Auto-Annotate

1. 01Six sourced feature corpora in one scan, each hit citing the record it came from: PlasMapper 3 plasmid parts, NCBI AMRFinderPlus resistance genes, FPbase fluorescent proteins, Swiss-Prot plasmid proteins (matched through translation), curated GenBank plasmid elements, and every family of Rfam 15.1 (structured RNA, covariance models, clans reported). The legacy entries that cite nothing say so
2. 02Two matching methods, reported per hit: exact, which finds only a sequence spelled byte for byte, and aligned, which finds a diverged variant and reports the identity and coverage that earned it
3. 03Matching on both strands and across the origin of a circular sequence
4. 04An acceptance policy taken from the corpus authority: at least 95% identity, over a coverage bar that depends on how long the feature is
5. 05Cre/loxP site detection
6. 06Gateway att site detection, with a greedy core-matching option for divergent sites
7. 07Per-source hit counts and a preview of every hit before it is applied
8. 08Hits applied as real annotations on the document, or dismissed
9. 09A detector that fails is reported by name, and the other sources still run

Recognition sites are detected by consensus. Library features are found either by exact match or by a BLASTN-class local alignment scored the way PlasMapper and pLannotate score theirs, checked against NCBI BLAST+ blastn run on the same sequences with the same scoring: over NCBI's own 147-record vector-screening corpus, 711 of the 720 alignments it reports are reproduced byte for byte on score, alignment length and identity count; over pBR322 against the whole 471-entry feature corpus, every one of blastn's 28 alignments is found at blastn's own coordinates. The panel runs on @ports/annotate.

Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.

How complete is the feature library?

Three corpora ship, and each hit names which one it came from and under what terms. PlasMapper 3 contributes 471 entries and the NCBI AMRFinderPlus resistance-gene set 9,786; a further 19 legacy entries have no recorded source, and are labelled that way rather than blended in, because a hit that cites a record and a hit that cites nothing are different claims.

Does it find variants of a feature?

Yes, and it tells you it did. Exact matching finds only a feature the sequence spells byte for byte, which misses a tag carrying different codon usage. Aligned matching finds the diverged copy and reports the percent identity and the coverage of the database entry that earned it, so a partial or imperfect hit is visible as one rather than presented as a clean call.

Why does the longest hit not always win?

When two library entries match the same stretch, only one can be annotated there, and the usual ranking prefers the hit covering more bases. On a real pBR322 that prefers a 906-base match to another plasmid origin at 71.9% identity over the 620-base match to its own origin at 99.7%. Switching on the corpus authority acceptance policy resolves the overlap among hits that clear 95% identity first, so the record that actually matches the sequence is the one that survives.

Does it overwrite my existing annotations?

No. Hits are previewed and applied as new annotations, and you choose which to keep.