Designing with the assistant
Designing a construct
One worked example, start to finish: a His-tagged protein domain in pET-28a for expression in E. coli. The shape is the same for anything else you build.
The brief
You have a gene. The full-length protein goes into inclusion bodies, so you want to express just the catalytic domain, purify it on nickel, and get soluble material out the other end.
That is three separate requirements — the right sequence, a tag you can purify with, and solubility — and only the first is a cloning problem. Saying all three up front is what stops you getting a construct that is correct and useless.
Set the scene
Put the backbone in the project
Import pET-28a(+) from the stock vectors, or bring in your own copy. Use yours if you have one — lab plasmids drift from the published maps, and you want to design against the DNA you will actually cut. Importing.Write down your constraints once
Put your host strain, your antibiotics and the enzymes you own into PLASMID.md. Every conversation in the project reads it, so you stop repeating yourself and stop getting plans built around an enzyme you do not have. Project instructions.Switch to Plan mode
For a build, always. The assistant cannot touch your files in Plan mode, so the first thing you get back is a strategy rather than a changed plasmid. Plan and Act.
Ask
I want to express the catalytic domain of my gene in E. coli BL21(DE3)
and purify it on nickel.
- Backbone: the pET-28a(+) in this project
- The full-length protein is insoluble, so I want the domain only
- N-terminal His6 is fine
- I have NdeI, XhoI, BamHI, EcoRI, HindIII
Work out the domain boundaries, plan the cloning, and tell me anything
about this that looks likely to fail.Note the last line. Asking for the failure modes is worth more than asking for the design — it is the difference between a plan that answers your question and a plan that survives contact with the bench.
Read the plan properly
What comes back is a document, not a construct. Read it in this order — it is roughly the order in which things go wrong.
| Check | What you are looking for |
|---|---|
| Where each part came from | An accession or a database record you can open. A part with no stated source is the single biggest risk in any plan. |
| The domain boundaries | Do they match the literature you would cite? Truncating a domain two residues short is a common and expensive mistake. |
| The enzymes | Are they yours, do they cut this backbone, and do they cut it once? The plan should say, from a search of your actual sequence. |
| The reading frame | Especially at the junction to the tag. See below. |
| The stated risks | Codon usage, solubility, whether the domain folds alone. If this section is empty, be suspicious. |
The junction is where builds fail
pET-28a already has a start codon, a His6 tag and a protease site built into it, upstream of the cloning region. Your insert is being fused onto the end of that, and the whole thing is translated as one protein from the vector's own start codon.
Which means the junction has to be in frame with what is already there. One base out and the ribosome reads a completely different protein and hits a stop codon thirty or forty codons later — the map looks perfect, every feature is drawn in the right place, and nothing expresses.
If something is wrong, refine
Use Refine rather than Cancel. Refining keeps the evidence already gathered and sends your correction back into the same piece of work. Cancelling starts from nothing.
Corrections that work well are specific:
Two things:
1. XhoI cuts twice in my backbone, not once — check it again against
the file in this project, not the published map.
2. The domain should start at residue 214, not 220. The construct in
PDB 3XYZ starts there and it is the one that crystallised.Approve, and let it build
Choosing Execute moves the work into building. The construct is assembled, saved as a file in your project, and checked.
If it hits a problem — an enzyme that turns out not to cut, a junction that will not work — it is expected to re-check the sequence and correct course rather than stopping and handing the problem back. What it will not do is silently substitute a different design for the one you approved.
Check the result yourself
The construct is now a real file. Open it and spend two minutes on it.
- Total length
- Does it match backbone plus insert, allowing for whatever the cloning strategy removed? A construct twice the size you expected has been assembled twice.
- The diff
- What changed against the parent plasmid, and which features were cut, moved or disrupted. This is the highest-value thirty seconds in the whole process. Reviewing a change.
- The translation
- From the start codon through to the stop. It should be your protein, with the tag on the end you asked for, and no internal stops.
- A diagnostic digest
- Simulate one now and you have a gel to check your colonies against later. Bench tools.
- The verification verdict
- Read it and understand what it does and does not claim. Verification.
Export and order
Export GenBank for a synthesis vendor or for your own records — it carries the annotations with it. Exporting and sharing covers the formats and what each one keeps.
Then write down what you did while it is fresh: an experiment entry in the same project, linked to the construct. Experiments and protocols.
The part that is not cloning
Come back to the original brief. You wanted soluble protein, and nothing so far has addressed that — a perfectly assembled, in-frame, correctly tagged construct can still deliver inclusion bodies.
This is worth asking about explicitly, because “express X” and “get working X out” are different problems. Things that belong in that conversation: codon usage for your host, whether the domain folds without its partners, whether disulfide bonds are needed and whether your strain's cytoplasm can form them, whether a different fusion partner or a periplasmic export signal would help, and what temperature and induction to try first.