Trust
Methodology
Six stages, each answering a question the next one depends on. This is the sequence behind all 1,000 catalog constructs, and the reason the platform is shaped the way it is.
6
stages
1,000
constructs
14.1 Mbp
designed DNA
HPLC
final check
01
Why a fixed sequence
The stages are ordered because each one constrains the next. Designing an enzyme before knowing whether the host can supply its substrate is wasted effort. Choosing a promoter before knowing which tissue expresses what is guesswork. Setting a heat target before testing the flux budget produces targets that cannot be met and disappointment that arrives a season later.
Running the stages in order is not bureaucracy. It is the cheapest arrangement, because the early stages are fast and free and the late ones are slow and expensive. Every design that dies at stage two is a design that did not consume a greenhouse.
02
The six stages
Read — transcriptomics
Establish what the host actually expresses, in which tissue, at what level. Capsaicinoid accumulation is tissue-specific, so promoter and targeting choices rest on this stage. Without it, later decisions are assumptions wearing the clothes of data.
Model — flux
COBRApy runs flux balance analysis against the host's genome-scale model with the capsaicinoid branch active, and Escher renders the solution. This is where a heat target is confirmed affordable or shown not to be, and where the bottleneck is located.
Design — ESM-2
Meta ESM-2 650M scans the controlling enzymes exhaustively; ESMFold, AlphaFold- and Boltz-2-class prediction fold the shortlist; AutoDock Vina and DiffDock supply receptor context against TRPV1 where it is relevant.
Edit — CRISPR
CHOPCHOP- and CRISPOR-style scoring proposes guide sites and Cas-OFFinder sweeps the host genome, so every guide carries its off-target evidence into the construct rather than acquiring it afterwards.
Blend — capsaicinoid ratios
Total heat and burn profile are separate targets. The acyl branch sets chain length and saturation, which determines the ratio across capsaicin, dihydrocapsaicin and the minor species — and therefore how the heat arrives and fades.
Validate — HPLC
Grown material is measured by HPLC and the rating compared against the prediction. This is the validation stage — the step that will hold the design predictions accountable. It is in progress in our greenhouse; no cultivar has completed it yet, so all heat values remain computed design targets.
03
Where the leverage is
- Pathway shape
- Two branches, one junction
- Dominant control
- Vanillylamine branch, ~90% of flux
- Primary lever
- PAMT activity
- Junction
- Capsaicin synthase, pun1 locus
- Ratio lever
- Acyl chain length and saturation
- Cheapest stage
- Flux modelling — CPU, free
- Slowest stage
- HPLC on grown material
Roughly ninety percent of flux control sitting on the vanillylamine branch is the fact that organizes the whole method. It means stage two does not merely say yes or no — it says where to spend stage three. Enzyme work spread evenly across the pathway produces modest results; enzyme work aimed at the branch the model identified produces movement.
The separation of total heat from burn profile is the second structural insight. They are set by different branches and are, within limits, independently adjustable. Designs that treat them as one variable tend to achieve neither cleanly.
04
What the method commits to
Three commitments hold across every stage. Predicted and measured values stay visually distinct wherever they appear together. Every computation is stored as a run with its inputs, parameters and pinned engine version, so results can be reproduced rather than recalled. And assumptions travel with their outputs — a flux map without its media and objective is not presented as a result.
The founding ten peppers are the worked reference for all of it: twenty-five constructs, twenty-four guide RNAs, 31,911 base pairs, every stage visible. The catalog has grown to 14.1 Mbp across 1,000 constructs and 65 organisms, and every one of them travelled this same sequence.
Ready to run it
Try this in the Splice Platform.
CPU engines are free to run. GPU engines run on prepaid credits.
Related