Databases

gmeta resolves every database under one root. The default is location-aware: a repo checkout uses <repo>/data/db (gitignored), an installed package uses the XDG data home (~/.local/share/gmeta/db) — databases are tens-of-GB external downloads and never live inside the package. Override with GMETA_DB_DIR or -D:

Tool	Expected under $DB_DIR/	Required files
kraken2 + bracken	kraken2/	*.k2d + databaseXXXmers.kmer_distrib
centrifuge	centrifuge/	<prefix>.1.cf.index*
kaiju	kaiju/	nodes.dmp, names.dmp, *.fmi
metaphlan	metaphlan/	mpa_v*/ or *.bt2l*
kmcp	kmcp/	*.uniki + taxid.map (+ taxdump/ with nodes.dmp)
sylph	sylph/	*.syldb / *.syl2db
host removal	host/	bowtie2 index prefix (e.g. grch38.1.bt2l)

Check what resolves: gmeta db list — a Rich table over all eight targets (ready/missing + the exact missing file); gmeta db list -f json is the stable machine contract. Every probe calls the same resolver the executor uses, so "ready" means the profiler would accept the path — including companion files like bracken's kmer_distrib or kaiju's names.dmp.

Fetch scripts

Databases are tens of gigabytes — gmeta db fetch never downloads. It writes commented, resumable bash scripts for you to review and run:

gmeta db fetch                    # scripts for every target → ./db-fetch/
gmeta db fetch kraken2 kmcp -O scripts/ --db-root /data/db

bash scripts/fetch-kraken2.sh     # wget -c → tar → verify layout
bash scripts/run_all.sh           # or run everything in order

Each script installs into $GMETA_DB_DIR (or the --db-root given at generation time) and verifies the layout gmeta's resolvers expect.

kmcp is special: its prebuilt databases ship on OneDrive/CowTransfer (no stable wget URL). fetch-kmcp.sh downloads the NCBI taxdump automatically, then exits with instructions for the manual database copy — re-run it after the manual step to finish verification.

URLs move with releases; every URL sits in a variable at the top of its script with upstream links in comments.

Reference genomes for 种基因组覆盖度

Coverage needs per-species reference FASTAs plus a manifest CSV:

accession,species_taxid,organism_taxid,scientific_name,file_path
GCF_000025985.1,817,272559,Bacteroides fragilis NCTC 9343,refs/GCF_000025985.1.fna

Relative file_path resolves against the manifest's directory. Without a manifest, --ref-dir is searched for {accession}*.fna / {taxid}*.fna.

Chinese name tables

src/gmeta/data/ ships the surveillance name tables (species_chinese_names.tsv, genus_chinese_names.tsv, taxid_chinese_mapping.csv); unmapped taxa fall back to their latin names. The tables ship inside the package (small, versioned with gmeta). Extend them by editing src/gmeta/data/*.tsv|csv and reinstalling; unmapped taxa fall back to their latin names.