Hmmbuild

# Force nucleotide alphabet when autodetection would be ambiguous
hmmbuild --dna dna_family.hmm dna_alignment.sto

# Save the annotated Stockholm alignment that hmmbuild actually used
hmmbuild -O annotated.sto family.hmm seed_alignment.sto

# Name a single model explicitly
hmmbuild -n KinaseDomain kinase.hmm kinase_alignment.sto

Recommended Workflow

1. Start from a real multiple-sequence alignment, not raw unaligned FASTA.
2. Inspect the alignment for bad fragments, frameshifts, or unrelated sequences before model building.
3. Decide whether consensus columns should be inferred automatically (--fast, default) or taken from reference annotation (--hand).
4. Build the HMM, and optionally capture the annotated Stockholm alignment with -O so you can review weights and chosen consensus columns.
5. Test the resulting model with hmmsearch or hmmstat, then run hmmpress if the model will become part of a scan database.

Guardrails

• The input must be an alignment, not a bag of unaligned sequences.
• Garbage in, garbage out: poor alignment quality produces misleading models.
• On this workstation, the current local binary is presently failing with a missing libopenblas.so.0; use the local man page conservatively until that library issue is fixed.
• msa_file may be - to read from stdin, but hmm_out cannot be - because stdout is used for other textual output.
• -n only applies when building a single alignment; with multi-alignment input, each alignment needs its own annotated name.
• --hand relies on reference annotation already present in the alignment.
• Keep model validation separate from model construction so thresholds can be tuned honestly.