Chemistries and platforms

Support matrix

Supported means a bundled, validated model selected by --enzyme, with defaults tuned for it and coverage in the test suite.

Chemistry	Platform	Status	Preset	Notes
Hia5 (m6A) Fiber-seq	PacBio	supported	--enzyme hia5 --seq pacbio	m6A on both strands (A+a, T-a)
Hia5 (m6A) Fiber-seq	Nanopore	supported	--enzyme hia5 --seq nanopore	m6A on the sequenced strand (A+a); strict ML ≥ 248
DddB DAF-seq	any	supported	--enzyme dddb	double-stranded deaminase; one strand per read
DddA DAF-seq	any	supported	--enzyme ddda	separate nucleosome and TF models, radial nucleosome recall, CpG-aware calling
EcoGII (m6A)	PacBio, Nanopore	development	custom -m only	a model file is shipped (ecogii_pacbio.json); not accepted by --enzyme, not validated; rejected by fiberhmm-consensus
CpG methyltransferase (SssI)	Nanopore	development	custom -m only	cpg_nanopore.json, observation mode cpg
GpC methyltransferase	—	development	custom -m only	observation mode gpc; no bundled model
Other enzymes	—	unsupported	train your own	see Training a model

The public preset list is fiberhmm/models/SUPPORTED_MODES.json. Model files for development chemistries are shipped for method development and carry no supported-workflow claim.

Some commands support a subset:

Command	Chemistries
fiberhmm-consensus, fiberhmm-transfer	ddda, dddb, hia5-pacbio, hia5-nanopore
fiberhmm-strand-rescue	ddda, dddb, hia5-nanopore (PacBio Hia5 already observes both strands)
fiberhmm-pair, fiberhmm-merge	DddA
fiberhmm-tag-m5c, fiberhmm-call-m5c	genome-wide DddA only
fiberhmm-dedup, fiberhmm-daf-snps, fiberhmm-daf-encode	DAF (DddA, DddB)

Default settings per chemistry

The resolved chemistry (--enzyme, the given or detected --seq, or the input's FIBERHMM-CHEMISTRY declaration for tools without --enzyme) sets these defaults. An explicit option always wins.

Setting	Hia5 PacBio	Hia5 Nanopore	DddB	DddA
Observation mode	pacbio-fiber	nanopore-fiber	daf	daf
HMM model	hia5_pacbio.json	hia5_nanopore.json	dddb_nanopore.json	ddda_nuc.json
TF-recall model	same	same	same	ddda_TF.json
ML threshold: call, apply	128	248	128	128
ML threshold: recall-tfs/-nucs, extract, qc	125	248	125	125
ML threshold: dedup, pair, merge (MM/ML dU)	—	—	128	128
TF interval cost --min-llr	5.0	5.0	5.0	5.0
Nucleosome recall	accessible-cut split	accessible-cut split	accessible-cut split	phase-aware radial
CpG-aware calling (--use-m5c)	off	off	off	on (call, pipeline, apply, posteriors, recall, pair/merge, qc)
Adjacent-target thinning (--daf-mask-runs)	—	—	off	2, keep-one
Strand-swap chimera filter	—	—	on	on
Duplicate marking and SNP screen (file input)	—	—	on	on
Declared platform when --seq is omitted	detected	detected	nanopore	pacbio

Notes:

• ML thresholds apply to calls read from MM/ML. Deaminations encoded as R/Y in the sequence or read from MD mismatches are binary and ignore the threshold. Hia5 Nanopore uses 248 because Dorado's m6A ML values are reliable only near the top of the scale; the bundled Hia5 Nanopore QC reference and the hia5-nanopore strand-rescue preset are calibrated at 248.
• Tools that re-read MM/ML from an existing BAM (recall-tfs, extract, qc) use 125 for non-Nanopore chemistries while call/apply use 128. This is a known inconsistency kept for compatibility; pass --prob-threshold to make them agree.
• fiberhmm-posteriors resolves its threshold like fiberhmm-call (248 for Hia5 Nanopore, 128 otherwise).
• Other tools keep their own defaults: fiberhmm-probs 128, fiberhmm-train 125, fiberhmm-utils transfer 128, fiberhmm-consensus 125 for Hia5 PacBio and 248 for Hia5 Nanopore.
• fiberhmm-call and fiberhmm-apply call primary and supplementary alignments (supplementary records on their aligned bases only); --primary calls primary records only, --no-primary every record.
• DAF: insertion and soft-clip bases (no reference counterpart) and SNP-masked sites are no evidence; unaligned stretches >= 50 bp are left uncalled (--no-daf-mask-unaligned to turn off).

Hia5 on PacBio and on Nanopore

A PacBio HiFi read reports m6A on both strands of the molecule: A's on the read strand (A+a) and A's on the opposite strand, which appear as T's in the read (T-a). The pacbio-fiber mode uses both, merging reverse- complement contexts. A Nanopore read reports m6A only on the strand that went through the pore (A+a), so nanopore-fiber sees only half the targets.

Consequences for Nanopore Hia5:

• sparser evidence per read: nucleosome recall splits fewer long protected blocks, and trimmed edge pieces shorter than --nuc-min-size are reported as 60–89-bp footprint-scale calls more often than on PacBio;
• the stricter ML threshold (248);
• strand rescue can recover footprints missed on one strand by using the other orientation (Strand rescue);
• in consensus, both alignment orientations of Hia5 are pooled into one channel.

Updated Nanopore Hia5 and DddB tables

3.0 ships re-calibrated Nanopore Hia5 and DddB emission tables. Re-run Nanopore Hia5 and DddB calls made with 2.x; fiberhmm-check lists the BAMs that need it. See Upgrading from 2.x.

DAF-seq: DddA and DddB

DAF-seq deaminates accessible cytosines. After amplification, a deamination on the strand that matches the reference reads as C→T ("CT" reads) and one on the opposite strand as G→A ("GA" reads). Each read carries one flavour and is informative only at reference C (CT) or reference G (GA). Its flavour, not its alignment orientation, decides which bases are targets.

• DddB: one model (dddb_nanopore.json) is used for the HMM and TF recall.
• DddA also deaminates inside nucleosomes, in a helically phased pattern. FiberHMM therefore uses separate models for the nucleosome HMM (ddda_nuc.json), TF recall (ddda_TF.json) and radial nucleosome refinement, and enables CpG-aware calling (HMM and recall) and keep-one masking by default. DddA is also the chemistry of scDAF duplex pairing and of the CpG-island methylation caller.

Where the deaminations come from, duplicate marking, SNP screening and the chimera filter are described in DAF-seq.