bio-protein-clustering-pangenomelisted

# Bio Protein Clustering Pangenome Cluster proteins into orthogroups and derive pangenome matrices. ## Instructions 1. After clustering, build and validate the complete small-to-large comparison bundle with: ```bash uv run --script skills/bio-protein-clustering-pangenome/scripts/build_pangenome_artifacts.py \ orthogroups.tsv --genomes genomes.tsv --marker-catalog marker_catalog.tsv \ --marker-hits marker_hits.tsv --ncrna ncRNA_census.tsv \ --out results/bio-protein-clustering-pangenome ``` The driver requires globally unique protein IDs, at least two reference genomes for a defensible median, and a fresh output directory. It persists marker and ncRNA censuses alongside copy-number, presence/absence, family-comparison, genome-frontier, and conserved-neighborhood artifacts. `fixtures/` is a runnable three-genome contract test. 2. Cluster proteins. Choose the tool by dataset size and goal: - Default for orthology inference up to a few hundred genomes: **OrthoFinder v3.1.5** (supports MSA-based gene trees; supersedes OrthoFinder v2 and OrthoMCL workflows). - Very large pangenomes where OrthoFinder is too RAM-heavy: **ProteinOrtho v6.3.6**. - Sequence clustering (not strict orthology) and similarity-search backbones: **MMseqs2 v18-8cc5c**. GPU search requires MMseqs2 v16 or newer plus a GPU-enabled build on CUDA Turing-or-newer hardware; full-speed kernels require Ampere or newer. Enable `--gpu` only for commands that expose it and record t