Pre-trained models
Models ship inside the package; --enzyme (and --seq for Hia5) selects
them, and -m is only for custom models.
| Chemistry | --enzyme | --seq | Model | Mode |
|---|---|---|---|---|
| Hia5 Fiber-seq, PacBio | hia5 | pacbio | hia5_pacbio.json | pacbio-fiber |
| Hia5 Fiber-seq, Nanopore | hia5 | nanopore | hia5_nanopore.json | nanopore-fiber |
| DddB DAF-seq | dddb | any | dddb_nanopore.json | daf |
| DddA DAF-seq | ddda | any | ddda_nuc.json (nucleosome HMM), ddda_TF.json (TF recall), ddda_nuc_profile.json (rotational profile) and ddda_nuc_refine.json (internal radial-nucleosome likelihoods) | daf |
fiberhmm-call --enzyme ddda uses all four DddA files in one pass. Each
preset has its own defaults: for example Nanopore Hia5 reads m6A at ML ≥ 248,
and DddA uses phase-aware radial nucleosome recall and CpG-aware calling
(CpGs masked in the HMM and in recall outside unmethylated islands).
See Chemistries and platforms.
Older tables, including the 2.x Nanopore Hia5 and DddB tables
(hia5_nanopore_gt_swapped_legacy.json, dddb_nanopore_gt_swapped_legacy.json), are kept in
fiberhmm/models/legacy/ to reproduce earlier results.
fiberhmm-utils inspect model.json # mode, context size, transitions, emission summary
Train your own model for another enzyme or with matched controls for your condition: see Training a model and Bundled models.