Biopharmaceutical manufacturing
Finding where a digital twin pays off in a recombinant protein process
A manufacturer of a recombinant therapeutic protein asks where a digital twin pays off first: upstream, refolding or purification.
The challenge
The process has many steps, and yield losses add up across them. Refolding is often the yield-critical step: it is sensitive to protein concentration, redox conditions, pH, temperature and time.
Results arrive offline after each step, and much of the know-how that keeps the process on track sits with experienced staff. In GMP manufacturing, any model that supports decisions also has to be defensible in front of QA.
The client
A biopharmaceutical manufacturer producing a recombinant therapeutic protein in E. coli, expressed as inclusion bodies, then refolded and purified in several chromatography steps.
Our approach
We start from literature and public sources, not from assumptions about the client’s process, and map candidate levers for each stage as hypotheses to test:
- Upstream: fermentation monitored against a golden batch, with soft sensors for biomass and inclusion-body formation
- Refolding: a hybrid model of refolding yield against conditions, to try conditions in silico before the lab
- Downstream: models of the chromatography steps, to see where product is lost
Estimated effects
35–70%
fewer refolding experiments
Range reported in published industry case studies of hybrid-model-guided experimentation.
≈ 3%
more product for every +1 percentage point of refolding yield
Illustrative: assumes a 30% refolding yield and unchanged downstream steps.
Hours, not weeks
to screen a refolding condition
A run on the model instead of a lab experiment.
Estimates to be checked against the manufacturer’s process data before a PoC variant is chosen, not results.
How we get to a PoC
-
01
Opening material
The process stages we identified, the possible levers and the data we need, written conditionally and based on the literature.
-
02
Process data
A P&ID, a process description and the data that exists today.
-
03
Hypotheses checked
Each lever scored on value, complexity and cost against the real data.
-
04
PoC variants
One or two recommended variants, delivered as a feasibility study (Stage 0) with a fixed scope and success criteria agreed up front.
-
05
Path to manufacturing
If the PoC succeeds, the model follows our qualification framework toward GMP use.
What we asked for
-
P&ID
Equipment and measurement points for the stages in scope.
-
Process description
Steps, set-points and the critical parameters and quality attributes per stage.
-
Available data
Which signals and lab results exist, in what form, and over how many batches.
What you get
-
Focused investment
Effort goes to the stage where a model pays off first, not to all of them at once.
-
Fewer lab experiments
Refolding conditions tried in silico before they are tried in the lab.
-
Earlier warning
A weak batch spotted during the run rather than after release testing.
-
A model QA can accept
A documented, staged path to a qualified model.
Everything above is a hypothesis built from the literature and public sources. Priorities are set only once the manufacturer has confirmed its process data.
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