Workflow
Scale heat to a target
Heat is not a dial you turn — it is a flux budget you negotiate with the host. This workflow moves a design along the ladder from 100 SHU to 16,000,000 SHU and tells you where the host stops cooperating.
6
tiers
70
series
100 – 16M
SHU ladder
HPLC
calibration
01
The goal
Every design has a heat target, and most of the interesting failures happen because a target was chosen without asking what it costs the plant. Capsaicinoids are expensive molecules. The precursors come out of the same pools the host uses for lignin, for amino acids, for everything else it needs to be a functioning organism. Push hard enough and the budget breaks before the chemistry does.
This workflow treats the Scoville ladder as a constraint-satisfaction problem. You name a target; the platform tells you whether the host can fund it, what the flux bottleneck is, and what the design would have to change to get there.
02
Stage by stage
Place the target on the ladder
Six tiers run from Pepper through Tier 5. Setting a target shows which tier it falls in and which of the 70 series already live near it, so you can see whether anything comparable has been designed before.
Test the budget
COBRApy runs flux balance analysis against the host model with the capsaicinoid branch active. The output is a feasible flux range, and if the target sits outside it the shortfall is reported in the same units as the model.
Find the bottleneck
Escher maps the solution so the limiting step is visible rather than inferred. In most hosts the vanillylamine branch is the constraint, which is why roughly ninety percent of flux control sits there.
Shape the burn
Total SHU is one number; the capsaicin-to-dihydrocapsaicin ratio and the minor capsaicinoids decide how the heat arrives and how long it lingers. The blend controls adjust that profile independently of the total.
Re-run and compare
Each adjustment produces a new solution saved alongside the old one. Comparing runs is how you find the point where additional heat starts costing more than it returns.
Anchor to measurement
Predictions are calibrated against HPLC heat ratings from material that has actually been grown. The platform keeps the distinction between a predicted rating and a measured one visible at every step.
03
Worked example
- Project
- SS-0733 — Tier 3 to Tier 4 revision
- Host organism
- Capsicum chinense
- Starting rating
- 410,000 SHU (HPLC, measured)
- Requested target
- 1,200,000 SHU
- Model verdict
- Feasible, bottleneck on the amine branch
- Primary lever
- PAMT activity
- Burn profile
- Slow build, long tail
Roughly a threefold increase, in a host that already carries the full pathway. The model called it feasible but located the ceiling on the amine side, which reframed the work: not a broad push on every enzyme, but focused attention on one branch and the promoter driving it.
The burn profile mattered as much as the number. A Tier 4 rating that arrives all at once reads as harsh; the same rating with a slow build and a long tail reads as structured. Those are different capsaicinoid ratios reaching the same total, and the designer treats them as separate targets because they are.
04
What you get
A heat-scaling run leaves behind a feasibility verdict with the model and media assumptions attached, the identified bottleneck, a predicted capsaicinoid ratio, and a comparison against the nearest measured catalog entries. When a target is infeasible you get the reason rather than a silent clamp.
Flux balance analysis is a CPU engine and runs free, so iterating on targets costs nothing but time. Credits only come into play if the revision sends you back to variant scoring or structure prediction.
Ready to run it
Try this in the Splice Platform.
CPU engines are free to run. GPU engines run on prepaid credits.
Related