Vaccine manufacturing
Early warning for vaccine fermentation batches
A vaccine manufacturer wants to see that a fermentation batch is drifting while there is still time to act, instead of learning it from an offline assay after the run.
The challenge
How much antigen a batch yields is decided in two steps: fermentation and inactivation. Both run to a fixed recipe and are checked by analysis after the fact.
Antigen titre is measured offline once fermentation is over. A weak batch therefore moves on: it occupies the vessel, purification, weeks of inactivation and quality-control time before anyone can act on it.
The client
A vaccine manufacturer producing protein antigens by bacterial fermentation, followed by a chemical inactivation step that runs for weeks.
Our approach
A golden-batch reference built from the manufacturer’s own historical runs. During each new run, the twin:
- compares temperature, pH, redox and agitation with the reference, live
- warns when the run starts to drift, and points to the variable that drifted
- forecasts the final antigen titre while the culture is still running
Estimated effects
6–8 weeks
of vessel and quality-control time freed per year
Illustrative: 20 batches a year, 1 in 10 weak, and a weak batch stopped at fermentation instead of occupying 3–4 weeks of inactivation and QC.
30–40%
fewer optimisation runs
Range reported in published industry case studies of model-guided bioprocess development; here it comes from the parameter ranking in variant B.
0
new sensors for variant A
Variant A works on signals the plant already records: temperature, pH and agitation, plus the final titre.
Estimates to be checked in Phase 0 on the manufacturer’s historical data, not results.
How the PoC runs
-
01
Phase 0 · Data assessment
On historical batches we measure how much earlier a warning would have come and how much titre varies between batches. That decides the variant.
-
02
Variant A · Minimal
Process curves and final titre per batch. You get an early warning of drift and the variable behind it.
-
03
Variant B · Standard
A, plus redox, in-run samples, media lots and recorded deviations. You also get a titre forecast during the run and a ranking of the parameters that drive it.
-
04
Variant C · Full
B, plus batches of a second product or strain. The reference is carried over to the next product, a proven pattern for the whole portfolio.
-
05
Optional second stage
A kinetic model of the inactivation step, so it ends on the measured state of the material rather than on the calendar. More exploratory, and only after the first stage shows the data can be joined with QC results.
What we asked for
-
Installation description
A P&ID or drawing of the fermenter with measurement points and ranges.
-
Sensor time series
Temperature, pH, redox, agitation, dosing and other signals recorded during cultivation.
-
Lab results
Antigen titre for each batch, and in-run measurements where they exist.
-
Batch history
Batch number, start and end, strain, media lot and recorded deviations.
What you get
-
Time to react
Correct the culture conditions instead of losing the batch.
-
Less work downstream
A weak batch no longer occupies inactivation and quality control.
-
Repeatability
One reference pattern for the next strains and products.
-
Fewer trials
A ranking of the parameters that decide the titre.
Data can come in any form, and no GxP validation is needed for the PoC. How much earlier the warning comes will be measured in Phase 0 on the manufacturer’s historical data, so we quote no number before that. The method, multivariate statistical process control, has been used in industrial fermentation for over 20 years and is described as a PAT tool for biopharmaceutical manufacturing.
Have a similar process?
Tell us about it. We’ll tell you where a model pays off first and what data you’d need.
Book a technical call