Abstract
Human height is a polygenic trait, influenced by a large number of genomic loci. In the pre-genomic era, height prediction was based largely on parental height. More recent predictions of human height have made great strides by integrating genotypic data from large biobanks with improved statistical techniques. Nevertheless, recent studies have not leveraged parental height, an added feature that we hypothesized would offer complementary predictive value. In this study, we assessed the predictive power of polygenic risk scores (PRS) combined with the traditional parental height predictors. Our study analyzed genotypic data and parental height from 1,071 trios from the United Kingdom Biobank and 444 trios from the Framingham Heart Study. We explored a series of statistical models to fully evaluate the performance of several PRS constructed together with parental information and proposed a model we call PRS++ that includes gender, parental height, and PRSs of parents and proband. Our estimate of height with an R2 of ∼0.82 is, to our knowledge, the most accurate estimate yet achieved for predicting human adult height. Without parental information, the R2 from the best PRS-driven model is ∼0.73. In summary, using adult height prediction as an example, we demonstrated that traditional predictors still play important roles and merit integration into the current trends of intensive PRS approaches.
Highlights
The prediction of human height has long been of great interest to the medical research community and as a model for complex trait prediction
Our study is the first to evaluate the predictive value of adult height, using all possible variables including parental height, the proband’s age and sex, genetic principal components, and millions of individual single nucleotide polymorphism (SNP)
We demonstrated the power of combining all these together to reach the most accurate prediction of adult height to our knowledge yet identified, with an R2 of 0.82
Summary
The prediction of human height has long been of great interest to the medical research community and as a model for complex trait prediction. During the 20th century, much of the adulthood height prediction has been based on parental information ref. In terms of variance explained and SNP heritability, three papers demonstrated increasing values: (Wright and Cheetham, 1999) in 2010, 45% of variance was explained by ∼300,000 common SNPs ref. (Yang et al, 2015); and (Yang et al, 2015) in 2017, 68.5% of SNP heritability based on the United Kingdom Biobank was estimated in ref. In terms of observed prediction accuracy, the squared correlation between phenotype and predictor (R2) ranges from 0.17 ref. A large R2 from the Lippert et al study could be attributed to two factors: (1) inclusion of gender in the prediction model; (2) based on a cohort of participants of diverse ancestry
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
