Genomic/Epigenomic features

The main type of features we used is the read counts from various sequencing experiments mentioned above. We firstly downloaded the alignment files (bam files) from ENCODE website. For the sake of consistency, we first convert the genomic coordinates to hg19 human reference genome assembly if the original mapping is for hg38. After that, we retrieved the genomic location information of each read from the alignment files and then count the number of reads located in each pre-defined 200bp bin across the whole human genome. For the experiment with multiple technical replicates, we merged the read counts into a single feature using their mean values. For other features, such as annotation scores, which are typically precomputed in base level, we assigned the score to the CpG locus according to the exact genomic coordinate as the feature.