2.8.0

No new reference data required: GTDB-Tk v2.8.0 uses the same R232 reference package as v2.7.x. The minor version bump reflects changes to classify_wf results (see below).

Major Changes:

• classify and classify_wf now apply the same species-assignment rule: the closest reference with AF ≥ --min_af is taken, and only that reference is checked against its ANI radius. Previously, classify_wf could assign a genome to a more distant species if the genome fell outside the radius of its closest reference. This only affects fuzzy species in condensed genera, whose ANI radius is above the default 95% (4 of 10,000 GlobDB genomes in our tests). These genomes are now placed in the reference tree instead.
• classify and classify_wf now report the same other_related_references: only references with AF ≥ --min_af, closest first, capped at 50. The assigned reference is no longer repeated in the list.
• New output file classify/ani_screen/<prefix>.ani_screen_unassigned_hits.tsv.gz: skani hits for genomes not assigned by the ANI screen.
• gtdbtk.json has a new unassigned_hits_file field in the ANI screen step. Resuming classify_wf in an output directory created by GTDB-Tk < 2.8.0 reruns the ANI screen. Older versions cannot resume from a 2.8.0 output directory.

Bug Fixes:

• (#717) Restored other_related_references and warnings in classify_wf for genomes not assigned to a species because their ANI falls below the closest representative's circumscription radius.
• AF values are now rounded identically in classify and classify_wf (floating-point noise and double rounding could differ by 0.001). AF in ani_summary.tsv is written at full precision.
• classify --genes no longer runs the skani ANI screen.
• Genomes that fail classification are now reported in the log instead of being silently dropped.
• Split mode: fixed the RED comparison used to choose the closest taxonomic rank between the backbone and class-level trees.
• Split mode: a malformed reference tree now raises an explicit error instead of printing debug output.