Fig. 4

Confusion matrices for DeepGRP predicting repeats of four different classes for hg19/chr1 and hg38/chr1. The gold standard annotation was computed by RepeatMasker. Values are averaged over five models and absolute counts are divided by the true number (with respect to the gold standard) of annotated base pairs per repeat class. Values smaller than 0.01 are omitted. The models used for this evaluation are the same as for the other evaluations, i.e. trained on hg19/chr11