Plasmids

A plasmid is a circle, but its map and FASTA are linear: they start at an arbitrary position (often the first base of ori). fiberhmm-pipeline takes the map itself as the reference and handles the two consequences, naming and the origin, so a plasmid run needs no hand-made FASTA, no rotated reference and no renaming:

fiberhmm-pipeline reads.fastq.gz --reference construct.dna --enzyme dddb -o run/

The reference from a map

Supported maps: SnapGene .dna, GenBank .gb/.gbk/.genbank, EMBL .embl/.emb. The pipeline writes run/<contig>.fa (+ .fai) and copies the map into run/. The contig is named the way FiberBrowser names the map when you load it, so the BAM and the map line up without renaming:

Map	Contig name	Example
SnapGene .dna	file name without extension, cut at the first space, characters outside A–Z a–z 0–9 _ . - replaced by _, leading/trailing _ removed	L-HH (v2).dna → L-HH
GenBank	the LOCUS name (same character rule, not stripped)	LOCUS pUC19 ... → pUC19
EMBL	the ID name	ID pX1; ... → pX1
FASTA (as a map)	the first header word (SnapGene character rule)	>pRef plasmid → pRef

The rule is fiberhmm.pipeline.reference.fiberbrowser_contig_name.

The map's topology decides whether the contig is circular: SnapGene's topology flag, circular/linear on the GenBank LOCUS or EMBL ID line. --topology circular makes every contig circular (for a plasmid given as a FASTA); --topology linear turns the origin handling off.

Reads through the origin

Whole-plasmid libraries are cut at random positions, so many reads run through the position where the map starts. minimap2 returns such a read in two pieces, one ending at the last base of the map and one starting at its first base. A linear pipeline keeps only the longer piece and trims the rest; in a Plasmidsaurus library that loses part of about a third of the reads, and coverage dips next to the origin. Rotating the reference (the "orirot" FASTAs some projects make) only moves the cut.

The pipeline joins the two pieces into one record instead, in the form the SAM specification prefers for circular references (SAMv1 §1.4, "Circular reference sequences"): the contig is declared @SQ ... TP:circular, POS lies within the contig, and the alignment may run past its end; a position p past the end means p - length. The two pieces are joined when

• they are on the same strand, one ends within 100 bp of the contig end and the other starts within 100 bp of its start, and
• they are consecutive on the read (a gap or overlap of at most 100 bp).

Bases the two pieces both claim keep the primary's placement; a gap becomes aligned, inserted or deleted bases, and MD/NM are recomputed against the reference, so the join is scored like any other part of the alignment. A molecule covers at most one full circle: anything beyond is concatemer sequence and is trimmed with the other unaligned arms. The joined record keeps the primary's flags and MAPQ; the supplementary record is dropped. Other supplementary records on a circular contig (concatemer copies of the same plasmid sequence) are dropped too, and DAF reads there are hard-clipped, so one molecule is not annotated twice; on linear contigs the pipeline keeps supplementary records and soft clips. --keep-soft-clips keeps soft clips on circular contigs too, and --hard-clip hard-clips on every contig.

fiberhmm-call then calls the whole molecule. FiberBrowser draws it continuously across the origin in a circular view (a view that runs past the end of the plasmid), and fiberhmm-extract splits each track row at the origin, so bigBed files stay valid.

--no-origin-merge keeps only the primary piece (the linear behaviour).

What it changes, on real data

Two DddB plasmid runs, against their earlier processing (minimap2, primary MAPQ ≥ 20, hard clip, fiberhmm-call, on a reference rotated to cut in the middle of ori):

	D plasmid, 11.4 kb	N1 plasmid, 12.2 kb
reads kept (primary, MAPQ ≥ 20)	1,072 (same)	1,271 (same)
reads joined across the origin	346 (32%)	1,159 (91%)
median aligned span	3,403 bp (was 3,251 bp)	12,179 bp, one full circle (was 11,765 bp)
coverage of the first kb (after the cut)	353 (was 212)	1,183 (was 693)
min/max coverage of 500 bp bins	0.83 (was 0.63)	—

The N1 library is a PCR product that runs around the whole construct, so the earlier processing cut 414 bp off nearly every molecule; joined, each read is one complete circle.

In the BAM header

@SQ SN:N1   LN:12179    M5:1276ace935ddb3051497a9b2b956a0c2 TP:circular
@CO FIBERHMM-REFERENCE:v1:contig=N1;length=12179;md5=1276ace935ddb3051497a9b2b956a0c2;topology=circular;source=N1.dna;source_format=snapgene;source_sha256=02bb9705b987b8df586ff0723c7d24de49f5683a85654c669a5bfc8e80243d88

M5 is the MD5 of the contig sequence (the SAM standard) and the @CO line names the map it came from and that file's SHA-256, so a viewer can pair a BAM with its map even after files are renamed (Header declarations).

Several constructs or a plasmid plus a genome

Keep one reference per run. When a construct shares sequence with the genome (a genomic insert, mini-white, an hsp70 promoter), a combined genome-plus-plasmid reference pushes the shared reads to MAPQ 0 and they are dropped. Run the plasmid and the genome separately; reads that do not map to the given reference are counted as unmapped in the log.