Automatic calibration and maintenance of Raman spectroscopy models for real-time prediction

Abstract

A method for monitoring and / or controlling a biopharmaceutical process includes determining a query point associated with a scan of the process by a spectroscopic system (e.g., a Raman spectroscopic system), and querying an observation database containing observation data sets associated with past observations of the biopharmaceutical process. Each of these observation data sets includes spectral data and corresponding actual analytical measurements. Querying the observation database includes selecting from the observation data sets those data sets that meet one or more relevance criteria about the query point as training data. The method also includes using the selected training data to calibrate a local model specific to the biopharmaceutical process. The local model (e.g., a Gaussian process model) is trained to predict analytical measurements based on spectral data input. The method also includes using the local model to predict analytical measurements of the biopharmaceutical process.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] Priority is claimed to U.S. Provisional Patent Application No. 62 / 749,359 filed on October 23, 2018, U.S. Provisional Patent Application No. 62 / 833,044 filed on April 12, 2019, and U.S. Provisional Patent Application No. 62 / 864,565 filed on June 21, 2019, each of which is hereby incorporated by reference in its entirety. Technical Field

[0003] The present application relates generally to the use of spectroscopic techniques such as Raman spectroscopy to monitor and / or control biopharmaceutical processes, and more particularly to the online calibration and maintenance of predictive models. Background Art

[0004] The stable production of biotherapeutic proteins by biopharmaceutical processes usually requires bioreactors to maintain balanced and consistent parameters (e.g., cell metabolism concentrations), which in turn requires strict process monitoring and control. In order to meet these needs, process analytical technology (PAT) tools are increasingly being used. Online monitoring of pH, dissolved oxygen, and cell culture temperature are some examples of traditional PAT tools that have been used in feedback control systems. In recent years, other process probes have been studied and deployed to continuously monitor more complex species, such as viable cell density (VCD), glucose, lactate, and other key cell metabolites, amino acids, titers, and key quality attributes.

[0005] Raman spectroscopy is a popular PAT tool that is widely used for online monitoring in biomanufacturing. Raman spectroscopy is an optical method that enables non-destructive analysis of chemical composition and molecular structure. In Raman spectroscopy, incident laser light is scattered inelastically due to molecular vibrational modes. The frequency difference between the incident photon and the scattered photon is called the "Raman shift", and the vector of the Raman shift and intensity level (referred to herein as the "Raman spectrum", "Raman scan" or "Raman scan vector") can be analyzed to determine the chemical composition and molecular structure of the sample. With the improvement of laser sampling and detector technology, the application of Raman spectroscopy in polymer, pharmaceutical, biomanufacturing and biomedical analysis has surged in the past three decades. Due to these technological advances, Raman spectroscopy has now become a practical analytical technique used inside and outside the laboratory. Since the first reported application of in-situ Raman measurements in biomanufacturing, it has been used to provide online real-time predictions of several key process states such as glucose, lactate, glutamate, glutamine, ammonia, VCD, etc. These predictions are usually based on calibration models or soft sensor models built in an offline environment based on analytical measurement results from analytical instruments. Partial least squares (PLS) and multiple linear regression modeling methods are commonly used to relate Raman spectra to analytical measurements. These models typically require filtering preprocessing of the Raman scans before calibration against the analytical measurements. Once the calibration model is trained, it can be implemented in a real-time environment to provide in-situ measurements for process monitoring and / or control.

[0006] Since biopharmaceutical processes are usually operated under strict constraints and regulations, Raman model calibration for biopharmaceutical applications is very necessary. The current state-of-the-art method for Raman model calibration in the biopharmaceutical industry is to first conduct multiple activity tests to generate relevant data for correlating Raman spectra with (multiple) analytical measurement results. These tests are both expensive and time-consuming, because, for example, each activity may last two to four weeks in a laboratory environment. In addition, only limited samples can be used for analytical instruments (for example, to ensure that laboratory-scale bioreactors maintain a healthy amount of viable cells). In fact, it is not uncommon to only be able to perform one or two measurements per day in online or offline analytical instruments. To further exacerbate the situation, current best practices produce calibration models related to specific processes, specific recipes or configurations of bioreactor culture media, and specific operating conditions. Therefore, if any of the above variables are to be changed, the model may need to be recalibrated based on new data. In fact, both Raman model calibration and model maintenance require a lot of resource allocation and are usually performed in an offline environment. Although methods for adapting models to new operating conditions have been proposed (for example, recursive methods, moving window methods, and time difference methods), these methods may not be sufficient to handle sudden process changes.

[0007] Many publications describe general Raman models for multiple molecules based on traditional chemometric methods (e.g., PLS modeling). However, these general models assume that these processes use similar (if not identical) culture medium formulations and / or operating process conditions. Culture medium and processes are usually platformed, with little or no difference. The disadvantage of this type of general model is that once the process deviates from the specification, or if the training data set contains a process range that is too large to attempt to consider the differences between different molecules (e.g., culture medium additives, process duration, and / or other process changes), the general model loses accuracy and precision. Therefore, these "general" models are only general within the strict limits described. See Biotechnolo.Prog.31(4) by Mehdizaheh et al.: 1004-1013, 2015 and Biotechnol.Prog.34(3) by Webster et al.: 730-737, 2018. Summary of the invention

[0008] The term "biopharmaceutical process" refers to a process used in biopharmaceutical manufacturing, such as a cell culture process for producing a desired recombinant protein. Cell culture is carried out in a cell culture vessel such as a bioreactor, under conditions that support the growth and maintenance of an organism engineered to express a protein. In the recombinant protein production process, process parameters such as culture medium component concentrations (including nutrients and metabolites (e.g., glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na+, K+ and other nutrients or metabolites)), culture medium status (pH, pCO2, pO2, temperature, osmotic pressure, etc.) and cell and / or protein parameters (e.g., viable cell density (VCD), titer, cell status, key quality attributes, etc.) are monitored to control and / or maintain the cell culture process.

[0009] In order to address some of the above-mentioned limitations of current best industrial practices, the embodiments described herein relate to systems and methods for improving conventional techniques for spectral analysis (such as Raman spectroscopy) for biopharmaceutical processes. Specifically, a "Just-In-Time Learning" (JITL) platform is used to build and maintain calibration models (e.g., Raman calibration models) in real time for biopharmaceutical applications. JITL is a nonlinear modeling platform based on local modeling and database sampling techniques. Unlike other machine learning methods, JITL generally assumes that all available observations are stored in a central database, and the model is dynamically built in real time based on queries using the most relevant data from the database. This allows the use of relatively simple local models to approximate complex process dynamics well. Under the JITL framework, the library can contain more than spectral data for a single process operating under specific operating conditions, and can also contain data for different processes, different culture medium configurations, and / or different operating conditions. This can significantly reduce the time required for calibration and maintenance models, especially for drugs in research and development that may have little or no past production history.

[0010] The JITL platform maintains a dynamic library that can be updated each time a new analytical measurement is available. In addition, to ensure that the local model adapts to new process conditions, the last available analytical measurement (e.g., for the product currently being monitored) can always be included in the training set used for local modeling. This allows the local model to adapt to new conditions more quickly or to new product lines with no history. Using this approach, both model calibration and model maintenance can be automated, and the time and expense (e.g., material and labor costs) associated with routine calibration in conventional systems can be greatly reduced. In addition, the ability to provide confidence bounds (or other confidence indicators, such as confidence scores) around model predictions can allow for robust monitoring and control strategies.

[0011] In some embodiments, a Gaussian process model is used for local modeling within the JITL framework. The Gaussian process model is a powerful statistical machine learning model that can efficiently capture complex nonlinear process dynamics and can easily adapt to almost any process changes. Compared with PLS, principal component regression (PCR), and other types of regression models, the Gaussian process model is a non-parametric method and is far more capable of capturing the complex correlation between Raman spectra and analytical measurement results from a limited data set. In addition, the Gaussian process model generally does not require filtering preprocessing of the Raman scan. Therefore, in some embodiments, the Gaussian process model is instead calibrated on the original Raman scan (on a logarithmic scale), which can save many steps in the model calibration / maintenance process. In addition, the Gaussian process model provides a confidence boundary around the prediction, which may be extremely difficult to obtain using a PLS model or a PCR model. The confidence boundary may be particularly useful for designing the optimal sampling strategy for analytical instruments and / or implementing closed-loop control (e.g., model predictive control or MPC), for example, to avoid changes based on unreliable predictions.

[0012] Although JITL is a nonlinear modeling framework, and although the above-described methods provide a certain adaptability by updating a dynamic library with the latest analytical measurements, JITL alone may not be sufficiently adaptive to account for time-varying process conditions (e.g., sudden changes in set points or other process conditions). Specifically, a local model calibrated using JITL may not be able to take advantage of the latest samples. For example, and particularly if there are recent and sudden changes in process conditions, the latest samples may not meet similarity criteria that are based entirely on "spatial" similarity (e.g., similarity of Raman scans). Also described herein are improved JITL techniques that can better utilize the information provided by the latest samples (independent of spatial similarity) and can therefore better adapt to time-varying process changes. Specifically, this article describes "adaptive" JITL (A-JITL) techniques and "spatiotemporal" JITL (ST-JITL) techniques for model calibration and maintenance.

[0013] Real-time model maintenance, in which local models can learn from recent analytical measurements and thereby quickly adapt to time-varying conditions, can be important to the success of JITL techniques. However, frequent access to analytical instruments / measurements (e.g., analyzing offline samples) tends to be resource intensive. In order to minimize such resource usage without excessively degrading model performance, a performance-based model maintenance protocol can be implemented in which the system schedules / triggers analytical measurements in response to determining that current model performance is unacceptable / unreliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Those skilled in the art will appreciate that the drawings described herein are included for illustrative purposes and are not intended to limit the present disclosure. The drawings are not necessarily drawn to scale, but rather focus on illustrating the principles of the present disclosure. It should be understood that in some cases, different aspects of the described embodiments may be enlarged or magnified to aid in understanding the described embodiments. In the drawings, similar reference numerals generally refer to functionally similar and / or structurally similar components throughout the various drawings.

[0015] Figure 1 is a simplified block diagram of an example Raman spectroscopy system that can be used to predict analytical measurement results for biopharmaceutical processes.

[0016] Figure 2 is a simplified block diagram of an example Raman spectroscopy system that can be used to predict analytical measurements of a biopharmaceutical process for closed-loop control of glucose concentration.

[0017] Figure 3 Depicted are experimental results of closed-loop control of glucose concentration using an example embodiment of the Raman spectroscopy system described herein.

[0018] Figure 4 Depicted is an example data flow that might occur when using just-in-time learning (JITL) techniques to analyze a biopharmaceutical process.

[0019] Figure 5 Depicted is an example data flow that may occur when using Adaptive JITL (A-JITL) technology to analyze a biopharmaceutical process.

[0020] Figure 6 Depicted is an example data flow that might occur when using spatiotemporal JITL (ST-JITL) techniques to analyze a biopharmaceutical process.

[0021] Figure 7 is a flow chart of an example method for analyzing a biopharmaceutical process. DETAILED DESCRIPTION

[0022] The various concepts introduced above and discussed in more detail below may be implemented in any of a variety of ways, and the concepts described are not limited to any particular implementation.For purposes of illustration, examples of implementation are provided.

[0023] Figure 1 is a simplified block diagram of an example Raman spectroscopy system 100 that can be used to predict analytical measurement results for a biopharmaceutical process. Figure 1 The system 100 is depicted as implementing Raman spectroscopy techniques, but it should be understood that in other embodiments, the system 100 may implement other spectroscopy techniques suitable for analyzing biopharmaceutical processes, such as, for example, near infrared (NIR) spectroscopy.

[0024] System 100 includes a bioreactor 102, one or more analytical instruments 104, a Raman analyzer 106 with a Raman probe 108, a computer 110, and a database server 112 coupled to the computer 110 via a network 114. The bioreactor 102 can be any suitable container, device, or system that supports a biologically active environment, which can include living organisms in a culture medium and / or substances derived from these living organisms (e.g., cell cultures). The bioreactor 102 can contain recombinant proteins expressed by cell culture, for example, such as for research purposes, clinical use, commercial sales, or other distribution. Depending on the monitored biopharmaceutical process, the culture medium can include specific fluids (e.g., "liquid culture medium") and specific nutrients, and can have target culture medium state parameters, such as target pH levels or ranges, target temperatures or temperature ranges, etc. The culture medium can also include organisms and substances derived from these organisms (such as metabolites and recombinant proteins). The contents and parameters / characteristics of the culture medium are collectively referred to as "media profiles" herein.

[0025] The analytical instrument(s) 104 can be any online, on-line and / or offline instrument(s) configured to measure one or more characteristics or parameters of the biologically active contents within the bioreactor 102 based on samples obtained from the biologically active contents. For example, the analytical instrument(s) 104 can measure one or more medium component concentrations, such as nutrient levels and / or metabolite levels (e.g., glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na+, K+, etc.) and medium state parameters (pH, pCO2, pO2, temperature, osmotic pressure, etc.). Additionally or alternatively, the analytical instrument(s) 104 can measure osmotic pressure, viable cell density (VCD), titer, critical quality attributes, cell state (e.g., cell cycle), and / or other characteristics or parameters associated with the contents of the bioreactor 102. As a more specific example, a sample may be collected, spun down, purified through multiple columns, and may pass through a first analytical instrument 104 (e.g., a high performance liquid chromatography (HPLC) or ultra high performance liquid chromatography (UPLC) instrument), and then through a second analytical instrument 104 (e.g., a mass spectrometer), where both the first and second analytical instruments 104 provide analytical measurements. One, some, or all of the analytical instrument(s) 104 may use destructive analytical techniques.

[0026] The Raman analyzer 106 may include a spectrometer device coupled to a Raman probe 108 (or in some embodiments, multiple Raman probes). The Raman analyzer 106 may include a laser source that delivers laser light to the Raman probe 108 via a fiber optic cable, and may also include a charge coupled device (CCD) or other suitable camera / recording device to record, for example, a signal received from the Raman probe 108 via another channel of the fiber optic cable. Alternatively, the laser source may be integrated into the Raman probe 108 itself. The Raman probe 108 may be an immersion probe or any other suitable type of probe (e.g., a reflection probe and a transmission probe).

[0027] The Raman analyzer 106 and the Raman probe 108 are collectively configured to non-destructively scan the bioactive content during the biopharmaceutical process within the bioreactor 102 by exciting, observing, and recording the molecular "fingerprint" of the biopharmaceutical process. The molecular fingerprint corresponds to the vibrational, rotational, and / or other low-frequency modes of the molecules within the bioactive content within the biopharmaceutical process when the contents of the bioreactor are excited by the laser light delivered by the Raman probe 108. As a result of this scanning process, the Raman analyzer 106 generates one or more Raman scan vectors, each of which represents intensity as a function of the Raman shift (frequency).

[0028] The computer 110 is coupled to the Raman analyzer 106 and the (multiple) analytical instruments 104, and is generally configured to analyze the Raman scan vectors generated by the Raman analyzer 106 in order to predict one or more analytical measurement results of the biopharmaceutical process. For example, the computer 110 can analyze the Raman scan vectors in order to predict the (multiple) analytical measurement results of the same type (multiple) performed by the (multiple) analytical instruments 104. As a more specific example, the computer 110 can predict the glucose concentration, while the (multiple) analytical instruments 104 actually measure the glucose concentration. However, although the (multiple) analytical instruments 104 can perform relatively infrequent "offline" analytical measurements of samples extracted from the bioreactor 102 (e.g., due to the limited amount of culture medium from the biopharmaceutical process and / or due to the high cost of performing such measurements, etc.), the computer 110 can perform relatively frequent "online" predictions of the analytical measurement results in real time. The computer 110 can also be configured to transmit the analytical measurement results performed by the (multiple) analytical instruments 104 to the database server 112 via the network 114, as will be discussed in further detail below.

[0029] exist Figure 1In the illustrated example embodiment, the computer 110 includes a processing unit 120, a network interface 122, a display 124, a user input device 126, and a memory 128. The processing unit 120 includes one or more processors, each of which may be a programmable microprocessor that executes software instructions stored in the memory 128 to perform some or all of the functions of the computer 110 as described herein. Alternatively, one, some, or all of the processors in the processing unit 120 may be other types of processors (e.g., an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc.), and the functions of the computer 110 as described herein may be implemented in part or in whole in hardware instead. The memory 128 may include one or more physical memory devices or units including volatile and / or non-volatile memory. Any suitable one or more memory types may be used, such as a read-only memory (ROM), a solid-state drive (SSD), a hard disk drive (HDD), etc.

[0030] The network interface 122 may include any suitable hardware (e.g., front-end transmitter and receiver hardware), firmware, and / or software configured to communicate via the network 114 using one or more communication protocols. For example, the network interface 122 may be or include an Ethernet interface. The network 114 may be a single communication network, or may include multiple communication networks of one or more types (e.g., one or more wired and / or wireless local area networks (LANs), and / or one or more wired and / or wireless wide area networks (WANs) (e.g., such as the Internet or an intranet)).

[0031] The display 124 may use any suitable display technology (e.g., LED, OLED, LCD, etc.) to present information to the user, and the user input device 126 may be a keyboard or other suitable input device. In some embodiments, the display 124 and the user input device 126 are integrated into a single device (e.g., a touch screen display). Typically, the display 124 and the user input device 126 may be combined to enable a user to interact with a graphical user interface (GUI) provided by the computer 110, for example, for purposes such as manually monitoring various processes being performed within the system 100. However, in some embodiments, the computer 110 does not include the display 124 and / or the user input device 126, or one or both of the display 124 and the user input device 126 are included in another computer or system communicatively coupled to the computer 110 (e.g., in some embodiments where predictions are sent directly to a control system implementing closed-loop control).

[0032] The memory 128 stores instructions for one or more software applications, including a Just-In-Time Learning (JITL) predictor application 130. The JITL predictor application 130, when executed by the processing unit 120, is generally configured to predict analytical measurements of a biopharmaceutical process in the bioreactor 102 by calibrating a local model 132 and analyzing Raman scan vectors generated by the Raman analyzer 106 using the local model 132. The JITL predictor application 130 may predict analytical measurements based on periodic times or other suitable times, depending on how often the Raman analyzer 106 generates such scan vectors. The Raman analyzer 106 itself may control when scan vectors are generated, or the computer 110 may trigger the generation of scan vectors by sending commands to the Raman analyzer 106. The JITL predictor application 130 may predict only a single type of analytical measurement (e.g., only glucose concentration) based on each scan vector, or may predict multiple types of analytical measurements (e.g., glucose concentration and viable cell density) based on each scan vector. In other embodiments, multiple different JITL predictor applications (e.g., each similar to JITL predictor application 130) each generate a different local model to predict different types of analysis measurements all based on the same scan vector. JITL predictor application 130 and local models 132 are discussed in further detail below.

[0033] The database server 112 may be remote from the computer 110 (e.g., such that a local setup may include only the bioreactor 102, the analytical instrument(s) 104, the Raman analyzer 106 with the Raman probe 108, and the computer 110), and may be configured as described above. Figure 1As seen in , an observation database 136 may be included or communicatively coupled to store observation data sets associated with past observations. Each observation data set in the observation database 136 may include spectral data (e.g., such one or more Raman scan vectors generated by the Raman analyzer 106) and one or more corresponding analytical measurements (e.g., such one or more measurements generated by (multiple) analytical instruments 104). Depending on the embodiment and / or scenario, past observations may have been collected for a variety of different biopharmaceutical processes, under a variety of different operating conditions (e.g., different metabolite concentration set points), and / or in a variety of different culture medium configurations (e.g., different fluids, nutrients, pH levels, temperatures, etc.). In general, it may be desirable to have the observation database 136 represent a wide variety of processes, operating conditions, and culture medium configurations. However, depending on the embodiment (as further discussed below), the observation database 136 may or may not store information indicating those processes, cell lines, proteins, metabolites, operating conditions, and / or culture medium configurations. In some embodiments, database server 112 is remotely coupled to multiple other computers similar to computer 110 via network 114 and / or other networks. This may be desirable in order to collect more observation data sets to store in observation database 136. However, in other embodiments, system 100 does not include database server 112, and computer 110 accesses local observation database 136 directly.

[0034] It should be understood that other configurations and / or components may be used instead. Figure 1 For example, different computers ( Figure 1 The computer 110 may be configured to include a computer 110 that is operable to perform measurement results provided by the analytical instrument(s) 104, a database server 112 that is operable to perform measurement results provided by the analytical instrument(s) 104, one or more additional computing devices or systems that may act as intermediaries between the computer 110 and the database server 112, some or all of the functions of the computer 110 as described herein may alternatively be performed remotely by the database server 112 and / or another remote server, and so on.

[0035] During run-time operation of the system 100, the biopharmaceutical process in the bioreactor 102 is scanned using the Raman analyzer 106 and the Raman probe 108 (i.e., a Raman scan vector of the biopharmaceutical process is generated), and the Raman scan vector(s) are then transmitted from the Raman analyzer 106 to the computer 110. The Raman analyzer 106 and the Raman probe 108 may provide scan vectors to support predictions (performed by the JITL predictor application 130) according to a predetermined monitoring cycle schedule (such as once per minute or once per hour, etc.). Alternatively, predictions may be made at irregular intervals (e.g., in response to a process-based trigger, such as a change in measured pH level and / or temperature), such that each monitoring cycle has a variable or indeterminate duration. Depending on the embodiment, the Raman analyzer 106 may send only one scan vector to the computer 110 per monitoring cycle, or may send multiple scan vectors to the computer 110 per monitoring cycle, depending on how many scan vectors the local model 132 accepts as input for a single prediction. For example, multiple scan vectors may improve the prediction accuracy of the local model 132 .

[0036] The query unit 140 of the JITL predictor application 130 uses the scan vector(s) received during a single monitoring cycle to generate query points that are used to query the observation database 136. In some embodiments, the query point (i.e., the data defining the query point) includes only data representing the Raman scan vector(s) received from the Raman analyzer 106 (e.g., including an intensity / frequency tuple for each scan vector). In other embodiments, the query point also includes one or more other types of information. For example, the query point may also include data representing operating conditions associated with the process (e.g., metabolite concentration set points in a control system, or laser wavelength and / or intensity associated with the Raman analyzer 106 or Raman probe 108, etc.), data representing the media configuration of the biopharmaceutical process media (e.g., fluid type, nutrient type or concentration, pH level, etc.), and / or other data (e.g., indicators of cell lines, proteins, or metabolites associated with the biopharmaceutical process).

[0037] Typically, the query point may include data representing the same vectors, parameters, and / or classifications used as input by the local model 132 (i.e., as a feature set for the local model 132). Using a variety of different data types for the feature set may improve the accuracy of the prediction of analytical measurements made by the local model 132. However, since each observation data set in the observation database 136 will typically need to include the same vectors, parameters, and / or classifications as the feature set, it may be preferred to limit the query points, and the feature set / inputs of the local model 132, to only include one or more Raman scan vectors. This may provide various benefits, such as allowing more information to be collected for storage in the observation database 136, and / or simplifying the collection of this information. For example, if only Raman scan vectors are used, then observation data sets may be included in the observation database 136 even if little or nothing is known about the process, cell line, protein, metabolite, operating conditions, and / or media configuration that existed when the observation data set was collected.

[0038] Then, the query unit 140 uses the generated query points to query the observation database 136. Figure 1 In the example embodiment of , query unit 140 accomplishes this by causing network interface 122 to transmit a query point (e.g., within a query message) to database server 112 via network 114 (which in turn causes database server 112 to retrieve the appropriate data from observation database 136). However, in embodiments where observation database 136 is instead included in computer 110 (or in a memory communicatively coupled to the computer), query unit 140 may instead query observation database 136 more directly. For ease of explanation, Figure 1 The remainder of the description will assume that observation database 136 is coupled to database server 112, such as Figure 1 However, one of ordinary skill in the art will readily appreciate how the communication paths may differ where the observation database 136 is instead local to the computer 110, or located at another suitable location within the system architecture.

[0039] After receiving the query point, the database server 112 uses the query point to select relevant observation data sets from the observation database 136 to be used as training data for the local model 132. Depending on the embodiment, the database server 112 may apply any suitable relevance criteria to identify which observation data sets are "relevant". In one embodiment, for example, the query point includes a single Raman scan vector, and the database server 112 determines whether the observation data set is relevant by calculating the Euclidean distance between the Raman scan vector of a given observation data set and the Raman scan vector of the query point. If the Euclidean distance is below a predetermined threshold (or below a variable threshold, such as a threshold calculated based on the average Euclidean distance between the query point scan vector and the scan vectors of all observation data sets, etc.), the observation data set is identified as a relevant observation data set. Those of ordinary skill in the art will understand how this approach can be easily extended to embodiments where the query point (and each observation data set) includes multiple Raman scan vectors. In some cases, using the Euclidean distance to select relevant observation data sets may be a suboptimal technique. However, if the local model 132 is a Gaussian process model (as discussed below), using Euclidean distance as the relevance criterion may be particularly advantageous. This is because Gaussian process models with radial basis functions or squared exponential kernels are themselves based on Euclidean distance. However, in other embodiments, other relevance criteria (e.g., angle-based criteria or correlation-based criteria, etc.) may be applied. It should be understood that in embodiments where the local model 132 also accepts other information (e.g., operating conditions, culture medium configurations, process data, cell line information, protein information, and / or metabolite information, etc.) as input / feature sets, more complex techniques may be used to identify "relevant" observation data sets. In some embodiments, the database server 112 selects only a predetermined number of relevant observation data sets in response to a single query, or selects no more than a certain maximum allowed number of relevant observation data sets to ensure that only a relatively small subset of all data sets in the observation database 136 are retrieved. However, in other embodiments, the database server 112 may select any number of relevant observation data sets as long as each such data set meets the relevance criteria.

[0040] In some embodiments, as described in more detail below (e.g., with reference to Figure 5 and Figure 6), select relevant observation datasets based not only on relevance to the query point in a “spatial” sense (e.g. similarity of Raman scan vectors), but also in a temporal sense (e.g. which datasets are the most recent, regardless of spatial similarity). These techniques can better exploit the fact that the most recent analytical measurements can provide useful information even if they correspond to different set points, etc.

[0041] After identifying the relevant observation data sets (each of which may or may not correspond to the same process conditions as the currently monitored biopharmaceutical process in the bioreactor 102), the database server 112 retrieves these data sets (e.g., Raman scan vectors and corresponding analytical measurement(s)) and transmits the retrieved data sets to the computer 110 via the network 114. The query unit 140 may then pass these relevant data sets to the local model generator 142, and the local model generator 142 uses these relevant data sets as training data to calibrate the local model 132. That is, the local model generator 142 uses the Raman scan vector(s) associated with each observation data set (and possibly other data) as a feature set, and uses the analytical measurement(s) associated with the same observation data set as labels for this feature set.

[0042] In some embodiments, as noted above, the local model generator 142 constructs a Gaussian process model to efficiently capture complex, nonlinear process dynamics and easily adapt to almost any process variation. Unlike PLS models and PCR models, the Gaussian process model uses a nonparametric approach and is able to capture complex nonlinear correlations between Raman scan vectors and analytical measurement results even with a very limited number of training samples. This may be particularly important when new products or processes correspond to only a limited number of data sets in the observation database 136. In this case, the Gaussian process model is typically able to extract the most information from those limited data sets in combination with other relevant data sets selected by the database server 112 from the observation database 136. However, in other embodiments, as long as the training time does not exceed the minimum expected duration of the monitoring cycle, the local model generator 142 can alternatively construct any other suitable type of machine learning model (e.g., recursive neural network, convolutional neural network, etc.). The local model generator 142 can also construct the local model 132 so that the local model 132 can output a credibility boundary, or some other suitable prediction confidence indicator (e.g., confidence score). At least compared to PLS models and PCR models, Gaussian process models are particularly suitable for providing confidence boundaries around predictions of analytical measurement results. Although various advantages of Gaussian process models relative to PLS models and PCR models have been described, it should be understood that in some embodiments, local model generator 142 can use PLS or PCR modeling methods to construct local model 132.

[0043] The local model generator 142 can build the local model 132 in an online, real-time manner, so that the prediction unit 144 can then use the trained local model 132 to predict one or more analytical measurement results of the biopharmaceutical process by processing the same Raman scan vector(s) that the query unit 140 used to generate the query point. In fact, in some embodiments, the query unit 140 can perform a new query and the local model generator 142 can generate a new version of the local model 132 each time the Raman analyzer 106 provides a new Raman scan vector (or a new set of Raman scan vectors) to the computer 110. However, in other embodiments, the query unit 140 performs new queries less frequently (and the local model generator 142 generates a new version of the local model 132 less frequently), such as once every 10 prediction / monitoring cycles, once every 100 prediction / monitoring cycles, etc.

[0044] The database maintenance unit 146 may also cause the analysis instrument(s) 104 to periodically collect one or more actual analysis measurements at a significantly lower frequency than the monitoring period of the Raman analyzer 106 (e.g., only once or twice a day, etc.). In some embodiments, the measurement(s) of the analysis instrument(s) 104 may be destructive and require permanent removal of the sample from the process in the bioreactor 102. At or near the time when the database maintenance unit 146 causes the analysis instrument(s) 104 to collect and provide the actual analysis measurements, the database maintenance unit 146 may also cause the Raman analyzer 106 to provide one or more Raman scan vectors. The database maintenance unit 146 may then cause the network interface 122 to send the Raman scan vector(s) and the corresponding actual analysis measurements to the database server 112 via the network 114 to be stored in the observation database 136 as a new observation data set. The observation database 132 may be updated according to any suitable timing, which may vary according to the embodiment. For example, if the (multiple) analytical instrument 104 outputs an actual analytical measurement result within a few seconds of measuring a sample, the observation database 132 can be updated with the new measurement result almost immediately when the sample is taken. However, in certain other embodiments, the actual analytical measurement result may be the result of processing performed by one or more of the (multiple) analytical instruments 104 over several minutes, hours, or even days, in which case the observation database 132 is not updated until such processing has been completed. In still other embodiments, new observation data sets can be added to the observation database 132 in an incremental manner as different analytical instruments in the analytical instruments 104 complete their respective measurements.

[0045] Thus, the observation database 136 provides a "dynamic library" of past observations that can be used by the local model generator 142 for model training. In some embodiments, the most recent analytical measurement(s) are always added to the observation database 136, and the local model generator 142 can always use the most recent observation data set(s) in the observation database 136 when calibrating the local model 132. This can allow the local model 132 to encode process information from the recent past and quickly adapt to new conditions, or quickly adapt to new process conditions for which there is no history. In addition, calibration and maintenance of the local model 132 can be automatic. In some embodiments, the adaptability of the local model 132 is further enhanced, for example as discussed below in conjunction with the A-JITL technique and the ST-JITL technique.

[0046] In some embodiments, the database maintenance unit 146 may cause the analysis instrument(s) 104 to collect and provide the actual analysis measurement(s) based on some other time or condition such as current model performance. For example, if the local model 132 outputs a confidence interval (e.g., a range of values ​​around a predicted value within which there is a 95% probability or confidence that the actual / measured value will fall), or some other confidence indicator along with the prediction (e.g., if the local model 132 is a Gaussian process model), and if the confidence indicator indicates a particularly unreliable prediction (e.g., if the interval / range exceeds a threshold width / range, etc.), the database maintenance unit 146 may trigger the collection of one or more actual analysis measurement results. As a more specific example, the database maintenance unit 146 may trigger the collection of the analysis measurement(s) in response to determining that the 95% confidence interval exceeds a predefined threshold. Optimal scheduling of analysis measurement results will be discussed in further detail below. After the measurement(s) have been made, the database maintenance unit 146 may cause the Raman analyzer 106 to generate one or more Raman scan vectors and cause the network interface 122 to provide the actual analysis measurement(s) and the corresponding Raman scan vector(s) to the database server 112 for storage as a new observation dataset in the observation database 132 (e.g., in the manner discussed above). The local model generator 142 may then utilize this most recent observation dataset when calibrating the local model 132, where appropriate (e.g., depending on relevance to the current query, or whether the embodiment always utilizes the most recent observation dataset).

[0047] Throughout the life cycle of a biopharmaceutical process in a bioreactor, some or all of the above processes may be repeated multiple times to continuously monitor the process using a local model whose calibration and maintenance are fully automatic and in real time. Depending on the embodiment and / or scenario, the analytical measurement(s) may be predicted for various purposes. For example, as part of a quality control process, certain parameters may be monitored (i.e., predicted) to ensure that the process is still within the relevant specifications. As another example, one or more parameters may be monitored / predicted to provide feedback in a closed-loop control system. For example, Figure 2 A system 150 is depicted that is similar to system 100, but that attempts to control glucose concentration in a biopharmaceutical process (i.e., attempts to match a predicted glucose concentration to a desired set point within some acceptable tolerance). It should be understood that in other embodiments, system 150 may alternatively (or also) be used to control a process parameter other than glucose level, or to control glucose level based on a prediction of one or more other process parameters (e.g., lactate level). Figure 2 The same reference numerals are used to indicate Figure 1 For example, Figure 2 The JITL predictor application 130 can be used with Figure 1 The JITL predictor application 130 is the same as the JITL predictor application 130 of FIG. 1 (wherein, for the sake of clarity, the individual elements of the JITL predictor application 130 are not shown in FIG. 1 ). Figure 2 ).

[0048] like Figure 2 As can be seen in the figure, within the system 150, the memory 128 also stores a control unit 152. The control unit 152 is configured to control the glucose pump 154, that is, to cause the glucose pump 154 ​​to selectively introduce additional glucose into the biopharmaceutical process within the bioreactor 102. The control unit 152 may include, for example, software instructions executed by the processing unit 120, and / or appropriate firmware and / or hardware. In some embodiments, the control unit 152 uses the glucose concentration as an input in a closed-loop architecture to implement a model predictive control (MPC) technique. In embodiments where the local model 132 provides a confidence boundary or other confidence indicator for each prediction (for example, in certain embodiments where the local model 132 is a Gaussian process model), the control unit 152 may also accept a confidence indicator as input. For example, the control unit 152 may only generate control instructions for the glucose pump 154 ​​based on the glucose concentration prediction having a sufficiently high confidence indicator (e.g., only based on predictions associated with confidence boundaries that do not exceed a certain percentage or absolute measurement result range, or only based on predictions associated with confidence scores that exceed a certain minimum threshold score, etc.), or may increase and / or decrease the weight of a given prediction based on the confidence indicator of the glucose concentration prediction, etc.

[0049] Figure 3 Depicted are experimental results 200 for an example implementation of using JITL techniques to calibrate and maintain a local Gaussian process model. Figure 3 In the plot of FIG. 1 , the horizontal dashed line 202 represents the glucose concentration set point, and the circle 204 represents the actual measurement result of the glucose concentration (e.g., by a line similar to Figure 1 104), solid line 206 represents a predicted measurement of glucose concentration (e.g., as predicted by a model similar to local model 132), and shaded area 208 represents a confidence boundary (95% confidence) associated with the predicted measurement. Figure 3 As can be seen in Figure 3, for a glucose concentration set point of 3 grams per liter (g / L), the predictions made using the JITL technique generally agree well with the analytical measurements.

[0050] The process of performing the query and building / calibrating the local model 132 will now be described in more mathematical detail with reference to a specific JITL embodiment in which the local model 132 is a Gaussian process model that uses a single Raman scan vector as input and predicts a single analytical measurement result:

[0051] make (or simply ) represents a set of ordered input and output data pairs such that is the input and is the output. In addition, assume Yes a -dimensional input vector, and is a scalar output. In fact, can be considered as a spectroscopic measurement (e.g. NIR or Raman), and The analytical measurements that are considered to be states of interest (e.g., glucose concentration or lactate concentration). Given a training dataset The goal of the spectral model calibration problem is to identify the relationship between the input and output of a model of the form:

[0052] b j =f(a j )+∈ j Equation (1)

[0053] in, is a spectral model, and is the measurement noise of zero-mean normal distribution, with variance σ 2 Unknown. The standard practice for model calibration is to assume that f(·) is linear and then use methods such as PLS to train the model. Rather than attributing any restrictions or fixed forms to f(·), it is assumed here that f(·) is an implicit function modeled as a Gaussian process such that

[0054]

[0055] represents a random sample from a Gaussian process, where the mean And the covariance function It is usually defined as follows:

[0056]

[0057]

[0058] and, represents the hyperparameters of the Gaussian process model. A Gaussian process is a collection of random variables where any finite number of random variables have a joint Gaussian distribution such that for a finite set of inputs We can write:

[0059]

[0060] The spectral model calibration problem then reduces to using To learn the implicit Gaussian process function For mathematical convenience and simplicity, it is assumed here that But this need not be the case in general, and the results here can be easily extended to The role of the covariance function in a Gaussian process is similar to that of the kernel used in a support vector machine (SVM). A common choice of covariance function is the Gaussian kernel and is given by:

[0061]

[0062] in, is the input pair {a i , a j}. If the set {a i , a j} are “close” to each other as defined by the Euclidean distance in equation (4), then the Gaussian kernel k θ (a i , a j ) are assigned higher relevance.

[0063] For the selected Gaussian kernel, equation (4) is a positive definite symmetric matrix such that In equation (4), the set is a set of hyperparameters. In fact, is the length scale parameter, and is the signal variance parameter. The choice of the Gaussian covariance function in equation (4) corresponds to the prior assumption that f is smooth and continuous. Therefore, by changing the hyperparameters of the covariance function, the "smoothness" of f can be changed. Here, a Gaussian process with a Gaussian covariance function is assumed. However, this is not necessarily the case in general.

[0064] Given The goal is to learn the hyperparameters of the Gaussian process, including any other unknown model parameters. For the Gaussian process in equation (1), the set of unknown parameters is The parameter learning step can be performed by maximizing the marginal likelihood (or evidence) function over the space of unknown parameters. For example, for the Gaussian process in equation (1), the marginal likelihood function is given by:

[0065]

[0066] in, is the marginal likelihood function, is the likelihood function given by:

[0067]

[0068] and, is the prior density function given in equation (3). For the Gaussian likelihood and prior density in equations (6) and (3), respectively, the integral in equation (5) has a closed form solution, so that the marginal likelihood function is given by:

[0069]

[0070] Now given equation (7), we can estimate

[0071]

[0072] Among them, γ*∈Γ is the optimal estimate. According to equation (7), we have:

[0073]

[0074] in, To solve the optimization problem in equation (8), the partial derivatives of equation (9) are determined with respect to γ ​​so that for all r = 1, 2, ..., n γ ,

[0075]

[0076] in, Given the marginal likelihood function in equation (7) and its derivative in equation (10b), equation (8) can be solved using gradient descent. Since equation (8) is generally a non-convex optimization problem with multiple local optima, care must be taken when solving the optimization problem. Here we assume that γ * is known or can be calculated by solving equation (8). In addition, unless otherwise stated, in order to reduce the burden of notation, it will be assumed here that γ is the optimal estimate γ * .

[0077] Once the Gaussian process spectral calibration model in Equation (1) is trained, it can be deployed for real-time prediction applications. As mentioned before, let is the training data set used to train the Gaussian process model, and let is a new test spectral signal. Then, the goal is to predict * The corresponding output Calculate b * The first step is to train the input set and test input a *Construct all training output sets as conditions And the test Gaussian process output f(a * ). This joint density is given as follows:

[0078]

[0079] in, Given equation (11), in the Bayesian framework, the Gaussian process output f(a * ). In other words, we seek the Gaussian process output f(a * ). Of course, f(a * ) only needs to include the posterior distribution of the training set In the probabilistic setting, this can be achieved by making the joint distribution in Eq. (11) To calculate f(a * ), giving the following equation:

[0080]

[0081] in, is the posterior distribution of the Gaussian process output, and Given by:

[0082]

[0083] and Given by:

[0084]

[0085] Given equation (12), output b * The predicted posterior distribution of can be calculated as follows:

[0086]

[0087] in, and They are given in equations (13) and (14) respectively. For a single test input The Gaussian process prediction in equation (15) gives the distribution of outputs with non-zero probability of realization. In real-time applications such as control and monitoring, one may be interested in point estimates rather than the entire distribution. Decision theoretic methods can be used to compute point estimates. It can be shown that for the Gaussian posterior distribution in equation (15), the mean function minimizes both the expected absolute risk function and the squared risk function, where is input a *In addition, for the choice As a prediction, an approximately 95% confidence interval is given by:

[0088]

[0089] The interval in equation (16) can be used to evaluate the quality of Gaussian process predictions and / or to design Gaussian process-based model predictive control or other robust monitoring strategies.

[0090] Now turning to selecting relevant samples (here, the observation dataset) in response to a query, the question is what is the optimal solution for a given query point? and including A central database / repository of input-output pairs (observation data sets) In time Select Include A local training set of samples Where D<<L. Assume is dynamic and may contain different entries during an activity. There are many methods for structure For the purpose of this analysis, the set The Euclidean distance between the spectra (e.g., Raman scan vectors) in Although Euclidean-based similarity measures in the JITL framework have been reported to be suboptimal in some cases, they can be a beneficial choice when using Gaussian process models. This is because the Gaussian process model itself is based on Euclidean distance. The Gaussian kernel is only valid in the set {a i , a j} are assigned a higher correlation only when the inputs in the query point are “close” to each other. Therefore, by creating a local training set where all inputs are “close” to the query point It is possible to ensure that the local Gaussian process model captures the maximum “relevance” to predict the output at the query point.

[0091] An example algorithm is provided in Algorithm 1 below, which formally outlines the algorithm used to obtain Creating a local training set Here's how to use this training set to train a Gaussian process model and use the trained model to make predictions:

[0092]

[0093] Algorithm 1

[0094] Now go to Figure 4, shows an example data flow 250 that may occur when using JITL techniques as described herein to analyze a biopharmaceutical process. For example, data flow 250 may occur in Figure 1 system 100, or Figure 2 150. In the data stream 250, spectral data 252 is provided by the spectrometer / probe. For example, the spectral data 252 may include a Raman scan vector generated by the Raman analyzer 106, or a NIR scan vector, etc. A query point 254 is generated (e.g., by the query unit 140) based on the spectral data 252, and is used to query a global data set 256, which may include, for example, all observation data sets in the observation database 136. Based on the query, a local data set 258 is identified within the global data set 256. As described above, the local data set 258 may be selected, for example, based on a relevance criterion (e.g., Euclidean distance).

[0095] The local data set 258 is then used (e.g., by the local model generator 142) as training data to calibrate the local model 260 (e.g., the local model 132). The local model 132 is then used (e.g., by the prediction unit 144) to predict outputs (analysis measurements) 262, such as medium component concentrations, medium status (e.g., glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na+, K+ and other nutrients or metabolites, pH, pCO2, pO2, temperature, osmotic pressure, etc.), viable cell density, titer, critical quality attributes, cell status, etc., and may also output confidence boundaries or another suitable confidence indicator.

[0096] While JITL-based local models (e.g., as in Algorithm 1 and data flow 250) provide a robust nonlinear modeling framework, this approach does not have an inherent mechanism for adapting to time-varying process changes. To address this shortcoming, some embodiments may use an "adaptive" JITL (A-JITL) strategy. As noted above, when new samples become available, these samples may be included in the In such embodiments (i.e., is a dynamic embodiment), It can be expressed as In one such embodiment, a moving time window approach is implemented in which newly acquired samples are added to the in and from Remove the oldest samples from the . Discarding the oldest samples may be beneficial because in the adaptive strategy, keeping The size of may be crucial to ensure the computational tractability of the entire JITL framework. However, a major issue with this approach is that simply discarding old samples may lead to information loss, as old samples may contain relevant information.

[0097] To avoid this information loss, in one embodiment, new samples are added to without removing any old / existing samples. Therefore, as new analytical measurement results become available, the central database Scales with the number of samples. In cell culture process applications, expanding the database may not cause any significant computational issues, because such processes are typically operated as batch processes with batch times lasting two to three weeks. This naturally limits the number of samples to be included. In addition, only a limited number of analytical measurements are typically sampled during batch cell culture processes (e.g., unlike the chemical industry where analytical measurements are sampled frequently). Therefore, the database is typically only moderately increased. without any significant impact on the computational stability of the entire JITL framework.

[0098] Although the new samples are included The continuous adaptation of Algorithm 1 (above) is very important, but the success of this method depends on the choice of local database The new samples in are used for local model calibration. Algorithm 1 (This algorithm is based on the Euclidean distance from Select the The samples that are not close to the query sample (e.g., line 6 of Algorithm 1) can be referred to as a “spatially correlated” approach because it only prioritizes samples that are correlated (close) in space. If the new samples are not close to the query sample, which may be the case when a sudden setpoint change (or other sudden process condition change) occurs, Algorithm 1 may not be able to include these samples. On the other hand, recursive methods (e.g., regularized partial least squares (RPLS), recursive least squares (RLS), and recursive N-way partial least squares (RNPLS)) are "time-dependent" because they prioritize the most recent measurements without considering spatial correlations. Updating the local model using recent samples allows recursive methods to successfully adapt to current process conditions.

[0099] One such embodiment, referred to herein as "adaptive" JITL (A-JITL), prioritizes samples that are correlated both spatially and temporally. Indicates that from the current experiment (i.e., query a * A set of L historical measurements available before the start of the experiment / process) and let Representing a set of n measurements available from the current experiment, samples can be redistributed as follows:

[0100]

[0101]

[0102] in, represents the central database, and represents a set of the most recent (latest) k measurement results. In some embodiments, contains the most recent k samples from the current experiment / process, and Includes samples from previous experiments / processes, and (potentially) samples from the current experiment / process that are earlier than the most recent k samples. The above equations (17a) and (17b) are for a given query a * For a query at another time, the dataset and may contain different samples, depending on the number of measurements available at this moment. For example, once the sample Available, from Remove And will Included in Then discard the sample Included in to prevent any information loss. Update with the most recent measurement results Ensure At least reflect some current conditions.

[0103] Given and The purpose is to choose As noted above, for A-JITL, both spatially correlated samples and temporally correlated samples are included in If we assume It can be broken down into:

[0104]

[0105] in, and are spatially related sets and temporally related sets respectively, then the goal is to select and First, assume Make Only unique samples are included. For design According to a distance-based (spatial) metric (such as a "similarity index" or "s-value") Select Dk samples from:

[0106] s i xsim(a i , a * )=exp(-||a i -a * ||). Equation (19)

[0107] For example, equation (19) can be used as the similarity measure in the (non-adaptive) JITL technique described above. Thus, for example, Select Dk samples with the largest s values ​​to include in In order to design If it is assumed that the most recent k samples from the current experiment / process are time-dependent, then in some embodiments can be defined as equal to It should be noted that The sample membership in s varies, and is determined based on the sampling time. Of course, depending on the scenario, The samples in may show larger s values. Regardless of the s value, is only assumed to be time-dependent. Similarly, is only spatially relevant, since, by construction, It is not time-dependent. It should be noted that and For a given query a * Defined, The samples in are selected based on their s values ​​calculated with respect to a*, and The samples in are chosen based on their sampling times calculated with respect to the sampling time of a*. For convenience, and The general definition is as follows:

[0108]

[0109]

[0110] in, and are the spatially correlated sample and the time-correlated sample from the Raman spectrometer, respectively, and and are the spatially correlated samples and the time-correlated samples from the analytical instrument, respectively, so that

[0111]

[0112]

[0113] Substituting equations (20a) and (20b) into equation (18) yields the set Generally expressed as in and Compared to the (non-adaptive) JITL techniques discussed above, local libraries / datasets Prioritize spatially and temporally correlated samples. Given and query a * , the Gaussian process model (eg, local model 132) in equation (1) can be calibrated. * The point estimate and credibility interval at can be calculated using equations (13) and (16), respectively, where and Given by:

[0114]

[0115]

[0116] in, and are respectively and The associated covariance function, and where yes and The covariance between .

[0117] An example algorithm that formally outlines the A-JITL technique is provided in Algorithm 2 below:

[0118]

[0119] Algorithm 2

[0120] Therefore, Algorithm 2 combines JITL (spatial correlation) with recursive learning (temporal correlation). For example, for The calibration of the local model 132 using Algorithm 2 is similar to the spatially correlated JITL, whereas for The calibration of the local model 132 using Algorithm 2 is similar to recursive learning. Therefore, by adjusting and It is possible to strike a proper balance between (non-recursive) JITL and recursive learning.

[0121] Now go to Figure 5 , shows an example data flow 300 that may occur when analyzing a biopharmaceutical process using the A-JITL technique as described herein. For example, the data flow 300 may be Figure 1 system 100, or Figure 2The system 150 of FIG. 300 is provided by a spectrometer / probe. In the data stream 300, spectral data 302 is provided by a spectrometer / probe. For example, the spectral data 302 may include a Raman scan vector generated by the Raman analyzer 106, or a NIR scan vector, etc. The query point 304 is generated (e.g., by the query unit 140) based on the spectral data 302, and is used to query a global data set 306, which may include, for example, all observation data sets in the observation database 136. The global data set 306 is logically divided into the most recent k entries 307A (e.g., all from the current experiment / process) and all entries 307B before the most recent k entries 307A (e.g., from previous experiments / processes, and possibly also from the current experiment / process). The value of k can be determined based on the number of samples at the query point 304. As used herein, the term "number of samples" can broadly refer to any indicator of time or relative time associated with a given sample / observation. Some of the entries 307B are added to the local dataset 308 based on spatial similarity (e.g., Euclidean distance) with the query point 304, while all entries 307A may be added to the local dataset 308 regardless of spatial similarity. For example, the local dataset 308 may be generated from the entries 307A and the entries 307B according to Algorithm 2.

[0122] The local data set 308 is then used (e.g., by the local model generator 142) as training data to calibrate the local model 310 (e.g., the local model 132). The local model 310 is then used (e.g., by the prediction unit 144) to predict outputs (analysis measurements) 312, such as medium component concentrations, medium status (e.g., glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na+, K+ and other nutrients or metabolites, pH, pCO2, pO2, temperature, osmotic pressure, etc.), viable cell density, titer, critical quality attributes, cell status, etc., and may also output confidence boundaries or another suitable confidence indicator.

[0123] If actual analytical measurements (e.g., measurements made by an analytical instrument such as one of analytical instrument(s) 104) are available, a new entry 314 is created and added to the global data set 306. For example, such measurements may be available based on periodic sampling (e.g., once or twice per day), and / or may become available in response to a trigger with variable timing (e.g., if a certain number of consecutive predictions have unacceptably wide confidence bounds, etc.), as discussed further below.

[0124] Although spatially correlated samples and temporally correlated samples are included The continuous adaptation of the A-JITL method discussed above is necessary, but the overall degree of adaptation achieved by A-JITL depends on The degree of effectiveness used for local model calibration. For the query sample / query point a * , spatially correlated samples Provides a function (f(a * ), f(a i )) is highly correlated. This is because for query a * , (a i , b i ) and the spatial correlation of (f(a * ), f(a i )) are based on (a i , a * ) is calculated by the Euclidean distance between them. Therefore, for the Euclidean-based similarity measure in equation (19) and the choice of the Euclidean-based kernel in equation (4), The samples in are expected to provide higher functional correlation. In contrast, the time-correlated samples Function may not be available This is because, as pointed out above, The samples in are not necessarily spatially correlated. As a result, if (a j , b j ) has a small spatial correlation, then (f(a * ), f(a j )) will be small. From a modeling perspective, it is undesirable to train the Gaussian process model in equation (1) with samples with small correlations, as this will lead to poor model performance. Mathematically, this can be proved as follows.

[0125] For query a * As well as the calibrated Gaussian process model of Algorithm 2, the model prediction can be calculated using Equation (13) Without loss of generality, if σ 2 = 0 (in the absence of noise), then equation (13) can be written as follows:

[0126]

[0127] if has negligible spatial correlation (i.e., with a * The value of s between is infinite), then Equation 4 yields Furthermore, by construction, due to with a * The distance is close to The result is and Substituting them into equation (23), we obtain

[0128]

[0129] It is clear from equation (24c) that the point estimate is Similarly, it can be shown that equation (16) is also related to For example, in equation (16) It can be calculated as follows:

[0130]

[0131]

[0132] From equations (25b) and (25c), it can be seen that several approximations are used, including and It is obvious from equations (20a) and (20b) that if With limited spatial correlation, Algorithm 2 cannot make good use of the set.

[0133] In some embodiments, to ensure The spatially correlated samples and the temporally correlated samples in can contribute, using a “space-time” JITL (ST-JITL) approach, and using the following space-time Raman model (e.g., as local model 132):

[0134] b i =g(a i , t i )+∈ i , Equation (26)

[0135] Where g: is the space-time Raman model and t i is a i The number of samples, and is a function with zero mean and unknown variance is a sequence of independent Gaussian random variables. Compared with equation (1), the spatiotemporal model of equation (26) depends on the spectral signal and its sampling time. As mentioned above, it is assumed that g is an implicit function modeled as a Gaussian process such that for any input (a, t),

[0136]

[0137] is a random function. For convenience, we assume that the mean function in equation (27) is zero, but this is not necessarily the case in general. In addition, for any input (ai , t i ) and (a j , t j ), covariance function r θ (a i a j t i t j ) can be defined as follows:

[0138] r θ (a i a j t i t j ) = k 空间 (a i , a j )+k 时间 (t i , t j ), Equation (28)

[0139] in, and They are (g(a i , t i ), g(a j , t j )) between spatial covariance and temporal covariance. It should be noted that for the query (a * , t * ), if the sample With negligible spatial correlation, k 空间 (a j , a * )≈0 but k 时间 (t j , t * )>0, so that equation (28) defines (g(a * , t * ), g(a j , t j )) has a non-zero correlation between them. Finally, it should be noted that equation (28) is a valid covariance function, since the sum of two independent kernels is also a kernel. Assume k 空间 and k 时间 is a Gaussian kernel such that for any input pair (a i , t i ) and (a j , t j ),

[0140]

[0141]

[0142] in, is the kernel parameter. Given equations (29a) and (29b), if (a i , t i ), (a j , t j ) are close to each other, then equation (28) is attributed to (g(a i , t i ), g(a j , t j )) are highly correlated. If and Respectively The number of state-dependent samples and time-dependent samples in , such that Then for the query (a * , t * ), the covariance function r in equation (28) θ can be written as:

[0143]

[0144]

[0145] It should be noted that, unlike the variables a and b, the role of t in equations (30a) and (30b) is only to improve In fact, given a, variable t has no effect on b. Therefore, if is defined as The number of samples corresponding to the samples in can be defined such that it satisfies the following equation:

[0146] |t i -t j |>>M, Equation (31a)

[0147] |t i -t * |>>N, Equation (31b)

[0148] |t i -t k |>>P, equation (31c)

[0149] For all i, j∈{1,...,Dk} and k∈{D-k+1,...,D}, where, is an arbitrarily large positive constant. In addition, if we assume and t * In order to make and For the above and It can be written as follows:

[0150]

[0151] Where, equation (32b) comes from equation (31a), which makes The off-diagonal terms in tend to zero. Similarly, the covariance and It can be calculated as follows:

[0152]

[0153]

[0154] Wherein, equation (33b) is based on equation (31b), and equation (33d) is based on equation (31c). Substituting equations (32b), (33b) and (33d) into equations (30a) and (30b) yields:

[0155]

[0156]

[0157] From equations (30a) and (30b), it can be easily confirmed that the covariance r θ Including k 空间 and k 空间 Given the covariance function of the spatiotemporal Raman model in equations (30a) and (30b), the kernel parameter θ and the noise variance σ can be estimated by maximizing 2 :

[0158]

[0159] in is the log marginal likelihood function, and r γ =r θ +I DxD Maximizing equation (35) over Γ yields the optimal estimate γ * For gradient-based optimizers, the gradient of Eq. (35) with respect to γ ​​can be computed in a manner similar to Eq. (10b). Given γ * , for the query (a * , t * ) can be calculated as:

[0160]

[0161]

[0162] where the covariance functions are given in equations (34a) and (34b). Similarly, the confidence bounds on the point estimates in equation (36a) are It can be calculated as follows:

[0163]

[0164]

[0165] in, It can be easily seen from equations (36a), (37a), and (37b) that both spatially correlated samples and temporally correlated samples contribute to the model prediction and the confidence bound calculation. Finally, substituting equations (34a) and (34b) into equations (36a) and (36b) yields the posterior mean and variance, respectively. It should be noted that, unlike the case of Algorithm 2, even in In the absence of spatial correlation, the model predictions in equation (36a) and the confidence intervals in equations (37a) and (37b) also depend on For example, when No spatial correlation (i.e. and When , equations (36a) and (36b) can be written as:

[0166]

[0167]

[0168] From the above, it can be seen that equations (38a) and (38b) still include the 空间 and k 时间 An example algorithm that formally outlines the ST-JITL technique is provided in Algorithm 3 below:

[0169]

[0170] Algorithm 3

[0171] It should be noted that for the case of β1 = 0, A-JITL and ST-JITL (in Algorithm 2 and Algorithm 3, respectively) can be the same. This is because, for β1 = 0, k 时间 = 0, so that r θ =k 空间 =k θ (As can be seen from equations (28) and (29b)).

[0172] Now go to Figure 6 , shows an example data flow 350 that may occur when analyzing a biopharmaceutical process using ST-JITL techniques as described herein. For example, data flow 350 may occur in Figure 1 system 100, or Figure 2 The system 150 of FIG. 350 is provided by a spectrometer / probe. In the data stream 350, spectral data 352 is provided by a spectrometer / probe. For example, the spectral data 352 may include a Raman scan vector generated by the Raman analyzer 106, or a NIR scan vector, etc. The query point 354 is generated (for example, by the query unit 140) based on the spectral data 352, and is used to query a global data set 356, which may include, for example, all observation data sets in the observation database 136. The global data set 356 is logically divided into the most recent k entries 357A (for example, all from the current experiment / process) and all entries 357B before the most recent k entries 357A (for example, from previous experiments / processes, and possibly also from the current experiment / process). The value of k can be determined based on the number of samples at the query point 354. For example, a local data set 358 can be generated from entries 357A and entries 357B according to Algorithm 3.

[0173] The local data set 358 is then used (e.g., by the local model generator 142) as training data to calibrate the local model 360 (e.g., the local model 132). The local model 360 is then used (e.g., by the prediction unit 144) to predict outputs (analysis measurements) 362, such as medium component concentrations, medium status (e.g., glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na+, K+ and other nutrients or metabolites, pH, pCO2, pO2, temperature, osmotic pressure, etc.), viable cell density, titer, critical quality attributes, cell status, etc., and may also output confidence boundaries or another suitable confidence indicator.

[0174] If actual analytical measurements (e.g., measurements made by an analytical instrument such as one of the analytical instrument(s) 104) are available, a new entry 364 (including its sample number) is created and added to the global data set 356. For example, such measurements may be available based on periodic sampling (e.g., once or twice per day), and / or may become available in response to a trigger with variable timing (e.g., if a certain number of consecutive predictions have unacceptably wide confidence bounds, etc.).

[0175] As noted above, analytical measurements can be scheduled / triggered based on the current and / or latest performance of one or more local models (e.g., local models 132, 260, 310, or 360) in order to maintain or improve prediction accuracy while reducing resource usage (e.g., usage of analytical instruments). For example, this technique can be used with A-JITL, ST-JITL, or pure JITL.

[0176] In one embodiment, the confidence interval is used to trigger model maintenance. Specifically, if the width of the confidence interval around a given model prediction (e.g., around the latest prediction made by the local model 132, 260, 310, or 360) (e.g., the distance between the confidence boundaries calculated using equation (16) or equations (37a), (37b)) is greater than a predefined threshold, the database maintenance unit 146 can generate a request message and cause the computer 110 to send the message to the (multiple) analysis instruments 104 to request the measurement results. Figure 3 In the example results of , for example, the database maintenance unit 146 may trigger new analysis measurement results near the end of the dates 12 / 08 / 17, 12 / 09 / 17, and 12 / 14 / 17, where the shaded area 208 indicates a wider confidence interval (i.e., b U -b L ).

[0177] In response to the request message, the analysis measurement(s) 104 perform(s) measurements and provide(s) measurement results to the computer 110. The database maintenance unit 146 may then send the measurement(s) and the corresponding Raman scan vector(s) received from the Raman analyzer 106 to the database server 112 for storage in the observation database 136. For example, the measurement(s) and scan vector(s) may be added to the library discussed above. (for pure JITL) or library (for A-JITL or ST-JITL).

[0178] Conversely, if the width of the confidence interval around a given model prediction is not greater than a predefined threshold, the database maintenance unit 146 may not request new analytical measurements, in which case the library in the observation database 136 remains unchanged. In embodiments where the analytical instrument(s) 104 include multiple instruments that measure different properties (such as medium component concentrations, medium status (e.g., glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na+, K+ and other nutrients or metabolites, pH, pCO2, pO2, temperature, osmotic pressure, etc.), viable cell density, titer, critical quality attributes, cell status, etc.), and separate local models are used to predict the values ​​of the various different properties, the scheduling process may be implemented separately for each predicted attribute and analytical instrument that measures this attribute, where each attribute may have a different confidence interval width threshold.

[0179] Mathematically, the database maintenance unit 146 can calculate the query point a under the following conditions: * Schedule / trigger (multiple times) new analysis measurements:

[0180] bU -b L > THR, equation (39)

[0181] where THR is a user-defined threshold. In some embodiments, THR can be adjusted by the user to suit a particular application or use case. For example, the user can set a relatively small THR value (used by the database maintenance unit 146) for applications where model reliability is critical, such that model / library maintenance operations occur more frequently. Generally, THR can be set to different values based on process criticality, based on the parameter being predicted (such as media component concentration, media state (e.g., glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na+, K+ and other nutrients or metabolites, pH, pCO2, pO2, temperature, osmotic pressure, etc.), viable cell density, titer, critical quality attributes, cell state, etc.), and / or based on the current time period (e.g., using a lower THR in the later stages of cultivation compared to the initial stage). The choice of THR represents a trade-off between model accuracy and resource (analytical instrument) usage, where a lower threshold tends to increase model accuracy at the cost of increased resource usage.

[0182] Variations of this scheduling protocol are also possible. For example, in one embodiment, the database maintenance unit 146 can apply one or more model performance criteria not only to the current (latest) prediction, but also to one or more other latest predictions (e.g., the latest N predictions, where N > 1). As an example of such an embodiment, the database maintenance unit 146 can calculate the average width of the confidence intervals for the latest N predictions (N ≥ 1), and then compare this average width with the threshold THR. As another example, the database maintenance unit 146 can identify the X largest confidence interval widths (X < Y) among the most recent Y predictions, and schedule / trigger a new analytical measurement only if each of these X widths is greater than the threshold THR.

[0183] Figure 7 is a flowchart of an example method 400 for analyzing a biopharmaceutical process (e.g., for monitoring and / or control purposes). For example, method 400 can be implemented by a computer such as Figure 1 or Figure 2 computer 110 (e.g., by the processing unit 120 executing the instructions of the JITL predictor application 130), and / or implemented by a server such as Figure 1 or Figure 2 database server 112.

[0184] At block 402, a query point associated with a scan of a biopharmaceutical process by a spectroscopic system (e.g., by the Raman analyzer 104 and the Raman probe 106 of the system 100 or the system 150) is determined. For example, the query point can be determined based at least in part on a spectral scan vector (e.g., a Raman scan vector or a NIR scan vector) generated by the spectroscopic system when scanning the biopharmaceutical process. Depending on the embodiment, the query point can be determined based on the raw spectral scan vector or after suitable filtering preprocessing of the raw spectral scan vector. In some embodiments, for example, the query point is also determined based on other information, such as a culture medium configuration associated with the biopharmaceutical process (e.g., fluid type, specific nutrients, pH level, etc.) and / or one or more operating conditions when analyzing the biopharmaceutical process (e.g., metabolite concentration set point, etc.).

[0185] At block 404, an observation database (e.g., observation database 136) is queried. The observation database can contain observation data sets associated with past observations of many biopharmaceutical processes. Each observation data set can include spectral data (e.g., Raman scan vectors or NIR scan vectors) and corresponding analytical measurements (or in some embodiments, two or more analytical measurements). For example, the analytical measurements can be culture medium component concentrations, culture medium states (e.g., glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na+, K+ and other nutrients or metabolites, pH, pCO2, pO2, temperature, osmotic pressure, etc.), viable cell density, titer, critical quality attributes, and / or cell states.

[0186] Block 404 may include selecting as training data those observation data sets from the observation data sets that satisfy one or more relevance criteria with respect to the query point. For example, if the query point includes a spectral scan vector, block 404 may include comparing the spectral scan vector to the spectral scan vector associated with each past observation represented in the observation database (e.g., by calculating (1) the Euclidean distance or other distance between the spectral scan vector on which the query point is based and (2) each spectral scan vector associated with the past observations, and then selecting as training data any spectral scan vectors from the spectral scan vectors associated with the past observations that are determined to be within a threshold distance of the spectral scan vector on which the query point is based).

[0187] At box 406, the selected training data is used to calibrate a local model specific to the biopharmaceutical process being monitored. At box 406, a local model (e.g., local model 132) is trained to predict analytical measurement results based on spectral data input (e.g., Raman spectral scan vectors or NIR spectral scan vectors). In some embodiments, the local model is a Gaussian process machine learning model.

[0188] At box 408, the local model is used to predict analytical measurements of the biopharmaceutical process. Box 408 may include using the local model to analyze spectral data (e.g., Raman scan vectors or NIR scan vectors) generated by the spectroscopic system when scanning the biopharmaceutical process. For example, box 408 may include predicting analytical measurements by using the local model to process the same scan vector or other spectral data on which the query point is based. Depending on the embodiment, the local model can be used to analyze raw spectral data (e.g., raw Raman scan vectors), or to analyze spectral data after appropriate filtering preprocessing of the raw spectral data. In some embodiments, box 408 also includes determining a confidence indicator (e.g., a credibility boundary, a confidence score, etc.) associated with the predicted analytical measurement of the biopharmaceutical process. In some embodiments, at box 408, the local model also predicts one or more additional analytical measurements.

[0189] In some embodiments, method 400 includes Figure 5 408. For example, method 400 may include an additional block in which at least one parameter of the biopharmaceutical process is controlled based, at least in part, on the analytical measurement predicted at block 408. Depending on the embodiment, the parameter may be of the same type as the predicted analytical measurement (e.g., controlling glucose concentration based on the predicted glucose concentration), or may be of a different type. For example, a model predictive control (MPC) technique may be used to control the parameter(s).

[0190] As another example, method 400 may include a first additional box in which actual analytical measurements of a biopharmaceutical process are obtained (e.g., as discussed above, obtained by or from one of the analytical instruments 104 in response to determining that a predicted analytical measurement, and possibly one or more earlier / most recent measurements, do not meet one or more model performance criteria); and a second additional box in which (1) spectral data generated by the spectral system when obtaining the actual analytical measurement and (2) the actual analytical measurement of the biopharmaceutical process are added to an observation database (e.g., by sending the spectral data and the analytical measurement to a database server (such as database server 112), or by directly adding the spectral data and the analytical measurement to a local observation database, etc.). In embodiments in which multiple types of analytical measurements are predicted, multiple actual analytical measurements may be obtained and added to the observation database.

[0191] As yet another example, method 400 may include one or more additional blocks, each similar to blocks 402 through 408. In each of these additional blocks, the local model may be calibrated by querying the observation database (or another observation database) and used to predict different types of analytical measurements.

[0192] Additional considerations related to the present disclosure will now be addressed.

[0193] The terms "polypeptide" or "protein" are used interchangeably throughout the text and refer to a molecule comprising two or more amino acid residues linked to each other by peptide bonds. Polypeptides and proteins also include macromolecules having one or more deletions, insertions and / or substitutions of amino acid residues of the native sequence, i.e., polypeptides or proteins produced by naturally occurring non-recombinant cells; or molecules produced by genetically engineered cells or recombinant cells and comprising one or more deletions, insertions and / or substitutions of amino acid residues having the amino acid sequence of a native protein. Polypeptides and proteins also include amino acid polymers in which one or more amino acids are chemical analogs of the corresponding naturally occurring amino acids and polymers. Polypeptides and proteins also include modifications including, but not limited to, glycosylation, lipid attachment, sulfation, γ-carboxylation of glutamic acid residues, hydroxylation, and ADP-ribosylation.

[0194] Polypeptides and proteins may have scientific or commercial significance, including protein-based therapeutics. Proteins include, among others, secreted proteins, non-secreted proteins, intracellular proteins, or membrane-bound proteins. Polypeptides and proteins may be produced by recombinant animal cell lines using cell culture methods, and may be referred to as "recombinant proteins". The expressed (multiple) proteins may be produced intracellularly or secreted into the culture medium, from which the proteins may be recovered and / or collected. Proteins include proteins that exert therapeutic effects by binding to targets, particularly those listed below, including targets derived therefrom, targets associated therewith, and modifications thereof.

[0195] Protein "antigen binding protein". "Antigen binding protein" refers to a protein or polypeptide that includes an antigen binding region or antigen binding portion that has a strong affinity for another molecule (antigen) to which it binds. Antigen binding proteins encompass antibodies, peptibodies, antibody fragments, antibody derivatives, antibody analogs, fusion proteins (including single-chain variable fragments (scFv) and double-chain (bivalent) scFv, mutant proteins, xMAbs, and chimeric antigen receptors (CAR)).

[0196] scFv is a single-chain antibody fragment that has the variable regions of the heavy and light chains of an antibody linked together. See U.S. Pat. Nos. 7,741,465, and 6,319,494 and Eshhar et al., Cancer Immunol Immunotherapy (1997) 45:131-136. scFv retains the ability of the parent antibody to specifically interact with the target antigen.

[0197] The term "antibody" includes glycosylated immunoglobulins and non-glycosylated immunoglobulins of any isotype or subclass, or an antigen binding region that competes with a complete antibody for specific binding. Unless otherwise indicated, antibodies include human, humanized, chimeric, multispecific, monoclonal, polyclonal, heteroIgG, XmAb, bispecific antibodies, and oligomers or antigen binding fragments thereof. Antibodies include IgG1 type, IgG2 type, IgG3 type, or IgG4 type. Proteins with antigen binding fragments or antigen binding regions are also included, such as Fab, Fab', F(ab')2, Fv, double antibodies, Fd, dAb, maximum antibodies (maxibody), single-chain antibody molecules, single domain VHH, complementary determining region (CDR) fragments, scFv, double antibodies, three antibodies, four antibodies, and polypeptides that contain at least a portion of an immunoglobulin sufficient to bind a specific antigen to a target polypeptide.

[0198] Also included are human, humanized, and other antigen binding proteins, such as human antibodies and humanized antibodies, which do not produce a significant deleterious immune response when administered to humans.

[0199] Also included are peptibodies, which are polypeptides comprising one or more biologically active peptides linked together, optionally via a linker, to an Fc domain. See US Pat. No. 6,660,843, US Pat. No. 7,138,370, and US Pat. No. 7,511,012.

[0200] Proteins also include engineered receptors such as chimeric antigen receptors (CARs or CAR-Ts) and T cell receptors (TCRs). CARs typically combine an antigen binding domain (such as a scFv) with one or more co-stimulatory ("signaling") domains and one or more activation domains in series.

[0201] Also included are bispecific T cell engagers Antibody constructs, which are recombinant protein constructs made of two flexibly linked antibody-derived binding domains (see WO 99 / 54440 and WO 2005 / 040220). One binding domain of the construct is specific for a selected tumor-associated surface antigen on a target cell; the second binding domain is specific for CD3 (a subunit of the T cell receptor complex on T cells). The constructs may also include the ability to bind context independent epitopes at the N-terminus of the CD3s chain (WO 2008 / 119567) to activate T cells more specifically. Half-life extension Constructs include fusions of small bispecific antibody constructs with larger proteins that preferably do not interfere with Therapeutic effects of antibody constructs. Examples of such further developments of bispecific T cell engagers include bispecific Fc-molecules, such as described in US 2014 / 0302037, US 2014 / 0308285, WO 2014 / 151910 and WO 2015 / 048272. An alternative strategy is to use human serum albumin (HAS) fused to a bispecific molecule or a fusion of a human albumin binding peptide only (see, e.g., WO 2013 / 128027, WO2014 / 140358). Another HLE The strategy involves fusing a first domain that binds to a target cell surface antigen, a second domain that binds to an extracellular epitope of human and / or macaque CD3 e chain, and a third domain that serves as a specific Fc pattern (WO 2017 / 134140).

[0202] Also included are modified proteins, such as proteins chemically modified by non-covalent bonds, covalent bonds, or both covalent and non-covalent bonds. Also included are proteins further comprising one or more post-translational modifications, which can be made by cell modification systems or by modifications introduced in vitro or otherwise by enzymes and / or chemical methods.

[0203] The protein may also include a recombinant fusion protein including, for example, a multimerization domain such as a leucine zipper, a coiled coil, the Fc portion of an immunoglobulin, etc. Also included are proteins comprising all or part of the amino acid sequence of a differentiation antigen (referred to as CD proteins) or its ligands or proteins substantially similar to any of these.

[0204] In some embodiments, the protein may include a colony stimulating factor, such as granulocyte colony stimulating factor (G-CSF). Such G-CSF agents include but are not limited to (filgrastim) and (Pegfilgrastim). Also included are erythropoiesis stimulating agents (ESAs), such as (Epoetin α), (darbepoetin alfa), (Epoetin delta), (methoxypolyethylene glycol-epoetin beta), MRK-2578, INS-22, (Epoetin ζ), (Epoetin β), (Epoetin ζ), (epoetin alfa), epoetin alfa Hexal, (Epoetin α), (Epoetine theta), (Epoetine theta), (Epoetin theta), Epoetin alpha, Epoetin beta, Epoetin zeta, Epoetin theta and Epoetin delta, Epoetin omega, Epoetin iota, tissue plasminogen activator, GLP-1 receptor agonist, and molecules of any of the foregoing or variants or analogs and biosimilars thereof.

[0205] In some embodiments, the protein may include proteins that specifically bind to one or more CD proteins, HER receptor family proteins, cell adhesion molecules, growth factors, nerve growth factors, fibroblast growth factors, transforming growth factors (TGF), insulin-like growth factors, osteoinductive factors, insulin and insulin-related proteins, coagulation and coagulation-related proteins, colony stimulating factors (CSF), other blood and serum proteins, blood group antigens; receptors, receptor-associated proteins, growth hormone, growth hormone receptor, T cell receptor; neurotrophic factors, neurotrophins, relaxins, interferons, interleukins, viral antigens, lipoproteins, integrins, rheumatoid factors, immunotoxins, surface membrane proteins, transporters, homing receptors, addressins, regulatory proteins and immunoadhesins.

[0206] In some embodiments, the protein may include a protein that binds to one or more of the following proteins, alone or in any combination: CD proteins (including but not limited to CD3, CD4, CD5, CD7, CD8, CD19, CD20, CD22, CD25, CD30, CD33, CD34, CD38, CD40, CD70, CD123, CD133, CD138, CD171, and CD174), HER receptor family proteins (including, for example, HER2, HER3, HER4, and EGF receptor), EGFRvIII, cell adhesion molecules (e.g., LFA-1, Mol, p150,95, VLA-4, ICAM-1, VCAM, and αv / β3 integrin), growth factors (including but not limited to, for example, vascular endothelial growth factor ("VEGF")); VEGFR2, growth hormone, thyroid stimulating hormone, follicle stimulating hormone, luteinizing hormone, growth hormone releasing factor, parathyroid hormone, mullerian-inhibiting substance (mullerian-inhibiting substance). substance), human macrophage inflammatory protein (MIP-1-α), erythropoietin (EPO), nerve growth factor (such as NGF-β), platelet-derived growth factor (PDGF), fibroblast growth factor (including, for example, aFGF and bFGF), epidermal growth factor (EGF), Cripto, transforming growth factor (TGF) (including TGF-α and TGF-β (including TGF-β1, TGF-β2, TGF-β3, TGF-β4 or TGF-β5)), insulin-like growth factor-I and insulin-like growth factor-II (IGF-I and IGF-II), des (1-3)-IGF-I (brain IGF-I) and osteoinductive factor, insulin and insulin-related proteins (including but not limited to insulin, insulin A chain, insulin B chain, proinsulin and insulin-like growth factor binding protein); (coagulation proteins and coagulation-related proteins, especially factor VIII, tissue factor, von Willebrand (von Willebrand) Willebrand factor, protein C, alpha-1-antitrypsin, plasminogen activators (such as urokinase and tissue plasminogen activator ("t-PA")), bombazine, thrombin, thrombopoietin and thrombopoietin receptors, colony stimulating factors (CSF) (including, among others, the following: M-CSF, GM-CSF, and G-CSF), other blood and serum proteins (including, but not limited to, albumin, IgE, and blood group antigens), receptors and receptor-associated proteins (including, for example, flk2 / flt3 receptors, obesity (OB) receptors, growth hormone receptors, and T-cell receptors);(x) neurotrophic factors, including but not limited to bone-derived neurotrophic factor (BDNF) and neurotrophin-3, neurotrophin-4, neurotrophin-5 or neurotrophin-6 (NT-3, NT-4, NT-5 or NT-6); (xi) relaxin A chain, relaxin B chain and pro-relaxin, interferons (including, for example, interferon α, interferon β and interferon γ), interleukins (IL) (e.g., IL-1 to IL-10, IL-12, IL-15, IL-17, IL-23, IL-12 / IL-23, IL-2Ra, IL1-R1, IL-6 receptor, IL-4 receptor and / or IL-13 receptor, IL-13RA2 or IL-17 receptor, IL-1RAP;(xiv) viral antigens, including but not limited to AIDS envelope virus antigens, lipoproteins, calcitonin, glucagon, atrial natriuretic peptide, pulmonary surfactant, tumor necrosis factor-α and tumor necrosis factor-β, enkephalinase, BCMA, IgKappa, ROR-1, ERBB2, mesothelin, RANTES (normal T cell expressed and secreted factor regulated by activation), mouse gonadotropin-related peptide, DNA enzyme, FR-α, inhibin and activin, integrin, protein A or D, rheumatoid factor, immunotoxin, bone morphogenetic protein (BMP), superoxide dismutase, surface membrane protein, decay accelerating factor (DAF), AIDS envelope, transporter, homing receptor, MIC (MIC-a, MIC-B), ULBP 1-6, EPCAM, addressin, regulatory protein, immunoadhesin, antigen binding protein, growth hormone, CTGF, CTLA4, eotaxin-1, MUC1, CEA, c-MET, claudin-18, GPC-3, EPHA2, FPA, LMP1, MG7, NY-ESO-1, PSCA, ganglioside GD2, ganglioside GM2, BAFF, OPGL (RANKL), myostatin, Dickkopf-1 (DKK-1), An g2, NGF, IGF-1 receptor, hepatocyte growth factor (HGF), TRAIL-R2, c-Kit, B7RP-1, PSMA, NKG2D-1, programmed cell death protein 1 and ligand, PD1 and PDL1, mannose receptor / hCGβ, hepatitis C virus, mesothelin dsFv [PE38 conjugate, Legionella pneumophila (lly), IFNγ, interferon gamma inducible protein 10 (IP10), IFNAR, TALL-1, thymic stromal lymphopoietin (TSLP), proprotein convertase subtilisin / Kexin PCSK9, stem cell factor, Flt-3, calcitonin gene-related peptide (CGRP), OX40L, α4β7, platelet-specific (platelet glycoprotein Iib / IIIb (PAC-1), transforming growth factor β (TFGβ), zona pellucida sperm-binding protein 3 (ZP-3), TWEAK, platelet-derived growth factor receptor α (PDGFRα), sclerostin, and biologically active fragments or variants of any of the foregoing.

[0207] In another embodiment, the protein comprises abciximab, adalimumab, adelimumab, aflibercept, alemtuzumab, alirocumab, anakinra, atacicept, basiliximab, belimumab, bevacizumab, biosozumab, blinatumomab, brentuximab, brodalumab, mocantuzumab, canakinumab, cetuximab, certolizumab, canakinumab, daclizumab , denosumab, eculizumab, edrecolomab, efalizumab, epratuzumab, etanercept, evolocumab, galiximab, ganitazumab, gemtuzumab, golimumab, ibritumomab tiuxetan, infliximab, ipilimumab, leridizumab, ruximab, lxdkizumab, mapatumumab, motesanib phosphate diphosphate), muromonab-CD3, natalizumab, nesiritide, nimotuzumab, nivolumab, ocrelizumab, ofatumumab, omalizumab, oprelleukin, palivizumab, panitumumab, pembrolizumab, pertuzumab, pexelizumab, ranibizumab, rituximab, rituximab, romiplostim, lomosozumab, sargramostim, tocilizumab, tositumomab, trastuzumab, ustekinumab, vedolizumab, visilizumab, volocizumab, zalumab, zalutumumab, and biosimilars of any of the foregoing.

[0208] Protein encompasses all of the foregoing, and further includes antibodies comprising 1, 2, 3, 4, 5 or 6 complementary determining regions (CDRs) of any of the above-mentioned antibodies. Such variants are also included, which include reference amino acid sequences with 70% or higher, particularly 80% or higher, more particularly 90% or higher, more particularly 95% or higher, specifically 97% or higher, more specifically 98% or higher, more specifically 99% or higher homology to the reference amino acid sequence of the protein of interest. Homogeneity in this regard can be determined using a variety of well-known and easily available amino acid sequence analysis software. Preferred software includes those implementing Smith-Waterman (Smith-Waterman) algorithm, which is considered to be a satisfactory solution to the problem of searching and aligning sequences. Other algorithms can also be used, particularly when speed is an important consideration. Common programs for alignment and homology matching of DNA, RNA, and polypeptides that can be used in this regard include FASTA, TFASTA, BLASTN, BLASTP, BLASTX, TBLASTN, PROSRCH, BLAZE, and MPSRCH, the latter being an implementation of the Smith-Waterman algorithm for execution on massively parallel processors manufactured by MasPar.

[0209] Some figures described herein show example block diagrams with one or more functional components. It should be understood that such block diagrams are for illustrative purposes, and the described and illustrated devices may have additional, fewer, or alternative components than those shown. In addition, in various embodiments, components (and the functions provided by the corresponding components) may be associated with any suitable components or otherwise integrated as a part thereof.

[0210] Embodiments of the present disclosure relate to non-transitory computer-readable storage media having computer codes for performing various computer-implemented operations on the non-transitory computer-readable storage media. The term "computer-readable storage medium" is used herein to include any medium capable of storing or encoding a series of instructions or computer codes for performing the operations, methods, and techniques described herein. The media and computer codes may be media and computer codes specifically designed and constructed for the purposes of the embodiments of the present disclosure, or they may be types known and available to those skilled in the art of computer software. Examples of computer-readable storage media include, but are not limited to: magnetic media, such as hard disks, floppy disks, and tapes; optical media, such as CD-ROMs and holographic devices; magneto-optical media, such as optical disks; and hardware devices that are specifically configured to store and execute program codes, such as ASICs, programmable logic devices ("PLDs"), and ROM and RAM devices.

[0211] The example of computer code includes machine code such as produced by a compiler and files containing higher-level codes executed by a computer using an interpreter or a compiler. For example, Java, C++ or other object-oriented programming languages ​​and development tools can be used to implement the embodiments of the present disclosure. Additional examples of computer code include encrypted code and compressed code. In addition, the embodiments of the present disclosure can be downloaded as a computer program product, which can be passed to a requesting computer (for example, a client computer or different server computers) from a remote computer (for example, a server computer) via a transmission channel. Another embodiment of the present disclosure can be implemented by replacing machine executable software instructions or combining them with hard-wired circuitry.

[0212] As used herein, the singular terms "a," "an," and "the" may include plural referents unless the context clearly indicates otherwise.

[0213] As used herein, the terms "connected," "connected," and "connector" refer to an operational coupling or link. Connected components may be coupled to each other directly or indirectly, such as through another set of components.

[0214] As used herein, the terms "substantially", "substantially", "substantially" and "approximately" are used to describe and explain small changes. When used in conjunction with an event or situation, these terms can refer to situations where the event or situation just occurs and situations where the event or situation approximately occurs. For example, when used in conjunction with a numerical value, these terms can refer to a range of variation of less than or equal to ±10% of that numerical value, such as less than or equal to ±5%, less than or equal to ±4%, less than or equal to ±3%, less than or equal to ±2%, less than or equal to ±1%, less than or equal to ±0.5%, less than or equal to ±0.1%, or less than or equal to ±0.05%. For example, if the difference between the values ​​is less than or equal to ±10% of the mean value of the value, such as less than or equal to ±5%, less than or equal to ±4%, less than or equal to ±3%, less than or equal to ±2%, less than or equal to ±1%, less than or equal to ±0.5%, less than or equal to ±0.1%, or less than or equal to ±0.05%, then two numerical values ​​can be considered to be "substantially" the same.

[0215] In addition, quantities, ratios, and other numerical values ​​are sometimes presented herein in a range format. It should be understood that this range format is used for convenience and brevity and should be flexibly interpreted to include the values ​​explicitly specified as the limits of the range, but also include all individual values ​​and subranges contained within that range, as if each value or subrange was explicitly specified.

[0216] Although the disclosure has been described and shown with reference to the specific embodiments of the disclosure, these descriptions and illustrations do not limit the disclosure. It should be understood by those skilled in the art that various changes can be made and equivalents can be substituted without departing from the true spirit and scope of the disclosure defined by the appended claims. These illustrations may not necessarily be drawn to scale. Due to manufacturing processes, tolerances and / or other reasons, there may be differences between the artistic reproduction in the disclosure and the actual device. There may be other embodiments of the disclosure that are not specifically shown. The specification (except for the claims) and the drawings should be regarded as illustrative rather than restrictive. Modifications may be made to adapt specific situations, materials, material compositions, technologies, or processes to the purpose, spirit and scope of the disclosure. All of these changes are intended to fall within the scope of the appended claims. Although the technology disclosed herein has been described with reference to specific operations performed in a specific order, it should be understood that these operations can be combined, subdivided, or reordered to form equivalent technologies without departing from the teachings of the disclosure. Therefore, unless specifically indicated herein, the order and grouping of operations are not limitations on the disclosure.

Claims

1. A computer-implemented method for monitoring and / or controlling a biopharmaceutical process, the method comprising: determining, by one or more processors, a query point associated with a scan of the biopharmaceutical process by the spectroscopic system; querying, by the one or more processors, an observation database comprising a plurality of observation data sets associated with past observations of a biopharmaceutical process, wherein each of the observation data sets comprises spectral data and corresponding actual analytical measurements, and wherein querying the observation database comprises selecting, from the plurality of observation data sets, an observation data set that satisfies one or more relevance criteria with respect to the query point as training data; calibrating, by the one or more processors and using the selected training data, a local model specific to the biopharmaceutical process, the local model being a Gaussian process machine learning model trained to predict analytical measurement results based on spectral data input; predicting, by the one or more processors, analytical measurements of the biopharmaceutical process, wherein predicting the analytical measurements of the biopharmaceutical process comprises using the local model to analyze spectral data generated by the spectroscopy system when scanning the biopharmaceutical process; and The local model is used to determine a confidence indicator associated with a predicted analytical measurement result of the biopharmaceutical process.

2. The computer-implemented method of claim 1, wherein: The spectroscopy system is a Raman spectroscopy system.

3. The computer-implemented method of claim 1 or 2, wherein: Determining a query point includes determining the query point based at least in part on a spectral scan vector generated by the spectroscopic system while scanning the biopharmaceutical process; as well as Selecting the observation data set that satisfies one or more relevance criteria with respect to the query point as training data includes comparing the spectral scan vector based on which the query point is determined with spectral scan vectors associated with past observations of the biopharmaceutical processes.

4. The computer-implemented method of claim 3, wherein: Determining a query point further includes: determining the query point based on a sample number associated with the spectral scan vector; and Selecting a data set of observation results that meets one or more relevance criteria with respect to the query point as training data includes: (i) comparing the spectral scan vector based on which the query point is determined with the spectral scan vectors associated with past observations of these biopharmaceutical processes; and (ii) comparing the number of samples associated with the query point with the number of samples associated with past observations of these biopharmaceutical processes.

5. The computer-implemented method of claim 4, wherein: Selecting a dataset of observations that satisfy one or more relevance criteria about the query point as training data includes: The most recent k observations of the dataset are selected to be included in the training data.

6. The computer-implemented method of any one of claims 3 to 5, wherein: Analytical measurements that predict the biopharmaceutical process include: The local model is used to analytically determine the spectral scan vector upon which the query point is based.

7. A computer-implemented method as claimed in any one of claims 3 to 6, wherein: Selecting a dataset of observations that satisfy one or more relevance criteria about the query point as training data includes: calculating a distance between (i) a spectral scan vector based on which the query point is determined and (ii) a spectral scan vector associated with past observations of the biopharmaceutical processes; and Any of the spectral scan vectors associated with the past observations that are within a threshold distance of the spectral scan vector on which the query point is based is selected as training data.

8. The computer-implemented method of any one of claims 1 to 7, wherein: Determine the query points including: The query point is determined based at least in part on one or both of: (i) a media configuration associated with the biopharmaceutical process, and (ii) one or more operating conditions under which the biopharmaceutical process is analyzed.

9. The computer-implemented method of any one of claims 1 to 8, wherein: Calibration of the local model specific to this biopharmaceutical process includes: The model is calibrated as a function of both the spectral data and the number of samples for a given observation data set.

10. The computer-implemented method of any one of claims 1 to 9, further comprising: At least one parameter of the biopharmaceutical process is controlled by the one or more processors and based at least in part on the predicted analytical measurements of the biopharmaceutical process.

11. The computer-implemented method of any one of claims 1 to 10, wherein: The predicted analytical measurement outcome for the biopharmaceutical process is media component concentration, media status, viable cell density, titer, critical quality attributes, or cell status.

12. The computer-implemented method of any one of claims 1 to 11, wherein: The predicted analytical measurements for this biopharmaceutical process are glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na + or K + concentration.

13. The computer-implemented method of any one of claims 1 to 11, wherein: The predicted analytical measurements for the biopharmaceutical process are pH, pCO2, pO2, temperature or osmolality.

14. The computer-implemented method of any one of claims 1 to 13, further comprising: The actual analytical measurement results of the biopharmaceutical process are obtained by analytical instruments; as well as (i) spectral data generated by the spectral system when obtaining the actual analytical measurement result and (ii) the actual analytical measurement result of the biopharmaceutical process are added to the observation result database by the one or more processors.

15. The computer-implemented method of claim 14, further comprising: determining, by the one or more processors, that at least the predicted analytical measurement result does not satisfy one or more model performance criteria, Wherein obtaining the actual analytical measurement result is performed in response to determining that at least the predicted analytical measurement result does not meet the one or more model performance criteria.

16. The computer-implemented method of claim 15, wherein: Determining that at least the predicted analytical measurement does not meet the one or more model performance criteria includes: generating a confidence interval associated with the predicted analytical measurement result; and This confidence interval is compared to a predefined threshold.

17. The computer-implemented method of any one of claims 1 to 16, wherein: The biopharmaceutical process is a cell culture process.

18. The computer-implemented method of claim 1, further comprising: analyzing the confidence indicator to determine that the confidence indicator exceeds a threshold; In response to the confidence indicator exceeding a threshold, triggering collection of one or more actual analytical measurements; and The local model is calibrated using the one or more actual analysis measurements.

19. A spectroscopic system for monitoring and / or controlling a biopharmaceutical process, the spectroscopic system comprising: one or more spectroscopic probes collectively configured to: (i) deliver source electromagnetic radiation to the biopharmaceutical process, and (ii) collect electromagnetic radiation while delivering the source electromagnetic radiation to the biopharmaceutical process; one or more memories that collectively store an observation database comprising a plurality of observation data sets associated with past observations of a biopharmaceutical process, wherein each of the observation data sets includes spectral data and corresponding actual analytical measurements; and One or more processors configured to: determining a query point associated with a scan of the biopharmaceutical process performed by the spectroscopic system; querying the observation database by at least selecting, from the plurality of observation datasets, an observation dataset that satisfies one or more relevance criteria with respect to the query point as training data; using the selected training data to calibrate a local model specific to the biopharmaceutical process, the local model being a Gaussian process machine learning model trained to predict analytical measurement outcomes based on spectral data input; predicting analytical measurement results of the biopharmaceutical process by at least analyzing spectral data generated by the spectroscopic system when scanning the biopharmaceutical process with the one or more spectroscopic probes using the local model; and The local model is used to determine a confidence indicator associated with a predicted analytical measurement result of the biopharmaceutical process.

20. The spectroscopy system of claim 19, wherein: The spectroscopy system is a Raman spectroscopy system.

21. The spectral system according to claim 19 or 20, wherein: The one or more processors are configured to: determining the query point based at least in part on a spectral scan vector generated by the spectroscopic system while scanning the biopharmaceutical process; as well as The training data is selected by at least comparing the spectral scan vector upon which the query point is determined to spectral scan vectors associated with past observations of the biopharmaceutical processes.

22. The spectroscopy system of claim 21, wherein: The one or more processors are configured to: determining the query point based in part on a sample number associated with the spectral scan vector; as well as The observation data set that satisfies one or more relevance criteria with respect to the query point is selected as training data in part by: (i) comparing the spectral scan vector based on which the query point is determined with spectral scan vectors associated with past observations of the biopharmaceutical processes; and (ii) comparing the number of samples associated with the query point with the number of samples associated with past observations of the biopharmaceutical processes.

23. The spectroscopy system of claim 22, wherein: The one or more processors are configured to select, as training data, a dataset of observations that satisfy one or more relevance criteria with respect to the query point by, in part: The most recent k observations of the dataset are selected to be included in the training data.

24. The spectral system according to any one of claims 19 to 23, wherein: The local model is a function of both the spectral data and the number of samples for a given observation dataset.

25. The spectral system according to any one of claims 19 to 24, wherein: The one or more processors are further configured to: At least one parameter of the biopharmaceutical process is controlled based at least in part on the predicted analytical measurements of the biopharmaceutical process.

26. The spectral system according to any one of claims 19 to 25, wherein: The predicted analytical measurement outcome for the biopharmaceutical process is media component concentration, media status, viable cell density, titer, critical quality attributes, or cell status.

27. The spectral system of any one of claims 19 to 26, wherein: The predicted analytical measurements for this biopharmaceutical process are glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na + or K + concentration.

28. The spectral system of any one of claims 19 to 26, wherein: The predicted analytical measurements for the biopharmaceutical process are pH, pCO2, pO2, temperature or osmolality.

29. The spectroscopy system of any one of claims 19 to 28, further comprising: an analytical instrument configured to obtain actual analytical measurements of the biopharmaceutical process, The one or more processors are further configured to add (i) spectral data generated by the spectral system when obtaining the actual analytical measurement result and (ii) the actual analytical measurement result of the biopharmaceutical process to the observation database.

30. The spectroscopy system of claim 29, wherein: The one or more processors are further configured to: determining that at least the predicted analytical measurement does not satisfy one or more model performance criteria; and The actual analytical measurement result is obtained from the analytical instrument in response to determining that at least the predicted analytical measurement result does not meet the one or more model performance criteria.

31. The spectroscopy system of claim 30, wherein: The one or more processors are configured to determine that at least the predicted analytical measurement result does not meet the one or more model performance criteria by at least: generating a confidence interval associated with the predicted analytical measurement result; and This confidence interval is compared to a predefined threshold.

32. The spectral system of any one of claims 19 to 31, wherein: The biopharmaceutical process is a cell culture process.

33. The spectroscopy system of claim 19, wherein: The one or more processors are configured to: analyzing the confidence indicator to determine that the confidence indicator exceeds a threshold; In response to the confidence indicator exceeding a threshold, triggering collection of one or more actual analytical measurements; and The local model is calibrated using the one or more actual analysis measurements.

34. A non-transitory computer readable medium storing instructions for monitoring and / or controlling a biopharmaceutical process, wherein: These instructions, when executed by one or more processors, cause the one or more processors to perform the method of any one of claims 1 to 18.

35. A bioreactor system comprising: a bioreactor chamber configured to house a biopharmaceutical process; one or more spectroscopic probes collectively configured to: (i) deliver source electromagnetic radiation to the biopharmaceutical process, and (ii) collect electromagnetic radiation while delivering the source electromagnetic radiation to the biopharmaceutical process; one or more memories that collectively store an observation database comprising a plurality of observation data sets associated with past observations of a biopharmaceutical process, wherein each of the observation data sets includes spectral data and corresponding actual analytical measurements; and One or more processors configured to: determining a query point associated with a scan of the biopharmaceutical process performed by the spectroscopic system; querying the observation database by at least selecting, from the plurality of observation datasets, an observation dataset that satisfies one or more relevance criteria with respect to the query point as training data; using the selected training data to calibrate a local model specific to the biopharmaceutical process, the local model being a Gaussian process machine learning model trained to predict analytical measurement outcomes based on spectral data input; predicting analytical measurement results of the biopharmaceutical process by at least analyzing spectral data generated by the spectroscopic system when scanning the biopharmaceutical process with the one or more spectroscopic probes using the local model; and The local model is used to determine a confidence indicator associated with a predicted analytical measurement result of the biopharmaceutical process.

36. The bioreactor system of claim 35, wherein: The one or more processors are configured to: determining the query point based at least in part on a spectral scan vector generated when scanning the biopharmaceutical process via the one or more spectral probes; as well as The training data is selected by at least comparing the spectral scan vector upon which the query point is determined to spectral scan vectors associated with past observations of the biopharmaceutical processes.

37. The bioreactor system of claim 36, wherein: The one or more processors are configured to: determining the query point based in part on a sample number associated with the spectral scan vector; as well as The observation data set that satisfies one or more relevance criteria with respect to the query point is selected as training data in part by: (i) comparing the spectral scan vector based on which the query point is determined with spectral scan vectors associated with past observations of the biopharmaceutical processes; and (ii) comparing the number of samples associated with the query point with the number of samples associated with past observations of the biopharmaceutical processes.

38. The bioreactor system of claim 37, wherein: The one or more processors are configured to select, as training data, a dataset of observations that satisfy one or more relevance criteria with respect to the query point by, in part: The most recent k observations of the dataset are selected to be included in the training data.

39. The bioreactor system of any one of claims 35 to 38, wherein: The local model is a function of both the spectral data and the number of samples for a given observation dataset.

40. The bioreactor system of any one of claims 35 to 39, wherein: The one or more processors are further configured to: At least one parameter of the biopharmaceutical process is controlled based at least in part on the predicted analytical measurements of the biopharmaceutical process.

41. The bioreactor system of any one of claims 35 to 40, wherein: The predicted analytical measurement outcome for the biopharmaceutical process is media component concentration, media status, viable cell density, titer, critical quality attributes, or cell status.

42. The bioreactor system of any one of claims 35 to 41, wherein: The predicted analytical measurements for this biopharmaceutical process are glucose, lactate, glutamate, glutamine, ammonia, amino acids, Na + or K + concentration.

43. The bioreactor system of any one of claims 35 to 41, wherein: The predicted analytical measurements for the biopharmaceutical process are pH, pCO2, pO2, temperature or osmolality.

44. The bioreactor system of any one of claims 35 to 43, further comprising: an analytical instrument configured to obtain actual analytical measurements of the biopharmaceutical process, The one or more processors are further configured to add (i) spectral data generated by the spectral system when obtaining the actual analytical measurement result and (ii) the actual analytical measurement result of the biopharmaceutical process to the observation database.

45. The bioreactor system of claim 44, wherein: The one or more processors are further configured to: determining that at least the predicted analytical measurement does not satisfy one or more model performance criteria; and The actual analytical measurement result is obtained from the analytical instrument in response to determining that at least the predicted analytical measurement result does not meet the one or more model performance criteria.

46. ​​The bioreactor system of claim 45, wherein: The one or more processors are configured to determine that at least the predicted analytical measurement result does not meet the one or more model performance criteria by at least: generating a confidence interval associated with the predicted analytical measurement result; and This confidence interval is compared to a predefined threshold.

47. The bioreactor system of any one of claims 35 to 46, wherein: The biopharmaceutical process is a cell culture process.

48. The bioreactor system of claim 47, wherein: The bioreactor system is configured to perform a cell culture process to produce a recombinant protein.

49. The bioreactor system of claim 35, wherein: The one or more processors are configured to: analyzing the confidence indicator to determine that the confidence indicator exceeds a threshold; In response to the confidence indicator exceeding a threshold, triggering collection of one or more actual analytical measurements; and The local model is calibrated using the one or more actual analysis measurements.

Images