Flux balance analysis

A heat target is a budget question wearing a costume. The host has a fixed carbon income and a long list of obligations it cannot default on, and the capsaicinoid branch is one claim among many. Flux balance analysis works out how much of that income can actually reach the end of the branch while the rest of metabolism still balances. You get a ceiling, a feasibility verdict, and — the part teams come back for — a reason.

Nothing has to be modelled first. All 65 organisms in the catalog ship with a genome-scale metabolic model already constrained for tissue context: the pathway native in the Capsicum species, present as designed insertions in the sixty-odd hosts that never had one. Select a construct and the host model, media defaults and pathway edits resolve from its record. Select nothing and you can paste a design of your own instead.

Construct

SS-0421 — Thorn Crown, Bramble series

Host

Raspberry (Rubus idaeus)

Target

25,000 SHU, Tier 3

Objective

Capsaicinoid branch flux, biomass floor 80%

Media

Fruit-tissue default

Solver

COBRApy

Runtime

Under 10 seconds

Verdict

Feasible

Rate-limiting step

PAMT, vanillylamine branch

Cost

Free — CPU engine

Raspberry is a good illustration precisely because it is an unlikely host. It makes no capsaicinoids at all, so the designed branch hangs off phenylpropanoid intermediates it already produces, and the interesting question is whether the amine supply can keep up. Here it can, at 25,000 SHU, with the biomass floor held. Push the target up and the same run tells you where it stops being true.

The named bottleneck is what changes behaviour. Everything downstream of PAMT in this solution has capacity to spare, which means an afternoon spent tuning capsaicin synthase would have bought nothing. That is a cheap thing to learn in ten seconds and an expensive thing to learn in a greenhouse.

A column of flux values is technically the answer and practically unreadable, so the result renders as an Escher map with the load visible on every edge. Thick edges carry material. Thin ones carry little. Edges at zero are inactive under this solution, and the shape of what remains tends to make the bottleneck obvious before you have read a single number.

One pattern recurs across hosts and is worth knowing in advance: roughly ninety percent of flux control sits on the vanillylamine branch, the amine side feeding PAMT. The acyl side is rarely where the constraint lives. Teams who have internalized that stop wasting scans on the wrong half of the pathway, and the map is how most of them internalize it.

Two solutions can be overlaid so the difference is the thing you look at. Widen a branch, shrink a competing sink, hold the budget — that reads in one glance and summarizes badly in prose, which is the whole argument for drawing it.

Cultivar designers run it first, before committing to a host. Enzyme teams run it to decide which protein is worth a variant scan. Anyone evaluating a target across several candidate organisms runs it dozens of times in an afternoon, which is the intended use and the reason it is priced the way it is.

Flux analysis is a CPU engine. It is free, it stays free, and no credits are consumed by running it a hundred times. Credits only enter the picture when a question turns structural and you move on to variant scoring or folding — by which point this engine has usually told you which protein to spend them on.