v2.10.4: auto-skip --deam on fiber-seq BAMs

Bug fix

fiberhmm-extract --all (default mode) now auto-skips --deam when the input BAM looks like fiber-seq.

What was happening

On vg-surjected fiber-seq CRAMs (pangenome-graph aligned, projected onto linear reference):

1. --all mode ran every track including --deam — even on fiber-seq data that has no deaminations (Hia5 is a methyltransferase).
2. --deam path-3 (MD walker) fell through on every read because paths 1 (MM/ML u code) and 2 (R/Y) had nothing to match.
3. Surjected BAMs frequently have MDs that disagree with CIGAR on ~97% of reads (documented upstream issue in vg surject).
4. pysam's get_aligned_pairs(with_seq=True) on bad MD raises AssertionError AND corrupts malloc state before propagating.
5. A later allocation in the worker segfaults. Python's finally block doesn't run on segfault, so buffered m6a/m5c BED writes are lost.
6. Result: --deam is empty, and m6a/m5c files are also empty or missing entirely.

Fix

Auto-skip --deam on BAMs where:

• MM/ML carries m6A (A+a / T-a) or 5mC (C+m) subtypes, AND
• no dU code (+u / +55797), AND
• no R/Y IUPAC in the query sequence

When triggered, prints:

Auto-skipping --deam: BAM looks like fiber-seq (MM/ML has m6A/5mC subtypes,
no dU code, no R/Y). Hia5 is a methyltransferase, not a deaminase --
no deamination calls to extract. Pass --deam explicitly to override.

Explicit --deam is still honored — the auto-skip only fires in --all / default mode.

Verified

Tested on a 5 Mb chr1 slice of a real surjected fiber-seq CRAM: 29,679 reads → 22.8M m6a + 4.3M m5c + 1.9M footprint + 1.9M msp positions, zero crashes, all bigBeds produced. Pre-fix this would have segfaulted mid-region and lost m6a/m5c writes.

Tests

294 pass. 8 new cover the fiber-seq signature detection across PacBio modern MM/ML, SAM 1.7 suffix variants (A+a., C+m?), DAF IUPAC, modkit dU, ChEBI 55797, and mixed cases.

Install

pip install --upgrade fiberhmm