What Assays Cannot Tell You About a Bioprocess
We engineer biological systems in one context and evaluate them in another. A genetic modification is designed at the bench, characterized under controlled conditions, and then introduced into a bioprocess where the environment, the competition, and the metabolic demands are fundamentally different. The assumption is that if the modification works in characterization, it will work at scale. When it does not, we treat that as a failure of the organism or the process — not as a failure of how we evaluated it.
But the evaluation itself is the problem. Characterization typically relies on endpoint assays and forced measurement — techniques that impose a controlled excitation on the system and read the response. This works when the question is narrow: does this gene express? Does this pathway activate under induction? It does not answer the question that matters for a bioprocess: how will this modification behave inside a living, shifting, competitive system over time?
That question cannot be answered reactively. The current paradigm waits until the engineered system is placed into the bioprocess, then monitors outputs — yield, growth rate, product titer — and adjusts. By the time a problem becomes visible in those outputs, the cause is upstream and already integrated into the system’s behavior. The data says the batch underperformed. It does not say why.
The obvious response is to test at smaller scale — a bench-top fermenter, a miniaturized reactor. This helps. Smaller scale reduces cost and cycle time. But if the measurement at that smaller scale is the same — endpoint assays, forced electrochemical techniques, output-level monitoring — then the scale changed but the observability did not. A smaller reactor running the same reactive measurement still tells you what happened. Not what is happening.
What is missing is not a smaller version of the same observation. It is observation that does not force a response from the system but captures the signal the system is already producing. Living systems generate continuous electrochemical and metabolic signal as a function of what they are doing. That signal contains structure — adaptation, stress response, population dynamics, metabolic shifts. Current measurement either filters it as noise or never captures it because the sensing method requires an applied excitation that overwrites it.
If that signal were captured continuously, without disrupting the biology, prediction becomes possible. Not prediction imposed from historical data — prediction that emerges from the system’s own trajectory. One guesses where the system will go. The other reads where it is heading.
This is the layer that bioprocess development is missing — not more measurement of the same kind, but a different relationship between the sensor and the biology. One where engineering decisions are informed by behavior, not just outcomes — while we are still in the position to change them.
— Pegah Farr