me and I can send you a draft.