Identification

Sequence Identification finds, for each selected sequence, the reference sequences it most closely matches in the platform's own reference database. It is a similarity search: a ranked list of hits per query with a short label summarising how clear the top of that list is. It does not assign a taxonomic rank or compute a confidence score, and it is a different system from the public service described in Identification engine; the two share no code, reference library or decision logic.

Inputs

Every record that survives the shared filters becomes a query, whatever its marker. The query ID is the record_id and the sequence is the stored nucleotide sequence. The length filter is applied; the marker you choose is not (see the parameters table).

The reference database is a BLAST database built from a daily export of BOLD Forensics' records: every marker, every sequence of at least 100 bases. An administrator builds it on demand, not on a schedule. Each entry carries the record's process ID, marker, BIN URI, taxonomy ID and public status; a hit's taxonomy is looked up by that ID. The job does not record which build it searched.

Parameters

Parameter	Form offers	Job applies
ID Databases	One choice, "COI Full Database (500 bp)"	Not applied. The single reference database holds every marker at 100 bases or more.
Marker	List of markers	Validated only (COI-5P, ITS, rbcL or matK). It restricts neither queries nor references.
Sensitivity	Low, Medium, High (default Medium)	Sets the minimum overlap of a hit: Low 250 bp, Medium 150 bp, High 100 bp.
Minimum Required Hits per Query	1–100 (default 10)	A query stops searching once it has this many accepted hits. It does not flag anything in the output.
Maximum Retrieved Hits per Query	10–1000 (default 100)	Caps accepted hits per query and TSV rows per query.
Minimum Identity	80–98% (default 95%)	Raises each tier's identity floor to this value where the floor is lower, and decides whether tier 2 runs.
Query Sequence Length Filter	100–4000 bp (default 400–1000)	Applied. Validation accepts only 100–2000 bp; a bound above 2000 fails the job at stage 1.
Exclude Records with Stop Codons	Checkbox	Read, but the filter compares the stored true/false flag as text and the comparison does not match. Leave it unticked.
Exclude Records Flagged as Misidentifications or Errors	Checkbox	Not applied.

Method

The search uses NCBI BLAST+ blastn in up to two tiers.

Tier	Word size	Identity floor	Runs for
1	48	96%	Every query
2	12	80%	Queries still short of Minimum Required Hits after tier 1

Your Minimum Identity interacts with the floors in two ways. A tier applies the higher of its floor and your value, so at the default of 95% tier 1 accepts hits at 96% and above and tier 2 at 95% and above, not 80%. The loop also stops after the first tier whose floor is at or below your value: at 96–98% only tier 1 runs, raised to your value; only at exactly 80% does tier 2 apply its own floor.

Before the search, each query is trimmed of leading and trailing ambiguity codes and gaps, and any internal ambiguity code (R, Y, K, M, S, W, B, D, H, V, N) or gap is deleted, not masked; the count removed is subtracted from BLAST's gap count. blastn runs with the tier's word size, at most 1000 target sequences, and BLAST's default task and e-value.

A hit is accepted when BLAST's percent identity for the local alignment reaches the tier threshold and the overlap (alignment length minus adjusted gaps) reaches the Sensitivity minimum. Query coverage is recorded but not used. Only the first alignment per reference sequence is kept. A p-distance per hit is 100 × mismatches ÷ overlap.

Hits are ranked by percent identity, then overlap, then bit score. The "top taxon" of a query is the lowest-rank name available for the top hit, from species up to phylum.

Outputs

• per_query_results.tsv: one row per accepted hit, at most Maximum Retrieved Hits per query: the matched record's process ID, marker, BIN URI, taxonomy ID and public status, the BLAST alignment statistics, p-distance, overlap and taxonomy.
• per_query_results.jsonl: one line per query, including queries with no hits, with the query sequence and all accepted hits.
• blast_result.json: the raw accepted hits.
• filtered_records.jsonl: the records that became queries.

The report has three tables. Run Summary counts queries by outcome and label. Identification Summary has one row per query: hit count, top match, top BIN, top taxon, top identity, the hits and taxa within the window, and a label. Top 10 Hits lists the ten highest-ranked hits per query with identity, p-distance, overlap and taxonomy.

Each query receives exactly one label. "The window" means hits within 2.0 identity points of the best hit.

Label	Meaning
no confident hit	No hit passed the thresholds in any tier that ran.
ambiguous top cluster (N taxa)	Hits within the window carry more than one reference taxonomy ID.
near-exact single taxon match	One taxonomy ID in the window; top identity 99% or higher.
strong single taxon match	One taxonomy ID in the window; top identity 98% or higher, below 99%.
single-taxon match below 98%	One taxonomy ID in the window; top identity below 98%.

"Taxon" means the reference record's own identification at whatever rank. A species-level and a genus-level record of the same genus count as two taxa, so an ambiguous label can arise from references identified to different depths rather than from competing species.

Limitations and interpretation

• No rank assignment, no confidence. The labels describe the shape of the hit list. Deciding what taxon a query belongs to, and at what rank, is left to you.
• Self-matches. The database is built from the same records you can select, and the job does not exclude a query's own record, so a query already in the database is expected to top its own list at or near 100%. This follows from the build; it has not been confirmed in a run.
• Library and marker. The library label does not describe the database, which holds every marker. As the marker choice filters neither side, a query can hit a reference of another marker.
• Identity is local. Percent identity covers the aligned region only; the overlap minimum, not query coverage, keeps out short high-identity fragments.
• Provenance. The result records neither the database build searched nor the BLAST+ version.
• A missing reference database or taxonomy export fails the job at stage 4.

Related

• The methods and their shared pipeline
• Identification engine
• Known limitations