Figure 5.

Here we plot the correlation (Pearson correlation coefficient 0.759, p=3.7e-13) between AUROCs computed using clinically curated and knowledge-base derived gold standards. Error bars for each AUROC couple are 95% Confidence Intervals computed using a bootstrap resampling. We observe that the two gold standards, despite significant disagreements (Table 1), ultimately provide evaluations with reasonable similarity. This result instills a confidence in both gold standards that could not be achieved with a single gold standard.