FiberHMM 3.0.1

Patch release. Outputs from FiberHMM 3.0 calls are unchanged; the fixes affect reverse-strand reads in BAMs called with FiberHMM 2.12 or earlier, or whose header lost its provenance.

Fixed

• fiberhmm-extract reads footprint tags in the frame the BAM records,
  by the same rule as recall, QC and consensus. Nucleosome, MSP and TF
  tracks from BAMs called with FiberHMM 2.12 or earlier were mirrored on
  reverse-strand reads; extract those BAMs again. Tracks from BAMs called
  with 3.0 are unchanged. New --input-frame auto|molecular|query sets the
  frame of legacy ns/nl/as/al; a merged BAM whose histories
  disagree now stops nucleosome and MSP extraction and asks for it.
• fiberhmm-extract --both-strand reads deam+/deam- groups by the same
  rule. On reverse-strand reads of BAMs without coord=molecular the
  both-strand region was mirrored; fiberhmm-pair output (forward
  consensus reads) is unchanged.
• The other readers of MA groups follow the same per-group frame rule.
  On reverse-strand reads of BAMs called with FiberHMM 2.12 or earlier (or
  whose header lost its @PG lines): fiberhmm-tag-m5c/-call-m5c read
  nucleosome and MSP footprints mirrored; consensus loading placed a .
  group wrongly when the same feature also had a +/- group; consensus
  export moved fibertools Ma footprints on headers without frame
  evidence; fiberhmm-pair placed nucleosome dyads mirrored; and
  fiberhmm-utils transfer --reference-bam read SEQ-frame ns/nl as
  molecular. Re-run those steps on such BAMs; BAMs called with 3.0 are
  unchanged. The same applies to fiberhmm-footprint-model input loading,
  fiberhmm-strand-rescue, -strand-rescue-annotate, -strand-rescue-audit,
  fiberhmm-tag-consensus and the native TF quality that consensus export
  copies onto a family.
• Writers keep the annotations they carry through at their original
  positions. fiberhmm-apply (--no-msps MSPs), -call
  (--no-legacy-tags) and -recall-tfs/-nucs (preserved MA groups such
  as tf_sr and deam, kept legacy tags, uncalled reads) declare molecular
  output and now convert what they carry from the input's frame; on reverse
  reads of FiberHMM 2.12 or @PG-stripped BAMs these were mirrored. On
  3.0 output their results are unchanged.
• fiberhmm-tag-m5c and fiberhmm-call-m5c --tag-mode locus on fibertools
  output keep its native Ma footprints: the MA they create (which wins
  over Ma) now carries them.
• Every command that reads or carries footprint annotations prints one
  warning per input file whose header has no FiberHMM/fibertools provenance
  (stripped @PG, no coord=molecular marker): its reverse-read coordinates
  cannot be confirmed. Results are unchanged.
• The QC that fiberhmm-call runs after calling follows --no-use-m5c and
  --cpg-mask-policy for the sampled reads it calls itself.
• Writing apply tags without footprints keeps the hl:i:1 marker on reads
  whose harmonized ns/nl stay in place (Python API; fiberhmm-apply
  output is unchanged).
• fiberhmm-transfer --parameters naming a parameter group the frozen
  catalog fixes exits 2 with one error line instead of a traceback.
• fiberhmm-call-m5c path conflicts exit 2 with a
  fiberhmm-call-m5c: error: line, as fiberhmm-tag-m5c does.
• fiberhmm-check describes the 2.x Nanopore Hia5 and pre-3.0 DddB tables
  as earlier emission tables. Its findings are unchanged.

Supported inputs

• Standalone pass-through tools (fiberhmm-dedup, -pair, -merge,
  -tag-m5c, -call-m5c, -tag-consensus, -strand-rescue-annotate) are
  supported on FiberHMM 3.0 output. On pre-3.0 or header-stripped BAMs,
  re-call with FiberHMM 3.0 first.

Documentation

• --input-frame help (recall-tfs/-nucs, tag-m5c, call-m5c --tag-input-frame) states the full frame rule.
• fiberhmm-probs no longer names the combined *_probs.tsv as
  fiberhmm-train input; -p takes the accessible and inaccessible tables.
• The fiberhmm-recall-tfs description no longer mentions --phase-nrl.
• CI tests Python 3.14.

pip install fiberhmm==3.0.1