mRNA Production Systems: From Published Process Models to System-Level Design

Over the past few years, mRNA manufacturing has moved from a largely experimental capability to an industrial reality. The COVID-19 response proved that mRNA drug substance could be produced at very large scale, and the next wave of applications is now pushing the same platform into oncology, individualized neoantigen therapies, and other small-batch therapeutic settings. For production system design, that shift matters. The process steps may look similar on paper, but the operating problem changes once batch size, product mix, turnaround time, and variability become central design constraints.

The purpose of this article is to make that production-system question explicit. We are not treating published techno-economic analyses as inputs to LineLab, and LineLab does not rely on SuperPro Designer or any other process simulation tool to generate its own results. Instead, we used a published mRNA vaccine study because the authors disclosed unusually high-quality process data and the underlying SuperPro models. Those files made it possible to inspect the assumptions behind the process: the equipment choices, process times, batch relationships, material assumptions, and routing logic. We then reconstructed the production system in LineLab from those public assumptions and evaluated what can be learned once flow, variability, queueing, and sensitivities are represented directly.

This distinction is important. Pharma already has a mature tradition of techno-economic modeling. SuperPro Designer is widely used for bioprocess modeling, equipment sizing, scheduling, and cost estimation. The point of this work is not that process-level techno-economic analysis is immature. The point is that at-scale production system design introduces questions that are adjacent to process modeling, but not identical to it: how assumptions interact, how variability propagates, where intermediate inventory accumulates, which buffers are economically useful, and which inputs actually determine cost and performance once the full system is considered.

Why the Kis et al. study was a useful starting point

The study by Kis et al. is a particularly strong foundation for this exercise because it provides far more than a high-level cost estimate. The authors assessed the techno-economic feasibility of producing RNA vaccines for global pandemic demand, modeled three mRNA vaccine cases, one self-amplifying RNA case, and a hypothetical next-generation saRNA case, and published the SuperPro Designer models used in the analysis [1]. The paper explicitly considered RNA amount per dose, production scale, titer, failure rate, reagent costs, labor, and QC/QA assumptions, and it found that RNA amount per dose had the largest effect on both annual production amount and cost per dose, with scale and titer also important but generally secondary [1].

The paper also contains process-level conclusions that are directly relevant to production system design. It identifies LNP formulation as a bottleneck for drug substance production, discusses the possibility that fill-finish becomes limiting when drug substance can be produced very quickly, and notes that scheduling gaps and QC timing were not considered in the study [1]. Those are not weaknesses in the paper. They are exactly the boundary between a high-quality process model and a production-system model.

In other words, this was not a case of filling in missing data with generic assumptions. It was a rare opportunity to start from a transparent, published biopharma process model and ask a different question: if we rebuild the system from the disclosed assumptions, what additional insight do we get from modeling it as a constrained production system?

What is known about the mRNA production process

The broad structure of the mRNA manufacturing process is now well established in the literature. A DNA template is used for in vitro transcription, the RNA is purified, and the purified RNA is formulated into lipid nanoparticles, followed by sterile filtration and fill-finish. The same general architecture appears across vaccine and therapeutic applications, although dose, batch size, quality requirements, and turnaround constraints can differ substantially [3][4].

The process is often described as a platform because the same unit operations can produce different RNA sequences once the template has been changed. Kis et al. make this point directly when discussing how switching to a new vaccine primarily requires changing the DNA template used for RNA synthesis [1]. This platform nature is one of the reasons mRNA is attractive, but it can also be misleading from a production system perspective. A platform process does not automatically imply a platform production system. A high-volume vaccine campaign, a personalized cancer vaccine workflow, and a distributed clinical-trial manufacturing network may share similar unit operations while requiring very different production architectures.

The published literature also makes clear that mRNA production economics are dominated by a small set of physical and material factors. Dose, titer, capping chemistry, nucleotide costs, LNP formulation materials, and single-use consumables all matter. However, a static cost breakdown is not the same as a design model. A reagent can be expensive without being the best optimization target. A buffer can appear costly while still being economically justified because it prevents blocking elsewhere. A reactor volume can look like a direct scale lever while having less influence than expected once downstream steps, batch timing, and flow constraints are considered.

From SuperPro assumptions to a LineLab production system

The LineLab reconstruction started from the public process assumptions, including the SuperPro model structure, equipment definitions, and process data. The objective was to preserve the published process logic while representing the system in a way that makes flow behavior visible. This included routing between operations, workstation capacity, batch relationships, process-time assumptions, material-cost assumptions, recurring and non-recurring costs, and intermediate inventory.

This is different from importing a prior techno-economic result. The published work helped identify the process assumptions. LineLab then solved the production system independently.

The practical advantage of this reconstruction is that assumptions are no longer buried inside a process model or reduced to a cost-per-dose output. They can be inspected through sensitivities, marginal costs, utilization, flow time, and inventory behavior. This is especially useful when an analysis includes many assumptions that are plausible but uncertain, such as process variability, yield, reagent prices, equipment scale, or titer.

Modeling uncertainty directly

Kis et al. already performed uncertainty and sensitivity analysis around several important inputs. The difference here is that LineLab can propagate uncertain inputs through a system model that includes constrained resources and flow interaction. In the model shown below, uncertainty was introduced in process-time variability, UTP cost, bioreactor working volume, and mRNA bioreactor titer.

Input	Min	Likely	Max	Unit
All Processes CV	10	15	20	%
100 mM UTP	1,125	1,762.8	6,751	$
Bioreactor Working Volume	20	30	50	L
mRNA Bioreactor Titer	2	5	7	count/L

The coefficient of variation across all processes was not taken from the SuperPro model. SuperPro campaign schedules are largely deterministic unless variability is represented externally. We therefore used public biopharma operations research and process engineering references to establish a reasonable variability range for process-time uncertainty. The purpose was not to claim an exact industrial value, but to test how sensitive the system is to realistic variation around nominal process times.

This table makes the modeling boundary explicit before the outcome distribution is introduced. It separates the public inputs inspected from the SuperPro model, the uncertainty assumptions added for the LineLab reconstruction, and the variables used for system-level sensitivity analysis.

Cost is a distribution under uncertainty

A deterministic techno-economic model typically produces a point estimate for cost per dose. That is useful for comparing scenarios, but it hides the question of robustness. Once uncertain inputs are propagated through the system, cost becomes a distribution.

This result is not a replacement for the paper’s cost estimate. It shows the additional information that becomes available once cost is treated as an outcome under uncertainty. The central estimate matters, but so does the shape of the distribution. The right tail is where unfavorable combinations of titer, material cost, process variation, and flow behavior begin to matter for investment decisions.

This is also where the public/private boundary becomes useful. The exact parameter ranges used in a confidential industrial project would clearly be proprietary. The modeling approach, the publicly disclosed process assumptions, and the fact that these uncertainties can be propagated through a system model are not.

Flow time, intermediate storage, and biopharma vocabulary

In general manufacturing, it is natural to talk about queueing and WIP. In biopharma, the same phenomenon often appears as intermediate hold, single-use buffer storage, material hold time, or controlled in-process inventory. The vocabulary matters because queueing is not an abstract line of parts sitting on a factory floor. It may require single-use bags, tubes, hold vessels, environmental controls, sampling, documentation, and sometimes explicit dwell-time limits.

That means buffering has an economic and quality dimension. Additional intermediate storage can reduce blocking and improve equipment utilization, but it can also add consumables, handling, and exposure time. Too little buffering can starve downstream steps or force upstream equipment to wait. Too much buffering can increase cost without improving throughput.

The flow-time view shows why nominal process durations are not enough to understand a production system. Flow time includes the effect of constrained resources and the interaction between process steps.

The WIP inventory view makes the same point from the perspective of in-process material. The issue is not simply that inventory exists, but that the amount and location of inventory become design variables. In a biopharma system, WIP is tied to containers, hold conditions, consumables, and sometimes stability constraints.

Bottlenecks and utilization under flow

The Kis et al. paper identifies LNP formulation as a bottleneck for RNA drug substance production and notes that the bottleneck can be addressed by operating multiple microfluidic mixing devices in parallel or using larger devices [1]. That conclusion remains important. The LineLab reconstruction does not contradict it. It gives a way to inspect how that constraint behaves when the surrounding system is represented explicitly.

The important distinction is that utilization alone should not be treated as the final answer. A resource can be highly utilized without being the most economically meaningful design lever, and a step can create local queueing without showing up as a top sensitivity. The more defensible interpretation is to look at utilization, flow time, WIP, and sensitivities together.

In this model, some queueing behavior appears large, but that does not automatically mean it is the first thing to fix. If the corresponding storage is relatively inexpensive or if the bottleneck is driven by another parameter, the cost-optimal design may include more intermediate inventory than intuition would suggest. This is exactly the kind of tradeoff that is difficult to resolve from a process flow diagram alone.

Cost structure and marginal interpretation

The original paper found that material costs dominate operating cost for the RNA vaccine cases, with consumables becoming especially important in the low-dose saRNA case because less RNA material is required per dose while single-use equipment remains central to the process [1]. CleanCap is also identified as a major material cost component across the modeled vaccine types [1].

Those findings are consistent with the LineLab reconstruction. However, the interpretation changes once the cost breakdown is connected to sensitivities and marginal costs.

A large cost category is not always the best optimization target. Some large costs scale mechanically with output and may be difficult to change without altering the process. Other costs may be smaller in absolute terms but more important because they control batch count, flow time, utilization, or storage requirements. For production system design, the relevant question is therefore not only where the money is spent, but which input changes the system when it moves.

A variable-cost view is useful because it grounds the discussion of UTP, CleanCap, and consumables in the modeled system rather than treating materials as a generic category. A capital-cost view adds a complementary perspective: equipment cost still matters, but at the modeled rate it may be less dominant than material and system-use assumptions.

Sensitivities in a system context

The sensitivity analysis is the center of the reconstruction because it shows which parts of the system are worth paying attention to. Rather than interpreting each chart in isolation, the sensitivity view connects cost, flow, and scale assumptions to decision-making.

This technical result connects the cost breakout to decision-making. Material costs matter, but not every material or consumable has the same leverage. Biological parameters such as titer and scale parameters such as working volume affect the system in less direct ways because they change batch counts, utilization, and the relationship between upstream and downstream resources.

One of the useful observations from the reconstruction was that some scenario differences that look important in a process model have a smaller system-level effect once the full configuration is considered. This does not mean that titer or dose is irrelevant. It means that their impact depends on the surrounding system architecture. A design that is constrained by one step may respond differently to titer than a design constrained by another. A material-cost reduction may have a different marginal value depending on whether the system is producing high-volume vaccine doses or individualized therapeutic batches.

This is also where LineLab’s morphological matrix approach becomes useful. In the original workflow, different reactor-volume alternatives and dose scenarios required separate model configurations. In LineLab, alternatives can be orchestrated within a structured comparison framework. The value is not only speed. It is that the comparison is performed against a consistent set of assumptions and can expose which architectural differences actually matter.

Why personalized mRNA therapies change the production problem

The vaccine case is a high-volume and relatively low-mix production problem. Personalized cancer vaccines and individualized neoantigen therapies are very different. Moderna and Merck’s mRNA-4157/V940 program is designed as an individualized neoantigen therapy in which a patient-specific set of tumor neoantigens is encoded into mRNA and administered with pembrolizumab [6][7]. BioNTech’s individualized programs, including autogene cevumeran, similarly depend on patient-specific design and manufacturing workflows [5][8].

The clinical literature around these therapies is already beginning to show the manufacturing implications. In a pancreatic cancer study of autogene cevumeran, vaccine production turnaround was reported as part of the end-to-end clinical workflow, with manufacturing feasibility treated as a central part of the platform [8]. BioNTech’s more recent individualized mRNA vaccine work reported average production turnaround from sample receipt to vaccine release, showing that the production system is part of the therapeutic concept rather than a downstream detail [9].

From a process perspective, the same unit operations remain recognizable: sequence design, template preparation, IVT, purification, formulation, release testing, and fill-finish. From a production system perspective, the operating regime changes. Instead of producing one product repeatedly, the system must coordinate many patient-specific products, each with its own release path, sequencing inputs, production order, and clinical deadline.

That shift changes what should be optimized. Large-batch economics become less important than turnaround time, schedule robustness, routing flexibility, and the ability to run many small batches without excessive idle time or excessive intermediate inventory. A system can have excellent nominal process economics and still fail operationally if patient-specific flows interfere with each other or if QC release timing becomes the practical bottleneck.

Prior art around small-batch and individualized manufacturing

The broader literature around personalized cancer vaccines and advanced therapies points toward several recurring manufacturing concepts. These include modular GMP production, closed processing, robotic or automated handling, parallel small-scale IVT, flexible purification, rapid LNP formulation, and regional or decentralized production models. These ideas are not unique to one company and should be treated as part of the public prior art around individualized manufacturing systems.

Multiply Labs is an example of a company exploring robotic automation for cell therapy manufacturing workflows, with the goal of reducing labor intensity and increasing parallelization in individualized production [10]. Although cell therapy is not the same as mRNA drug substance production, the production-system problem is related: many small, high-value, patient-specific workflows must be coordinated under strict quality and timing constraints. Work adjacent to engineered immune-cell therapies, including regulatory T-cell engineering, also illustrates why manufacturing architecture matters when the product is individualized and process-sensitive [11].

For mRNA specifically, reviews of therapeutic mRNA delivery and mRNA-LNP cancer vaccines describe the process logic and formulation challenges, but generally do not provide production-system models that evaluate routing, queueing, WIP, or resource interaction at rate [3][4][12]. That leaves a clear space for system-level modeling. The prior art establishes that the therapeutic platform exists, that individualized workflows are clinically relevant, and that manufacturing turnaround is a real constraint. The production-system question is how to design the operating architecture that can make these therapies practical at scale.

What this case study validates

This case study is useful because it sits at the intersection of a mature modeling tradition and an emerging production-system problem. The underlying paper is strong precisely because it was built with SuperPro Designer, disclosed its assumptions, and made the model files available. That made it possible to reconstruct the system without relying on confidential industrial data.

The LineLab reconstruction validates several things. First, public process models can be translated into system-level production models without using proprietary data. Second, the resulting model can expose sensitivities, marginal costs, utilization, WIP, and flow-time behavior that are difficult to interpret from process-level cost estimates alone. Third, the same modeling framework can be extended from high-volume vaccine production toward high-mix individualized therapy production, where the need for system-level analysis becomes even stronger.

It also provides a useful boundary for future work with industrial partners. Public information can support a detailed discussion of mRNA process structure, production-system modeling, uncertainty propagation, and general design tradeoffs. Confidential partner data would refine the inputs, parameter ranges, actual process constraints, and proprietary operating assumptions. The capability to model these systems does not depend on receiving confidential information first.

Closing thoughts

The mRNA manufacturing literature already contains strong process models and high-quality techno-economic analyses. The opportunity is not to replace that work, but to extend it into production system design.

That extension matters because the next generation of mRNA applications will place more pressure on flow behavior than the first wave of pandemic vaccine production did. Personalized cancer vaccines, individualized neoantigen therapies, and other small-batch mRNA products require systems that can handle variability, routing complexity, release timing, and intermediate storage economics without losing control of cost or turnaround time.

The central design question is therefore not only whether an mRNA process can be run. It is how the production system should be structured so that the process can be run repeatedly, flexibly, and economically under real operating conditions.

---

References

1. Kis, Z. et al. “Resources, Production Scales and Time Required for Producing RNA Vaccines for the Global Pandemic Demand.” Vaccines (2021).
   https://www.mdpi.com/2076-393X/9/1/3

2. Intelligen. “SuperPro Designer.”
   https://www.intelligen.com/superpro-designer/

3. Pardi, N. et al. “mRNA vaccines — a new era in vaccinology.” Nature Reviews Drug Discovery (2018).
   https://www.nature.com/articles/nrd.2017.243

4. Kowalski, P. S. et al. “Delivering the Messenger: Advances in Technologies for Therapeutic mRNA Delivery.” Molecular Therapy (2019).
   https://www.cell.com/molecular-therapy-family/molecular-therapy/fulltext/S1525-0016(19)30377-3

5. BioNTech. “Oncology Pipeline.”
   https://www.biontech.com/int/en/home/pipeline-and-products/pipeline.html

6. Moderna. “Pipeline.”
   https://www.modernatx.com/pipeline

7. Weber, J. S. et al. “Individualised neoantigen therapy mRNA-4157 (V940) plus pembrolizumab versus pembrolizumab alone in resected melanoma.” The Lancet (2024).
   https://pubmed.ncbi.nlm.nih.gov/38246194/

8. Rojas, L. A. et al. “Personalized RNA neoantigen vaccines stimulate T cells in pancreatic cancer.” Nature (2023).
   https://www.nature.com/articles/s41586-023-06063-y

9. Sahin, U. et al. “Individualized mRNA vaccines evoke durable T cell responses.” Nature (2026).
   https://www.nature.com/articles/s41586-025-10004-2

10. MIT News. “Multiply Labs brings robotic automation to cell therapy manufacturing.”
    https://news.mit.edu/2021/multiply-labs-cell-therapy-0512

11. RegMedNet. “Developing therapeutic regulatory T cells: an interview with Leonardo Ferreira.”
    https://www.regmednet.com/developing-therapeutic-regulatory-t-cells-an-interview-with-leonardo-ferreira/

12. Cao, L. et al. “Key Clinical Frontiers of mRNA-Loaded Lipid Nanoparticles in Cancer Vaccines.” International Journal of Nanomedicine (2025).
    https://www.dovepress.com/key-clinical-frontiers-of-mrna-loaded-lipid-nanoparticles-in-cancer-va-peer-reviewed-fulltext-article-IJN