HMM_TRAIN(1)

✨ HMM_TRAIN ✨

okay bestie, pull up a chair, because we're about to talk about hmm_train — that's the User Commands girlie from the HMM family, and she's here to help you estimate transition probabilities. 💅

---

🌸 NAME 🌸

hmm_train — Estimate the transition probabilities of an HMM, based on multiple

-g <gff_fname_list> [OPTIONS] > out.hmm

like, that's the whole vibe. you feed her stuff, she spits out an HMM to stdout. no cap. 🎀

---

🦋 DESCRIPTION 🦋

So what does she actually do? She estimates the transition probabilities of an HMM. And she does it based on:

• multiple alignments
• sequence annotations
• and a category map

that's it, that's the girl. three ingredients, one iconic output. ✨

---

💖 OPTIONS 💖

🌈 required options

these are not optional, bestie. you literally cannot skip these. it's giving "mandatory main character energy."

• -m <msa_fname_list> — the list of multiple sequence alignment files. 🧚 heads up though: currently, in testing mode, this list must be of length one. so like, one file. just the one. don't get cute with it.
• -c <category_map_fname> — the file that defines the mapping of feature types to category numbers. it's the translator, the middleman, the one who knows everybody. 🫶
• -g <gff_fname_list> — your files in GFF, defining the sequence features that get used when labeling sites. the frame of reference of the feature indices is determined feature-by-feature, according to the 'seqname' attribute. and!!! the filenames must correspond in number and order to the elements of <msa_fname_list>. so like, keep them lined up. matching sets. a coordinated outfit. 👑

---

🌟 alignment options

• -M <msa_length_list> — mutually exclusive with -m, so pick a lane. this one assumes alignments of the specified lengths (that's a comma-separated list, bestie) and does not attempt to map the coordinates in the specified GFFs — it just assumes they're already in the desired coordinate frame. the slay here is that it lets an HMM be trained directly from GFFs, without alignments. not permitted with -I though, so don't try it. 🙈
• -i PHYLIP|FASTA|MPM|SS — the alignment format. default is SS. pick your fighter. 💀
• -R <tag> — before estimating those transition probabilities, this groups features by <tag> (think "transcript_id" or "exon_id", iykyk) and then reverse complements the segments of the alignment corresponding to groups on the reverse strand. ⚠️ the groups must be non-overlapping — see refeature --unique. no overlap, no chaos.

---

🎀 indel options

• -I <indel_cat_list> — model indels for the specified categories. okay so listen: for there to be nonzero probability for the states corresponding to a specified category range, the indels must be:
  • "clean" (nonoverlapping) 🧼
  • assignable by parsimony to a single branch in the phylogenetic tree 🌳
  • and have lengths that are exact multiples of the category range size 📏

Avoid -G with this option. and if you're using it in training mode, it requires -T. don't forget, bestie, I'm serious. 😭

• -t <tree_fname> — use the specified tree topology when training for indels. give her a tree, she'll give you the whole forest. 🌲✨
• -n <nseqs> — train an indel model for <nseqs> sequences, even though the training alignment has a different number. bold. iconic. all (non-trivial) gap patterns are assumed to be equally frequent. periodt.

---

💜 other options

• -q — proceed quietly, with no updates to stderr. she's not gonna text you the whole time. 🤫
• -h — print this help message and exit. the classic "actually, tell me the rules" move. 🫶

---

fr fr, that's every flag, every default, every warning. nothing skipped, nothing merged, nothing invented. the vibe is mine, the facts are hers. 💖

---

hmm_train 1.4 · May 2016

and that's the whole page, bestie — version 1.4, dated May 2016, brought to you by the User Commands girlies. go train that HMM, you absolute legend. 💅✨🌈