All analysis was done using a bisulfite sequnecing data analysis pipeline DNMTools developed in the Smith lab at USC.

Mapping reads from bisulfite sequencing: Bisulfite treated reads are mapped to the genomes with the abismal program. Input reads are filtered by their quality, and adapter sequences in the 3' end of reads are trimmed. This is done with cutadapt. Uniquely mapped reads with mismatches/indels below given threshold are retained. For pair-end reads, if the two mates overlap, the overlapping part of the mate with lower quality is discarded. After mapping, we use the format command in dnmtools to merge mates for paired-end reads. We use the dnmtools uniq command to randomly select one from multiple reads mapped exactly to the same location. Without random oligos as UMIs, this is our best indication of PCR duplicates.

Estimating methylation levels: After reads are mapped and filtered, the dnmtools counts command is used to obtain read coverage and estimate methylation levels at individual cytosine sites. We count the number of methylated reads (those containing a C) and the number of unmethylated reads (those containing a T) at each nucleotide in a mapped read that corresponds to a cytosine in the reference genome. The methylation level of that cytosine is estimated as the ratio of methylated to total reads covering that cytosine. For cytosines in the symmetric CpG sequence context, reads from the both strands are collapsed to give a single estimate. Very rarely do the levels differ between strands (typically only if there has been a substitution, as in a somatic mutation), and this approach gives a better estimate.

Bisulfite conversion rate: The bisulfite conversion rate for an experiment is estimated with the dnmtools bsrate command, which computes the fraction of successfully converted nucleotides in reads (those read out as Ts) among all nucleotides in the reads mapped that map over cytosines in the reference genome. This is done either using a spike-in (e.g., lambda), the mitochondrial DNA, or the nuclear genome. In the latter case, only non-CpG sites are used. While this latter approach can be impacted by non-CpG cytosine methylation, in practice it never amounts to much.

Identifying hypomethylated regions (HMRs): In most mammalian cells, the majority of the genome has high methylation, and regions of low methylation are typically the interesting features. (This seems to be true for essentially all healthy differentiated cell types, but not cells of very early embryogenesis, various germ cells and precursors, and placental lineage cells.) These are valleys of low methylation are called hypomethylated regions (HMR) for historical reasons. To identify the HMRs, we use the dnmtools hmr command, which uses a statistical model that accounts for both the methylation level fluctations and the varying amounts of data available at each CpG site.

Partially methylated domains: Partially methylated domains are large genomic regions showing partial methylation observed in immortalized cell lines and cancerous cells. The pmd program is used to identify PMDs.

Allele-specific methylation: Allele-Specific methylated regions refers to regions where the parental allele is differentially methylated compared to the maternal allele. The program allelic is used to compute allele-specific methylation score can be computed for each CpG site by testing the linkage between methylation status of adjacent reads, and the program amrfinder is used to identify regions with allele-specific methylation.

For more detailed description of the methods of each step, please refer to the DNMTools documentation.