Soft Sensors in Biopharmaceutical Manufacturing: The Complete Guide

Start with the fundamentals below, then see how hybrid soft sensors work and how teams estimate titer, glucose and quality risk. If you’re choosing a platform, use the soft sensor buyer’s guide or explore the BioMedAna Soft Sensors solution.

Introduction: What If You Could See What Is Happening Between Lab Measurements?

Imagine you are managing a monoclonal antibody (mAb) production run.

Your bioreactor continuously displays dissolved oxygen, pH, temperature, agitation, and gas flow. You can see how these parameters change throughout the process.

But some of the variables you care about most—glucose, lactate, viable cell density, and product titer—are measured only periodically.

You collect samples, send them for analysis, and receive the results later.

What happens between those measurements?

Is glucose declining faster than expected? Are the cells becoming less productive? Is the process beginning to behave differently from successful historical batches?

Traditionally, these questions often require waiting for the next analytical result.

Soft sensors offer a different approach: they estimate difficult-to-measure variables using the process information already available, helping manufacturing teams see emerging changes sooner.

For biopharmaceutical manufacturers, this creates opportunities to improve process monitoring, investigate variability, assess productivity, and make more informed decisions across upstream and downstream operations.

This guide explains soft sensors from the perspective of the people who use them—not the engineers who develop the underlying algorithms.

What Is a Soft Sensor in Biopharmaceutical Manufacturing?

A soft sensor, also called a virtual sensor, is a software-based capability that estimates a process variable using other available measurements and a predictive model.

In simple terms, a soft sensor helps estimate something you cannot conveniently measure continuously.

For example:

A glucose soft sensor estimates glucose concentration between offline laboratory tests.

A titer soft sensor estimates product concentration during cell culture.

A biomass soft sensor estimates cell growth or biomass using available process signals.

A chromatography soft sensor estimates breakthrough behavior during purification.

Soft sensors do not necessarily require new physical instrumentation. Depending on the application, they may use existing bioreactor data, process historians, laboratory results, or Process Analytical Technology (PAT) measurements.

Some soft sensors use machine learning. Others use mathematical relationships based on biology, chemistry, or physics. More advanced approaches combine both.

The fundamental objective remains the same: provide timely, scientifically useful information that would otherwise be unavailable or delayed.

Why Are Soft Sensors Becoming Important in Biologics Manufacturing?

Biological processes are dynamic and inherently variable.

Cells respond to changes in nutrients, oxygen availability, pH, temperature, raw materials, and other process conditions.

These changes can influence cell growth, metabolism, product concentration, and product quality.

At the same time, many biological measurements still rely on offline laboratory testing.

This creates three practical challenges.

Limited visibility between measurements

A manufacturing team may know the glucose concentration at the most recent sample, but not how quickly it has changed since that sample was collected.

A soft sensor can provide an updated estimate using current process information.

Delayed detection of unusual process behavior

An unexpected change in metabolism or productivity may not become apparent until the next laboratory result.

Soft sensors can provide additional trends that help scientists identify potential changes earlier.

Difficulty comparing batches and operating conditions

Two batches may have similar temperature and pH profiles but different biological responses.

Estimated growth, metabolic, or productivity indicators can provide another dimension for understanding these differences.

This is particularly useful for Process Development, MSAT, and manufacturing teams investigating batch variability.

The goal is not to eliminate laboratory testing. It is to make better use of the information available between tests.

What Can Soft Sensors Predict in a Bioprocess?

Different soft sensors address different manufacturing challenges.

The most useful applications depend on the modality, process, available data, and decisions the manufacturing team needs to make.

Upstream bioprocessing

For a mammalian cell culture process, these estimates can provide a more complete view of what is happening inside the bioreactor.

For microbial fermentation, biomass, substrate consumption, and oxygen uptake may be particularly relevant.

Downstream bioprocessing

Soft sensors are not limited to upstream cell culture.

They can also support downstream purification operations.

The suitability of each application depends on the available measurements and whether a reliable relationship with the target variable can be demonstrated.

Product quality prediction

Soft sensors may also estimate selected quality attributes or quality-risk indicators.

Examples include glycosylation-related attributes, charge variant distributions, aggregation risk, and certain impurity measurements.

These applications are typically more challenging because product quality depends on complex biological and manufacturing interactions.

A quality prediction should therefore be clearly distinguished from a measured analytical result and should not automatically replace an approved quality-control method.

For a closer look at biologics monitoring, see how soft sensors estimate titer, glucose and selected quality risks. The hybrid AI guide explains where biological and physical relationships enter those estimates.

How Would You Use a Soft Sensor During an Actual Manufacturing Run?

Consider a typical fed-batch mAb manufacturing process.

You are monitoring a bioreactor during a multi-day production run.

Without a soft sensor

Your manufacturing dashboard shows pH, temperature, dissolved oxygen, agitation, and feed rates.

Glucose, lactate, and titer are measured offline.

Between samples, your understanding of biological performance depends on historical trends, operating parameters, and scientific judgment.

You may not know whether the current batch is following the expected productivity trajectory until additional results become available.

With a soft sensor

Alongside the conventional process measurements, your dashboard displays estimated glucose, lactate, and product titer.

You can compare these estimates against historical batch trajectories and the most recent laboratory measurements.

Suppose the predicted glucose concentration is decreasing faster than expected.

The soft sensor does not necessarily instruct you to increase the feed rate. Instead, it provides additional information for assessing whether the current process is behaving as intended.

You might review the prediction against laboratory measurements, assess uncertainty, examine related process trends, or investigate the feeding strategy.

Similarly, if estimated titer begins diverging from historically successful batches, MSAT can examine whether the difference is associated with cell growth, metabolite behavior, or operating conditions.

The soft sensor acts as an additional source of process intelligence, helping the team identify questions sooner and make better-informed decisions.

What Does a Soft Sensor Dashboard Actually Show?

An effective soft sensor dashboard should make predictions easy to interpret.

For most users, the essential information is straightforward.

Current estimated value: What does the model estimate the glucose, titer, or other variable to be right now?

Historical trend: How has the estimate changed during the batch?

Measured versus predicted: How closely does the model agree with available laboratory results?

Expected operating behavior: Is the current trajectory consistent with historical or scientifically established expectations?

Prediction uncertainty: How much confidence should be placed in the estimate under the current conditions?

Model status: Is the soft sensor operating within its evaluated range, and are the required input measurements reliable?

Example: A titer soft sensor

Imagine an illustrative manufacturing dashboard displaying:

Latest laboratory titer: 2.4 g/L.

Current estimated titer: 2.8 g/L.

Recent prediction trend: Increasing.

Model status: Within evaluated operating conditions.

Next scheduled laboratory measurement: Pending.

The 2.8 g/L value is an estimate, not a newly measured assay result.

Its usefulness depends on the model's demonstrated performance and the uncertainty associated with the estimate.

The scientist can use this information to understand how productivity may be evolving between measurements.

A strong user interface should make it immediately clear which values are measured and which are predicted.

Explore the BioMedAna soft sensor applications and demo.

Can Soft Sensors Really Predict Product Titer and Quality?

Yes, soft sensors can predict selected product and process variables, but performance differs significantly between applications.

For example, estimating glucose concentration using suitable Raman spectroscopy data may be more straightforward than predicting a complex glycosylation profile from routine process measurements alone.

Research has demonstrated the potential of soft sensors for bioprocess monitoring.

A 2025 study published in Pharmaceutics investigated Raman spectroscopy-based prediction of nutrients, metabolites, IgG titer, and cell density in 10-liter CHO cell cultures.

The researchers reported strong predictive performance for glucose and IgG titer, while estimates for some other variables were less accurate.

This illustrates an important point: one successful soft sensor does not mean every biological or quality variable can be predicted equally well.

For end users, the right question is not simply, "Does the platform support titer prediction?"

It is, "How accurately can the platform estimate titer for our process, using our available measurements, under the conditions in which we intend to use it?"

How Do You Know Whether a Soft Sensor Prediction Can Be Trusted?

This is one of the most important questions for any manufacturing user.

A soft sensor should not be trusted simply because its predictions appear on a sophisticated dashboard.

Before using predictions to support operational decisions, teams should understand five things.

Accuracy against actual measurements

How closely do predictions agree with independent laboratory results?

The evaluation should represent the processes and operating conditions where the soft sensor will be used.

Consistency across batches

Does the model perform reliably on different batches, rather than only the historical data used to develop it?

Confidence and limitations

Does the system communicate uncertainty and identify when a prediction may be unreliable?

Process applicability

Was the soft sensor evaluated for the relevant cell line, product, equipment, manufacturing scale, and operating range?

Ongoing performance monitoring

Can users see whether the model continues to agree with laboratory measurements over time?

A reliable soft sensor solution should make these aspects visible and understandable.

A prediction without context, uncertainty, or evidence of accuracy can create false confidence.

What Happens When a Soft Sensor Prediction Is Wrong?

All predictive models have limitations.

A soft sensor may become less reliable because of an instrument problem, missing data, a change in raw materials, a process disturbance, or conditions outside the range used to evaluate the model.

For example, a glucose soft sensor calibrated using one set of media formulations may perform differently after a significant media change.

Similarly, a titer model evaluated at one manufacturing scale may require additional assessment before being used at another scale.

A well-designed soft sensor application should detect or flag situations that may compromise prediction reliability.

Depending on the application, the system should be able to:

Indicate when required input data is missing or unreliable.

Show when operating conditions fall outside the evaluated model range.

Flag significant disagreement between estimates and reference measurements.

Communicate uncertainty and model status.

Support investigation and appropriate fallback to approved monitoring methods.

For GMP manufacturing, operational procedures should define how users respond to unreliable predictions.

The objective is not to promise perfect predictions. It is to ensure that users understand when estimates are suitable for their intended purpose.

How Can Soft Sensors Help Process Development, MSAT, and Technology Transfer?

Soft sensors can provide value throughout the bioprocess lifecycle.

Process Development: Understand process behavior

Process Development scientists can use soft sensors to gain more frequent insight into cell growth, metabolism, productivity, and process responses.

For example, estimated glucose and lactate trajectories may help scientists understand differences between feeding strategies.

Soft sensor estimates can also support the analysis of experimental data and the identification of process relationships.

MSAT: Investigate batch variability sooner

MSAT teams often investigate differences between successful and underperforming batches.

Soft sensors can provide additional process trajectories to help explain those differences.

For example, two batches may show similar dissolved oxygen profiles but different estimated oxygen uptake or metabolic behavior.

These observations may help MSAT identify hypotheses for further investigation.

Technology Transfer: Compare process behavior across sites

Transferring a process between facilities introduces potential differences in equipment geometry, mixing, oxygen transfer, control strategies, and operating conditions.

Soft sensors can support comparisons between source-site and receiving-site process behavior.

However, predictions must be evaluated for the receiving site, particularly when scale or equipment changes materially affect the underlying process relationships.

Manufacturing: Monitor performance between measurements

Manufacturing teams can use soft sensor estimates to assess process trajectories, recognize emerging changes, and support timely review of operating conditions.

Quality: Strengthen process understanding

Quality professionals can use qualified soft sensor outputs as additional evidence in monitoring, trending, and investigations.

The suitability of a prediction for a GMP quality decision depends on its intended use and associated validation and governance requirements.

What Is the Difference Between a Soft Sensor and a Digital Twin?

A soft sensor primarily estimates a variable that is difficult or impractical to measure continuously.

A digital twin represents aspects of a physical process through models and data, potentially supporting simulations, forecasts, and scenario analysis.

Consider a bioreactor.

A soft sensor estimates the current product titer based on available process information.

A digital twin may evaluate how the process could respond to changes in feeding strategy, temperature, or other operating conditions.

When used together, they can provide a more complete understanding of current and possible future process behavior.

Soft sensors help answer: "What is happening now?"

Digital twins can help answer: "What might happen next if conditions change?"

The distinction is not absolute: some soft sensors provide forecasts, and some digital twins perform state estimation. The technologies frequently complement each other.

Do Soft Sensors Require New Equipment or Large Datasets?

Not necessarily.

Some soft sensors work with existing process measurements and historical laboratory results.

Others benefit from instruments such as Raman spectroscopy, NIR spectroscopy, dielectric sensors, or gas analyzers.

The requirements depend on the variable being predicted.

For example, an oxygen uptake estimator may use existing gas analysis and bioreactor measurements, while a Raman-based glucose soft sensor requires suitable spectroscopic data.

Data requirements also vary.

Some applications can benefit from mechanistic models that incorporate established scientific relationships. Others require substantial representative training data.

The important considerations are whether the available data contains sufficient information about the target variable, whether the model performs reliably on independent data, and whether that evidence covers the intended operating conditions.

A manufacturer does not necessarily need a large new digital infrastructure project to begin exploring soft sensors.

A focused feasibility assessment of one high-value variable is often a sensible starting point.

Can Soft Sensors Be Used in GMP Manufacturing?

Yes, soft sensors can be used in GMP manufacturing when appropriate controls, evidence, and procedures support the intended application.

There is an important distinction between using a soft sensor for process visibility and using it to make a critical manufacturing or quality decision.

For example, displaying an estimated glucose trend for additional monitoring is different from automatically changing a feed rate or replacing an approved product-release assay.

The level of assurance should reflect the potential consequences of incorrect predictions.

Users should expect the solution to support relevant requirements such as:

Defined intended use and operating limitations.

Documented performance and appropriate validation.

Traceability of model versions and prediction data.

Reliable access controls and auditability.

Procedures for changes, model drift, and recalibration.

Human review and appropriate manufacturing or quality approvals.

FDA's Process Analytical Technology framework supports scientifically grounded innovations in process understanding and control.

Soft sensors should be implemented within the organization's established quality system and applicable regulatory expectations.

They do not automatically eliminate the need for laboratory testing or formal qualification of analytical methods.

What Should You Look for When Evaluating a Soft Sensor Solution?

When evaluating soft sensor software for biologics manufacturing, the most important questions are practical.

Does it support your process and modality?

A solution designed for mammalian cell culture may not automatically suit microbial fermentation, chromatography, or cell therapy.

Can it use your existing data?

Assess whether it can work with your manufacturing equipment, PAT instruments, laboratory measurements, and historical batch data.

Does it provide scientifically credible predictions?

Look for appropriate model evaluation, measured-versus-predicted comparisons, and clearly described limitations.

Can users understand its outputs?

A useful solution should communicate estimates, uncertainties, trends, and model applicability without requiring users to understand the underlying algorithms.

Does it fit into existing workflows?

Soft sensor estimates are more useful when connected to process monitoring, batch comparison, MSAT investigation, and process simulation.

Does it support governance and lifecycle management?

Regulated manufacturers need traceability, controlled model versions, appropriate approvals, and ongoing performance monitoring.

Can it extend beyond one soft sensor?

Consider whether the solution can support additional applications across upstream, downstream, process development, and manufacturing as needs evolve.

The best solution is not necessarily the one offering the largest number of prediction models.

It is the one that can demonstrate reliable, useful predictions for your processes and fit into the way your teams work.

If you are building a shortlist, compare six options in the bioprocess soft sensor platform buyer’s guide.

How Should Your Organization Get Started?

The most effective approach is to start with a specific process question.

For example:

"Can we estimate product titer between laboratory measurements?"

"Can we detect unexpected glucose consumption earlier?"

"Can we understand why one batch is behaving differently from another?"

"Can we better anticipate chromatography breakthrough?"

Once the business or scientific question is clear, identify the measurements already available and the offline data that can be used to evaluate predictions.

Choose an application where the information could improve a meaningful decision or investigation.

Evaluate the soft sensor using representative historical data, then assess its performance prospectively before expanding its operational role.

Start with a manageable use case, demonstrate measurable usefulness, and expand based on evidence.

How BioMedAna Brings Soft Sensors into Connected Bioprocess Workflows

BioMedAna is a modality-native AI platform for biopharmaceutical development and manufacturing.

Its approach combines biology, physics, mechanistic modeling, process data, statistical methods, and AI/ML to support predictive bioprocess applications.

Rather than treating soft sensors as isolated algorithms, BioMedAna connects them to broader scientific and manufacturing workflows.

Bio Hub connects and contextualizes process measurements, laboratory data, PAT signals, and batch histories.

Bio Twin supports scientific, mechanistic, and hybrid modeling for soft sensing, process prediction, and simulation.

Bio Agents help users investigate process behavior, compare batches, assess trends, and prepare evidence-backed scientific summaries for human review.

Bio OS provides governance, traceability, access controls, and lifecycle management across connected workflows.

For end users, this means the opportunity to move beyond a single predicted measurement toward a more complete understanding of process behavior.

A titer estimate becomes more useful when it can be compared against historical batches, interpreted alongside metabolism and cell growth, and used as evidence within an MSAT or process development workflow.

BioMedAna supports a scientific, explainable, and human-governed approach to AI in biomanufacturing.

Conclusion: Better Visibility. Better Understanding. Better Decisions.

Biopharmaceutical manufacturers already collect significant amounts of process data.

The challenge is that many important biological and quality variables remain difficult to measure continuously.

Soft sensors help bridge the gap between available process measurements and the information scientists and manufacturing teams need.

By providing estimates of cell growth, metabolites, product titer, purification behavior, and selected quality indicators, soft sensors can support earlier process understanding and more informed decisions.

Their value depends on reliable predictions, scientific credibility, clear limitations, and integration into real manufacturing workflows.

The future of soft sensors is not simply about predicting more variables.

It is about giving bioprocess professionals the information they need, when they need it, to understand and improve manufacturing performance.

Discover how BioMedAna combines scientific modeling, AI, and connected bioprocess data to support predictive process monitoring across upstream and downstream manufacturing.