Database profiles

Profile	Reference content	Additional content	Search indexes
curated	SILVA 138.2 NR99 and PR2 5.1.1	None	Separate BLAST v5 indexes for 16S rRNA genes and 18S rRNA genes.
img	SILVA 138.2 NR99 and PR2 5.1.1	IMG sequences from the eukcensus 16S rRNA gene and 18S rRNA gene collections	Separate BLAST v5 indexes for 16S rRNA genes and 18S rRNA genes.

RF00177 routes to the 16S rRNA gene index. RF01960 routes to the 18S rRNA gene index. The mapping is stored in config/model_markers.json.

Database packages use a version independent of the SSUextract application. The version is stored in each profile's manifest.json and printed before a pipeline run.

Installation and update checks

pixi run setup lists each profile with the version and compressed download size from config/database_catalog.json. The selected path and profile are stored as database_path and database_profile in config/local.config.

The installer accepts a Zenodo release only when all of these values agree:

• the configured concept record and the latest release record;
• the exact four-file release inventory and Zenodo MD5 values;
• the release manifest, archive sizes, and archive SHA-256 values;
• the two entries in SHA256SUMS.

Archive extraction takes place in a staging directory. The profile becomes the installed profile after its manifest files and BLAST indexes validate. A failed replacement restores the installed profile.

Archive downloads use a 60-second network read timeout and stop after five consecutive transient failures. A partial archive is keyed by profile, database version, archive SHA-256 digest, and host. A later setup run resumes it only when the server returns a matching HTTP byte range. A rejected range or failed checksum permits one clean restart for that condition. The installer checks the final byte count and SHA-256 digest before extraction. It removes the verified download cache after using the archive.

Interactive terminals redraw a width-bounded progress line with written bytes, transfer rate, and estimated time remaining. Non-interactive logs include the start, completion, and intermediate progress every 10 seconds.

pixi run ssuextract and pixi run example check Zenodo before starting Nextflow. The check has a five-second total timeout. Interactive runs offer to install an available release. Non-interactive runs and runs with an explicit --database_path report the update and continue with the installed profile. A valid installed profile remains usable when the check times out, Zenodo is unavailable, or the remote release contract fails validation.

Runtime files

Path	Contents
manifest.json	Profile identity, artifact paths, sizes, and SHA-256 digests.
provenance.json	Source versions, build contracts, software versions, and source-tree fingerprints.
blast/16S.*	BLAST v5 nucleotide index for 16S rRNA genes.
blast/18S.*	BLAST v5 nucleotide index for 18S rRNA genes.
tables/sequences.parquet	Content-addressed sequence identifiers, lengths, hashes, and marker membership.
tables/preferred_taxonomy.parquet	Selected sequence taxonomy, centroid names and taxonomy, and explicit cross-domain conflict state.
tables/source_records.parquet	Normalized source record provenance.
tables/taxonomy_assignments.parquet	Native and derived taxonomy evidence.
tables/img_location.parquet	IMG taxon identifier and valid latitude/longitude values.

The optional manifest key runtime_calibration names a JSON file in the profile. That file must also appear in the artifact list with its size and SHA-256 hash. Runtime calibration uses schema 3, calibration_use_case=runtime_query, classification_policy=runtime_lca_v1, and stratum_basis=predicted_candidate_source_set_and_lca_domain. The root status must be calibrated when rank caps exist and failed when none exist. The positive integer max_targets must match the run's --max_blast_targets. Its reference digest covers the BLAST files and preferred-taxonomy table. Rank caps or identity rules apply to the predicted marker, source set, and domain.

The schema-3 img_centroid calibration has a separate purpose: it limits labels assigned during IMG database construction. It cannot substitute for runtime query calibration. Released v1.0.2 profiles lack runtime calibration; nonexact query calls remain unclassified while reference evidence stays available.

Centroid calibration can supply optional rank_rules for each predicted source and domain. Rules specify consecutive ranks from domain to the rank cap and nondecreasing minimum candidate identities. Every source route must pass. The classifier uses the lowest cap and highest required identity across those routes. Applying corrected centroid calibration to stored IMG labels requires a new profile build.

Raw source FASTA files, source project descriptions, contacts, email addresses, comments, and cluster tables are not distributed in a runtime profile. The IMG profile retains only the centroid names required to interpret cluster-derived taxonomy.

Curated profile counts

The curated profile contains 609,298 unique exact sequences from 683,597 source records. The 16S rRNA gene index contains 416,021 sequences; the 18S rRNA gene index contains 193,282. Exact sequences assigned across domains remain marked as conflicts.