Fig. 4 | Nature Communications

Fig. 4

From: Efficient cross-trait penalized regression increases prediction accuracy in large cohorts using secondary phenotypes

Fig. 4

Scatter plots of actual HGT vs. predicted HGT of 20K testing samples from either NHS/HPFS/PHS cohort or UK Biobank which were estimated by 30K or 437K training samples from UK Biobank using LDpred + MTAG and Lasso + CTPR. We provided prediction R2 as well as sex-specific prediction R2 (R2m: PA for men, R2w: PA for women). We showed scatter plots of actual HGT vs. predicted HGT of 20K testing samples from NHS/HPFS/PHS cohort or UK Biobank data to give practical insight on how differently Lasso + CTPR and LDpred + MTAG generate predicted outcomes using 30K or 437K training samples from UK Biobank. Prediction R2 for men is constantly smaller than prediction R2 for women and the use of the different cohort as a validation set produced worse predicted values than the use of the same cohort. Lasso + CTPR generated better predicted HGT than LDpred + MTAG for all analyses

Back to article page