Conversion Details (C→T, G→A)

Bisulfite sequencing relies on chemical conversion of unmethylated cytosines to uracil (read as thymine after PCR). bwa-mem3 --meth models this with an in-memory read transformation applied to every read before the alignment kernel sees the bases.

What gets converted

Paired-end bisulfite reads follow a strand convention:

R1 (read 1): every C in the base sequence is replaced with T. This models the OT (original top) and CTOB (complementary to original bottom) strands as they appear after bisulfite treatment and PCR.
R2 (read 2): every G in the base sequence is replaced with A. This models the OB (original bottom) and CTOT strands.

Single-end mode uses the R1 (C→T) rule for all reads.

The doubled reference built by bwa-mem3 index --meth (or bwameth.py) contains two projections of each chromosome:

f-prefixed contigs (e.g. fchr1): the chromosome with every C replaced by T.
r-prefixed contigs (e.g. rchr1): the reverse complement of the chromosome with every G replaced by A.

Converted R1 reads are therefore alignable to f-prefixed contigs and converted R2 reads to r-prefixed contigs. The contig prefix records which strand hypothesis was used and feeds the YD:Z: tag directly.

Where conversion happens

Read conversion runs inside src/fastmap.cpp in the meth_mode ingest block, immediately after sequence parsing and before any alignment work. The transformation is applied to the in-memory bseq1_t.seq buffer; the original FASTQ file is never rewritten.

Before the bases are modified, the original sequence is recorded in the read’s comment buffer as:

YS:Z:<l_seq bases>\tYC:Z:<direction>

where <direction> is CT for R1 (C→T) and GA for R2 (G→A). These fields pass through the alignment kernel untouched and are emitted as SAM aux tags in the output BAM. See SAM tags: YS, YC, YD for the per-tag reference.

Sequence restoration in the BAM SEQ field

Methylation callers such as MethylDackel identify methylated cytosines by examining the BAM SEQ field at each CpG site. They need to see real C/T bases — not the uniformly-converted T/A bases that were used for alignment.

meth_mem_aln_to_bam (in src/meth_bam.cpp) restores the original sequence from YS:Z: before writing the BAM record:

The YS:Z: payload is located at the start of the bseq1_t.comment field (offset +5 past the YS:Z: header bytes).
For forward-aligned records (!p.is_rev), the pre-c2t bases are copied directly into the BAM SEQ buffer.
For reverse-aligned records (p.is_rev), the bases are reverse-complemented using the standard TGCAN table before being placed in SEQ.
If YS:Z: is absent (e.g. when running with an external c2t converter that does not emit it), the code falls back to the converted sequence in s->seq, with the same RC flip logic.

Warning — Soft-clip and supplementary trimming

When computing the SEQ range for supplementary alignments, the qb/qe boundaries account for soft-clip or hard-clip operations at the CIGAR ends. The YS:Z: restoration applies over the same trimmed range so SEQ length always matches the emitted CIGAR.

QUAL field handling

The QUAL field is taken directly from the original FASTQ (bseq1_t.qual) over the same [qb, qe) range and is never modified by the c2t process. Quality scores correspond to the original base calls, not the converted ones.

Relationship to the reference index

bwa-mem3 index --meth ref.fa writes ref.fa.bwameth.c2t, which applies the same C→T / G→A projection to the reference sequence. The resulting file is compatible with what bwameth.py index produces, so the same doubled-reference FASTA can be used interchangeably with either tool across tested versions.