Designing with the assistant
Verification
A verification pass reads the finished construct and reports what it could establish about it. Reading the verdict properly is the last thing between you and a synthesis order.
What it looks at
Verification works from the bases, not from the labels on the map. In broad terms it asks:
- Is the construct structurally sane?
- Right total length, no accidental duplication of the backbone, nothing assembled twice, elements in a sensible order.
- Is everything essential present, exactly once?
- An origin, a selection marker, the promoter, your insert. Present twice is as much of a problem as absent.
- Does the coding sequence actually code?
- Reading frame intact from the start codon, no unintended internal stops, the protein it translates to being the one you asked for.
- Do the parts relate to each other correctly?
- A promoter has to actually drive your transgene, not merely exist somewhere on the plasmid. A ribosome binding site has to be positioned to be used.
It reports what it finds, with the coordinates and identities it found them at, so each item is something you can go and look at yourself.
There are three outcomes, not two
This is the part worth reading twice, because “did not pass” covers two very different situations.
| Outcome | What it means | What to do |
|---|---|---|
| Passed | The check ran and the bases support the claim. | Nothing. |
| Failed | The check ran and the bases contradict the claim. Something is genuinely wrong with the construct. | Fix it before going further. This is a real defect. |
| Could not be established | The check ran but could not reach a verdict — most often because the element is not one the recognisers cover. | Confirm it yourself. It is not evidence of a problem, and it is not evidence of correctness either. |
Reading the verdict
If everything passed
The construct is consistent with what you asked for, as far as the checks reach. That is a real result and worth having. It is not a promise that the experiment will work — nothing computational can tell you whether your protein folds, whether your promoter is strong enough in your cell line, or whether the ligation will go.
If something failed
Take it seriously and read the specific finding. A failure names what it found and where. Common ones are worth recognising on sight:
| Finding | Usually means |
|---|---|
| An unexpected stop codon inside a coding sequence | The junction is out of frame, or the insert went in on the wrong strand. |
| An essential element present more than once | A fragment was assembled twice, or a whole backbone was duplicated. |
| An essential element missing | The cloning boundaries cut something out that you needed. |
| The promoter does not drive the transgene | They are both present but not arranged so that one transcribes the other. |
Each finding names a way to address it. Following that is usually faster than starting over — the construct is often one boundary correction away from being right.
If something could not be established
The construct can still be perfectly good, and you can still use it. What you must not do is treat the pass rate as though those items had passed.
When a construct has unestablished items, the report says so explicitly and carries the list. Anything built on top of it — a summary, a protocol — carries that caveat forward too. It is never described as verified.
The right move is to check those specific items yourself. Usually that is one translation, one alignment against the source record, or one look at the map.
Findings that were already there
Some findings are about the plasmid you started from rather than anything you did. A quirk that was in the parent backbone before your edit is reported as such and does not count against your build — you are being told about the molecule, not blamed for it.
Before you order DNA
A short checklist. Synthesis is the point where mistakes start costing real money and weeks.
- Read the verdict, not the chat message
- The conversation summarises. The report is the thing with the findings in it.
- Resolve every failure
- Not most of them.
- Personally check every unestablished item
- These are the ones that reach a vendor unnoticed.
- Translate the full coding sequence one last time
- Start to stop. It is thirty seconds and it catches the expensive class of error.
- Run a feasibility check
- Ask whether the sequence is manufacturable — extreme GC, long repeats and homopolymer runs are what vendors reject or surcharge.
- Save the map
- Export GenBank and keep it with the order. Exporting.
What verification cannot tell you
It is a sequence-level check. It cannot tell you whether your protein will fold, whether the promoter is strong enough in your cell line, whether the enzyme is active, whether your ligation will work, or whether the biology of the experiment makes sense. Those are still yours — and still the bench's.
How the assistant works is the wider version of that argument.