Fig. 2: AutoComplete provides accurate imputations across a range of simulation settings. | Nature Genetics

Fig. 2: AutoComplete provides accurate imputations across a range of simulation settings.

From: Deep learning-based phenotype imputation on population-scale biobank data increases genetic discoveries

Fig. 2

a, Average Pearson’s r2 imputation accuracy across phenotypes for a range (1–50%) of simulated missingness (bars denote 95% CIs obtained through 100 bootstraps). b, Comparisons of imputation accuracy per phenotype between AutoComplete (AC) and SoftImpute (SI; next best). Blue dots indicate a significant difference in accuracy (two-sided t-test with\(P < 2.17\times 1{0}^{-4}\) and \(P < 1.34\times 1{0}^{-4}\), adjusted for the number of phenotypes, for cardiometabolic and psychiatric disorder phenotypes). c, Relative improvements in imputation accuracy for binary-valued phenotypes between AutoComplete and each compared method (percentages thresholded at 200% for clarity). Boxes indicate the first, median and third quartiles, and whiskers extend to 1.5× the interquartile range. The psychiatric disorders dataset contained 372 phenotypes and the cardiometabolic dataset contained 230 phenotypes.

Back to article page