Command-line reference

Every FiberHMM command and every option it accepts, with its default. The tables below are generated from the commands' own argument parsers, so they match --help exactly. Behaviour that does not fit in a flag description is explained on the workflow pages; defaults that depend on the chemistry are collected in Chemistries and platforms. Every command also accepts -h/--help and --version (prints fiberhmm <version>).

Command	What it does	Guide
fiberhmm-pipeline	Reads + reference (FASTA or plasmid map) to a called BAM: align, call, QC	Quick DAF-seq, Plasmids
fiberhmm-call	Call nucleosomes, MSPs and TF footprints (HMM + recall) in one pass	Calling
fiberhmm-apply	HMM nucleosomes/MSPs only	Calling
fiberhmm-recall-tfs, fiberhmm-recall-nucs	Re-call TFs (and nucleosomes) on a called BAM	Re-calling
fiberhmm-qc	Bounded QC report for one or more BAMs	QC
fiberhmm-extract	Calls to BED12 / bigBed	Extracting tracks
fiberhmm-dedup	DAF PCR-duplicate marking/collapse	DAF-seq
fiberhmm-daf-encode	Stamp R/Y deamination codes into a DAF BAM	DAF-seq
fiberhmm-daf-snps	Recurrent C→T/G→A SNP mask for DAF-seq	DAF-seq
fiberhmm-pair, fiberhmm-merge	Pair, merge and jointly re-call scDAF duplexes	Duplex
fiberhmm-tag-m5c, fiberhmm-call-m5c	DddA CpG-island methylation	DAF-seq
fiberhmm-consensus	Footprint classes across molecules	Consensus
fiberhmm-transfer	Apply frozen classes to new data	Transfer
fiberhmm-footprint-model	Footprint population model	Footprint model
fiberhmm-strand-rescue (+ -annotate, -audit)	Two-strand TF rescue and edge normalization	Strand rescue
fiberhmm-tag-consensus	Consensus-state slots on tf_sr calls	Tag consensus
fiberhmm-posteriors	Per-position HMM posteriors	Posteriors
fiberhmm-probs, fiberhmm-train	Build a custom model	Training
fiberhmm-utils	Model, header and bigBed utilities	Training, Extracting
fiberhmm-check	Whether existing outputs need re-running with this version	Checking outputs
fiberhmm-nfr	NFR variants and co-accessibility (experimental preview)	NFR detection

Generated from each command's argparse definition by python tools/gen_cli_reference.py; do not edit by hand. Hidden compatibility options are omitted. auto means the value is resolved at run time as the description says.

fiberhmm-pipeline

Flag	Default	Description
reads	required	Read files: FASTQ (.fastq/.fq, optionally .gz), unaligned BAM, a BAM aligned to --reference, raw Nanopore POD5 files (basecalled with dorado first), or a directory of them (a directory with POD5 files, searched recursively, is raw input). All reads given form one sample.
--reference	required	Reference FASTA, or a plasmid map (.dna, .gb/.gbk/.genbank, .embl) converted to a FASTA whose contig is named as FiberBrowser names the map.
--enzyme	required	Chemistry: ddda / dddb (DAF-seq) or hia5 (Fiber-seq). Choices: ddda, dddb, hia5.
-o / --outdir	required	Output directory.
--sample	—	Sample name for output files and the read group (default: the first input's name without extensions). One plain file name: no path separators, spaces or leading '.'/'-'.
-c / --cores	4	minimap2 threads and fiberhmm-call worker processes (default 4).
--seq	—	Sequencing platform. Default: detected from the reads. ddda/dddb: a BAM's FIBERHMM-CHEMISTRY declaration or @RG PL/@PG records; reads with no record are Nanopore when aligned here, and keep fiberhmm-call's default when called as given. hia5: also the MM tags of the first reads (T-a = PacBio, A+a only = Nanopore), an error when nothing settles it. Sets the minimap2 preset (map-ont / map-hifi), the read group's PL and fiberhmm-call's --seq. Choices: nanopore, pacbio.
--force-chemistry	off	Run although the reads contradict --enzyme/--seq (a FIBERHMM-CHEMISTRY declaration naming another enzyme or platform; m6A-tagged reads without deaminations for ddda/dddb; deaminated or untagged reads for hia5). Default: refuse.
--topology	auto	Reference topology. auto (default): a plasmid map's own topology, FASTA contigs linear. circular: every contig is circular (a plasmid FASTA). Choices: auto, circular, linear.
--region	—	Keep only reads overlapping this region (1-based, inclusive; repeatable). The first region is the one outputs.json asks FiberBrowser to open.
--min-mapq	20	Keep primary alignments with at least this MAPQ (default 20); also passed to fiberhmm-call and fiberhmm-qc.
--keep-soft-clips	off	Keep unaligned read arms as soft clips everywhere. Default: kept on linear contigs (DAF calling treats them as no evidence); for ddda/dddb hard-clipped on circular contigs (concatemer arms beyond one full circle).
--hard-clip	off	Hard-clip unaligned read arms on every contig (the earlier ddda/dddb default).
--no-origin-merge	off	Do not join the two pieces of reads that run through the origin of a circular reference (keep the primary piece).
--aligner	auto	The minimap2 program on PATH or the mappy module (default: auto, program first). Choices: auto, minimap2, mappy.
--min-read-length	—	Minimum aligned read length to call (default 1000).
--dedup	auto	DAF PCR-duplicate detection (default auto: on for file input). Choices: auto, on, off.
--dedup-mode	flag	flag (default): mark duplicates 0x400 and keep them; collapse: keep one read per duplicate cluster. Choices: flag, collapse.
--snp-screen	auto	DAF recurrent-SNP screen and mask (default auto: after a depth preflight). Choices: auto, on, off.
--snp-mask	—	DAF: your own BED of SNP sites to exclude.
--daf-mask-unaligned / --no-daf-mask-unaligned	on	DAF: treat insertion and soft-clip bases as no evidence and leave unaligned stretches >= 50 bp uncalled (default on).
--chimera-filter / --no-chimera-filter	on	DAF: skip strand-swap chimeric reads (default on).
--alignments	primary-supplementary	Alignment records to keep and call (default primary-supplementary: a read's primary and supplementary records, i.e. both sides of a split alignment, on linear contigs; secondary records are dropped). primary: the primary record only; all: as primary-supplementary here (the aligner step never keeps secondary records) and passed to fiberhmm-call. Choices: primary, primary-supplementary, all.
--primary	—	Same as --alignments primary.
--no-primary	—	Same as --alignments all.
--prob-threshold	—	ML threshold override, 0-255 (default: chemistry preset).
--use-m5c / --no-use-m5c	auto	DddA CpG-aware calling: CpGs masked in the first-pass HMM and recall outside ddda_ucg islands (default: on for ddda).
--cpg-mask-policy	—	DddA CpG mask policy for the HMM and recall (default unmethylated-only). Choices: unmethylated-only, methylated-only.
--call-args	—	Other fiberhmm-call options, quoted as one string (e.g. --call-args "--with-scores --min-llr 6").
--call-mode	auto	auto (default): fiberhmm-call's resumable region-parallel mode for genome-scale data (>=20,000 reads over at least --cores regions), streaming for targeted runs (one amplicon or plasmid), which it calls faster. Choices: auto, streaming, resumable.
--no-qc	off	Skip fiberhmm-qc.
--basecaller-info	—	Record the basecaller of reads whose files do not say (FASTQ), or correct it: key=value pairs, e.g. "program=dorado version=0.9.6 basecall_model=dna_r10.4.1_e8.2_400bps_sup@v5.0.0" (keys: program, version, basecall_model, modbase_models). Takes precedence over the input headers.
--modbase-model	—	The modification model that called the MM/ML tags (e.g. dna_r10.4.1_e8.2_400bps_sup@v5.0.0_6mA@v2; repeatable or comma-separated; 'none' for none). Takes precedence over the input headers.
--dorado	—	The dorado program (or its install folder). Default: $FIBERHMM_DORADO, PATH, then the usual install locations.
--dorado-model	—	dorado basecalling model: fast/hac/sup[@vX.Y.Z], a model complex such as sup,6mA, or a model folder (default sup).
--dorado-modified-bases	—	dorado --modified-bases codes (default: 6mA for hia5, none for ddda/dddb, whose deaminations are read from the sequence); 'none' to basecall without a modification model.
--dorado-modbase-models	—	dorado --modified-bases-models: comma-separated model names or paths (instead of --dorado-modified-bases).
--dorado-device	auto	dorado --device: auto (default), metal, cuda:all, cuda:0, cpu.
--dorado-batchsize	—	dorado --batchsize (default: dorado chooses).
--dorado-models-dir	—	Where dorado keeps downloaded models (default $FIBERHMM_DORADO_MODELS_DIR or ~/.fiberhmm/dorado_models).
--dorado-args	—	Other dorado basecaller options, quoted as one string (e.g. --dorado-args "--min-qscore 8").
--tracks	off	Also extract nucleosome/MSP/TF/deamination (or m6A) tracks into OUTDIR/tracks (bigBed; BED without bedToBigBed).
--redo	—	Redo this step and the later ones although complete (also needed to change the inputs or settings of an existing OUTDIR; discards an interrupted call's resumable state). 'all' is every step after basecalling; only --redo basecall basecalls again. Choices: all, basecall, align, call, qc, tracks.
--progress-json	—	Append JSON-lines progress events to FILE ('-' for stdout).
-v / --verbose	off	Echo the output of fiberhmm-call, -qc and -extract.
-q / --quiet	off	No progress messages on stderr.

fiberhmm-call

Flag	Default	Description
-i / --input	required	Input BAM. Use "-" for stdin (streaming mode).
-o / --output	required	Output BAM path or "-" for stdout (unsorted).
-m / --model	—	Apply HMM model JSON. If omitted, bundled model for --enzyme/--seq is used.
--recall-model	—	Separate model for TF LLR tables. Default: reuse apply model.
--enzyme	—	Bundled enzyme preset. Choices: ddda, dddb, hia5.
--seq	—	Sequencing platform. For Hia5 it selects the model; when omitted it is detected from the input (MM specs: PacBio T-a vs Nanopore A+a only; header records) and the run stops if the evidence conflicts. A given --seq that the reads contradict is refused (see --force-seq). For dddb/ddda it only sets the declared platform. Choices: pacbio, nanopore.
--force-seq	off	Use the given --seq even when the input's MM specs or header say the reads come from the other platform (normally refused: the wrong platform model changes the calls, e.g. ~100x more TF calls for Nanopore reads called as PacBio), and run Hia5 even when the first reads carry no m6A calls.
--basecaller-info	—	Basecaller provenance to record when the input does not say (or says wrongly): key=value pairs, e.g. "program=dorado version=0.9.6 basecall_model=dna_r10.4.1_e8.2_400bps_sup@v5.0.0" (keys: program, version, basecall_model, modbase_models). Default: read from the input header (@RG DS > @PG CL > per-read RG).
--modbase-model	—	Modification model that called the input's MM/ML tags (repeatable or comma-separated; 'none' for none). Overrides the header.
--replace-chemistry	off	Replace, instead of reconcile with, the input BAM's FIBERHMM-CHEMISTRY declaration (re-calling a BAM with a deliberately different chemistry).
--reference	—	Reference FASTA for DAF-seq BAMs that lack both R/Y IUPAC encoding and MD tags. When present, acts as a fallback source for deamination-site detection (R/Y codes and a usable MD tag take precedence). Must match the BAM's assembly and be faidx-indexed.
-k / --context-size	—	Context size override. Default: from model.
--edge-trim	10	Bases to mask at edges (default 10)
--min-mapq	0	Min mapping quality (default 0)
--prob-threshold	—	Min MM/ML modification probability 0-255. Default: chemistry preset -- 248 for Hia5 Nanopore (--seq nanopore, given or detected), 128 otherwise. R/Y- and MD-encoded DAF input is binary and ignores it.
--min-read-length	1000	Min aligned read length (default 1000 — matches fiberhmm-apply)
--msp-min-size	0	Min MSP size (default 0)
--nuc-min-size	85	Min footprint size to count as nucleosome (default 85)
--with-scores / --scores	off	Write the HMM posterior-mean nq score of baseline nucleosomes (with --no-recall-nucs; nucleosome recall writes its own LLR-based nq). No aq is written. --scores is the fiberhmm-apply spelling.
-r / --circular	off	Enable circular molecule mode (3x tile internally, emit wrapped MA/AQ/AN annotations).
--process-unmapped / --no-process-unmapped	auto	Call unmapped reads that carry SEQ + MM/ML. Default: automatic -- on for stdin, unindexed and unaligned (uBAM) input, off (pass-through) for indexed aligned BAMs. A run that skips >90% of records as unmapped fails unless --no-process-unmapped is given.
--alignments	primary-supplementary	Alignment records to call (default primary-supplementary): primary and supplementary records, the parts of a split read (e.g. both sides of a structural variant); secondary records, alternative placements of the same bases, pass through uncalled. On a supplementary record only its aligned bases are called. "primary" calls primary records only; "all" also calls secondary records. Hard-clipped records whose MM/ML cannot match SEQ are always skipped (hard_clipped_mm). Choices: primary, primary-supplementary, all.
--primary	—	Same as --alignments primary.
--no-primary	—	Same as --alignments all.
--min-llr	—	Native LLR cost per TF interval in joint decoding (default: enzyme preset; not a calibrated FDR threshold).
--min-opps	3	Min informative target positions per TF call (default 3).
--unify-threshold	90	v2 nucs with nl < this may be demoted to tf+ (default 90).
--emission-uplift	—	Emission power transform. Default: enzyme preset.
--use-m5c / --no-use-m5c	auto	DddA CpG-aware calling: CpG observations are excluded from the first-pass HMM and from nucleosome and TF recall, except inside confident unmethylated island calls (MA ddda_ucg from fiberhmm-tag-m5c) the input already carries. The same mask applies in fiberhmm-apply, -posteriors and -pair/-merge. Default: on for --enzyme ddda, off otherwise; --no-use-m5c for an ablation.
--cpg-mask-policy	—	With CpG-aware calling (HMM and recall): keep CpGs only inside ddda_ucg islands (default unmethylated-only), or mask only ddda_mcg spans (methylated-only, the former behaviour). DddA only; ignored with a warning otherwise. Choices: unmethylated-only, methylated-only.
--no-legacy-tags	off	Skip ns/nl/as/al, emit only MA/AQ.
--downstream-compat	off	Skip MA/AQ; write TF calls into legacy ns/nl track.
--recall-nucs / --no-recall-nucs	auto	Split over-merged nucleosomes + resolve platform-aware edges (emits nuc.QQQ) and promote nucleosome-sized TF leaks to nuc. ON by default for all enzymes (DddA uses phase-aware radial inference, others the accessible-cut Kadane split). Use --no-recall-nucs for baseline HMM nucleosomes (nuc.Q).
--split-min-llr	4.0	Min accessible-run LLR to split a nucleosome; for DddA, the molecule-local linker-residue configuration LLR (default 4.0).
--split-min-opps	3	Min informative positions in a nucleosome-splitting cut or DddA linker residue (default 3).
--ddda-derived-tf-max-edge-gap	12	DddA phase-aware radial recall only: TF calls exposed solely by nucleosome refinement must have a deamination hit within BP on both sides (default 12). Original HMM-accessible TF scan space is unchanged. Use -1 to disable the safeguard.
--keep-chimeras	off	DAF only: keep strand-swap chimeric reads (C->T in one segment + G->A in another). Default: filter them out and report the count.
--chimera-min-seg	5	DAF chimera: min same-strand deamination events per segment to call a swap (default 5).
--chimera-purity	0.8	DAF chimera: min same-strand purity per segment (default 0.8).
--daf-mask-runs	—	DAF only: thin targets lying in same-strand runs of >= N original C (CT) or G (GA) bases (CC/GG and longer at N=2; see --daf-run-policy). Adjacent conversions are coupled and do not follow the per-site emission model. Default: 2 with keep-one for --enzyme ddda (duplex-validated), off otherwise; 0 disables.
--daf-run-policy	keep-one	With --daf-mask-runs: keep the 5'-most target of each run (keep-one, default) or remove the whole run (drop). Choices: keep-one, drop.
--daf-mask-unaligned / --no-daf-mask-unaligned	on	DAF only: treat query bases with no reference counterpart (CIGAR insertions and soft clips) as no evidence, like SNP-masked sites, and leave unaligned stretches of >= 50 bp uncalled (default on). Deaminations are read-versus-reference mismatches, so these bases can never carry one; --no-daf-mask-unaligned restores the pre-3.0 encoding, which counts them as protected.
--daf-insert-consensus	auto	DAF only: give insertions carried by many reads real evidence. A pre-pass clusters CIGAR insertions >= 50 bp by breakpoint, builds a deamination-aware consensus of each cluster with enough carriers (a column that is sometimes C and sometimes T on C->T reads is a C; G/A likewise) and re-encodes each carrier's inserted bases against it; other insertions stay masked. auto (default): on for file input with --daf-mask-unaligned; off: mask only. Choices: auto, off.
--daf-insert-min-carriers	20	Carriers an insertion needs for --daf-insert-consensus (default 20).
--daf-snp-mask	—	DAF only: 0-based BED of recurrent C>T/G>A SNP sites to exclude from deamination observations. MD is preserved.
--daf-call-snps	auto	DAF only: force two-pass recurrent opposite-conversion SNP masking. By default file-based DddA/DddB runs screen automatically after a bounded depth preflight.
--no-daf-call-snps	auto	Disable automatic recurrent SNP screening.
--daf-snp-output-prefix	—	Output prefix for --daf-call-snps (default: qc/<output BAM stem>.daf_snps).
--daf-snp-min-fraction	0.2	Minimum opposite-direction fiber fraction for --daf-call-snps (validated default 0.2).
--daf-snp-min-depth	5	Minimum fiber depth in each conversion-direction class for --daf-call-snps (validated default 5; all thresholds must pass in both classes).
--daf-snp-min-alt-fibers	5	Minimum alternate fibers in each direction for --daf-call-snps (validated default 5).
--daf-snp-min-dominant-events	5	Minimum dominant conversions to classify a fiber for --daf-call-snps (default 5).
--daf-snp-min-dominant-purity	0.8	Minimum conversion-direction purity for --daf-call-snps (default 0.80).
--daf-snp-min-amplicon-reads	20	Minimum aligned reads required to discover and plot an amplicon consensus in SNP QC (default 20).
--ddda-mcg	off	Deprecated integrated per-CpG mode; retained only to emit a clear migration error. Run fiberhmm-call, then fiberhmm-tag-m5c (whole CpG islands), then fiberhmm-call again on the tagged BAM.
--dedup	auto	DAF (ddda/dddb) only: force PCR duplicate detection (already automatic for file-based DddA/DddB calls). Detect by deamination-pattern fingerprint (see fiberhmm-dedup) and similar alignment ends BEFORE footprinting. The integrated default is nondestructive: retain every read and set 0x400 plus di/ds cluster tags. Amplicon/UMI-less DAF libraries can be heavily PCR-duplicated and coordinate dedup does not apply. Requires a file input (not stdin). Ignored for fiber-seq (hia5).
--no-dedup	auto	Disable automatic DddA/DddB duplicate marking.
--dedup-min-jaccard	0.95	Deamination-set Jaccard threshold for --dedup (default 0.95). Use 0.90 for optional more aggressive matching; lower thresholds can merge distinct molecules.
--dedup-flag-only	off	Deprecated compatibility spelling for the nondestructive integrated default (0x400 + di/ds; retain every read).
--dedup-collapse	off	With --dedup: destructively collapse each duplicate cluster to one representative. Default: mark and retain all reads.
--dedup-min-deam	10	With --dedup: reads with fewer deamination calls are not fingerprinted and pass through (default 10).
--dedup-prob-threshold	—	With --dedup: min ML probability for MM/ML-native dU calls, 0-255 (default: the calling --prob-threshold, 128). R/Y and MD inputs are binary and ignore it.
--dedup-ignore-strand	off	With --dedup: cluster reads across deamination flavours (C->T with G->A reads). Default: only reads of the same flavour can be duplicates.
--dedup-max-end-diff	50	With --dedup: maximum difference at both aligned reference ends for duplicate matching (default 50 bp).
--dedup-stats-tsv	—	With --dedup: write a cluster_id<TAB>n_reads table.
-c / --cores	4	Worker processes (0 = all CPUs; default 4).
--chunk-size	500	Reads per worker chunk (default 500; streaming mode only).
--io-threads	8	htslib I/O threads per stage (default 8).
--max-reads	0	0 = no limit (default; streaming mode only)
--qc / --no-qc	on	Run bounded signal/periodicity/footprint QC after a file output (default: on; use --no-qc to disable).
--qc-sample-reads	2000	Target random-window QC sample size (default 2000).
--qc-seed	20260824	Deterministic QC sampler seed (default 20260824).
--qc-min-mapq	20	Minimum mapping quality for the QC sample (default 20).
--qc-output-prefix	—	QC output prefix (default: qc/<BAM stem> beside output BAM).
--region-parallel	off	Process genomic regions in parallel (one worker per region). Scales linearly with --cores up to chromosome count. Requires coordinate-sorted + indexed input BAM. Recommended for full-genome runs; use streaming for stdin/unaligned.
--region-size	10000000	Region size in bp for --region-parallel (default 10 Mb).
--skip-scaffolds	off	Skip scaffold/contig chromosomes in region-parallel mode.
--chroms	—	Only process these chromosomes (region-parallel mode).
--resume	off	Continue an interrupted --region-parallel run: regions finished in the work directory are reused, missing or partial regions are rerun, then the output is merged and published as usual. Refused if the input BAM or any effective parameter changed. Implies --region-parallel for an indexed, aligned input; with no work directory present a new run starts. Streaming (stdin, unindexed) runs cannot resume.
--work-dir	—	Region-parallel work directory holding finished region BAMs and the resume manifest (default: .<output name>.fiberhmm-work beside the output). Kept when a run fails or is interrupted; removed after a successful publish.
--keep-work-dir	off	Keep the region-parallel work directory after a successful run.
--progress-json	—	Write machine-readable progress, one JSON object per line, to stderr (no value) or append to FILE: start, region (regions done/total, reads, reads/s, ETA), merge, done and stopped events (schema fiberhmm.progress.v1).

fiberhmm-apply

Flag	Default	Description
-i / --input	required	Input BAM file with modification calls, or "-" for stdin (unaligned and unindexed BAMs are streamed)
-m / --model	—	Path to trained HMM model (.json, .npz, or .pickle). If omitted, the bundled model for --enzyme/--seq is used.
-o / --outdir	required	Output directory, or "-" to write BAM to stdout (for piping)
--enzyme	—	Auto-select a supported bundled chemistry model. Use --seq pacbio|nanopore for Hia5. Choices: ddda, dddb, hia5.
--seq	—	Hia5 sequencing platform. When omitted it is detected from the input (MM specs: PacBio T-a vs Nanopore A+a only; header records); conflicting evidence stops the run, and a given --seq that the reads contradict is refused (see --force-seq). Ignored for dddb/ddda. Choices: pacbio, nanopore.
--force-seq	off	Use the given --seq even when the input's MM specs or header say the reads come from the other platform (normally refused: the wrong platform model changes the calls, e.g. ~100x more TF calls for Nanopore reads called as PacBio), and run Hia5 even when the first reads carry no m6A calls.
-k / --context-size	—	Context size (auto-detected from model if not specified)
--cores / -c	1	Number of CPU cores (0=auto, default: 1)
--io-threads	4	Number of htslib decompression/compression threads for BAM I/O (default: 4)
--streaming	off	Use streaming pipeline mode (works with unaligned/unindexed BAMs and stdin). Recommended for unaligned data or when reading from pipes.
--chunk-size	500	Reads per compute chunk in streaming mode (default: 500)
--min-mapq / -q	0	Minimum mapping quality; reads below this are written to output unchanged without footprint/nucleosome tags. Default 0 (call on all mapped reads). Pass a positive value to filter.
--prob-threshold	—	Minimum MM/ML probability (0-255) to call a modification. Default: chemistry preset -- 248 for Hia5 Nanopore (--seq nanopore, given or detected), 128 otherwise. R/Y- and MD-encoded DAF input is binary and ignores it.
--min-read-length	1000	Minimum aligned read length in bp; shorter reads are written to output unchanged without footprint/nucleosome tags. Set to 0 to attempt calling on all reads regardless of length (default: 1000)
-t / --train-reads	—	TSV file of read IDs used in training (to exclude)
--alignments	primary-supplementary	Alignment records to call (default primary-supplementary): primary and supplementary records, the parts of a split read (e.g. both sides of a structural variant); secondary records, alternative placements of the same bases, pass through uncalled. On a supplementary record only its aligned bases are called. "primary" calls primary records only; "all" also calls secondary records. Hard-clipped records whose MM/ML cannot match SEQ are always skipped (hard_clipped_mm). Choices: primary, primary-supplementary, all.
--primary	—	Same as --alignments primary.
--no-primary	—	Same as --alignments all.
--process-unmapped / --no-process-unmapped	auto	Process unmapped reads that have sequences and modification tags. Default: automatic -- on for stdin, unindexed and unaligned (uBAM) input. A run that skips >90% of records as unmapped fails unless --no-process-unmapped is given.
--edge-trim / -e	10	Bases to trim from read edges (default: 10)
-r / --circular	off	Enable circular mode (tiles reads 3x)
--scores	off	Compute per-footprint confidence scores (slower but more informative)
--msp-min-size	0	Minimum size for MSP regions in bp. Default 0 (emit every accessible run; matches fibertools, which does not impose an MSP size filter at this stage). Pass a positive value to filter.
--nuc-min-size	85	Minimum footprint size (bp) to count as nucleosome-sized for MSP boundary detection. Only footprints >= this size split MSPs; smaller footprints are absorbed (default: 85)
--no-msps	off	Do not write MSP tags (as/al/aq) to output BAM. Useful for Fiber-seq where MSPs are computed differently by fibertools
--stats	off	Generate summary statistics and plots
--stats-sample	10000	Number of reads to sample for statistics (default: 10000)
--stats-seed	42	Random seed for sampling (default: 42)
--output-posteriors	—	Export HMM posteriors to file (H5 or TSV)
--debug-timing	off	Show per-read timing breakdown
--daf-mask-runs	—	Thin DAF targets in same-strand runs of >= N original C (CT) or G (GA) bases (N=2: CC/GG and longer). Default: 2 with keep-one for DddA (duplex-validated), off otherwise; 0 disables.
--daf-run-policy	keep-one	With --daf-mask-runs: keep each run's 5'-most target (default) or drop the run. Choices: keep-one, drop.
--daf-mask-unaligned / --no-daf-mask-unaligned	on	DAF only: treat insertion and soft-clip bases as no evidence and leave unaligned stretches of >= 50 bp uncalled (default on, as in fiberhmm-call).
--use-m5c / --no-use-m5c	auto	DddA CpG-aware calling, as in fiberhmm-call: CpG observations are hidden from the HMM except inside confident unmethylated islands (MA ddda_ucg from fiberhmm-tag-m5c) the input carries. Default: on for --enzyme ddda, off otherwise.
--cpg-mask-policy	—	With CpG-aware calling: keep CpGs only inside ddda_ucg islands (default unmethylated-only), or mask only ddda_mcg spans (methylated-only). DddA only; ignored with a warning otherwise. Choices: unmethylated-only, methylated-only.
--harmonize-legacy-tags	off	Write ns/nl as the nucleosomes that MA would hold (footprints outside MSPs), so legacy tags tile with as/al like fiberhmm-call output. Default: ns/nl list every HMM footprint, which fiberhmm-recall-tfs/-nucs need; they refuse harmonized input.

fiberhmm-recall-tfs

Flag	Default	Description
-i / --in-bam	required	Input BAM tagged by fiberhmm-apply (has ns/nl/as/al). Use "-" for stdin.
-o / --out-bam	required	Output BAM with MA/AQ + refreshed legacy tags. Use "-" for stdout (for piping to ft fire, samtools, etc).
-m / --model	—	FiberHMM model JSON. If omitted, the bundled model for --enzyme/--seq is used automatically.
--enzyme	—	Enzyme preset: auto-selects the bundled model and min-llr/emission-uplift defaults (ddda, dddb, hia5). Choices: ddda, dddb, hia5.
--seq	—	Hia5 sequencing platform. When omitted it is taken from the input's FIBERHMM-CHEMISTRY declaration or detected from its MM specs (PacBio T-a vs Nanopore A+a only); conflicting evidence stops the run, and a given --seq that the reads contradict is refused (see --force-seq). Ignored for dddb/ddda. Choices: pacbio, nanopore.
--force-seq	off	Use the given --seq even when the input's MM specs or header say the reads come from the other platform (normally refused: the wrong platform model changes the calls, e.g. ~100x more TF calls for Nanopore reads called as PacBio), and run Hia5 even when the first reads carry no m6A calls.
--replace-chemistry	off	Replace, instead of reconcile with, the input BAM's FIBERHMM-CHEMISTRY declaration. By default a custom --model inherits the input's enzyme/platform when its observation mode matches, and a conflicting --enzyme/--seq is refused.
--daf-mask-runs	—	DAF only: thin targets lying in same-strand runs of >= N original C (CT) or G (GA) bases (CC/GG and longer at N=2; see --daf-run-policy). Adjacent conversions are coupled and do not follow the per-site emission model. Default: 2 with keep-one for --enzyme ddda (duplex-validated), off otherwise; 0 disables.
--daf-run-policy	keep-one	With --daf-mask-runs: keep each run's 5'-most target (default) or drop the run. Choices: keep-one, drop.
--daf-mask-unaligned / --no-daf-mask-unaligned	on	DAF only: treat CIGAR insertion and soft-clip bases as no evidence and leave unaligned stretches of >= 50 bp uncalled (default on, as in fiberhmm-call).
--daf-insert-consensus	auto	DAF only: re-encode insertions carried by enough reads against their deamination-aware consensus, as in fiberhmm-call (default auto: file input). Choices: auto, off.
--daf-insert-min-carriers	20	Carriers an insertion needs for --daf-insert-consensus (default 20).
--min-llr	—	Override native LLR cost per TF interval in joint decoding (nats; default: enzyme preset; not an FDR threshold).
--prob-threshold	—	Min MM/ML probability 0-255 for re-reading modification calls. Default: chemistry preset -- 248 for Hia5 Nanopore (--seq nanopore, or the input's declared chemistry), 125 otherwise. R/Y- and MD-encoded DAF input is binary and ignores it.
--min-opps	3	Min informative target positions per call (default 3)
--emission-uplift	—	Power transform on emission probabilities. Default 1.0 (identity). Use a pre-uplifted model file (e.g. ddda_TF.json) for DddA rather than setting this.
--use-m5c / --no-use-m5c	auto	Enable CpG-aware DddA recall. By default, retain CpGs only inside confident ddda_ucg MA spans. Enabled automatically for DddA, disabled for other enzymes; use --use-m5c explicitly with a custom DddA model.
--cpg-mask-policy	—	CpG-aware policy: retain CpGs only inside confident unmethylated islands (default unmethylated-only), or reproduce the former behavior that masks only ddda_mcg spans. DddA only; ignored with a warning otherwise. Choices: unmethylated-only, methylated-only.
--unify-threshold	90	v2 nucs with nl < this are scanned + may be demoted to tf+ if overlapped by a recaller call (default 90)
--input-frame	auto	Coordinate frame of the input ns/nl/as/al tags. "auto" (default) follows the @PG PP chain: the last footprint writer decides (fibertools predict-m6a/add-nucleosomes/fire -> molecular; FiberHMM call/apply/recall -> molecular when declared coord=molecular, else query); no writer -> molecular if coord=molecular is declared, else query (FiberHMM <= 2.12); a header with no writer and no declaration (stripped @PG) is read as query, except reads whose MA has a "." nuc/msp/tf group (molecular). Merged histories whose writers disagree, with no declaration, stop the run at the first read with these tags. A wrong frame mirrors reverse-strand calls. fibertools Ma tags are always molecular. Choices: auto, molecular, query.
--no-legacy-tags	off	Skip refreshed ns/nl/as/al -- emit only MA/AQ.
--downstream-compat	off	Downstream-compatibility mode: skip MA/AQ entirely and write TF calls INTO the legacy ns/nl tag alongside nucleosomes (sorted by start). Use for older tools that do not understand the Molecular-annotation spec. Loses per-TF quality scoring (tq/el/er) -- positions and lengths only.
-c / --cores	1	Worker processes (0 = all CPUs; default 1).
--chunk-size	1024	Reads per worker chunk (default 1024). Larger values reduce IPC overhead; decrease if RAM is constrained (each chunk holds reads in memory).
--io-threads	4	htslib BAM compression threads (default 4).
--context-size	—	Override context size. Default: read from model.
--max-reads	0	0 = no limit (default)
--recall-nucs / --no-recall-nucs	off	Enable nucleosome recall before TF recall. (Default on for fiberhmm-recall-nucs.)
--split-min-llr	4.0	Min accessible-cut LLR to split a footprint; for DddA, the molecule-local linker-residue configuration LLR (default 4.0)
--split-min-opps	3	Min informative positions for a split cut or DddA linker residue (default 3)
--ddda-derived-tf-max-edge-gap	12	DddA phase-aware radial recall only: require TF scan space opened solely by nucleosome refinement to have a deamination hit within BP on both sides (default 12; -1 disables).
--nuc-min-size	85	Min refined nucleosome size; smaller footprints are demoted to accessible/MSP (default 85)
--msp-min-size	0	Min re-derived MSP size to keep (default 0)

fiberhmm-recall-nucs

Flag	Default	Description
-i / --in-bam	required	Input BAM tagged by fiberhmm-apply (has ns/nl/as/al). Use "-" for stdin.
-o / --out-bam	required	Output BAM with MA/AQ + refreshed legacy tags. Use "-" for stdout (for piping to ft fire, samtools, etc).
-m / --model	—	FiberHMM model JSON. If omitted, the bundled model for --enzyme/--seq is used automatically.
--enzyme	—	Enzyme preset: auto-selects the bundled model and min-llr/emission-uplift defaults (ddda, dddb, hia5). Choices: ddda, dddb, hia5.
--seq	—	Hia5 sequencing platform. When omitted it is taken from the input's FIBERHMM-CHEMISTRY declaration or detected from its MM specs (PacBio T-a vs Nanopore A+a only); conflicting evidence stops the run, and a given --seq that the reads contradict is refused (see --force-seq). Ignored for dddb/ddda. Choices: pacbio, nanopore.
--force-seq	off	Use the given --seq even when the input's MM specs or header say the reads come from the other platform (normally refused: the wrong platform model changes the calls, e.g. ~100x more TF calls for Nanopore reads called as PacBio), and run Hia5 even when the first reads carry no m6A calls.
--replace-chemistry	off	Replace, instead of reconcile with, the input BAM's FIBERHMM-CHEMISTRY declaration. By default a custom --model inherits the input's enzyme/platform when its observation mode matches, and a conflicting --enzyme/--seq is refused.
--daf-mask-runs	—	DAF only: thin targets lying in same-strand runs of >= N original C (CT) or G (GA) bases (CC/GG and longer at N=2; see --daf-run-policy). Adjacent conversions are coupled and do not follow the per-site emission model. Default: 2 with keep-one for --enzyme ddda (duplex-validated), off otherwise; 0 disables.
--daf-run-policy	keep-one	With --daf-mask-runs: keep each run's 5'-most target (default) or drop the run. Choices: keep-one, drop.
--daf-mask-unaligned / --no-daf-mask-unaligned	on	DAF only: treat CIGAR insertion and soft-clip bases as no evidence and leave unaligned stretches of >= 50 bp uncalled (default on, as in fiberhmm-call).
--daf-insert-consensus	auto	DAF only: re-encode insertions carried by enough reads against their deamination-aware consensus, as in fiberhmm-call (default auto: file input). Choices: auto, off.
--daf-insert-min-carriers	20	Carriers an insertion needs for --daf-insert-consensus (default 20).
--min-llr	—	Override native LLR cost per TF interval in joint decoding (nats; default: enzyme preset; not an FDR threshold).
--prob-threshold	—	Min MM/ML probability 0-255 for re-reading modification calls. Default: chemistry preset -- 248 for Hia5 Nanopore (--seq nanopore, or the input's declared chemistry), 125 otherwise. R/Y- and MD-encoded DAF input is binary and ignores it.
--min-opps	3	Min informative target positions per call (default 3)
--emission-uplift	—	Power transform on emission probabilities. Default 1.0 (identity). Use a pre-uplifted model file (e.g. ddda_TF.json) for DddA rather than setting this.
--use-m5c / --no-use-m5c	auto	Enable CpG-aware DddA recall. By default, retain CpGs only inside confident ddda_ucg MA spans. Enabled automatically for DddA, disabled for other enzymes; use --use-m5c explicitly with a custom DddA model.
--cpg-mask-policy	—	CpG-aware policy: retain CpGs only inside confident unmethylated islands (default unmethylated-only), or reproduce the former behavior that masks only ddda_mcg spans. DddA only; ignored with a warning otherwise. Choices: unmethylated-only, methylated-only.
--unify-threshold	90	v2 nucs with nl < this are scanned + may be demoted to tf+ if overlapped by a recaller call (default 90)
--input-frame	auto	Coordinate frame of the input ns/nl/as/al tags. "auto" (default) follows the @PG PP chain: the last footprint writer decides (fibertools predict-m6a/add-nucleosomes/fire -> molecular; FiberHMM call/apply/recall -> molecular when declared coord=molecular, else query); no writer -> molecular if coord=molecular is declared, else query (FiberHMM <= 2.12); a header with no writer and no declaration (stripped @PG) is read as query, except reads whose MA has a "." nuc/msp/tf group (molecular). Merged histories whose writers disagree, with no declaration, stop the run at the first read with these tags. A wrong frame mirrors reverse-strand calls. fibertools Ma tags are always molecular. Choices: auto, molecular, query.
--no-legacy-tags	off	Skip refreshed ns/nl/as/al -- emit only MA/AQ.
--downstream-compat	off	Downstream-compatibility mode: skip MA/AQ entirely and write TF calls INTO the legacy ns/nl tag alongside nucleosomes (sorted by start). Use for older tools that do not understand the Molecular-annotation spec. Loses per-TF quality scoring (tq/el/er) -- positions and lengths only.
-c / --cores	1	Worker processes (0 = all CPUs; default 1).
--chunk-size	1024	Reads per worker chunk (default 1024). Larger values reduce IPC overhead; decrease if RAM is constrained (each chunk holds reads in memory).
--io-threads	4	htslib BAM compression threads (default 4).
--context-size	—	Override context size. Default: read from model.
--max-reads	0	0 = no limit (default)
--recall-nucs / --no-recall-nucs	on	Enable nucleosome recall before TF recall. (Default on for fiberhmm-recall-nucs.)
--split-min-llr	4.0	Min accessible-cut LLR to split a footprint; for DddA, the molecule-local linker-residue configuration LLR (default 4.0)
--split-min-opps	3	Min informative positions for a split cut or DddA linker residue (default 3)
--ddda-derived-tf-max-edge-gap	12	DddA phase-aware radial recall only: require TF scan space opened solely by nucleosome refinement to have a deamination hit within BP on both sides (default 12; -1 disables).
--nuc-min-size	85	Min refined nucleosome size; smaller footprints are demoted to accessible/MSP (default 85)
--msp-min-size	0	Min re-derived MSP size to keep (default 0)

fiberhmm-qc

Flag	Default	Description
-i / --input	required	One or more FiberHMM-compatible BAM/CRAM files; -i may be repeated
-o / --output-dir	—	QC output directory (default: qc/ beside the input BAMs)
--mode	auto	Observation mode (default: infer from @PG/tags) Choices: auto, daf, pacbio-fiber, nanopore-fiber.
--enzyme	auto	Enzyme (default: infer from @PG) Choices: auto, hia5, ddda, dddb.
--reference-profile	auto	Empirical QC reference (default: assay-aware auto selection) Choices: auto, none, ddda, dddb, hia5_nanopore, hia5_pacbio.
--reference	—	Indexed FASTA fallback for raw DAF BAMs lacking MD/R/Y
--sample-reads	2000	Target bounded sample size (default 2,000)
--seed	20260824	Deterministic sampler seed (default 20260824)
--min-mapq	20	Minimum mapping quality (default 20)
--prob-threshold	—	Minimum MM/ML probability, 0-255 (default: 248 for Hia5 Nanopore, the threshold its QC reference is calibrated at; 125 otherwise)
--min-opportunities	200	Minimum target sites per read for rate QC (default 200)
--snp-mask	—	Applied DAF SNP-mask BED to summarize (single input only)
--snp-report	—	fiberhmm-daf-snps JSON to plot (single input only)
--state-source	auto	Where the in-MSP/outside-MSP split comes from: the BAM's FiberHMM calls when the sample carries them, else a bounded light call with the bundled model of the declared chemistry (auto, default); calls only (tags); always re-call (light-call); or skip (none) Choices: auto, tags, light-call, none.
--min-msp-bp	85	Shortest MSP counted as accessible for the in-MSP rate; shorter gaps (linkers) count as outside-MSP (default 85; the packaged references are calibrated at this value)
--light-call-reads	400	Most sampled reads the light call runs on (default 400)
--light-call-seconds	60.0	Wall-time budget of the light call (default 60 s)
--fail-on-qc	off	Exit 2 when the final status is FAIL

fiberhmm-check

Flag	Default	Description
PATH	—	BAM/CRAM/SAM, <prefix>.qc.json, posteriors .tsv(.gz)/.h5, a fiberhmm-consensus output directory, or a fiberhmm-pipeline output directory
--json	off	Print one JSON document (schema fiberhmm.advisory_check.v1) with a fiberhmm.advisory_report.v1 report per path
--scan-records	—	BAM records to scan for read-level evidence (duplicate, pairing and call tags); 0 reads the header only (default: 5000)
--no-sidecars	off	Do not also check the QC report fiberhmm-call writes beside a BAM (qc/<name>.qc.json)
--list	off	List every known advisory and exit

fiberhmm-extract

Flag	Default	Description
-i / --input	required	Input tagged BAM file
-o / --outdir	—	Output directory (default: same as input)
-c / --cores	1	Number of CPU cores
--nucleosome / --footprint	off	Extract nucleosomes (ns/nl or MA nuc tags). (--footprint is a deprecated alias.)
--msp	off	Extract MSPs (as/al tags)
--tf	off	Extract TF/Pol II footprints from MA/AQ tag (tf.QQQ). Requires a BAM called by fiberhmm-call or fiberhmm-recall-tfs (not --downstream-compat). See --min-tq for the quality floor.
--min-tq	50	Minimum TF quality (tq) to extract. 0-255 scale where tq = min(255, round(LLR * 10)). Default 50 (LLR >= 5 nats, ~148:1 likelihood ratio), the per-interval cost every bundled preset calls with, so it keeps essentially every call. Set 0 for every emitted call, 100+ for a stricter view.
--m6a	off	Extract m6A positions
--m5c	off	Extract DddA MA ddda_mcg spans, or native MM/ML 5mC positions
--deam	off	Extract DAF-seq deamination calls. Priority: (1) MM/ML-native dU calls (mod code "u" or ChEBI 55797); (2) IUPAC R/Y codes in the query sequence (fiberhmm-daf-encode output); (3) MD-tag ref mismatches as a fallback for raw DAF BAMs. First non-empty source wins per read. blockMod: 0 = R/GA-dea, 1 = Y/CT-dea, matching FiberBrowser flavor codes.
--both-strand / --bothstrand	off	Extract the deam+ / deam- intersection of paired DddA consensus reads as a BED12 coverage overlay.
--all	off	Extract all tag types (default if none specified)
--bed-only	off	Output BED only (no bigBed)
--keep-bed	off	Keep BED files when creating bigBed
-q / --min-mapq	0	Min mapping quality (default: 0, no filtering)
-p / --prob-threshold	—	Min probability for native MM/ML m6a/m5c/dU calls (0-255). Default: 248 when the BAM declares Hia5 Nanopore (FIBERHMM-CHEMISTRY header), 125 otherwise. Not applied to DddA MA ddda_mcg spans or to R/Y/MD deaminations (binary).
--no-scores	off	Omit scores from output
--block-scores	off	Append per-block quality as extra BED column(s) (BED12+N). nucleosome -> blockNq/blockEl/blockEr, msp -> blockAq, m6a/native-m5c -> blockMl (DddA ddda_mcg spans -> 0), tf -> blockTq/blockEl/blockEr. bigBed uses -type=bed12+N and the matching autoSQL schema so FiberBrowser/UCSC can surface per-feature quality without a sidecar database.
--circular-groups	off	Append circId/circPart/circParts/molStart/molLength columns and preserve MA/AN groups for circular wrapped nucleosome, MSP, and TF features.
--haplotype-fields	off	Append scalar hp and ps columns copied from the source BAM HP/PS tags after every other optional BED field. Missing or non-integer tags are written as -1. Off by default to preserve the existing BED/bigBed schema byte-for-byte.
--sample-name	—	Sample/dataset identifier to embed in the autoSQL description of every output bigBed ("Sample: <name>. ..."). Default: BAM basename stem. Lets downstream tools match a bigBed to its source without filename parsing.
--input-frame	auto	Coordinate frame of legacy ns/nl/as/al tags (read when a read has no MA nuc/msp). "auto" (default) follows the @PG provenance, as fiberhmm-recall-tfs: the last footprint writer decides (fibertools, and FiberHMM declaring coord=molecular -> molecular; FiberHMM <= 2.12 -> query); a header with no evidence (stripped @PG) is read as query, except reads whose MA has a "." nuc/msp/tf group (molecular). Merged histories that disagree stop the run. MA follows its own header rule. Choices: auto, molecular, query.
--region-size	10000000	Region size for parallel
--skip-scaffolds	off	Skip scaffold chromosomes
--chroms	—	Comma-separated chromosomes to process
--sort-mem / -S	1G	Memory buffer for the BED sort, passed to sort -S (e.g. 4G, 8G, or 50% on GNU sort). Bigger = fewer temp-file merge passes = faster. Default 1G; pass an empty string to disable. The sort also runs under LC_ALL=C regardless, which alone is a large speedup.
--sort-parallel	0	Threads for the BED sort (GNU coreutils only; ignored on BSD/macOS sort). 0 = use --cores. Feature-detected, so safe to leave on.

fiberhmm-dedup

Flag	Default	Description
-i / --input	required	Input DAF-seq BAM (indexed not required)
-o / --output	required	Output BAM
--min-jaccard	0.95	Min deamination-set Jaccard to call two reads the same molecule (default 0.95). Use 0.90 for optional more aggressive matching; lower thresholds can merge distinct molecules.
--flag-only	off	Keep all reads and only mark duplicates: set the 0x400 duplicate flag on non-representatives and di/ds cluster tags on every member of a duplicate cluster. Default: collapse each cluster to one representative read.
--min-deam	10	Reads with fewer than this many deamination calls are not fingerprintable; passed through untouched (default 10).
--max-end-diff	50	Maximum difference at both aligned reference ends for two reads to be duplicates (default 50 bp).
--ignore-strand	off	Cluster across deamination flavours. Default: only reads with the same dominant flavour (C->T vs G->A, i.e. the same template strand) can be duplicates; alignment orientation is never used.
-p / --prob-threshold	128	Min ML probability for MM/ML-native dU calls (0-255; default 128, the same as fiberhmm-call). R/Y- and MD-encoded input is binary and ignores it.
--num-hashes	32	MinHash signature width (default 32).
--bands	8	LSH bands; rows = num-hashes / bands (default 8 -> rows 4). More bands = higher recall, more candidate pairs.
--seed	7	MinHash RNG seed (default 7).
--stats-tsv	—	Write a cluster_id<TAB>n_reads table to this path.
--io-threads	4	htslib BAM compression threads for output (default 4).

fiberhmm-daf-encode

Flag	Default	Description
-i / --input	required	Input BAM file, or "-" for stdin
-o / --output	required	Output BAM path, or "-" for stdout
--reference	—	Reference FASTA (fallback if MD tag is missing)
-q / --min-mapq	20	Minimum mapping quality (default: 20)
--min-read-length	1000	Minimum aligned read length in bp (default: 1000)
--io-threads	4	htslib I/O threads for BAM compression/decompression (default: 4)
--strand	auto	Force conversion strand: CT (+ strand), GA (- strand), or auto (per-read consensus, default: auto) Choices: CT, GA, auto.

fiberhmm-daf-snps

Flag	Default	Description
-i / --input	required	Input aligned BAM with MD tags
-o / --output-prefix	—	Output prefix (default: <BAM stem>.daf_snps)
--min-fraction	0.2	Minimum mismatch fraction in each direction (validated default 0.2)
--min-depth	5	Minimum depth in each conversion-direction class (validated default 5; all thresholds must pass in both classes)
--min-alt-fibers	5	Minimum mismatch-supporting fibers in each direction (validated default 5)
--min-dominant-events	5	Minimum dominant conversions per classifiable fiber (default 5)
--min-dominant-purity	0.8	Minimum dominant-direction purity (default 0.80)
--min-mapq	20	Minimum mapping quality (default 20)
--min-amplicon-reads	20	Minimum aligned reads required for an amplicon consensus (default 20)
--reference	—	Indexed FASTA fallback for BAMs lacking MD

fiberhmm-pair

Flag	Default	Description
-i / --input	required	Coordinate-sorted, indexed FiberHMM-called DddA BAM
-o / --output	required	Output BAM: joint duplex molecules by default; paired source reads with --stop-after pair
-r / --reference	—	Matching indexed FASTA. Required by the default sequence-free score; optional with --sequence-only when MD+CIGAR is available
--sequence-only	off	Accept only direct A/T sequence-supported pairs; disable the sequence-free model
--pairs-tsv	—	Write selected-pair evidence to this TSV
--receipt-json	—	Write a machine-readable pairing receipt
--pairs-only / --paired-only	off	Write only paired records: merged duplex molecules (default), or paired source reads with --stop-after pair. Otherwise unpaired reads pass through unchanged.
--stop-after	recall	Last stage to run: pair (tag pairs only), merge (both-strand molecules without re-calling) or recall (default) Choices: pair, merge, recall.
--from-paired	off	Input is already pair-tagged (mt/mp from an earlier --stop-after pair run): skip pairing and start at merge. Pairing-stage options (--reference, --sequence-only, --pairs-tsv, --model, --min-* ...) are rejected here.
--model	—	Override the bundled frozen sequence-free model JSON
--call-layer	auto	Nucleosome calibration for the sequence-free model (default auto) Choices: auto, input-ma, rotational-recall.
--min-margin	1.0	Minimum two-sided sequence-free model margin (default 1.0)
--null-floor	0.0	Virtual null model score for a lone candidate (default 0.0)
--min-overlap	1500	Min genomic overlap bp (default 1500)
--min-nucs	4	Min nucleosome dyads within the overlap, each read (default 4)
--min-sequence-bases	500	Min shared reference-A/T bases for a sequence edge (default 500)
--min-component-discordance-rate	0.02	Min rejected-edge difference rate to constrain a 2x2 (default 0.02)
--max-sequence-pair-rate	0.01	Max difference rate on a sequence-selected pair (default 0.01)
--min-sequence-margin	0.002	Min sequence preference/assignment margin (default 0.002)
-p / --prob-threshold	128	Min ML probability for MM/ML-native dU calls (0-255; default 128, the same as fiberhmm-call). R/Y- and MD-encoded input is binary and ignores it.
--max-component	10000	Safety ceiling for a complete overlap component (default 10000)
--io-threads	4	htslib compression threads for output (default 4)
--no-index	off	Do not index a paired-source output
--ddda-derived-tf-max-edge-gap	12	Edge-evidence requirement for TF calls exposed only by DddA nucleosome refinement (default 12; -1 disables)
--use-m5c / --no-use-m5c	auto	Joint recall: DddA CpG-aware calling, as in fiberhmm-call -- CpG observations are excluded from the joint HMM and from recall except inside the source reads' ddda_ucg islands (fiberhmm-tag-m5c). Default: on; --no-use-m5c for an ablation.
--cpg-mask-policy	—	With CpG-aware calling: keep CpGs only inside ddda_ucg islands (default unmethylated-only), or mask only ddda_mcg spans (methylated-only), as in fiberhmm-call. Choices: unmethylated-only, methylated-only.
--daf-mask-runs	—	Thin DAF targets in same-strand runs of >= N original C (CT) or G (GA) bases (N=2: CC/GG and longer). Default: 2 with keep-one for DddA (duplex-validated), off otherwise; 0 disables.
--daf-run-policy	keep-one	With --daf-mask-runs: keep each run's 5'-most target (default) or drop the run. Choices: keep-one, drop.

fiberhmm-merge

Flag	Default	Description
-i / --input	required	BAM from fiberhmm-pair (mt/mp tags)
-o / --output	required	Output consensus BAM (sorted + indexed)
--pairs-only	off	Emit only consensus reads (default: also pass through unmerged reads)
--recall	off	Re-call footprints on each both-strand consensus read (HMM, nucleosome and TF recall over both strands; writes ns/nl/as/al + MA nuc/msp/tf). Reads the deam+/deam- regime and uses C and G targets jointly.
--enzyme	ddda	Model preset for --recall (default ddda)
--ddda-derived-tf-max-edge-gap	12	With --recall, require TF calls exposed solely by DddA radial nucleosome refinement to have a deamination hit within BP on both sides (default 12; -1 disables).
--use-m5c / --no-use-m5c	auto	With --recall: DddA CpG-aware calling, as in fiberhmm-call (CpGs excluded from the joint HMM and from recall except inside the source reads' ddda_ucg islands). Default: on for --enzyme ddda.
--cpg-mask-policy	—	With CpG-aware calling: keep CpGs only inside ddda_ucg islands (default unmethylated-only), or mask only ddda_mcg spans (methylated-only), as in fiberhmm-call. Choices: unmethylated-only, methylated-only.
-p / --prob-threshold	128	Min ML probability for MM/ML-native dU calls (0-255; default 128, the same as fiberhmm-call). R/Y- and MD-encoded input is binary and ignores it.
--io-threads	4	htslib compression threads (default 4)
--daf-mask-runs	—	Thin DAF targets in same-strand runs of >= N original C (CT) or G (GA) bases (N=2: CC/GG and longer). Default: 2 with keep-one for DddA (duplex-validated), off otherwise; 0 disables.
--daf-run-policy	keep-one	With --daf-mask-runs: keep each run's 5'-most target (default) or drop the run. Choices: keep-one, drop.

fiberhmm-tag-m5c

Flag	Default	Description
-i / --input	required	R/Y-encoded, FiberHMM-tagged coordinate BAM
-o / --output	required	Output BAM
-r / --reference	required	Reference FASTA
--enzyme	required	Required chemistry assertion. Only DddA DAF-seq is supported. Choices: ddda.
--posterior	0.99	Whole-island posterior needed for a call: >= this is methylated (ddda_mcg), <= 1 - this unmethylated (ddda_ucg) (default 0.99).
--min-other	10	Minimum non-CpG observations on the molecule's island overlap, the internal accessibility baseline (default 10).
--cpg-islands	—	Optional BED3 of merged, non-overlapping CpG islands. Default: infer islands from the reference sequence of the contigs the reads map to (header-only contigs need not be in the FASTA).
--write-cpg-islands	—	Optional BED or BED.GZ containing the exact island catalog used.
--min-island-cpg	15	Minimum CpG observations for a whole-island call (default 15).
--calls-tsv	—	Optional whole-island audit table containing methylated, unmethylated and uninformative molecule/island overlaps.
--cpg-island-window	200	Reference inference window in bp (default 200).
--cpg-island-step	10	Reference inference step in bp (default 10).
--cpg-island-min-gc	0.5	Minimum GC fraction for inferred islands (default 0.50).
--cpg-island-min-oe	0.6	Minimum CpG observed/expected for inferred islands (default 0.60).
--five-prime-factors	—	Comma-separated A,C,G,T factors; default calibrated DddA values
--estimate-factors	off	Estimate 5' factors from this BAM instead of using calibrated values
--factor-sample-reads	5000	Reads sampled by --estimate-factors (default 5000).
--input-frame	auto	Frame of legacy ns/nl tags when MA is absent. auto follows the @PG provenance, as fiberhmm-recall-tfs --input-frame auto (last footprint writer; no writer: molecular if coord=molecular is declared, else query); a header with no evidence (stripped @PG) is read as query, except reads whose MA has a '.' nuc/msp/tf group (molecular); merged histories that disagree are read as query Choices: auto, molecular, query.
--io-threads	4	htslib BAM compression threads (default 4).

fiberhmm-call-m5c

Flag	Default	Description
-i / --input	required	One or more coordinate-sorted, indexed DAF FiberHMM BAMs
-r / --reference	required	Indexed reference FASTA
-o / --output	required	Output BED6, or - for stdout
--region	required	contig[:start-end], 1-based display syntax
--enzyme	required	Required chemistry assertion. Only DddA DAF-seq is supported. Choices: ddda.
--window	1000	Aggregate window size in bp (default 1000).
--chunk-bp	5000000	Bound observation collection in this many bp; the HMM still runs once across the complete region
--min-other	10	Minimum non-CpG observations for a molecule's per-window baseline (default 10).
--min-cpg	10	Minimum CpG observations for an informative window (default 10).
--posterior	0.99	Window posterior needed to join a called domain (default 0.99).
--max-gap	1000	Maximum gap in bp bridged between called windows of the same state (default 1000).
--five-prime-factors	—	Comma-separated A,C,G,T factors; default calibrated DddA values
--estimate-factors	off	Estimate 5' factors from the complete supplied region (explicit high-memory audit option)
--tag-bam	—	Optionally add whole-island annotations to this BAM
--tag-output	—	Output BAM for --tag-bam (required with --tag-bam)
--tag-mode	island	island: one molecule state per complete CpG island (default); locus: copy aggregate validation domains Choices: island, locus.
--cpg-islands	—	Optional merged BED3 island catalog; default: infer from reference
--write-cpg-islands	—	Optional BED/BED.GZ recording the exact catalog used
--read-posterior	0.99	Whole-island posterior threshold (default: 0.99)
--read-min-island-cpg	15	Minimum CpGs per molecule-island call (default: 15)
--tag-input-frame	auto	Frame of tag-BAM legacy ns/nl when MA is absent; auto as in fiberhmm-tag-m5c --input-frame Choices: auto, molecular, query.
--io-threads	4	htslib BAM threads (default 4).

fiberhmm-consensus

Flag	Default	Description
--max-discovery-molecules	—	Molecules per dataset and chemical channel for discovery; 0 = all molecules; 2000 is a good cap for deep data (thousands of reads per site) (default 0)
--max-quantification-molecules	—	Molecules per dataset and chemical channel for quantification; 0 = all molecules; 2000 is a good cap for deep data (thousands of reads per site) (default 0)
--sample-seed	—	Stable molecule hash seed (default 1)
--schema	off	Print every parameter group with defaults and help as JSON (cr.engine selects the engine; the default is lattice_recaller)
--bam	—	Repeat for separate datasets; chemistry comes from BAM @CO metadata
--datasets	—	JSON list of
--evidence	—	Saved evidence.json.gz, including pooled evidence
--resume	—	Start a NEW run (in a new --output) from a finished run directory: reuses its evidence.json.gz, manifest.json and fit_cache to rerun stages. To finish an interrupted multi-window run in place, use --continue
--bed	—	BED3 windows for independent runs; BED6 with equal widths for CL-CR
--region	—	Alternative CHROM:START-END, explicitly 0-based half-open
--pool-loci	off	CL-CR: pool BED6 windows in their provided orientations
--chemistry	—	Explicit missing-metadata declaration for --bam; conflicts fail Choices: ddda, dddb, hia5-pacbio, hia5-nanopore.
--parameters	—	JSON parameter groups; see --schema
--engine	—	Consensus engine (default lattice_recaller; staged_native_families is the deprecated Monte Carlo engine). Overrides cr.engine from --parameters or --resume Choices: lattice_recaller, staged_native_families.
--consolidation-bp	—	staged_native_families only: shared-family edge allowance (default 10; 5 gives finer grouping)
--stop-after	—	staged_native_families only: last stage to compute (the lattice recaller runs in one pass) Choices: native, parents, consolidated, resolved.
--start-at	native	staged_native_families only: consolidation restarts from saved native fits Choices: native, consolidation.
--cores	—	Worker processes (sets compute.cores; default 4)
--robust	—	lattice_recaller only, optional and slower: also rerun discovery under N other read orders and mark which classes are robust to read order (supported in every one of the N+1 orders, or the share set by recaller.order_robust_fraction; classes.tsv columns order_robustness and robust). Takes about N+1 times as long; the classes themselves are unchanged. Sets recaller.order_replicates (default 0 = off)
--cache	—	staged_native_families only: persistent exact native-fit cache directory
--json-progress / --progress-json	off	Structured progress on stderr (JSON lines)
--daf-mask-runs	—	DAF only: thin targets in same-strand runs of >= N original C (CT) or G (GA) bases in lattices and native replay (2 = CC/GG and longer; 0 = off). Default: per dataset chemistry, DddA keep-one on runs >= 2 (duplex-validated), DddB off
--daf-run-policy	keep-one	With --daf-mask-runs: keep each run's 5'-most target (default) or drop the run Choices: keep-one, drop.
--no-bam	off	Save frozen results/reports without materializing family-tagged BAMs
--bam-scope	regions	Export whole alignments overlapping analyzed windows (default), or the full source BAM Choices: regions, full.
--bam-grouping	datasets	One BAM per logical dataset (default) or original source file Choices: datasets, files.
--bam-recaller-layer	off	lattice_recaller: also write the optional tf_recaller MA layer (the recaller's own per-molecule class calls at every prevalence tier; bytes tq,fi,tier,q0,lr,rr) to exported BAMs. Off by default; the calls are always in result.json.gz
--output	—	New or empty result directory (with --continue: the interrupted run's directory)
--continue	off	Finish an interrupted --bam + --bed/--region run in its existing --output directory: windows whose completion marker matches are kept, missing or partial windows are rerun, and the aggregate outputs (regions.json, report.html, BAMs) are rebuilt. Refused if the BAMs, windows or parameters differ from the run's consensus_run.json (--cores, --window-jobs and --json-progress may change). Not the same as --resume, which starts a new run from saved evidence
--window-jobs	0	Independent BED windows analysed at the same time (default 0 = automatic: up to --cores windows, each with an equal share of --cores; one at a time for staged_native_families, whose per-window compute.maximum_matrix_mb budget is a hard limit). 1 = one window at a time, each using every core

fiberhmm-transfer

Flag	Default	Description
--freeze-run	—	Completed consensus result directory: a lattice_recaller run (any frame) or an oriented staged_native_families CL-CR run; export its classes/models without refitting
--models	—	frozen_classes.json.gz (lattice_recaller), frozen_models.json.gz (staged), or the --freeze-run output directory
--bam	—	Target BAM; repeat for separate datasets (needs --bed)
--datasets	—	JSON list of dataset_id and BAM paths
--evidence	—	Saved oriented native evidence.json.gz (one window or pooled evidence)
--bed	—	Equal-width, explicitly oriented BED6 target windows
--chemistry	—	Explicit missing-metadata declaration for --bam, as in fiberhmm-consensus Choices: ddda, dddb, hia5-pacbio, hia5-nanopore.
--parameters	—	BAM preparation parameters; target calls are replayed, families are never fitted
--chip-bed	—	Optional independent ChIP peaks, joined only after scoring
--no-bam	off	Do not write family-tagged BAMs
--bam-scope	regions	Export alignments overlapping the target windows (default), or the full source BAM Choices: regions, full.
--bam-grouping	datasets	One BAM per logical dataset (default) or per source file Choices: datasets, files.
--json-progress	off	Structured progress on stderr
--cores	—	Worker processes for scoring (default: this machine's consensus default)
--dataset-map	—	Use the frozen per-channel boxes and spots of source dataset SOURCE for target dataset TARGET (needed when several source datasets share the target chemistry)
--include-training-molecules	off	Score molecules the catalog was trained on (default: exclude them, as for staged families); use for self-application checks
--output	required	New or empty output directory

fiberhmm-nfr

Flag	Default	Description
--bam	—	Input BAM (repeat for separate datasets); chemistry from BAM @CO metadata
--datasets	—	JSON list of
--evidence	—	Saved evidence.json.gz (e.g. a fiberhmm-consensus run's), instead of loading BAMs
--region	—	CHROM:START-END analysis window (0-based, half-open); use this or --bed with BAMs
--bed	—	BED3 windows for independent NFR runs, written in BED order to window_XXXXXX directories
--window-jobs	0	Concurrent BED windows (0 = up to --cores; 1 = sequential). Workers are reused across windows
--nfr	—	NFR region inside the window (repeat for several). Default: detected from the share of callable reads inside a >= 175-bp gap (see --detect-method)
--chemistry	—	Explicit chemistry for --bam when the BAM header does not declare one Choices: ddda, dddb, hia5-pacbio, hia5-nanopore.
--parameters	—	JSON: loader groups as for fiberhmm-consensus (input, compute, ...) and an "nfr" group with any engine parameter (see --schema)
--stringency	—	Prediction strength needed for k (default 0.9; 0.85 admits finer variants)
--mode	—	variants (discovered, default) or depth (Timer 175/300/500-bp width states) Choices: variants, depth.
--elements	—	Co-accessibility elements: each variant (default) or whole NFRs Choices: variants, nfrs.
--classes	—	A fiberhmm-consensus (lattice recaller) output directory: its supported footprint classes become elements (variant x class pairs; unit_ids must come from the same datasets in the same order). Repeatable; class ids get the prefix R1:, R2: ... when more than one is given
--no-internal-footprint-labels	off	Do not label a split whose internal protection matches a supported class as "full NFR with internal footprint"
--pairs	—	Pairs tested: with at least one NFR element (default) or also class x class Choices: nfr, all.
--within-clusters	—	Also report the effect within K masked k-means clusters (default off)
--combo	—	Comma-separated element ids (3-8) for combination patterns (default: automatic)
--robust	—	Rerun discovery under N hashed read orders (default 0)
--bootstrap	—	Bootstrap replicates for prevalence intervals (default 200)
--detect-method	—	NFR detection: raw (default; share >= --min-fraction), enrichment (also gates on fold and q), or fixed (share above --detect-threshold) Choices: raw, enrichment, fixed.
--min-fraction	—	Raw/enrichment: minimum share of callable reads inside a >= 175-bp gap (default raw 0.10; enrichment 0.05)
--clip-gaps-to-region / --no-clip-gaps-to-region	auto	Assign each gap to the NFR region with largest overlap (ties: leftmost; default off)
--min-fold	—	Enrichment detection: share over the window's background at least this (default 2)
--detect-q	—	Enrichment detection: Benjamini-Hochberg q of the binomial test (default 0.01)
--min-open-reads	—	Raw/enrichment detection: reads inside a >= 175-bp gap at least this (default 10)
--detect-threshold	—	Fixed detection: share of callable reads inside a >= 175-bp gap above this (default 0.15; given alone, it selects the fixed rule)
--cores	—	Worker budget (compute.cores; default 4); BED runs reuse workers across independent windows
--max-discovery-molecules	—	Molecules per dataset and chemical channel for discovery; 0 = all molecules; 2000 is a good cap for deep data (thousands of reads per site) (default 0)
--max-quantification-molecules	—	Molecules per dataset and chemical channel for quantification; 0 = all molecules; 2000 is a good cap for deep data (thousands of reads per site) (default 0)
--sample-seed	—	Stable molecule hash seed (default 1)
--schema	off	Print the engine parameters with defaults as JSON and exit
--json-progress	off	Structured progress on stderr (JSON lines)
--output	—	New or empty output directory

fiberhmm-footprint-model

Flag	Default	Description
-i / --input	required	Recalled BAM containing ordinary tf and msp MA groups
-o / --output-prefix	required	Output filename prefix (not a directory)
--region	—	Analyze a zero-based, half-open region; repeatable and requires a BAM index. A bare CONTIG selects that complete contig.
--genome	—	Genome/assembly label recorded in provenance (for example dm6)
--genomewide	off	Assert that scanning the complete BAM represents a genome-wide analysis
--bigbed	off	Also create indexed population and FiberBrowser BigBeds
--bed-to-bigbed	—	Path to UCSC bedToBigBed; implies --bigbed
--force	off	Replace existing artifacts for this output prefix
-q / --min-mapq	0	Minimum alignment MAPQ (default: 0)
--include-duplicates	off	Include records carrying the BAM duplicate flag (excluded by default)
--smoothing-sigma	3.0	Footprint-center Gaussian sigma in bp (default: 3)
--peak-distance	15	Minimum distance between center modes (default: 15)
--assignment-radius	10	Maximum center-to-mode assignment radius (default: 10)
--edge-compatibility	12	Maximum within-family diameter for each boundary (default: 12)
--minimum-geometry-support-per-stratum	3	Molecules per stratum needed for that stratum to vote on canonical geometry (default: 3)
--minimum-geometry-support	3	Descriptive geometry-ready threshold (default: 3)
--minimum-population-support	3	Descriptive population-ready threshold (default: 3)
--minimum-mapped-fraction	0.95	Required mapped fraction for geometry, MSP projection, and site denominators (default: 0.95)

fiberhmm-strand-rescue

Flag	Default	Description
-i / --bam	required	Post-TF/post-nuc BAM; repeat only to pool shards or compatible timepoints from one inference cohort
--preset	required	Choices: ddda, dddb, hia5-nanopore.
--region	required
--model	—
--nuc-model	—	Protected/accessibility model used only for nucleosome edge evidence
--prob-threshold	—
--control-flank	2000
--min-mapq	20
--min-support	10
--minimum-geometry-support	3
--source-boundary-margin	10
--center-radius	10
--peak-distance	15
--tf-edge-compatibility	12	Maximum within-family diameter in bp for each TF boundary; co-centered calls outside this bound form distinct TF models
--max-boundary-mad	12.0
--min-local-enrichment	2.0
--local-background-radius	250
--max-auto-sites	0	Optional TF-family cap after scoring; 0 keeps the exhaustive locus map
--site	—	Externally seed one zero-based half-open START-END site; repeatable. Support and canonical edges are recomputed from ordinary cohort calls.
--forced-sites-only	off	Analyze only --site geometries rather than unioning automatic sites
--nuc-min-support	—	Required ordinary nuc calls on one strand (default: --min-support)
--nuc-center-radius	25
--nuc-source-boundary-margin	20
--nuc-edge-assignment-radius	48
--nuc-max-boundary-mad	24.0
--max-auto-nuc-sites	0	Optional nucleosome-family cap after scoring; 0 keeps all families
--nuc-site	—	Seed one zero-based half-open nucleosome population; repeatable. Edges are relearned from ordinary nuc calls and no length ceiling applies.
--forced-nuc-sites-only	off	Refine only --nuc-site populations rather than automatic nuc sites
--skip-nuc-edge-refinement	on	Do not run independent one-for-one nucleosome edge normalization (default; retained for command-line compatibility)
--independent-nuc-edge-refinement	off	Experimental population nucleosome-edge normalization. This is not the TF-conditioned consensus-nuc reconciliation stage.
--strand-min-source-support	—	Required source-strand ordinary TF calls (default: --min-support)
--strand-min-source-enrichment	1.5	Minimum source-strand focal enrichment over local background
--strong-posterior	0.95
--review-posterior	0.5
--tf-class-pseudocount	0.5	Symmetric per-configuration pseudocount for localized single/composite site-consensus priors
--tf-class-locus-gap	30	Maximum gap joining atomic footprint states into one consensus locus
--tf-class-max-span	250	Maximum span in bp of one localized site-consensus locus
--tf-class-max-sites	10	Maximum atomic states enumerated in a complete site-consensus action set; larger loci are reported as skipped
--maximum-sites-per-decision	8
--accessible-site-gap	220	Maximum gap joining TF sites into one MSP-origin decision
--control-shifts	``	Optional comma-separated target-coordinate shifts. Source priors remain at the true sites; controls are diagnostic, not an FDR null.
--max-reads	0
--molecule-collapse	auto	Collapse amplified DAF PCR families; auto enables for DddA/DddB Choices: auto, on, off.
--molecule-min-jaccard	0.95
--molecule-min-deam	10
--per-molecule-efficiency	on	Calibrate hard-call efficiency from each molecule's MSPs (default)
--global-efficiency	off	Use the model-wide accessible hard-call rate for every molecule
--efficiency-pseudo-count	20.0
--efficiency-min-opportunities	20
--report-layout	auto	Report action storage: auto spills at the bounded v4 limits; stream forces v5 BGZF action sidecars Choices: auto, inline, stream.
--diagnostics	aggregate	Per-call diagnostic storage (aggregate is the production default) Choices: aggregate, stream.
--proposal-tsv	—
-o / --output	required

fiberhmm-strand-rescue-annotate

Flag	Default	Description
--report	required
-i / --bam	—	Report input BAM to materialize; default is every report BAM
-o / --output	—	Output BAM; requires exactly one selected input
--output-dir	—	Directory receiving one regional BAM per selected input
--region	—
--minimum-posterior	0.0	Optional output-size filter; default 0 preserves every decision
--allow-input-drift	off
--io-threads	1

fiberhmm-strand-rescue-audit

Flag	Default	Description
-i / --bam	required
-o / --output	—
--max-errors	100

fiberhmm-tag-consensus

Flag	Default	Description
-i / --input	required	Input BAM with tf_sr.QQQ
-o / --output	required	New sorted/indexed BAM
-a / --assignments	required	v1/v2 family assignment TSV
--force	off	Replace an existing output

fiberhmm-posteriors

Flag	Default	Description
-i / --input	required	Input BAM file
-m / --model	—	FiberHMM model file. If omitted, uses the bundled model for --enzyme/--seq.
-o / --output	required	Output file (.tsv.gz for TSV, .h5/.hdf5 for HDF5)
--format	auto	Output format (default: auto-detect from extension) Choices: auto, hdf5, tsv.
--enzyme	—	Auto-select a bundled enzyme model. Choices: ddda, dddb, hia5.
--seq	—	Hia5 platform; detected from the input when omitted, as in fiberhmm-call (and checked against the reads when given). Ignored for dddb/ddda. Choices: pacbio, nanopore.
--force-seq	off	Use the given --seq even when the input's MM specs or header say the reads come from the other platform (normally refused: the wrong platform model changes the calls, e.g. ~100x more TF calls for Nanopore reads called as PacBio), and run Hia5 even when the first reads carry no m6A calls.
--edge-trim / -e	10	Bases to trim from read edges (default: 10)
--prob-threshold	—	Min ML probability (0-255) for an MM/ML modification call. Default: chemistry preset, as in fiberhmm-call -- 248 for Hia5 Nanopore (--seq nanopore, given or detected), 128 otherwise.
--keep-chimeras	off	Do not drop DAF strand-swap chimeric reads
--chimera-min-seg	5	DAF chimera: min same-strand deamination events per segment (default: 5)
--chimera-purity	0.8	DAF chimera: min same-strand purity per segment (default: 0.8)
--daf-snp-mask	—	0-based BED of reference positions whose conversions are ignored (e.g. the mask fiberhmm-call used)
--daf-mask-runs	—	Thin DAF targets in same-strand runs of >= N original C (CT) or G (GA) bases (N=2: CC/GG and longer). Default: 2 with keep-one for DddA (duplex-validated), off otherwise; 0 disables.
--daf-run-policy	keep-one	With --daf-mask-runs: keep each run's 5'-most target (default) or drop the run. Choices: keep-one, drop.
--use-m5c / --no-use-m5c	auto	DddA CpG-aware calling, as in fiberhmm-call: CpGs are hidden from the HMM except inside confident unmethylated islands (MA ddda_ucg) the input carries. Default: on for --enzyme ddda, off otherwise.
--cpg-mask-policy	—	With CpG-aware calling: keep CpGs only inside ddda_ucg islands (default unmethylated-only), or mask only ddda_mcg spans (methylated-only). DddA only; ignored with a warning otherwise. Choices: unmethylated-only, methylated-only.
--cores / -c	4	Worker processes (0 = all CPUs; default: 4)
--region-size	5000000	Region size in bp for parallel processing (default: 5,000,000)
--skip-scaffolds	off	Skip scaffold/contig chromosomes
--chroms	—	Only process these chromosomes
--streaming	off	Read the BAM once, in file order, in a single process (no index needed; --cores is not used). Default: region-parallel over an indexed BAM.
--batch-size	1000	Fibers per HDF5 write batch (default: 1000)
-v / --verbose	off	Also print model/region details and a progress bar (the summary is always printed, to stderr)

fiberhmm-probs

Flag	Default	Description
--accessible / -a	required	BAM file(s) from accessible/naked DNA (dechromatinized, MTase-treated)
--inaccessible / -u	required	BAM file(s) from inaccessible/untreated samples (native chromatin)
-o / --output	required	Output directory; its name is the table prefix. Writes tables/<name>_accessible_<base>_k<k>.tsv, tables/<name>_inaccessible_<base>_k<k>.tsv and a combined tables/<name>_<base>_k<k>_probs.tsv
-k / --context-sizes	3 4 5 6	Context size(s) to compute (bases on each side: 3=7mer, 6=13mer). Single value or list. Default: 3 4 5 6 (default: [3, 4, 5, 6])
--mode	pacbio-fiber	Analysis mode: pacbio-fiber (PacBio), nanopore-fiber (Nanopore), daf (DAF-seq) (default: pacbio-fiber) Choices: pacbio-fiber, nanopore-fiber, daf, gpc, cpg.
-n / --max-reads	100000	Maximum reads to process per sample type (0 = all) (default: 100000)
-s / --seed	42	Random seed for sampling (default: 42)
-q / --min-mapq	20	Minimum mapping quality (default: 20)
-p / --prob-threshold	128	Minimum ML probability for modification call (0-255) (default: 128)
--min-read-length	1000	Minimum aligned read length (default: 1000)
-e / --edge-trim	10	Bases to exclude at read edges (default: 10)
--save-interval	10000	Save intermediate results every N reads (default: 10000)
--stats	off	Generate summary statistics and QC plots
--verbose / -v	off	Show detailed filter statistics per BAM file

fiberhmm-train

Flag	Default	Description
-i / --input	—	Input BAM file(s) for training (not required with --base-model)
-p / --probs	required	Accessible and inaccessible probability files (.tsv or .probs.pkl)
--base-model	—	Use transitions from existing model with new emissions (skip training)
-o / --outdir	required	Output directory
--mode	pacbio-fiber	Analysis mode: pacbio-fiber (PacBio), nanopore-fiber (Nanopore), daf (DAF-seq) (default: pacbio-fiber) Choices: pacbio-fiber, nanopore-fiber, daf, gpc, cpg.
-k / --context-size	3	Context size (bases on each side): 3=7mer, 5=11mer, etc. (default: 3)
-c / --iterations	10	Training iterations (random initializations) (default: 10)
-r / --read-count	500	Total reads to sample for training (default: 500)
-s / --seed	42	Random seed (default: 42)
-e / --edge-trim	10	Edge masking (default: 10)
-q / --min-mapq	20	Min mapping quality (default: 20)
--prob-threshold	125	Min ML probability (default matches ft-extract) (default: 125)
--min-read-length	1000	Min aligned length (default: 1000)
-a / --prob-adjust	1.0	Accessible probability adjustment factor (default: 1.0)
--use-hmmlearn	off	Use hmmlearn instead of native implementation (for legacy compatibility)
--stats	off	Generate training statistics and example plots
--n-examples	5	Number of example reads to plot (with --stats) (default: 5)
--daf-mask-runs	—	Thin DAF targets in same-strand runs of >= N original C (CT) or G (GA) bases (N=2: CC/GG and longer). Default: 2 with keep-one for DddA (duplex-validated), off otherwise; 0 disables.
--daf-run-policy	keep-one	With --daf-mask-runs: keep each run's 5'-most target (default) or drop the run. (default: keep-one) Choices: keep-one, drop.

fiberhmm-utils

fiberhmm-utils convert

Flag	Default	Description
input	required	Input model file (.pickle, .pkl, or .npz)
output	required	Output JSON file

fiberhmm-utils inspect

Flag	Default	Description
model	required	Model file to inspect (.json, .npz, .pickle)
--full	off	Print full emission probability table

fiberhmm-utils transfer

Flag	Default	Description
--target / -t	required	Target BAM file (e.g., DAF-seq)
-o / --output	required	Output directory
--mode	daf	Analysis mode for target data (default: daf) Choices: pacbio-fiber, nanopore-fiber, daf, gpc, cpg.
--reference-bam / -rb	—	Reference BAM with footprint tags (ns/nl)
--accessibility-priors / -ap	—	Pre-computed P(accessible|context) TSV
-k / --context-sizes	3 4 5 6	Context size(s) (default: [3, 4, 5, 6])
-n / --max-reads	100000	Max reads to process (0 = all) (default: 100000)
-q / --min-mapq	20	Min mapping quality (default: 20)
-p / --prob-threshold	128	Min ML probability for modification call (default: 128)
--min-read-length	1000	Min aligned read length (default: 1000)
-e / --edge-trim	10	Bases to exclude at read edges (default: 10)
--min-observations	100	Min observations per context for regression (default: 100)
--stats	off	Generate diagnostic plots

fiberhmm-utils adjust

Flag	Default	Description
model	required	Input model file (.json)
--state	required	Which state(s) to adjust Choices: accessible, inaccessible, both.
--scale	required	Multiplier for emission probabilities
-o / --output	required	Output model file (.json)

fiberhmm-utils ma-types

Flag	Default	Description
bam	required	BAM to update in place
--types	—	Logical MA name(s), comma- or space-separated (no strand/quality suffixes)
--scan	off	Exhaustively scan every alignment and discover non-empty MA types
--io-threads	4	BAM compression and index threads (default: 4)

fiberhmm-utils fix-bigbed

Flag	Default	Description
inputs	required	Input bigBed file(s)
--sample-name	—	Explicit sample name to embed (sanitized to a dot/space-free token). Default: derived per file from the filename (stem minus the _<layer> suffix).
--in-place	off	Overwrite the input bigBed(s) in place.
-o / --output	—	Output path (single input only). Default: write <name>.fixed.bb alongside each input.

NFR detection and diagnostics

NFR raw detection uses min_fraction directly (default 0.10), with at least 10 open reads, 20 callable reads and 60 bp of region width. Background, fold and q remain descriptive. Explicit enrichment requests keep the prior 0.05 share default and fold/q gates; detect_threshold alone selects fixed.

clip_gaps_to_region defaults to false. When true, a gap belongs only to its largest-overlap region (ties: leftmost), with its original edges and width. The same assignment applies to discovery, quantification and frozen-catalogue transfer (catalogue.assignment_regions).

The run's detection.opportunity_diagnostic summarizes complete long gaps outside NFRs versus overlapping NFRs, using native unit.positions (marked and unmarked opportunities). It reports counts, median opportunities per 100 bp and the share with fewer than 10 opportunities per gap. Partial gaps and missing opportunity arrays are excluded and counted separately; empty groups have null statistics. Native masks may exclude otherwise markable bases. This diagnostic does not show that gaps occurred by chance and never gates detection. enrichment_test means the test was computed; enrichment_gates means its fold/q criteria controlled detection.