Figure 1From: Dataset size and composition impact the reliability of performance benchmarks for peptide-MHC binding predictionsCross-validations tend to over-estimate predictive performances against blind data sets. For each predictive method, a mean of its distribution of differences in performances between cv_rnd and blind (i.e. cv_rnd - blind) and a standard error of the mean are shown. Hence, a positive mean indicates over-estimation in performance by cv_rnd. Predictive performances are reported as AROC values (Area under Receiver Operating Characteristics curve). Scores of 0.5 and 1.0 indicate a random pattern and perfect prediction, respectively.Back to article page