Fig. 3

Within-experiment differences. a Numbers of selected experiments and files to study within experiment differences. Only experiments that contain released files together with revoked files were selected. b Predictive probabilities across experiments summarize the good predictive power in each data subset. c Benchmark of mouse ChIP-seq files in selected experiments illustrates the good predictive power of the models. P_low: predictive probability to be of low quality. d Density curves of within-experiment differences between predictive probabilities for low- and high-quality files. Average differences are large (range from 0.5 to 1). The low-quality probabilities P_low shown in panels b, c, and d were computed by the optimal specialized models trained on the corresponding data subset but without the files from the experiment under investigation