Platform
The pipeline
Read, model, design, edit, blend, validate. Six stages, each one a screen you can stop at, inspect, revise and re-run — with a clear contract for what goes in and what comes back out.
6
stages
70
catalog series
14.1 Mbp
designed DNA
Versioned
at every stage
01
Why a pipeline and not a button
A single button that turns a heat number into finished DNA would be easier to market and much harder to trust. Design work involves judgement, and judgement needs places to stand. So the platform breaks the job into six stages, each with a defined input, a defined output and a screen where a human can disagree with the software before the next stage inherits the decision.
Stages are re-runnable in place. Change an assumption at stage two and everything downstream is marked stale rather than silently recomputed, which means you always know which parts of a project reflect the current thinking and which are left over from an earlier one. The 14.1 Mbp of designed DNA in the catalog all came through these six stages in this order.
02
The six stages
You supply sequence, it returns structure
Paste a sequence, upload a FASTA, or pick a reference from the 65 organisms in the catalog. Biopython-backed annotation returns coding regions, regulatory elements, feature tracks and a clean record you can browse rather than scroll.
You supply a target, it returns a ceiling
A COBRApy metabolic model of the host, with the capsaicinoid branch grafted on. Flux balance analysis returns a feasibility verdict and an upper bound on achievable heat, drawn as an Escher pathway map you can click through reaction by reaction.
You supply enzymes, it returns a ranked shortlist
Meta ESM-2 scores substitutions across the pathway enzymes and returns candidates ordered by predicted effect, with folded structures rendered in Mol* for the ones worth inspecting. Ranking, not verdicts — the shortlist is yours to prune.
You supply a locus, it returns an edit plan
Guide candidates scored for on-target efficiency, then swept across the host genome for off-target matches. Guides with crowded neighbourhoods are flagged before they reach a construct, and the evidence stays attached to the plan.
You supply parts, it returns a construct
DNA Chisel adapts codon usage to the host, Primer3 proposes verification primers, and the assembly step turns a parts list into a real record with junctions, orientation and a predicted heat rating for the whole.
You supply rules, it returns a report
The compliance pass checks the finished design against the project's own constraints and returns a pass, a warning or a block with the reason named. Only then does the export drawer open.
03
What a run record looks like
- Project
- SS-0421 — Tier 2 tomato
- Stage 1
- Annotated — CPU, free, 4 s
- Stage 2
- Feasible to 3,100 SHU — CPU, free
- Stage 3
- 182 variants scored — GPU, Standard
- Stage 4
- 6 guides, 2 flagged — CPU, free
- Stage 5
- Construct 11.4 kb — CPU, free
- Stage 6
- Pass with one warning — CPU, free
Read the cost column and the shape of the product becomes obvious. Five of the six stages run on CPU and cost nothing, so iteration is genuinely free: revise the target, re-model, reassemble, revalidate, as many times as the problem takes. Only stage three reaches for a GPU, and only when you ask it to.
Every stage writes its result with a version stamp and the engine build that produced it. Re-run the same project six months later and you can tell whether a changed answer came from your edit or from an updated engine — a distinction that is impossible to recover after the fact if nobody recorded it at the time.
04
Running it at scale
The same six stages drive batch work. Point the batch runner at a list of targets and it walks each one through the pipeline, parallelising where the stages allow and collecting the results into a single table you can sort by feasibility, predicted heat or construct size. Teams use this to triage a season of ideas in an afternoon.
Everything the interface does, the API does, which makes the pipeline a step in your own system rather than a destination. Submit at stage one, collect at stage six, or enter partway through with sequence you already have. Enterprise workspaces add scoped API keys, webhook callbacks on stage completion, reserved GPU capacity and SSO.
Ready to run it
Try this in the Splice Platform.
CPU engines are free to run. GPU engines run on prepaid credits.
Related