Should the next intervention be more harvest buffer, parallel chromatography, greater UF capacity, shorter release time or a different fermentation rate?
Independent model study · recombinant insulin API
When continuous insulin fermentation meets finite downstream capacity
Resolve whether harvest buffering, enzymatic modification, chromatography, UF/DF or drying limits accepted API output before adding fermentation capacity.
Yeast cultureContinuous harvestCell removalModificationChromatographyUF/DFDrying
Decision model
A bounded question with a complete plant consequence.
Media and seed through continuously harvested production culture, recovery, modification, chromatography, UF/DF, drying, bulk hold, clean utilities, wastewater and QC release.
Equations and accounting rules
d(CᵢV)/dt = FᵢₙCᵢ,ᵢₙ − FₕCᵢ + rᵢV; constant volume requires Fₕ = FᵢₙdVₕ/dt = Fₕ − F recovery for harvest-tank inventoryOUR = qO₂XV ≤ OTR = kLa(C* − Cᴸ)Vm load ≤ DBC × V resin × utilisation; overall yield = product of step yieldsBase, alternative and stress cases
- Base train versus +20% precursor titer
- Double harvest-buffer capacity
- Parallel chromatography or shorter QC hold
- Stress: lower UF flux plus extended CIP
Engineering brief
Model the complete decision, not an isolated unit operation.
The EMA publicly describes insulin icodec active-substance manufacture as yeast fermentation followed by recovery, enzymatic modification, acylation, chromatography, ultrafiltration and drying. That architecture is unusually detailed public evidence for a transparent capacity model.
This study couples a continuous broth source to discrete downstream campaigns. It is not a digital twin of Kalundborg: vessel sizes, titers, yields, cycle times and bottlenecks remain explicit assumptions until authorised site data replaces them.
01 · Model basis
What the Acatian model needs to resolve
Inputs, mechanisms, limits and outputs remain reviewable on one declared basis.Inputs
Fermenter volume, feed and harvest profiles
Biomass, precursor titer, OUR and heat release
Step yields, resin DBC, UF flux and dryer rate
CIP/SIP, buffer, utility, hold and release times
Mechanisms
d(CᵢV)/dt = FᵢₙCᵢ,ᵢₙ − FₕCᵢ + rᵢV; constant volume requires Fₕ = Fᵢₙ
dVₕ/dt = Fₕ − F recovery for harvest-tank inventory
OUR = qO₂XV ≤ OTR = kLa(C* − Cᴸ)V
m load ≤ DBC × V resin × utilisation; overall yield = product of step yields
Constraints
Harvest overflow or recovery starvation
Oxygen-transfer and heat-removal margin
Chromatography cycles and buffer availability
UF flux, dryer rate and validated hold times
Outputs
Accepted API kg/year and batch lead time
Harvest inventory, WIP and queue loss
DSP utilisation and chromatography cycles
Utility peaks, COGS and intervention value
02 · Acatian workflow
Build it in six controlled steps
Each step creates a reviewable object, not a hidden spreadsheet assumption.- 01
Declare public facts and assumed plant basis.
- 02
Build fermentation and continuous-harvest balances.
- 03
Add recovery, modification and purification yields.
- 04
Schedule finite downstream equipment and holds.
- 05
Reconcile clean utilities, waste and annual output.
- 06
Compare interventions and expose decision-reversing data gaps.
03 · Decisions
Questions the model should answer
Does more fermentation produce more accepted API?
Which downstream cycle first creates a queue?
How much harvest buffer is genuinely useful?
Which missing input can reverse the investment choice?
04 · Evidence boundary
Validate before the result carries weight
Validation
Require component residual below the declared tolerance, no negative inventory, no hold violation and independent checks of harvest rate, step recovery, resin capacity, UF flux and elapsed time.
Limitations
No proprietary Novo Nordisk strain, titer, resin, equipment dimension, schedule or site bottleneck is asserted. The model does not establish GMP validation or process safety.
Public evidence
What the company context supports—and what it does not.
Sources establish the public process architecture. They do not reveal private operating parameters, site performance or an Acatian relationship.Frequently asked questions
Practical modelling questions
Is this the named organisation's real plant model?
No. It is an independent hypothetical Acatian study based only on the cited public process architecture. It claims no affiliation, endorsement, deployment, confidential data or actual plant performance.
Where do the numerical inputs come from?
Every input must be marked as a public fact, literature estimate, transparent engineering assumption or authorised customer input. The public article does not invent private operating values.
Can the model be calibrated to a real facility?
Yes, when the operator supplies authorised process, equipment, schedule and utility evidence and agrees the intended use, acceptance criteria and validation plan.