Fig. 5.

The differences between the test set AUROC medians over the 100 data splits for the original GLMnet method and other approaches. The AUROC was calculated for the test data sets in the discovery cohort. The results are presented for each class separately. Positive values indicate, that the compared (new) method had higher AUROC median than the original method and negative values indicate that the original method had higher AUROC median. The results are presented for thinning with total read counts ${10}^4,{10}^5$ and ${10}^6$ in figures A, B and C, respectively. The red lines indicate the means over the eight classes