Filtration Optimisation: Commercially Competitive with 92% Fewer Experiments
With
8.6x
increase in product recovery
55%
reduction in production costs
37.4%
reduction in processing time
2.1x
fewer experiments vs DoE
At a glance
Multus Biotechnology needed to reduce the cost of recombinant growth factor purification to reach commercially competitive pricing. Filtration processes were the primary cost driver, but manual optimisation could only test a handful of conditions per round — not enough to find the best operating parameters when upstream output varies between batches.
​
New Wave Biotech modelled Multus's filtration processes on the Bioprocess Foresight platform, running four iterative cycles of virtual experimentation and targeted physical testing. The platform adapted parameters to each batch's specific feedstock and optimised multiple purification processes simultaneously — something Multus hadn't expected to be possible. Multus reached their target unit economics without the years of manual iteration that conventional approaches would require.
The Problem
Growth factors are among the most expensive inputs in biomanufacturing. Whether producing recombinant proteins for cell culture media, biologics, or food-grade ingredients, the downstream purification step determines both unit economics and product quality.
​
Yet downstream processing remains the least optimised part of the production chain. DSP typically accounts for 60–80% of production costs, but receives a fraction of the R&D attention given to upstream processes. Most companies optimise manually, testing a handful of conditions and selecting the best option from a narrow set.
​
This is not unique to growth factors. Any company purifying recombinant proteins, enzymes, or other large molecules through filtration faces the same set of challenges:
1. Membrane fouling is unpredictable
Filtration performance degrades as proteins interact with membrane surfaces. Generic models don’t capture the dynamics specific to each product, equipment setup, and operating condition. Without process-specific models, companies either over-engineer their membranes or accept yield losses they can’t quantify.
2. Batch-to-batch variation compounds the problem
Upstream output varies between batches, but DSP parameters are typically fixed. Running the same parameters on different feedstocks leaves yield on the table and introduces inconsistency that becomes expensive at scale.
3. Multiple products multiply the optimisation burden
Companies purifying multiple protein products face the same optimisation problem repeated across product lines. Each new product starts from scratch, with no systematic way to transfer what was learned from the last one.
The Partnership
Multus Biotechnology designs and supplies cell culture media for biomanufacturers across advanced therapies, biologics, and cellular agriculture. Their AI-driven MediOP™ platform accelerates media development and supply, and their products include recombinant growth factors whose production costs directly impact the affordability of their customers’ processes.
​
Multus needed to optimise the downstream purification of these growth factors. Filtration processes formed a major part of their R&D cost base, and with commercialisation ramping up, reducing unit costs was critical to reaching price points that would make their products viable at scale.
​
New Wave Biotech was identified as a way to validate and improve purification system performance, optimise across multiple product lines simultaneously, and co-develop key capabilities around batch-adaptive DSP.
What we did
Over a 12-month engagement, we designed a work package around Multus’s specific needs: target production scale, product quality requirements, and known challenges including protein fouling behaviour.
​
After modelling Multus’s processes on the Bioprocess Foresight platform and uploading existing experimental data to establish a performance baseline, we ran four iterative cycles of optimisation and experimentation.
In each cycle, the platform simulated 10,000 process configurations and recommended optimised parameter sets for Multus to test physically. After each experiment, the new data was uploaded and the model refined. Each cycle made predictions more accurate and recommendations more targeted.
​
Note: The Multus engagement ran 10,000 scenarios per cycle across 4 cycles. Platform capability has since improved to test 100,000+ configurations per cycle.
Three capabilities proved particularly valuable
-
Hybrid modelling with minimal data: The platform combines mechanistic modelling with AI, which meant it didn’t need large datasets to deliver accurate results. The full optimisation was achieved with just eight datasets. This was particularly important for capturing phenomena like membrane fouling, which is normally difficult to predict and where generic models fall short.
-
Simultaneous multi-step optimisation: The platform optimised multiple steps of the downstream process at the same time, rather than sequentially. This was something Multus had not expected to be possible, and it significantly accelerated their overall development timeline.
-
Batch-adaptive parameters: Because upstream output varies between batches, the platform generated DSP parameters specifically adapted to each batch’s characteristics. Not just optimised for the process in general, but tuned to the specific feedstock in each run.
How the optimisation works
-
Model: The baseline process is modelled on our platform with target optimisation goals defined.
-
Calibrate: Existing experimental data is uploaded to tune the model to the specific equipment & operating conditions.
-
Optimise: The platform simulates thousands of scenarios per cycle, recommending optimal parameter sets with clear trade-offs between yield, cost, and purity.
-
Experiment: The team runs targeted experiments based on the platform’s recommendations.
-
Refine: New experimental data is uploaded, the model improves, and the next cycle narrows in further.
Each cycle makes the model smarter. The result is fewer, more targeted experiments — reaching the right answer faster rather than replacing experiments entirely.
Results
Over 12 months and four optimisation cycles, Multus transformed the economics of their growth factor purification:
8.6x
increase in product recovery
55%
reduction in production costs
37.4%
reduction in processing time
2.1x
fewer experiments vs DoE
Production costs reached commercially competitive levels. Product concentration more than doubled. The 92% reduction in experiments compared to a traditional Design of Experiments (DoE) approach meant the full optimisation was achieved with just eight datasets.
“With New Wave we have been able to create an 8X increase in yield and speed up R&D within a year, which is impressive. As a young company, it was really important for us to have strong support and New Wave were always on hand to answer questions and tackle challenges. Together we were really able to test the capability of the platform and learned a lot in the process.”
— Brandon Ma, Senior Scientist, Multus
What this enabled
-
Commercially viable unit economics: costs reduced to levels that support scaled production and competitive pricing
-
Early elimination of unviable routes: the platform identified which processes could and couldn’t reach target performance before committing resources to scale-up
-
Batch-adaptive DSP: parameters tuned to each batch’s specific upstream output, not just optimised for the process in general
-
Accelerated R&D timeline: four optimisation cycles within a year, with only eight datasets required
Relevance beyond growth factors
The downstream challenges in this project — membrane fouling, batch variation, process route lock-in — are not unique to growth factor purification. They apply wherever large molecules are purified through filtration, centrifugation, or related unit operations.
​
The Bioprocess Foresight platform now covers 18 unit operations across the full downstream processing train, with integrated Techno-Economic Analysis and ISO 14040/44-aligned Life Cycle Assessment. Companies producing recombinant proteins, industrial enzymes, yeast-derived proteins, fermentation-derived food ingredients, or lipid products face structurally similar DSP decisions — and the same optimisation approach applies.
"Combining AI and real-world data collection to produce cheaper and more scalable growth media ingredients helps us put our customers one step closer to making widely available and affordable cultivated meat a reality.”
— Cai Linton, CEO and Co-Founder, Multus
