Extended Data Figure 8. Quality metrics for Basenji convolutional neural network accessibility predictions.

a. Precision-recall (PRC, left) and receiver-operating characteristic (ROC, right) curves measuring favorable performance (as measured by positive predictive value, sensitivity, true positive rate, and false positive rate) of Basenji accessibility predictions for cMN7 e10.5. AU denotes area under curve. Dotted lines represent the baseline classification rate.

b. Scatterplot depicting Basenji accessibility predictions vs. true scATAC sequencing coverage for cMN7 e10.5. Each point represents a 128 bp test bin whose sequence was excluded from training. Measured and predicted coverage are positively correlated (Pearson’s R = 0.833).

c. Boxplot summarizing area under PRC (AUPRC) and ROC (AUROC), and Pearson’s R for all samples and replicates. Quality metrics are consistent across samples. Data points depicted in (a) and (b) are highlighted in red. Centre line – median; box limits – upper and lower quartiles; whiskers – 1.5 x interquartile range.