Drop all-gap alignment columns per trajectory

[rchikhi]

Summary

• For forwards trajectories (trajectory.py): drop columns where every node on the root-to-tip path has - or N.
• For pairwise trajectories (pairwise_trajectory.py): drop columns where both tips have - or N.

Why

With diverse-phylum alignments (bac120 cyano with 2470 species), MAFFT creates an alignment column for every lineage-specific insertion. For any given tip/pair, columns from other lineages' insertions are - throughout the trajectory and carry zero signal.

Example: a cyano TIGR00810 trajectory pre-fix is 841 chars × 12 frames, with 72% gaps per frame. The first ~180 characters of every frame are identical pure-gap blocks that only exist because some other cyano lineage has an insertion there.

Post-fix, the same trajectory is 237 chars × 12 frames, 0% gaps, and every mutation is still visible.

What's preserved

• Any column where at least one node on the path has a real base is kept, so lineage-specific insertion/deletion events are preserved (a deletion base→- on some branch stays visible because the pre-deletion ancestors have a base).
• Hamming distances in headers are unchanged (trimmed columns contribute 0 to Hamming by construction).

Impact

Dataset	Before	After
cyano TIGR00810 trajectory (typical tip)	841 chars, 72% gap	237 chars, 0% gap
spike-xs trajectory	2055 chars, ~0% gap	2055 chars, ~0% gap (unchanged)

Viral datasets (spike, RdRp, cytb, flu, n450) are unaffected — their alignments have almost no insertion columns so there are no dead-gap columns to drop.

Test plan

• Regenerate cyano trajectories from the clean auspice JSONs and verify expected length reduction
• Regenerate bacteroidota trajectories and verify expected length reduction
• Spot-check that per-frame mutations remain the same (only positions shift due to column removal)
• Confirm trajectories for viral datasets remain essentially unchanged

🤖 Generated with Claude Code