A new “polygenic risk score” (PRS) paper is making some waves, Polygenic Prediction of Weight and Obesity Trajectories from Birth to Adulthood. Since it is open access I suggest you read it.
But basically, they took ~2 million common variants (there are about ~100 million common variants in the world population) in ~300,000 individuals in 4 cohorts, and used it to predict weight. A genome-wide polygenic score statistic. The correlation with BMI of the score is 0.29. This is pretty modest. But it seems to me that the biggest and most important finding is that it seems to capture a lot of the people at the tails of the distribution.
I’m becoming more and more convinced that the best things these PRS scores can do in the near-term is to identify people who are possibly at these tails. In a complex trait context, the tails are where for diseases a lot of the people who are going to have issues later in life exist. People with BMI in the range 25-30 may have a modest increase in risks, but someone who is very obese, with BMI above 35, is at much greater risk. Over 40% of the people in the top decile here were obese. Only 10% of people in the bottom decile were.
This research comes out of the context of earlier work on the heritability of BMI. It’s around 0.75 or so. That means it runs in families. Combined with the fact that in the recent past, or in other nations, there is a great variation in median size and distribution, one can intuit that genetic dispositions and environmental context both help explain the variation we see around us. The modern American environment is clearly obesogenic. When most of the American population were involved in physical jobs on farms the environmental context was very different.
Over the next few years, there risk scores for BMI will get better, and expand to other populations. One thing that some people are pointing out is that we know it’s heritable, so why not just look at your family? As many of you know, Mendelian segregation means that siblings may have quite different risk profiles on the genomic level. Polygenic risk score prediction is I think going to be extremely interesting and informative in the case of traits which are known to be found within families across generations (e.g., autism), but don’t seem to impact everyone. Perhaps we’ll find for a given characteristic expression is random, due to some life event or cofactor such as infection. Or perhaps we’ll find that differences among siblings have some genetic basis in variants inherited from parents?