Deeper Process Visibility: Advanced Monitoring Technologies for Modern Biologics Manufacturing

• PAT is enabling real-time manufacturing insight. Integrated sensor platforms, chemometric modeling, and control strategies allow manufacturers to monitor critical process variables and quality attributes continuously rather than relying solely on offline testing.

• Spectroscopic and sensor technologies are expanding process visibility. Techniques like Raman spectroscopy, mid-infrared monitoring, dielectric biomass sensing, and optical probes enable in situ measurement of metabolites, viable cell density, protein concentration, and other key parameters.

• Inline analytics are improving both upstream and downstream control. Continuous monitoring of metabolic activity in cell culture and protein concentration during purification steps provides earlier detection of process deviations and supports faster process development.

• Advanced analytics are essential for interpreting sensor data. Soft sensors, machine learning models, and predictive analytics help translate complex measurement streams into actionable process variables and support the development of digital twin–style process models.

• Remaining challenges include integration and reliability. Detecting contaminants in real time, ensuring sensor robustness, and integrating multiple analytical streams into unified control strategies remain key hurdles for fully data-driven biologics manufacturing.

Biologics manufacturing relies on living systems whose behavior can shift quickly in response to subtle changes in their environment. Mammalian cell cultures, microbial fermentation systems, and other biological production platforms operate through complex metabolic networks that influence growth, productivity, and product quality throughout a manufacturing run. As a result, successful biologics production depends on careful monitoring of both process conditions and the attributes of the molecules being produced. Developers must track multiple critical quality attributes (CQAs) while simultaneously maintaining control over variables such as nutrient concentrations, metabolite levels, cell density, and product titer. These parameters are tightly interconnected, meaning that shifts in one part of the process can propagate through the system and affect final product quality.

Historically, however, much of the information needed to manage these processes has been obtained through offline analytical testing. Samples are withdrawn from the bioreactor, transported to analytical instruments, and analyzed using laboratory methods. While these approaches provide valuable information, they introduce delays between sampling and measurement. Important variables such as biomass, substrate concentration, and product titer are therefore often assessed after the conditions that produced them have already evolved, limiting the ability of operators to respond immediately to process deviations.¹ In complex cell culture systems where metabolic conditions can change rapidly, this delay can constrain both process understanding and operational control.

Process analytical technology (PAT) frameworks have emerged to address these limitations by promoting the use of integrated analytical tools capable of monitoring processes in real time. PAT approaches combine advanced sensing technologies with chemometric modeling and data analytics to provide continuous insight into process behavior, supporting more proactive control strategies and enabling manufacturers to align production more closely with quality-by-design principles.²

Recent advances in sensor technologies, spectroscopic analytics, and computational modeling are beginning to transform how biologics processes are monitored and controlled. Instead of relying solely on periodic sampling, manufacturers increasingly have the ability to observe key process variables continuously, providing deeper visibility into cell culture dynamics and downstream purification performance. This shift toward real-time process insight is helping to support more responsive, stable, and data-driven biologics manufacturing systems.

The push toward deeper process visibility in biologics production has been shaped in large part by the development of PAT. PAT represents a systematic approach to designing, analyzing, and controlling pharmaceutical manufacturing processes through timely measurements of critical quality and performance attributes. Rather than treating analytical testing as a separate downstream activity, PAT integrates measurement directly into the manufacturing workflow, allowing process conditions and product characteristics to be monitored continuously as production unfolds.²

At its core, PAT brings together three interconnected components: advanced measurement systems, data analysis through chemometric modeling, and control strategies capable of translating analytical insight into operational decisions. Measurement technologies generate data on chemical, physical, or biological parameters within the process stream. Chemometric models interpret these complex data sets to estimate variables that may not be directly measurable. Control systems then use this information to adjust process conditions in ways that maintain consistent performance and product quality. This integration of sensing, modeling, and control allows manufacturers to move beyond reactive quality testing toward a more proactive approach to process management.

A range of analytical technologies can function within the PAT framework. Spectroscopic tools are among the most widely applied because they can provide rapid, non-destructive measurements that are compatible with inline or online deployment. Chromatographic systems can also be adapted for near-real-time monitoring of product attributes and impurity profiles during purification steps. In addition, biosensors capable of detecting specific molecular targets are emerging as complementary tools for monitoring critical quality attributes in both upstream and downstream processes.

The increasing sophistication of data analytics is further expanding the capabilities of PAT-enabled manufacturing. Chemometric modeling allows complex sensor signals to be translated into meaningful process variables, enabling the development of predictive models that anticipate process behavior under changing conditions. In some cases, these models can be integrated with digital representations of the manufacturing process — often referred to as digital twins — to simulate system responses and guide operational decisions. Such capabilities support the broader goal of real-time release strategies, in which product quality can be verified through continuous monitoring and process control rather than relying exclusively on end-of-batch testing.^2,3

By providing the analytical and computational infrastructure needed to interpret continuous data streams, PAT creates the organizational framework that allows advanced sensors and inline analytics to be effectively integrated into biologics manufacturing operations.

Among the analytical technologies incorporated into PAT strategies, spectroscopic methods have become some of the most widely adopted tools for real-time monitoring in biologics manufacturing. Their appeal stems from a combination of analytical flexibility, speed, and compatibility with in situ measurement. Spectroscopy allows operators to observe chemical and biological changes within a process stream without requiring sample removal or extensive preparation, making it well suited for integration into continuous monitoring workflows.

Several spectroscopic techniques are now routinely explored for bioprocess monitoring. Ultraviolet–visible (UV/Vis) spectroscopy can provide information about protein concentration and other absorbing species in solution. Near-infrared (NIR) spectroscopy has been applied to the analysis of nutrients, metabolites, and product concentrations within cell culture systems. Mid-infrared (MIR) methods offer more detailed molecular information by probing fundamental vibrational modes of chemical bonds. Raman spectroscopy has emerged as one of the most versatile tools for monitoring complex cell culture environments because it can detect multiple metabolites simultaneously within highly scattering biological media. Fluorescence spectroscopy provides another complementary approach, particularly when specific fluorophores or intrinsic cellular signals can be monitored to track metabolic activity or protein expression.

These technologies share several characteristics that make them well suited for integration into biologics production environments. Spectroscopic measurements are typically non-destructive, meaning that they do not alter or consume the sample being analyzed. This feature is particularly valuable in bioreactor systems where maintaining sterile conditions and avoiding process perturbation are essential. Spectroscopic methods also produce results rapidly, often in seconds or minutes, allowing frequent measurements that can reveal dynamic changes in process conditions. Many spectroscopic sensors can be deployed through immersion probes or flow cells, enabling inline or online measurements that capture the chemical state of the process stream continuously.

Within PAT frameworks, spectroscopic data are frequently paired with chemometric modeling techniques that translate complex spectral signals into quantitative estimates of process variables. Multivariate calibration models can link spectral features to concentrations of metabolites, nutrients, or product molecules, allowing manufacturers to infer biochemical states that would otherwise require more time-consuming analytical methods. By combining these modeling approaches with inline sensors, manufacturers gain access to continuous streams of process information that can support automated control strategies and improved process understanding.

The result is a shift in how bioprocess monitoring is performed. Instead of relying exclusively on periodic laboratory testing, spectroscopic technologies enable manufacturers to observe multiple biochemical variables directly within the production environment. This capability provides a foundation for more responsive process control, improved reproducibility, and deeper insight into the biological systems that drive modern biologics manufacturing.

Among the spectroscopic techniques applied in bioprocess monitoring, Raman spectroscopy has emerged as one of the most widely adopted tools for analyzing complex cell culture environments. Its versatility stems from its ability to generate detailed molecular fingerprints based on vibrational scattering of light, allowing multiple chemical species to be detected within highly heterogeneous biological systems. Over the past decade, Raman methods have been implemented across the biopharmaceutical life cycle, supporting activities ranging from early process development to monitoring of production processes under current good manufacturing practice (cGMP) conditions.⁴

One of the primary advantages of Raman spectroscopy in bioprocessing is its compatibility with direct in situ measurement. Raman probes can be inserted into bioreactors as immersion probes or incorporated into flow-cell configurations, enabling continuous measurement of the culture environment without requiring sample withdrawal. Because the technique can operate through transparent reactor windows or sterile probe housings, it can be deployed in ways that preserve the integrity of closed bioprocess systems while still providing continuous analytical feedback.⁵

This capability has made Raman spectroscopy particularly valuable for monitoring metabolic activity in mammalian cell cultures used for biologics production. Cell metabolism depends on a network of nutrient consumption and metabolite generation that changes throughout the course of a culture run. Traditional analytical methods often measure these parameters through periodic offline assays, which provide snapshots of process conditions at discrete time points. Raman spectroscopy offers a different approach by enabling repeated measurements of key metabolic indicators directly inside the bioreactor. These measurements can then be interpreted using chemometric models that translate spectral features into quantitative estimates of biochemical variables.

A range of important process variables can be monitored using Raman-based methods. Studies have demonstrated the ability to measure nutrient concentrations such as glucose and glutamine, as well as metabolites including glutamate and lactate. In addition to tracking metabolic intermediates, Raman models have been developed to estimate viable cell density and product titer, providing insight into both cellular activity and production performance during culture runs.⁵ Because these measurements can be collected frequently throughout the process, Raman monitoring can reveal dynamic changes in metabolic state that might otherwise go undetected between offline sampling intervals.

For these reasons, Raman spectroscopy is often regarded as one of the leading PAT tools for upstream bioprocess monitoring. Its ability to observe multiple analytes simultaneously using a single analytical platform provides a comprehensive view of the biochemical environment within the bioreactor. When integrated with multivariate modeling and automated control systems, Raman monitoring can support strategies that adjust nutrient feeds, manage metabolic byproducts, and stabilize culture conditions. These capabilities allow manufacturers to move toward a more responsive and data-driven approach to managing cell culture metabolism in biologics production.^4,5

While spectroscopic techniques provide insight into metabolic activity and molecular composition within cell culture systems, dielectric sensing offers a complementary method for monitoring the physical presence of viable cells. Radio-frequency impedance, commonly referred to as capacitance-based sensing, measures the electrical properties of a culture medium to estimate the volume of living cells suspended within it. Because viable cells possess intact membranes that behave as electrical capacitors, changes in electrical impedance within the reactor can be correlated with the concentration of living biomass in the culture.⁶

This approach allows capacitance sensors to provide continuous measurements of viable cell biovolume without requiring sampling or staining procedures. Traditional cell-counting techniques often rely on offline microscopy, automated cell counters, or biochemical assays that require removal of samples from the bioreactor. These methods can introduce delays and may be influenced by sample handling or dilution steps. Dielectric measurements, in contrast, can be performed directly within the process environment, enabling frequent measurements that track how viable cell populations change over the course of a production run.⁶

The ability to observe biomass in real time makes dielectric sensors particularly valuable in applications where cell concentration plays a central role in process performance. Perfusion culture systems, for example, rely on maintaining stable cell densities while continuously exchanging media. Capacitance measurements can support automated control strategies by providing feedback on viable cell concentrations, helping maintain optimal growth conditions. The technology has also proven useful in situations where traditional offline counting becomes unreliable, such as cultures containing cell aggregates or high-density suspensions that are difficult to quantify accurately using conventional methods.

In addition to supporting process development, capacitance-based monitoring has been implemented in cGMP production environments as an additional tool for observing cell culture behavior. By providing direct measurements of viable biomass, dielectric sensors offer a valuable complement to spectroscopic methods that track metabolic indicators. Together, these analytical approaches provide a more complete view of the biological system inside the bioreactor, enabling manufacturers to manage both cellular growth and metabolic activity with greater precision during biologics production.

In addition to spectroscopic and dielectric sensing technologies, optical fiber–based sensors are emerging as promising tools for monitoring key environmental conditions inside bioreactors. These sensors rely on optical signals transmitted through fiber-optic cables to detect chemical or physical changes in the surrounding medium. Because the sensing elements can be positioned directly within the culture environment, optical fiber systems enable in situ monitoring of critical process parameters without requiring sample removal or complex instrumentation inside the reactor vessel.⁷

Optical fiber sensors offer several practical advantages for bioprocess monitoring. Their small size allows them to be integrated into reactor systems with minimal disruption to existing equipment or process configurations. High sensitivity enables detection of subtle changes in environmental conditions that influence cellular behavior. Fiberoptic systems can also transmit signals over long distances, allowing measurement hardware to remain outside the reactor or even outside cleanroom environments while still collecting real-time data from within the process vessel. In addition, the optical nature of these sensors supports multiplexing approaches in which multiple sensing elements can be incorporated along a single fiber, enabling simultaneous monitoring of several parameters within the same system.

Current development efforts have focused primarily on using optical fiber sensors to measure fundamental bioreactor parameters such as pH and dissolved oxygen. Both variables play central roles in regulating cell metabolism, enzyme activity, and protein expression during biologics production. Continuous monitoring of these parameters is therefore essential for maintaining stable culture conditions and preventing shifts that could negatively affect cell viability or product quality. Optical fiber–based sensors can detect changes in pH or oxygen concentration through optical signals generated by specialized indicator chemistries or fluorescence-based detection mechanisms embedded within the fiber probe.⁷

As these technologies mature, optical sensing platforms may expand beyond core environmental measurements to support monitoring of additional biochemical indicators within cell culture systems. Their ability to operate continuously within the reactor environment, combined with their compact footprint and compatibility with sterile manufacturing systems, positions optical fiber sensors as an important component of future process monitoring strategies. When integrated with other PAT tools, such as spectroscopy and capacitance-based biomass monitoring, optical sensors contribute to a more comprehensive picture of the biological and chemical dynamics that govern modern biologics manufacturing.

Although much of the early adoption of PAT focused on upstream cell culture monitoring, increasing attention is now being directed toward analytical visibility during downstream purification. Downstream processing involves a sequence of complex unit operations, including affinity capture, chromatography, and filtration steps, that must consistently deliver high product purity while preserving protein integrity. These operations have traditionally relied heavily on offline analytical measurements to evaluate performance, often requiring samples to be collected and analyzed after a purification step has already progressed. Inline analytical technologies are beginning to change this paradigm by allowing critical process information to be observed continuously during purification operations.

Spectroscopic techniques have proven particularly useful in this context because they can monitor biochemical properties of process streams without interrupting the flow of material through the purification system. By collecting spectral data directly from the process stream, these methods can provide continuous information about molecular composition, metabolite content, or product concentration. When paired with multivariate modeling approaches, spectroscopic signals can be translated into quantitative estimates of process variables and product attributes, enabling operators to monitor purification performance in near real time.⁸

One example of this approach is the use of inline variable-pathlength spectroscopy during affinity capture steps. Affinity chromatography is a central operation in many biologics purification processes, and determining the dynamic binding capacity (DBC) of the chromatography resin is critical for optimizing loading conditions. Traditionally, DBC has been determined through offline titer measurements, which require sample collection and subsequent analytical testing. Inline variable-pathlength technology enables rapid measurement of protein concentration directly within the purification system, allowing DBC to be estimated in real time during process development. Studies have shown that this method can produce results comparable to conventional offline measurements while significantly reducing the time required to evaluate chromatography performance.⁹

Spectroscopic monitoring has also been applied to filtration steps, such as ultrafiltration and diafiltration (UF/DF), which are widely used to concentrate proteins and exchange buffer systems during downstream processing. Mid-infrared (MIR) spectroscopy has been implemented in inline configurations using flow cells to measure protein concentration directly within the filtration stream. Calibration models developed from MIR spectral data can accurately predict protein concentrations when compared with established offline analytical methods. This capability allows operators to track protein concentration continuously during UF/DF operations, providing improved visibility into process progression and enabling more precise control of concentration targets.⁸

By providing real-time insight into purification performance, inline monitoring technologies reduce dependence on delayed laboratory assays and allow process engineers to observe how downstream operations evolve during production. As these analytical capabilities continue to develop, they are expected to support faster process development, improved process control, and more efficient optimization of downstream manufacturing workflows for biologics.

As biologics have grown more structurally complex, the analytical strategies used to characterize them have also evolved. Traditional quality control approaches often rely on multiple orthogonal assays to evaluate individual product attributes such as glycosylation patterns, sequence variants, oxidation states, or fragmentation. While these assays provide valuable information, they can require significant analytical effort and may generate fragmented data sets that must be interpreted across several separate workflows. Advances in mass spectrometry–based analytics have begun to address these limitations by enabling more integrated approaches to product characterization.

One such approach is the multi-attribute method (MAM), which applies liquid chromatography–mass spectrometry (LC–MS) peptide mapping to monitor multiple product quality attributes simultaneously. In this workflow, therapeutic proteins are enzymatically digested into peptides, which are then separated chromatographically and analyzed using high-resolution mass spectrometry. The resulting data sets allow site-specific attributes (e.g., posttranslational modifications or sequence variants) to be quantified across the molecule within a single analytical method. This capability provides a comprehensive view of product quality at a level of molecular detail that is difficult to achieve with conventional assays.¹⁰

A key advantage of MAM lies in its ability to consolidate multiple measurements within a unified analytical framework. Instead of relying on separate methods to measure individual attributes, the LC–MS peptide-mapping workflow can capture many of these attributes simultaneously. In practice, this means that a single MAM analysis can quantify several site-specific modifications while also detecting unexpected or emerging variants through new-peak monitoring. By reducing the number of independent assays required to characterize a product, MAM has the potential to streamline analytical workflows and improve consistency across development and manufacturing activities.¹¹

Another important feature of MAM is its flexibility across the product life cycle. The same analytical framework can be applied during early process development, where detailed characterization of product variants is needed, as well as during later stages of development and commercial manufacturing. Studies have shown that MAM workflows can support applications ranging from development studies and comparability assessments to product release and stability testing. This continuity allows data generated at different stages of the product life cycle to be interpreted within a consistent analytical context, strengthening the connection between development knowledge and manufacturing control strategies.¹²

By enabling simultaneous monitoring of multiple product attributes within a single workflow, MAM represents a shift toward more integrated analytical approaches in biologics development. When combined with other real-time monitoring technologies and advanced data analysis methods, mass spectrometry–based strategies like MAM contribute to a broader trend toward more comprehensive and information-rich analytical frameworks across modern biologics manufacturing.

As biologics manufacturing systems incorporate increasing numbers of sensors and analytical technologies, the challenge shifts from data collection to data interpretation. Continuous monitoring tools can generate large volumes of process data describing metabolic activity, environmental conditions, and product characteristics. Extracting actionable insight from these data streams requires analytical frameworks capable of translating complex measurements into meaningful process variables. In this context, modeling and machine-learning approaches have become essential components of modern process monitoring strategies.

One widely used approach is the development of soft sensors. Unlike physical sensors, which measure variables directly, soft sensors rely on mathematical models to estimate quantities that are difficult or impractical to measure in real time. These models use combinations of sensor inputs, such as spectroscopic signals, temperature measurements, or biomass indicators, to infer variables such as nutrient concentrations, metabolite production, or product formation. By integrating multiple signals into predictive models, soft sensors can extend the analytical capabilities of existing instrumentation and provide continuous estimates of key process parameters.¹³

Developing reliable soft sensors, however, presents several technical challenges. Bioprocesses often progress through multiple phases during a production run, including growth, production, and stationary phases, each with distinct metabolic characteristics. Process durations can vary between runs, making it difficult to apply fixed modeling assumptions across different batches. In addition, errors or drift in underlying sensor measurements can propagate through predictive models, leading to inaccurate estimates if faults are not detected and corrected. These factors require careful model validation and monitoring strategies to ensure that predictive systems remain robust during routine manufacturing operations.

Recent advances in machine learning are beginning to expand the capabilities of soft-sensor approaches. Automated machine learning methods can analyze complex datasets and identify predictive relationships between sensor signals and process variables with reduced manual model development. Such techniques have been explored for real-time monitoring of biochemical variables in biomanufacturing environments and may ultimately contribute to the development of digital representations of manufacturing processes. These digital models, often referred to as digital twins, could simulate process behavior and support predictive control strategies that anticipate changes before they affect product quality.¹⁴

As sensor technologies continue to expand the amount of process data available, advanced analytics will play a central role in converting those measurements into operational knowledge. The integration of predictive models with real-time monitoring systems offers a pathway toward more adaptive, data-driven manufacturing environments capable of responding dynamically to changing process conditions.

Despite significant progress in the development and deployment of advanced sensing technologies, several technical and operational challenges continue to limit the full realization of real-time bioprocess monitoring. Biologics manufacturing environments remain complex, and translating analytical innovation into reliable production tools requires addressing both measurement limitations and broader system-integration issues.

One important challenge involves the detection of biological contaminants during manufacturing. Viral particles, microbial contaminants, and mycoplasma represent critical safety risks in biologics production, yet reliable real-time monitoring technologies for these threats remain limited. Many detection methods for these contaminants still rely on laboratory-based assays that require sample collection and extended analysis times. As a result, even facilities that employ advanced PAT tools for monitoring metabolic conditions and product attributes may still depend on traditional testing approaches for certain critical safety assessments.¹⁵

The integration of advanced analytics also presents ongoing difficulties. While machine learning and other data-driven approaches offer powerful tools for interpreting complex sensor datasets, implementing these methods within regulated manufacturing environments requires careful validation and transparency. Models must demonstrate consistent performance across different batches, process conditions, and equipment configurations. Ensuring that predictive models remain reliable as processes evolve or scale can require continuous evaluation and updating, adding complexity to process monitoring strategies.

Sensor reliability represents another important consideration. Many predictive monitoring systems rely on signals from multiple physical sensors, and errors or drift in these measurements can propagate through analytical models. Faults in a single sensor may therefore compromise the accuracy of derived process variables unless robust fault-detection strategies are implemented. Designing monitoring systems that can recognize and compensate for sensor failures remains an important area of development within PAT-enabled manufacturing.

Finally, the integration of multiple analytical technologies into unified process-control systems remains a complex engineering challenge. Modern biologics facilities may deploy spectroscopic sensors, dielectric biomass monitors, optical probes, and other analytical tools simultaneously. Combining the resulting data streams into coherent control strategies requires sophisticated data management and modeling frameworks capable of interpreting diverse measurements in real time. Addressing these challenges will be essential for translating advances in sensing and analytics into fully integrated, data-driven biologics manufacturing systems.

Biologics manufacturing is undergoing a gradual but meaningful transformation in how process information is generated and used. Historically, much of the analytical insight needed to manage complex cell culture and purification systems has come from offline measurements performed after samples were removed from the process stream. While these methods remain essential for many forms of product characterization and quality testing, they inherently provide delayed snapshots of conditions that may already have changed by the time results are available. The growing adoption of advanced sensors and inline analytics is beginning to shift this paradigm toward more continuous observation of manufacturing processes.

Technologies like spectroscopic monitoring, dielectric biomass sensing, optical probes, and inline spectroscopic measurements in downstream purification now allow manufacturers to observe critical aspects of production as they occur. These tools provide continuous insight into variables that influence cell metabolism, nutrient consumption, product formation, and purification performance. As a result, operators gain improved visibility into the dynamic behavior of biological production systems, enabling earlier recognition of deviations and more informed operational decisions.

At the same time, advances in computational modeling and data analytics are expanding the value of these measurements. Predictive models and soft-sensor approaches can translate complex analytical signals into estimates of variables that are difficult to measure directly, while automated machine learning techniques offer new ways to extract patterns from large process datasets. When combined with continuous sensor data, these analytical capabilities can support more proactive control strategies that anticipate changes rather than simply reacting to them.

These developments point toward a future in which biologics manufacturing operates with far greater process awareness than has historically been possible. As sensor technologies, analytical methods, and predictive modeling tools continue to mature and integrate, they have the potential to support manufacturing environments that are increasingly adaptive, automated, and guided by real-time process intelligence.

1. Mishra, Abhishek, et al. “Spectroscopic Advances in Real Time Monitoring of Pharmaceutical Bioprocesses: A Review of Vibrational and Fluorescence Techniques.” Spectrosc. J. 3: 12 (2025).

2. Sathiyapriyan, Pavithra, et al. “Current PAT Landscape in the Downstream Processing of Biopharmaceuticals.” Anal. Sci. Adv. 6: e70013 (2025).

3. Prasad, Akanksha. “Leveraging Process Analytical Technology for Real-Time Control in Biopharmaceutical Manufacturing.” BioProcess International. 27 Jan. 2026.

4. Esmonde-White, Karen A, Maryann Cuellar, and Ian R Lewis. “The role of Raman spectroscopy in biopharmaceuticals from development to manufacturing.” Analytical and Bioanalytical Chemistry. 414: 969–991 (2022).

5. Graf, Alexander, et al. “Automated Data Generation for Raman Spectroscopy Calibrations in Multi-Parallel Mini Bioreactors.” Sensors. 22: 3397 (2022).

6. Carvell, John P and Jason E Dowd. “On-line Measurements and Control of Viable Cell Density in Cell Culture Manufacturing Processes using Radio-frequency Impedance.” Cytotechnology. 50: 35–48 (2006).

7. Cui, Guoqiang, et al. “Optical Fiber pH and Dissolved Oxygen Sensors for Bioreactor Monitoring: A Review.” Sensors. 26: 10 (2026).

8. Milewska, Adrianna, et al. “In-line monitoring of protein concentration with MIR spectroscopy during UFDF.” Engineering in Life Sciences. 23: e2200050 (2022).

9. Bhangale, Rashmi P, et al. “Application of inline variable pathlength technology for rapid determination of dynamic binding capacity in downstream process development of biopharmaceuticals.” Biotechnol. Prog. 38: e3236 (2022).

10. Yang, Feng, et al. “Mass spectrometry-based multi-attribute method in protein therapeutics product quality monitoring and quality control.” mAbs. 15: 2197668 (2023). /

11. Carvalho, Sofia B, et al. “Multi attribute method implementation using a High Resolution Mass Spectrometry platform: From sample preparation to batch analysis.” PLoS One. 17: e026711 (2022).

12. Sitasuwan, Pongkwan, et al. “Enhancing the multi-attribute method through an automated and high-throughput sample preparation.” mAbs. 13: 1978131 (2021).

13. Brunner, Vincent, et al. “Challenges in the Development of Soft Sensors for Bioprocesses: A Critical Review.” Front. Bioeng. Biotechnol. 9: 722202 (2021). /

14. Kim, Sun-Jong, et al. “AutoML‐Driven Soft Sensors for Real‐Time Monitoring of Amino Acids in Mammalian Perfusion Cultures.” Biotechnol. Bioeng. 122: 3123–3138 (2025).

15. Maruthamuthu, Murali K, et al. “Process Analytical Technologies and Data Analytics for the Manufacture of Monoclonal Antibodies.” Trends Biotechnol. 38: 1169–1186 (2020).