biologic process monitoring
Biologic Process Monitoring: How to Build a Measurement Plan for Research and Process Development
Orient the monitoring problem around its intended use
A monitoring plan is useful when each measurement supports a defined process question, not when it simply produces more data. In biologic development, that question may concern process understanding, detection of drift, operational control, or a decision that must be supported by documented evidence. Good bioprocess control requires an adequate monitoring methodology, and the methodology has to match the state, attribute, or parameter you need to understand.
Quality by Design begins with predefined objectives and emphasizes product and process understanding and control using sound science and quality risk management. A Quality by Design approach identifies the Quality Target Product Profile, Critical Quality Attributes, critical material attributes, and critical process parameters, then links critical material attributes and critical process parameters to Critical Quality Attributes. That sequence gives you a practical starting point: define the intended use before selecting a sensor, sampling route, analytical method, or model.
Use the following chain to frame the work:
- Decision or question: What do you need to understand, detect, control, or decide?
- Measurand: Which process variable, biological attribute, chemical attribute, or quality-relevant outcome could answer it?
- Access and timing: Where can you sample or measure, and how quickly must information become available?
- Evidence standard: What reference method, calibration coverage, uncertainty estimate, qualification, or validation is appropriate for the intended use?
- Response: What alert, investigation, adjustment, hold, or escalation follows from the result?
The chain prevents a measurement from becoming an isolated data stream. It also makes scope, cost, and implementation questions easier to separate from the analytical question itself.
Separate monitoring objectives before choosing variables
Monitoring objectives determine what level of information you need. Process understanding may call for broad observations across development runs; drift detection requires a way to recognize departure from expected behavior; control requires timely information connected to an operating response; and a formal decision requires defined evidence, limits, and records. These objectives can overlap, but they should not be treated as interchangeable.
Real-time monitoring provides information during the process, gives direct insight into bioprocess states, supports early detection of drift from desired process-parameter values, and allows action while the process is running. That capability matters only when the timing of the result can influence the intended response. If a result will be reviewed after the relevant process opportunity has passed, a high-frequency stream may add complexity without changing the decision.
A Critical Process Parameter is a process variable whose variability affects a Critical Quality Attribute and should therefore be monitored or controlled to help ensure the desired quality. Start by asking whether a candidate variable has a defined relationship to the attribute or process behavior under consideration. Then state whether the measurement is intended for explanation, detection, control, or a decision. That purpose determines the required timing, precision, data review, and response pathway.
Before choosing a variable, write one sentence that answers each question:
- What process state or change are you trying to observe?
- What would count as a meaningful deviation?
- How much time would remain to respond after detection?
- Which decision would use the result?
- What evidence would make the result suitable for that decision?
A plan that cannot answer these questions is not ready for method comparison.

Build the variable set from process behavior and quality relevance
A variable set should reflect the process and its quality questions rather than a universal panel. Organize candidate measurements into four groups:
- Operating conditions: composition, pH, temperature, pressure, agitation, aeration, feed conditions, or other variables that describe the process environment.
- Biological state: observations related to cell or organism behavior, growth, viability, metabolism, or the state of the biocatalyst.
- Chemical attributes: concentrations or other chemical properties that describe substrates, metabolites, products, impurities, or the surrounding matrix.
- Quality-relevant attributes: measurements connected to the product profile or to a Critical Quality Attribute.
The final set should be narrowed by the process question, the available sampling route, and the decision that the data must support.
The biocatalyst is sensitive to environmental conditions; even a small uncontrolled change in composition, pH, temperature, or pressure can alter cell metabolism and change process efficiency and productivity. This does not mean that every condition must be measured continuously. It means that you should identify which conditions can plausibly explain the process behavior you need to understand, then connect those conditions to an appropriate observation and response.
pH illustrates why a process variable needs a quality context. Keeping pH in the correct working range affects cell viability and product quality. Too low a pH can negatively influence monoclonal-antibody glycosylation and can result in loss of final-product bioactivity. If pH appears in your plan, state the working range, the process stage, the measurement conditions, and the quality or biological question that makes the range relevant; do not treat the variable as meaningful without its process context.
Defining Critical Quality Attributes supports selecting, developing, and optimizing methods for measuring a biologic throughout its lifecycle. Use the CQA discussion to distinguish a direct quality measurement from a process variable that may influence quality. The distinction helps you avoid presenting a correlated process signal as though it were a direct product attribute.

Compare measurement routes by when and where information appears
Measurement route determines how a sample reaches the analytical method and how quickly information becomes available. PAT may use on-line, in-line, off-line, or at-line analysis, depending on where sampling and measurement occur. Compare the routes by access, timing, sample handling, and the decision they can realistically support.
The four routes can be distinguished at a high level:
| Measurement route | Where measurement occurs | Timing implication | Main planning question |
|---|---|---|---|
| Off-line | A transported sample is measured manually in another laboratory | The result follows sampling and laboratory analysis, so a delay is introduced | Can the decision wait for the analytical result? |
| At-line | A manually collected sample is measured next to the process | The result remains dependent on sampling and manual measurement, but the measurement is near the process | Can the sampling and measurement schedule provide useful visibility? |
| On-line | A process sample is cycled to an analytical tool | The route can provide real-time data while adding sample-transfer and conditioning requirements | Can the sample path preserve representativeness and timing? |
| In-line | The analytical tool measures in the bioprocess | The route can provide information as the process occurs, subject to installation and method conditions | Can the tool operate in the process environment and measure the intended quantity? |
Off-line monitoring can provide high-precision measurements, but it introduces a delay between sampling and analysis. Physical and chemical Critical Process Parameters are commonly monitored with in-line or on-line process sensors or at-line process analyzers. Treat those statements as starting points, not automatic method selections: your process conditions, sample properties, required response time, and evidence standard still determine whether a route is suitable.
For each candidate route, record the sampling point, transport path, conditioning steps, measurement frequency, expected delay, maintenance requirements, and decision enabled by the result. This comparison exposes a common mismatch: a method may be analytically capable but operationally too slow, or fast enough but unable to obtain a representative sample.
Test whether the sample represents the process
Sample representativeness depends on sampling location, mixing, aggregates, settling, carryover, matrix effects, and sample conditioning. Treat the sampling interface as part of the measurement method rather than as plumbing around the instrument. A result can be precise for the collected sample and still fail to describe the process stream if the collection point or handling path changes what reaches the detector.
Aseptic sampling protects the culture from microbial contamination, while carryover from an earlier sample can influence a later measurement even when the culture remains uncontaminated. Miniaturized sampling systems may preserve bioprocess sterility while providing an analytical system with a liquid sample. When you compare such systems, consider cell sensitivity to shear stress and the conductivities and dielectric constants of the cells and cultivation medium.
Sample conditioning can determine whether the analytical principle sees the intended material. For online HPLC, cells and debris require sample conditioning because they are incompatible with most HPLC columns. Document filtration, dilution, separation, temperature control, mixing, hold time, and any surface or tubing contact that could change the sample before analysis.
Use this sampling review before comparing instruments:
- Identify the process location and explain why it represents the material of interest.
- Describe mixing, settling, aggregates, and expected concentration gradients.
- Define the collection frequency and the time between collection and measurement.
- Identify contamination, carryover, fouling, shear, and matrix risks.
- Specify every conditioning step and its effect on the target measurement.
- Compare the sample presented to the method with the sample used by the reference method.
A sampling design is ready for feasibility work only when these conditions are explicit.
Distinguish an observed signal from an actionable estimate
A spectrum is not the same as a model-derived concentration estimate. The spectrum is an observed signal; the estimate depends on calibration data, preprocessing, model structure, reference results, and the process conditions represented in the model. Compare estimates with a reference method and examine sample preparation and handling before treating an estimated value as a process measurement.
Calibration samples should span the expected concentration range, including its extremes, and process conditions should match the model’s evidence. Model calibration is important to validation and future prediction, while validation statistics can be interpreted only when the calibrated model and calibration data are sound. Calibration and validation data requirements also depend on model type; a mechanistic model requires a much smaller dataset than a hybrid or chemometric model.
Evaluate model predictiveness, uncertainty, interpretability, and process-specific behavior. Measurement uncertainty should be fit for purpose; a Singapore technical guide dated 3 November 2025 states that every medical laboratory is responsible for estimating it. For your plan, define what uncertainty means for the intended decision and identify the conditions under which the estimate should be withheld, investigated, or reported with a qualification.
One fermentation study collected near-infrared spectra at 20 g/L biomass over 20 h and found no signal drift between near-infrared predictions and off-line analytical responses attributable to optical-window fouling. That result belongs to the reported study conditions. Use it as an example of a fouling question to test, not as a general performance expectation for every vessel, matrix, biomass concentration, or operating duration.

Check whether the evidence travels with the process
A laboratory result supports the conditions tested; it does not automatically support every pilot, pressurized, or plant condition. Before transferring an analytical method or model, compare the laboratory and process environments, including scale, pressure, temperature, matrix, sampling path, concentration range, vessel geometry, and operating strategy. Any material difference becomes a boundary to investigate.
Calibration-model transferability varies by analyte. The study reports good transferability for glucose, lactate, glutamine, and glutamate models, while transferability for viable cell density and titer was more difficult. The study also reports clone-related bias in titer models and describes a strategy that did not calibrate generic titer models while succeeding for glucose and lactate. Treat transferability as an analyte- and process-specific question rather than as a property of spectroscopy in general.
Batch-retraining models can perform poorly when new feeding strategies and process conditions produce cell-growth behavior that differs significantly from historical data. A change in feed, clone, medium, scale, equipment, or operating regime should therefore trigger a boundary review. Do not assume that historical model performance describes a changed process.
Use a transfer review with four explicit tests:
- Do calibration samples span the expected concentration range, including its extremes?
- Do the process conditions match the model’s evidence?
- Does the sample path preserve the conditions represented in calibration?
- Does the model remain interpretable and sufficiently predictive for the new decision?
If an answer is uncertain, define a feasibility or qualification activity rather than treating the result as established process knowledge.
Set the evidence threshold for the intended decision
Exploratory monitoring and decision-support monitoring have different evidence thresholds. Exploratory data can help you understand relationships, identify candidate variables, and shape later experiments.
Single-lot monitoring reviews data associated with a lot before release; a data point beyond a control limit can cause a nonconformance that must be accounted for in lot disposition. Across-lot monitoring analyzes process-performance data for trends and can use statistical process-control tools such as run rules, control rules, or run tests. State which of these uses applies before selecting charts, review frequency, or alert logic.
Alert limits indicate potential drift from acceptable operating limits and may require frequent monitoring. Action limits indicate apparent drift and require predetermined action, including investigation. Specifications define a range of values; exceeding it makes the process, product, equipment, environment, or utilities unacceptable for use. Keep alerts, actions, and specifications distinct in the monitoring plan because they carry different responses.
Validation records should preserve training sets, versioned coefficients, and exact runtime settings so the basis for a decision can be reconstructed. A described validation process tests accuracy, precision, specificity, range, and robustness to assess whether a method is reliable, reproducible, and suitable for its intended purpose. Define the intended use before assigning a qualification or validation pathway.
The first set of run rules was published in 1956, and those rules divided a control chart into six segments using one, two, and three standard deviations above and below the centerline. The historical detail is less important than the planning principle: trend rules need defined interpretation criteria and a predetermined response.
Turn the evaluation into a usable technical brief
A technical brief should let a qualified process, analytical, engineering, quality, or instrument team understand the question without reconstructing the process from scattered notes. Write the brief in the same order that the measurement will be made and the result will be used.
Include these items:
- Process question and intended decision: State whether the purpose is understanding, drift detection, control, lot review, trend review, qualification, or another defined use.
- Process context: Describe organism or cell system, medium or matrix, vessel, scale, operating mode, feed strategy, temperature, pressure, pH, aeration, agitation, and relevant process stages.
- Target variables and ranges: List each measurand, expected range, concentration extremes, units, sampling frequency, and timing requirement.
- Sampling conditions: Identify the location, sample path, mixing state, conditioning, sterility controls, carryover risks, and hold time.
- Reference method: Name the reference measurement, its reproducibility, sample preparation, and the conditions under which it is considered suitable.
- Model evidence: State the model type, calibration population, preprocessing, validation design, uncertainty, interpretability, and known boundaries.
- Scale and transfer evidence: Identify which scales, clones, matrices, vessels, and operating strategies are represented in the data.
- Data and decision path: Describe storage, visualization, alert thresholds, action thresholds, review ownership, and the response to missing or suspect data.
Crossvalidation splits an original dataset into training and validation datasets and uses the validation data to predict responses. It is often applied to smaller datasets, such as sample size < 10, during initial model development and validation. Include samples at different levels, such as scales, in training and validation datasets to support model reliability and assess sources of variation. Crossvalidation does not remove the need to define representative samples, an appropriate reference method, and a process boundary.
At Paeonia Innovations, we provide continuous measurement through an OrionIR 0.1 mL flow cell generating one spectrum per second. If that capability is part of your feasibility question, include the process stream, sampling arrangement, required measurands, expected ranges, and intended decision rather than treating acquisition frequency as proof of application fit.
Define how monitoring results will change action
A monitoring plan becomes operational only when it defines what happens after each meaningful result. Establish protocols that define how information will inform decisions, including thresholds for alerts and actions based on observed trends. Real-time monitoring can use fault detection, classification, or soft sensors, and multivariate spectroscopic models can monitor glucose and lactate in bioreactors.
Map each condition to a response without implying that the measurement alone authorizes the response:
- Expected result: Continue the defined process and retain the record.
- Alert condition: Increase review or monitoring as specified by the protocol and assess whether the signal represents drift.
- Action condition: Follow the predetermined action, including investigation where required.
- Uninterpretable result: Hold the result for the affected decision and investigate the measurement path, sample, model, or process.
- Potential specification exceedance: Follow the applicable disposition and escalation procedure rather than making an ad hoc decision from the data stream.
Alert limits may require frequent monitoring, while action limits require predetermined action, including investigation.
Multivariate monitoring can assess process and equipment performance, detect deviations in product quality, and initiate material discharge or segregation. Out-of-specification discharge can alternatively be managed with soft sensors and residence-time-distribution modelling. These are possible control architectures, not universal recommendations; select an architecture only after defining the material flow, decision authority, model evidence, and applicable quality controls.
Use analytics software that supports real-time visualization and trend analysis for bioprocess monitoring, and consolidate information from multiple sources so relevant details remain available together. The system design should identify missing data, late data, model status, alert ownership, and the record needed to reconstruct a decision.
Seek qualified evaluation when evidence or process behavior changes
Escalate the evaluation when the measurement path, model evidence, or process behavior moves beyond the defined boundary. PAT can use chemometric or statistical models to combine many weak signals into a prediction, and measurement systems should be verified before capability is declared. Measurement uncertainty should be fit for purpose. A result that appears precise but has an unexamined sample, model, or process limitation should not be treated as decision-ready.
Use residual monitoring as part of model maintenance, including review of the spectrum, sample, process conditions, calibration boundary, and reference result. Do not convert the signal into a new concentration estimate without resolving what changed.
Seek qualified process, analytical, quality, engineering, or regulatory evaluation when:
- data are missing, unstable, or inconsistent with the reference method;
- spectral residuals rise or the signal enters an unfamiliar region;
- feed strategy, clone, medium, scale, pressure, vessel, or sampling conditions change;
- a result may affect product quality, safety, compliance, lot disposition, or another consequential decision; or
- the analytical tool cannot be connected to the existing oversight and record systems.
Ensure new analytical tools are compatible with existing oversight systems to avoid data silos and support continuous data flow.
Do not rely on unverified monitoring data for a critical decision or override established controls because a dashboard appears current. Preserve the original signal, sample and instrument conditions, model version, reference result, and decision record while the issue is investigated.
Prepare the technical brief by process stage
For biologic process work, organize the technical brief by development stage and scale: record the variables or attributes of interest, sampling conditions, reference method, and transfer conditions before comparing analytical routes. The practical next step is to assemble a concise technical brief rather than begin with an instrument category.
Prepare the process question, operating context, expected ranges, sampling conditions, reference method, calibration and validation data, model type, transfer conditions, data-system requirements, and intended decision. State which parts are exploratory and which parts may support a qualified or regulated use. A qualified professional should determine individual suitability for your process, method, quality system, and decision context.
Use that brief to define the questions that remain open: whether the sample is representative, whether the method can deliver information on the required timeline, whether the model evidence covers your process, and what response follows from each result. This gives the next discussion a technical scope and a clear boundary.
What should you clarify before discussing biologic process monitoring?
How long does a biologic process monitoring project typically take?
There is no defensible standard timeline. Duration depends on the intended decision, sampling integration, method development, calibration data, model evaluation, data review, qualification, validation, and the number of process conditions or scales that must be represented. Separate feasibility, method development, integration, and qualification or validation milestones instead of assuming one project duration.
What drives the cost of biologic process monitoring?
Cost is project-dependent rather than a fixed instrument price. Major drivers include the measurement route, sampling and conditioning hardware, process integration, reference-method work, calibration and model development, data-system integration, maintenance, data review, qualification, and validation. A useful comparison separates the analytical instrument from the complete measurement chain and the work required to produce decision-ready results.
Can real-time monitoring replace an established reference method?
Not automatically. A real-time signal or model-derived estimate should be compared with a suitable reference method, and its calibration, uncertainty, sample handling, and process boundaries should be evaluated. The model cannot be treated as decision-ready merely because it produces results quickly; the reference method and evidence supporting the intended process conditions remain important.
What should be reassessed when a biologic process changes?
Reassess the model and measurement boundary when the clone, medium, feed strategy, scale, vessel, pressure, sampling path, or other operating conditions change. Historical models can perform poorly when new conditions produce different cell-growth behavior. Check calibration coverage, sample representativeness, model residuals, uncertainty, and reference-method agreement before using the data for a consequential decision.