An Imported Feature Is Unverified Until Coordinates Match

An implementation checklist for validating plasmid annotations is a coordinate-truth walk, not a prettier map. An imported feature is unverified until its location matches the bases it names. Prefer keeping a feature when span, strand, and translation agree. Reject a pretty label on the wrong ORF or leftover backbone.

The INSDC feature table treats a feature as a key plus a location on a sequence. Map feature completeness still owns which objects must be named. Plasmid provenance still owns source, passage, and license. Database entry fields still own intake. This page only asks whether the spans are true. Cores inherit GenBank files that look finished. Import copies the table. Validation reads the table against the sequence. Presence is not truth. A handoff that only asks whether annotations exist can pass a CDS shifted by one base.

An Imported Feature Is Unverified Until Coordinates Match

The search asks how to validate plasmid annotations. The outcome is a map on which every kept feature has a location that can be read back from the sequence. INSDC location syntax exists because a feature is not a floating name. It is a span, a join, or a complement on numbered bases. If the AmpR label sits over a promoter, or a CDS translation does not match the named protein, the annotation is false even if the graphic is handsome.

Do not treat auto-annotation as validation. Auto-annotation is a hypothesis generator. It proposes keys and spans. A person or a documented rule must still accept or reject each proposal against the sequence that will be used at the bench. Do not treat a vendor color scheme as evidence. Color is display. Coordinates are the claim.

Walk the Coordinate Checks

Run the walk on the exact sequence version that will be cloned, prepped, or submitted. Do not validate a related accession and then edit the file.

1. Freeze the sequence identity. Record topology, length, and the file version. If two people open different lengths, every later span is undefined. Checkpoint: a second opener reports the same length and circular or linear state.
2. List candidate features. Take imported keys, auto-annotated keys, and names from the request. Do not delete leftovers yet. Checkpoint: the list includes both intended objects and suspects.
3. For each coding feature, read the stated location against the sequence. Confirm start, end, strand, and that the translation of that span matches the named product. Checkpoint: a one-base shift is a fail, not a style note.
4. For each noncoding feature that matters to the next experiment — origin, marker, promoter, primer binding site — confirm the span covers the motif you intend to use. Checkpoint: a marker name over empty sequence is a fail.
5. Mark leftovers. Features that cannot be confirmed stay unlabeled or marked unverified. Do not hide them by making the map look clean. Checkpoint: a reviewer can see what was rejected.
6. Record the validator, the date, and the sequence version. Checkpoint: a later edit of the same file does not silently inherit the old validation.

If any checkpoint fails, stop. Fix the span or drop the feature. Do not pass the map because the rest of the graphic looks familiar. Open versus proprietary file formats still own whether another tool can open the file. That is not this walk.

A worked fail is a 5.4 kb cloning vector imported from a vendor GenBank file. The map shows a CDS named “insert” from 412–1602. Translation of that exact span yields a truncated peptide because the true start is 409 on the circular file the core actually preps. The graphic still looks complete. Validation fails at the translation checkpoint. The fix is to move the span, not to recolor the arrow. A second fail is a leftover KanR feature from a previous backbone sitting over empty sequence after a marker swap. Completeness would still list a marker. Truth removes or relabels it. Those two fails are why this walk exists beside the feature-name checklist.

Expected Result and Verification

The walk is done when a later reader can open the same sequence version and reconstruct every kept feature from coordinates without asking the annotator. Verification is not a prettier screenshot.

• Every kept CDS translates to the named product on the stated span.
• Every kept origin, marker, and promoter covers the motif the next experiment will use.
• Every leftover is marked unverified or removed with a reason.
• The sequence length and topology used in the walk are written on the record.

Common failures are an imported CDS from a linear view placed on a circular backbone, a marker copied from a related plasmid that used a different cassette, and a primer feature left on the wrong strand. Those failures look small on a zoomed-out map. They are enough to order the wrong oligo. If verification cannot be completed, the map is not validated. It is decorated.

Use the Map After Truth Not as Proof of Truth

Use a map after the spans are true, not as a certificate that they are true. After the walk, a workspace such as ZettaGene can display named features on the sequence you just checked. Official simulation coverage is restriction, Gibson, and homologous alignment. Golden Gate is not a Zetta feature, and a simulation does not replace coordinate validation. ZettaGene is the annotation surface being checked or the place the checked map is held. It is not a validator certificate and not a reason to skip the walk.

If the core still ships maps because they “came from GenBank,” you do not have a validation process. You have an import habit. Import copies. Validation matches. Write the match. Then share the map.

Circular versus linear numbering is a third fail class. A feature copied from a linearized GenBank view can wrap the origin on the circular file the core preps. The label still says ori or AmpR. The span now crosses a join the linear file never had. Validation reads the circular coordinates, not the screenshot. If the origin is a join, write the join. If the tool cannot show a join, write the two intervals by hand on the record. A pretty unbroken arrow across the cut site is not a location. It is a drawing.

Frequently Asked Questions

Is an imported GenBank feature already validated?

No. Import copies a span. Validation checks that span against sequence. A copied name can sit on the wrong bases.

Does a complete feature list prove the annotations are true?

No. Completeness names the objects. Truth is coordinate match. A complete wrong table is still wrong.