[iskandr]

Part of openvax/varcode#722. Bumps to 0.52.0.

Why

Most T2T VCFs come from the T2T Consortium's chm13v2.0_maskedY_rCRS analysis set, which its README recommends for short-read variant calling. That set's chrM is the rCRS (NC_012920.1), not CHM13's own mitochondrion (CP068254.1), and Ensembl's rapid-release T2T annotation has no mitochondrial genes. So chrM variants from those files could not be annotated.

The rCRS is GRCh38's MT (J01415.2 = NC_012920.1 in GRCh38's assembly report), so Ensembl GRCh38's MT annotation describes that chrM exactly.

Change

t2t-chm13 now includes Ensembl GRCh38 release 115's MT annotation: 37 genes, of which 13 are protein coding, 22 are tRNAs and 2 are rRNAs.

Contig name. The genes are on a contig named NC_012920.1. It can't be confused with MT, Ensembl's name for CHM13's own mitochondrion. CHM13's own begins at what the rCRS numbers 577, and both are 16,569 bp.

Files. The three files (GTF, cDNA+ncRNA and protein FASTAs, 12 KB in all) are packaged in orfdata/data/genomes/. They are pinned by SHA-256 under a new mitochondria entry of the pin. PinnedGenome.mitochondria is generic, so another assembly's alternative mitochondrion would be a registry entry.

Builder. tools/build_rcrs_mitochondrion.py rebuilds the files from the five Ensembl 115 source files, each pinned by SHA-256 and checked against Ensembl's CHECKSUMS. It checks:

• that Ensembl's MT DNA is NC_012920.1's sequence;
• that every transcript's exons spell its cDNA or ncRNA record;
• that every CDS, translated with table 2, gives its protein.

Opening the genome. The GTF that PyEnsembl indexes is the downloaded Ensembl GTF with the MT GTF's gzip stream appended; gzip readers read the two as one file. The packaged FASTAs are copied beside the index, so the genome still opens and unpickles after ORFdata is upgraded. NC_012920.1 is declared mitochondrial in reference_metadata, so its transcripts translate with table 2.

Contig table. contig_aliases("chm13v2.0_maskedY_rCRS") maps chrM to NC_012920.1 (GenBank J01415.2). Files with CHM13's own mitochondrion keep chrM → MT, which stays unannotated.

Reinstalling. Downloaded files are reused because the store digest is unchanged. The index digest changes, so the next load_genome rebuilds the index once (about 1.1 GB). The CHANGELOG says the old index can be deleted.

Validated on the real genome

I installed t2t-chm13 from Ensembl into a scratch cache: 3m37s, 236,688 transcripts.

• The real genome has 37 genes on NC_012920.1, 13 of them protein coding, and no MT or chrM contig.
• MT-ND1 has translation_table == 2, compartment == "mitochondrial", and the expected protein. The table survives pickling.
• 22:21798486 still lies in RIMBP3B and RIMBP3C.
• Proteome.build(genome) gives 104,970 reference proteins, the documented 104,957 plus the 13 MT proteins; the docs are updated.

Tests

tests/test_genomes.py:

• The packaged files against NCBI's NC_012920.1 (vendored in tests/data/t2t/), offline: all 37 transcripts spell their cDNA, all 13 CDSs translate to their proteins with table 2 (and differently with table 1), and the gene count and names are as expected.
• The _rCRS names' contig table maps chrM to NC_012920.1; no CHM13-own name lists it.
• A slice genome carrying the packaged mitochondrion puts MT-ND1 on NC_012920.1 with table 2 after install, load and pickling. The FASTAs are copied beside the index, the merged GTF's size is checked, and force=True rebuilds.
• Adding a mitochondrion to an installed pin opens offline with a new index.

Verification

• ./lint.sh: passes.
• pytest tests -n 8: 1531 passed, 6 skipped.
• mkdocs build --strict: passes.

https://claude.ai/code/session_01JHJhwfx4um52Nczrv3G814