Fig. 6: Prediction of phenotypes in the MCS.
From: Family-based genome-wide association study designs for increased power and robustness
We computed PGIs for BMI, height and educational attainment (EA) using summary statistics produced by different estimators applied to UKB data (Methods and Table 2). We use GCSE grades as the outcome for the EA PGI because MCS samples are too young to have completed their education. A GCSE (general certificate of secondary education) is an academic qualification based on exams taken at age 16 by nearly all students in England. The outcome here is the average of a transformation of English and Mathematics GCSE grades to normally distributed Z-scores (Methods). Both the sib-differences and robust estimators were applied to UKB data without ancestry restrictions, whereas the other estimators were applied to the white British ancestry subsample of the UKB. Phenotypes and PGIs were normalized to have variance 1, so that the ‘population effect’ of the PGI corresponds to its partial correlation with the phenotype, and the ‘direct effect’ of the PGI, which is the regression coefficient controlling for parental PGIs (Methods), also corresponds to a partial correlation coefficient. a, ‘Population effect’ of PGIs in European ancestry (EUR) subsample. b, ‘Direct effect’ of PGIs in EUR subsample. c, ‘Population effect’ of PGIs in South Asian (SAS) ancestry subsample. Error bars give 95% confidence intervals. EUR and South Asian ancestry (SAS) subsamples were defined in reference to 1000 Genomes24 superpopulations (Methods). EUR sample sizes: 5,285 for BMI, 5,285 for height and 4,145 for EA. SAS sample sizes: 685 for BMI and height and 615 for EA. We did not estimate direct effects of the PGIs in the SAS sample due to its small size.
