Going further
GeneLoops
A GeneLoop is a longer piece of work you set running and come back to — the kind of survey you would give a rotation student a week to do.
What they are
An ordinary conversation is a back-and-forth. A GeneLoop is a single brief that runs through several phases on its own: reading the literature, analysing your starting construct, proposing variants, screening them computationally, and writing up what it found.
It produces a report and a set of proposals, not a finished construct. The output is something to read and argue with — the same as what you would get from a colleague who spent two days on the question.
Start one from GeneLoops in the Home panel. Running loops are grouped as Running, Recent and Completed.
The templates
Eight starting points, in three groups. Each has a research prompt and a set of grounding rules, both of which you edit before launching — the template is a starting point, not a form.
| Group | Templates |
|---|---|
| Expression & optimisation | Expression optimization loop · Promoter screening |
| Construct design | Resistance marker swap · Vector backbone comparison · Insert design & cloning strategy |
| Quality & validation | Synthesis feasibility screen · Protein variant design · Literature deep dive |
Writing the brief
A GeneLoop runs for a while without checking in, so the brief carries more weight than a chat message does. The two fields:
- The research prompt
- What you want to know. Be specific about the outcome — “find promoters that give lower expression than T7 in BL21” beats “look at promoters”.
- Grounding rules
- The constraints it must respect. Your host, your available parts, approaches you have already ruled out and why. This is what stops a loop coming back with a beautiful plan built on something you cannot use.
When one is worth it
Good fits
Open-ended survey questions with a lot of ground to cover. Which backbone suits this experiment; what is known about this protein's solubility; which promoter strengths are available in this host; what variants have been reported at this position.
Poor fits
Anything with one right answer you already half know. “Does EcoRI cut this plasmid?” is a tool call, not a research loop — open Enzyme Search and you have the answer in two seconds.
Building a specific construct is also better as a planned design conversation, where you can correct the strategy before anything is built.
Reading the output
Treat it as a literature review from a capable but overconfident colleague. The specific things to check:
- Every citation
- Open them. A reference that does not resolve to a real paper is a reference you cannot use, and the failure mode of language models here is well documented. Checking a citation.
- Every part it proposes
- Does the accession resolve to the thing it says? A variant name and a variant sequence are not the same claim.
- Whether the conclusion follows
- A ranked list of variants is a hypothesis to test at the bench, not a result. Computational screening narrows what to try; it does not tell you what works.
A note on scheduled automations
There is an Automations area next to GeneLoops for recurring scheduled work. It is a preview: what you create there is not saved and nothing is ever scheduled or run. Do not build a workflow around it yet. GeneLoops, which you launch yourself, do run.