LAST-TRAIN(1)

💖 last-train(1) 💖

okay bestie sit DOWN because this one is a User Commands page and it's literally the glow-up tool for your alignments ✨

---

🎀 NAME 🎀

last-train — she tries to find suitable score parameters for aligning the given sequences. that's the whole job. that's her assignment and she ate 💅

---

🌸 SYNOPSIS 🌸

last-train [options] lastdb-name sequence-file(s)

---

🦋 DESCRIPTION 🦋

Try to find suitable score parameters for aligning the given sequences. no cap, that's it — she takes your lastdb-name and your sequence-file(s) and goes hunting for the score parameters that actually fit your data, so your alignments don't look like they were picked by a random number generator 🙈

---

✨ OPTIONS ✨

• -h, --help — show this help message and exit. the classic escape hatch, iconic behavior
• -v, --verbose — show more details of intermediate steps. she'll narrate the whole thing for you, main character commentary

💜 Training options:

• --revsym — force reverse-complement symmetry
• --matsym — force symmetric substitution matrix
• --gapsym — force insertion/deletion symmetry
• --fixmat=NAME — fix the substitution probabilities to e.g. BLOSUM62. like, lock it in, no vibes-based changes
• --pid=PID — skip alignments with > PID% identity (default: 100). too-similar alignments can go, we don't need them
• --postmask=NUMBER — skip mostly-lowercase alignments (default=1)
• --sample-number=N — number of random sequence samples (default: 20000 if --codon else 500). yes the default literally switches depending on whether --codon is on, keep up bestie 😭
• --sample-length=L — length of each sample (default: 2000)
• --scale=S — output scores in units of 1/S bits
• --codon — DNA queries & protein reference, with frameshifts. the fancy one 🧚

🍓 Initial parameter options:

• -r SCORE — match score
• -q COST — mismatch cost
• -p NAME — match/mismatch score matrix
• -a COST — gap existence cost
• -b COST — gap extension cost
• -A COST — insertion existence cost
• -B COST — insertion extension cost
• -F LIST — frameshift probabilities: del-1,del-2,ins+1,ins+2 (default: 1-b,1-b,1-B,1-B). note how those defaults are literally built out of your -b and -B values, so cute, so dependent 🫶

🌈 Alignment options:

• -D LENGTH — query letters per random alignment (default: total sample length)
• -E EG2 — maximum expected alignments per square giga
• -s STRAND — 0=reverse, 1=forward, 2=both (default: 2 if DNA and not lastdb -S2, else 1). another shapeshifter default, iykyk
• -S NUMBER — use score matrix: 0=as-is, 1=on query forward strands (default: 1)
• -C COUNT — omit gapless alignments in COUNT others with > scoreper-length
• -T NUMBER — type of alignment: 0=local, 1=overlap (default: 0)
• -R DIGITS — lowercase & simple-sequence options
• -m COUNT — maximum initial matches per query position (default: 10)
• -k STEP — use initial matches starting at every STEP-th position in each query (default: 1)
• -P THREADS — number of parallel threads
• -X NUMBER — N/X is ambiguous in: 0=neither sequence, 1=reference, 2=query, 3=both (default=0)
• -Q NAME — input format: fastx, sanger (default=fasta)

heads up bestie: several of these defaults are not fixed numbers — they actively change based on other flags you pass. read them before you assume, fr fr 👑

---

okay that's the whole page, and that's the vibe check passed 💖 she's last-train version 1654, dated August 2026, and yeah the title at the top said LAST-TRAIN(1) but we already covered that, periodt ✨💅🦋