COGs And Biology, Two Endpoints Needlessly At Odds In Advanced Therapy

By Arnaud Deladeriere, Ph.D., Cell&Gene Consulting Inc.

Over the past decade, cell and gene therapy has moved from early academic programs to a growing number of commercial-stage products. As the field matures, attention has increasingly shifted toward industrialization: manufacturing scale, operational efficiency, and cost of goods.

Advanced therapies will only reach broad patient populations if they can be manufactured reliably and at a sustainable cost. However, an important question is beginning to surface alongside this industrial focus: are we optimizing the right variables and, more importantly, are we collecting the right data to be able to do so?

The seventh Cell&Gene Foundry roundtable explored this question through the lens of translational science. The discussion examined how clinical data, product characterization, and manufacturing development can be more closely integrated to inform future programs. Participants also considered how current development models, inherited in part from traditional pharmaceutical paradigms, may not fully reflect the complexity of living cell therapies.

Four themes emerged from the discussion: operational silos that limit learning across development stages, structural limitations in current development models, the potential role of data-driven analytical tools and artificial intelligence, and the need to rethink how economic performance is measured in cell therapy programs.

The Problem: Misplaced Priorities And Structural Data Silos

The focus on reducing COGS is rational. Without credible pathways to sustainable margins, advanced therapies will struggle to attract long-term investment. However, the group questioned whether the industry’s operational focus has narrowed too early around manufacturing cost per run, without sufficiently integrating what determines clinical success.

In many organizations, three parallel streams operate with limited feedback loops:

• Manufacturing teams focus on meeting release specifications and ensuring operational reliability.
• Product development optimizes yield, speed, and robustness within predefined constraints.
• Translational science analyzes clinical outcomes, patient characteristics, and biological correlates of response.

Each function performs well within its scope. The issue arises when these streams do not converge into a coherent, iterative learning system.

Product development may optimize a phenotype it believes to be desirable. Manufacturing may consistently release product that meets specification. Meanwhile, translational analysis may identify correlations between specific product attributes and patient response that never fully inform upstream product decisions. In some cases, translational functions are structurally aligned with medical teams rather than CMC, reinforcing separation rather than integration.

The result is a missing feedback loop. The industry generates large amounts of data but does not consistently close the loop between patient outcomes, product characterization, and process design.

This misalignment has economic consequences. Several early commercial CAR-T manufacturers built capacity based on projected patient volumes that did not materialize. Facilities were designed for scale before product and market realities were sufficiently validated. Capacity was later reduced and capital redeployed. The lesson is not that scale was misguided, but that scale often preceded biological precision.

At a deeper level, the group questioned whether traditional pharmaceutical development logic has been inappropriately overlaid onto living cell therapies. In small molecules, the molecule is the molecule. In cell therapy, the product is dynamic, patient-dependent, and sensitive to subtle process variables. Treating it as a static entity introduces flawed assumptions.

The Root Cause: A Development Model Under Strain

Beneath the operational silos lies a broader structural issue: the development model itself.

Several recurring patterns were identified.

Recycled science and institutional memory loss

Biological insights in T cell biology are frequently rediscovered rather than systematically integrated into development programs. Mechanisms such as Fas-ligand mediated cell death, T cell exhaustion, and the impact of starting material quality are not new. Yet they are often reframed as new discoveries when revisited in different contexts.

This reflects both fragmentation and talent attrition. As capital retreated from the sector over the past several years, experienced professionals moved to adjacent modalities such as ADCs, radiopharmaceuticals, or biologics. Institutional knowledge dissipated. As new companies form, similar mistakes risk being repeated.

The impact of starting material quality

CAR-T therapies have historically been deployed in late-line settings. Patients entering treatment may have undergone multiple prior therapies. Their T cells are often exhausted or senescent.

The biological principle is straightforward: a healthier starting material tends to yield a healthier final product. Yet the economic implications are less frequently discussed. Poor starting material can drive longer expansion times, higher failure rates, and increased variability. All of these increase the cost per successful outcome.

At the same time, patient selection is constrained by trial design, ethical considerations, and recruitment realities. Sponsors cannot always choose ideal patients. This creates tension between biological optimization and practical enrollment.

Funding-driven acceleration

The pressure to reach the clinic rapidly remains a defining feature of cell therapy startups. Milestones such as IND submission are often tied directly to financing events. Eighteen- to 24-month timelines from concept to clinic are not uncommon.

Under these constraints, companies rely on informed assumptions rather than exhaustive optimization. Cytokine concentrations, seeding densities, culture vessels, and expansion strategies may be selected based on precedent and feasibility rather than deep comparative data sets. Once clinical programs begin, changes become more difficult and comparability requirements limit flexibility. The train has effectively left the station.

This dynamic reinforces the use of a “good enough” process rather than a data-optimized one.

The CMC capability gap

Another structural challenge lies in organizational design. Historically, some academic spinouts prioritized scientific founders while underweighting early CMC leadership. In advanced therapies, this imbalance is particularly costly. Development is not a downstream afterthought. It is central to product definition.

Boards and investors are increasingly recognizing this. Experienced CGT CMC leaders who have navigated prior failures bring critical pattern recognition. However, the pool of such individuals remains limited.

The overarching conclusion was clear: the current development model, borrowed from traditional modalities and compressed by funding realities, does not align well with the biological complexity of living cell therapies.

The Opportunity: Data-Optimized Scalability And The Role Of AI

If the model is under strain, what replaces it?

The discussion converged on the concept of data-optimized scalability. Instead of scaling first and refining later, scalability should be designed from the outset to integrate biological and clinical data.

In practice, this means several shifts.

Integrating multidimensional data

Cell therapies generate complex data sets: starting material characteristics, manufacturing parameters, phenotypic markers, cytokine profiles, persistence metrics, patient disease state, prior lines of therapy, and clinical outcomes. The interactions between these variables are multidimensional and often nonlinear.

Expecting small teams to intuitively navigate hundreds of interacting parameters is unrealistic. Even assembling large cross-functional groups does not eliminate cognitive limitations. This is where advanced analytics and AI platforms may provide leverage.

Emerging platforms are beginning to structure data sets across more than 100 defined variables, connecting process inputs to product attributes and clinical endpoints. These systems do not replace biological reasoning. Rather, they enable systematic exploration of relationships that would otherwise remain obscured.

For example:

• How does cryopreservation of leukapheresis material or an in-process intermediate affect the downstream phenotype?
• How do different prior treatment burdens impact manufactured drug quality beyond CQAs?
• Which combinations of memory phenotype, CD4:CD8 ratio, and exhaustion markers most strongly correlate with durable response?

These are not questions that can be resolved by isolated experiments alone. They require structured, cross-study data integration. Additionally, they require the entirety of the process to be completed, as often a change in the process has an impact on the biology of the final product. This increases cost and development time, but it is critical to fully defining process impacts.

AI as accelerator, not replacement

Importantly, AI does not replace human judgment. Predictions still require experimental validation and biological plausibility. The role of data-driven tools is to narrow the experimental field, reduce unnecessary iterations, and highlight promising parameter spaces.

In a funding-constrained environment, this can provide meaningful leverage. Instead of investing heavily in broad exploratory experimentation, teams can focus resources on targeted validation of data-driven hypotheses.

Enabling smarter patient selection

Data integration also informs patient selection. Not all products are equally suited for all patient populations. A manufacturing approach optimized for patients with limited prior therapy (e.g., a healthy donor) may not translate to heavily pretreated populations.

If predictive models can link starting material characteristics to likely manufacturability and response, sponsors may be able to define more precise inclusion criteria. This improves not only biological coherence but also economic efficiency.

Implications for in vivo approaches

The discussion briefly extended to in vivo cell engineering strategies. While these platforms are often positioned as distinct from ex vivo autologous approaches, the biological lessons from two decades of T cell manufacturing remain relevant. Delivery challenges may dominate current in vivo efforts, but translational learning will eventually need to inform these modalities as well.

In short, data-optimized scalability reframes development from cost-minimization to knowledge-maximization under economic constraints.

A New Metric: From COGS Per Run To COGS Per Successful Patient Outcome

The final theme addressed a more provocative question: are we measuring the wrong economic variable?

Today, COGS per manufacturing run is a dominant benchmark. In some cases, this metric is simplified further to raw material cost, even though true cost of goods includes labor, facilities, quality systems, testing, storage, logistics, taxes, and overhead. Even then, this captures only part of the story.

Manufacturing success is not guaranteed. Not every leukapheresis yields a releasable product, and not every released product yields a meaningful clinical response. When viewed through this lens, the relevant metric becomes cost per successful patient outcome.

If a program invests substantial capital in manufacturing but produces only a limited number of durable responses, the effective cost per successful outcome can be dramatically higher than the nominal COGS per run.

This reframing has several implications.

COGS per releasable run as an interim step

Before shifting to clinical outcomes, a practical intermediate metric is COGS per releasable run. This incorporates manufacturing failure rates and emphasizes predictability of manufacturability. It encourages deeper understanding of starting material variability and process robustness.

Linking cost to efficacy

Ultimately, the economic sustainability of cell therapy programs depends less on the cost of manufacturing a single dose and more on the probability that the dose will deliver a meaningful clinical outcome. When viewed through this lens, the connection between translational science and manufacturing strategy becomes central.

Translational data provides the critical bridge between product characteristics and patient outcomes. Clinical trials generate large data sets on phenotype, functional markers, persistence, and patient response. When systematically analyzed and fed back into development, these data begin to reveal patterns.

Certain product attributes correlate with improved persistence or durability. Certain patient characteristics influence manufacturability or treatment success. Over time, these signals can inform both product design and process development.

If this feedback loop is actively integrated into development, it allows teams to progressively refine the product itself. Process parameters can be adjusted to enrich for desirable cell states. Product characterization can evolve toward markers that are more predictive of efficacy. Patient inclusion criteria can be refined to focus on populations most likely to benefit. Each iteration increases the probability that the therapy delivered to the patient has the biological characteristics associated with clinical activity.

The financial implications of this approach are significant. In many programs today, large amounts of capital are deployed to manufacture and administer products that ultimately do not produce a durable clinical response. When response rates remain low, the effective cost per successful patient outcome becomes extremely high, even if the nominal manufacturing cost per dose appears reasonable. From an investment perspective, the key variable is therefore not simply how cheaply a product can be produced but how reliably that product translates into clinical benefit.

Translational learning offers a mechanism to improve that reliability. By using clinical and product data to inform subsequent process design, the development program gradually reduces the number of suboptimal manufacturing runs that generate products with limited therapeutic value. Over time, this approach shifts resources away from producing large numbers of biologically heterogeneous doses and toward generating fewer, more consistently effective products.

In practical terms, this means that improvements in translational integration can have a multiplicative economic effect. A modest increase in response rate can dramatically reduce the effective cost per successful outcome, because the same development and manufacturing investment produces a larger number of meaningful clinical responses. In that sense, translational science becomes a central lever for improving both therapeutic performance and the overall efficiency of capital deployed in advanced therapy programs.

The role of reimbursement models

Performance-based reimbursement models could further accelerate this shift. If payment becomes more directly tied to patient outcomes rather than product delivery, incentives align more clearly around efficacy and durability.

Such models introduce complexity, including risk-sharing and long-term follow-up requirements. However, they may serve as external drivers that reinforce internal discipline around translational integration.

Closing Reflections

The conversation at Cell&Gene Foundry No. 7 highlighted a pattern.

The industry has generated extraordinary scientific innovation. Yet the operating model often separates biology, manufacturing, and clinical learning into parallel tracks. Cost reduction has been prioritized, sometimes before product definition has matured.

A more sustainable path may lie in rebalancing priorities:

• Integrating translational insights into process design earlier
• Elevating experienced CMC leadership within young companies
• Leveraging advanced analytics to navigate multidimensional data sets
• Measuring economic success in relation to patient outcomes, not batch economics

None of these shifts are simple. They require collaboration, data sharing, cultural change, and investor alignment. But as capital becomes more disciplined and expectations more realistic, the environment may now be more receptive to this reframing.

In advanced therapies, we need to think about the product beyond the terms of what leaves the cleanroom. The product is what endures in the patient. Any economic model that fails to reflect that reality will eventually need revision.

Key Takeaways

1. Cost reduction alone will not solve the scalability challenge in CGT.
Lower manufacturing costs are necessary for broader access, but focusing solely on COGS per run risks overlooking the biological characteristics that determine clinical efficacy.

2. Translational science must close the loop between the patient and the manufacturing process.
Clinical data contains critical insights into which product attributes and patient characteristics drive therapeutic success. Integrating this knowledge into process and product development is essential for improving outcomes.

3. The current development models remain poorly adapted to living cell therapies.
Frameworks inherited from traditional modalities often prioritize speed-to-clinic and manufacturing scale before sufficient biological understanding has been established.

4. Data integration and AI-driven tools may enable a more rational development model.
Advanced analytical platforms can help connect complex data sets across starting material, manufacturing parameters, product phenotype, and patient outcomes, supporting more informed product design and process optimization.

5. The industry may need to rethink how it measures economic success.
Programs should increasingly consider the cost per successful patient outcome, linking economic efficiency directly to clinical efficacy.

About The Author:

Arnaud Deladeriere, Ph.D., is principal consultant at Cell&Gene Consulting Inc. Previously, he was head of MSAT and Manufacturing at Triumvira Immunologics, and before that, manufacturing manager at C3i. He received his Ph.D. in biochemistry from the University of Cambridge.