Fig. 4

Per-read performance of 5mC prediction in different genomic contexts. A F1 score across four datasets at genic and intergenic regions, CpG islands, shores, and shelves. Promoter is 2000 bp up and down the transcription start site (TSS). B F1 score across four datasets at regions of different CG densities (20%, 40%, 60%, 80%, 100%). C F1 score across four datasets at repetitive regions. We consider short interspersed nuclear elements (SINE), long interspersed nuclear elements (LINE), long terminal repeats (LTR), DNA transposons, and “Others” for other repetitive regions. A–C Evaluation across four datasets using BS-seq as ground truth