Method for modelling one or more bioprocesses in a bioreactor

Abstract

The invention relates to a method (100) for modelling one or more bioprocesses (2) in a bioreactor (3), comprising the steps of: a) obtaining (101) a trained machine learning model (4), wherein the machine learning model is trained based on measurement data (6) acquired during one or more bioprocesses in one or more bioreactors; b) defining (102) one or more time series parameters (1, 7); and c) modelling (106) one or more other time series parameters (1, 8) of one or more bioprocesses in a bioreactor using the trained machine learning model and the one or more time series parameters defined in step b).

Description

[0001] Title: Method for modelling one or more bioprocesses in a bioreactor

[0002] FIELD

[0003] The present disclosure relates to a method, in particular a computer- implemented method, for modelling one or more bioprocesses in a bioreactor, a computer system, comprising one or more computer system processors, and memory storing instructions that are executable by the one or more computer system processors to perform the computer-implemented method, a computer software program, comprising instructions for performing one or more steps of the method, a computer system, comprising one or more computer system processors for executing the computer software program and memory storing the computer software program, as well as a non-transitory computer-readable medium comprising the computer software program.

[0004] BACKGROUND

[0005] Bioprocess optimization currently takes lots of time, effort and money. Often, many real-life experiments are needed to find optimal bioprocess conditions. Traditional “statistical” design of experiments, for example, requires a lot of experiments to be done.

[0006] EP 4116 403 A1 discloses methods for monitoring, controlling, optimizing and simulating a bioprocess comprising a cell culture in a bioreactor are provided. The methods comprise obtaining the values of one or more state variables of a state space model at one or more maturities, and predicting the value of one or more critical quality attributes of a product of the bioprocess using a machine learning model trained to predict the value of the one or more critical quality attributes based on input variables comprising values of the one or more state variables or variables derived therefrom, at one or more maturities.

[0007] US 2022 / 0284269 A1 discloses analyzing a fermentation process that occurs in a bioreactor. Such a fermentation process may involve microbes consuming a substrate, and producing various metabolites. A computing device may train and execute one or more machine learning models to analyze such a fermentation process. Such a machine learning model may be configured to determine a current fermentation state of such a fermentation process as one example.

[0008] US 2023 / 0168667 A1 discloses real-time quality control of a culture for bioproduction is facilitated using machine learning. In this approach, real-time process data for a set of parameters for a current production run is received. Based on this process data, a prediction is made using an instance of a machine learning model that has been trained on process data from past production or development runs.

[0009] US 2022 / 0411749 A1 discloses a method for optimizing a process (PROC) to produce a biochemical product (P) defined by a quality attribute, the process being controlled by an actuation parameter (C) and being monitored to get a measured value (T). The method includes training a predictive model (PRED) on a training database; and deploying the trained predictive model (PRED) to provide a correction actuation parameter (dC) when a predicted quality attribute (pQA) is out of a targeted quality attribute interval (QAmin, QAmax).

[0010] US 2022 / 0259532 A1 discloses a computer implemented method performed by a controller (C) configured to control a bioprocess comprised in a bioreactor (BR), the method comprising obtaining (410) measurement results by pen forming spectroscopy of a bioprocessing fluid (FL) comprised in the bioreactor (BR), generating bioprocessing parameters using the measurement results, one or more bioprocessing target parameters and one or more trained models, and, controlling the bioprocess using the generated bioprocessing parameters.

[0011] WO 2021 / 164957 A1 discloses methods for monitoring, controlling and simulating a bioprocess comprising a cell culture in a bioreactor. The methods comprise obtaining values of one or more process conditions for the bioprocess at one or more maturities, and determining the specific transport rates of one or more metabolites in the cell culture using the values obtained as input to a machine learning model trained to predict the specific transport rates of the one or more metabolites at a latest maturity of the one or more maturities or a later maturity based at least in part on the values of one or more process conditions for the bioprocess at the one or more preceding maturities.

[0012] WO 2022 / 187818 A1 discloses analyzing a fermentation process that occurs in a bioreactor. Such a fermentation process may involve microbes consuming a substrate, and producing various metabolites. A computing device may train and execute one or more machine learning models to analyze such a fermentation process. Such a machine learning model may be configured to determine a current fermentation state of such a fermentation process as one example. As another example, a machine learning model may be configured to predict metabolite production of a fermentation process based on historical fermentation data and a window of control decisions for the fermentation process.

[0013] WO 2019 / 129891 A1 relates to a method for predicting the metabolic condition of a cell culture of cells of a specific cell type. The method comprises providing a metabolic model of a cell of the specific cell type and carrying out the following steps at each of a plurality of points in time during the cultivation of the cell culture: receiving measured concentrations of several extracellular metabolites and a measured cell density in the culture medium; inputting the received measured values as input parameter values into a trained machine learning program logic - MLP, predicting extracellular flows of the extracellular metabolites at a future point in time by means of the MLP, carrying out a metabolic flow analysis for calculating the intracellular flows at the future point in time on the basis of the predicted extracellular flows and the stoichiometric equations of the metabolic model.

[0014] US 2021 / 0395675 A1 discloses methods for the development and integration of multiple apparatuses and methods for achieving administration of stem cell therapies include precision manufacturing of tissue scaffolds and / or bioreactor substrates. The nano / microscale fiber material extrusion typifying the electrospinning process is married with the fiber alignment and layering characteristic of an additive manufacturing process. The method generates porous fibrous 3-D meshes with precision controlled structures from biopolymer melts and solutions and gels, blends, and suspensions with and without cells. A method of tracking the migration histories and shapes of stem cells on scaffold surfaces relies on immunofluorescent imaging and automated algorithms based on machine learning. The combination of the precision manufacturing method and the method of cell tracking and cell shape statistics, along with understanding of the intimate relationship between the cell shape / phenotype and scaffold architecture leads to an integrated method for cultivating and harvesting cells having desired phenotypes.

[0015] The abovementioned methods, however, are limited to e.g. simulating specific transport rates of one or more metabolites in the cell culture (WO 2021 / 164957 A1), simulating a fermentation process (WO 2022 / 187818 A1), predicting the metabolic condition of a cell culture of cells of a specific cell type (WO 2019 / 129891 A1) or tracking the migration histories and shapes of stem cells on scaffold surfaces (US 2021 / 0395675 A1). Thus, only specific aspects of bioprocesses are modelled and / or only specific types of bioprocesses are modelled.

[0016] An object of the present disclosure is thus to provide a method for modelling one or more bioprocesses in a bioreactor, wherein optimal bioprocess parameters can be achieved with fewer experiments compared to e.g. traditional statistical approaches.

[0017] SUMMARY

[0018] According to the present disclosure, a method for modelling one or more bioprocesses in a bioreactor, such as a single-use or multi-use bioreactor, is provided, comprising the steps of: a) obtaining a trained machine learning model, wherein the machine learning model is trained based on measurement data acquired during one or more bioprocesses in one or more bioreactors; b) defining one or more time series parameters; and c) modelling one or more other time series parameters of one or more bioprocesses in a bioreactor using the trained machine learning model and the one or more time series parameters defined in step b).

[0019] By using the abovementioned method a digital “twin” of a bioreactor can be created, so an end user may run “virtual” experiments to speed up bioprocess optimization. Thus, optimal bioprocess parameters can be achieved with fewer experiments compared to e.g. traditional statistical approaches. Basically, machine learning / AI is “packaged” in such a way that a user, such as a biologist, is able to use it. This has been done by using machine learning / AI to generate a “virtual bioreactor” which can be used to run “virtual cultivations”. The biologist hereby does not have to think about “models”, “parameters” and “simulations”, but simply about (virtual) bioreactors and (virtual) cultivations. By packaging machine learning / AI into a “virtual bioreactor” and “virtual cultivations”, machine learning / AI is made accessible and usable for biologists. Normally, (optimization of) a real-life bioprocess, in particular a cell cultivation process, such as a Chinese hamster ovary (CHO) cell cultivation, may take multiple weeks. The above method, however, allows the biologist to perform a virtual cultivation in a matter of minutes. The time series approach helps the biologist to obtain optimal insight into “what is going on” in the bioreactor time-wise. It should be noted that in the present patent application machine learning and artificial intelligence (Al) are sometimes used interchangeably, i.e. in the present patent application these terms are considered to be equal (although strictly speaking machine learning (ML) is a specific branch of artificial intelligence (Al)).

[0020] Basically, the one or more time series parameters may be used as input parameters for the modelling / predicting and the one or more other time series parameters may be used as output parameters for the modelling. Preferably, the one or more times series parameters are different from the other time series parameters. The method may in principle be used to model any type of relation between the one or more time series parameters (input) and the one or more other time series parameters (output).

[0021] In the context of the present patent application, “time series” generally means a series of data points indexed (or listed or graphed) in time order. Most commonly, a time series is a sequence taken at successive equally spaced points in time. The modelling may thus employ time series forecasting to predict future values based on the trained machine model and previously observed values. Time series data may also have non-equal intervals - time-series data may also be ‘log on change’. This holds for both the model inputs and outputs. The true nature of time series data is that it is (at least) 2D: a parameter consists of (at least) a magnitude value and an accompanying time value.

[0022] In the context of this patent application, a bioreactor generally relates to a manufactured or engineered device or system that supports a biologically active environment. A bioreactor may also be broadly regarded as an artificial (so nonnatural) environment for cell growth. Such a bioreactor usually comprises a vessel in which a biological process or “bioprocess” is carried out which involves organisms or biochemically active substances derived from such organisms. This process can either be aerobic or anaerobic.

[0023] A bioprocess may generally be defined as a biological process that is used in the creation of a material or product.

[0024] A bioreactor may also refer to a device or system designed to grow cells or tissues in the context of cell culture. Such devices are being developed for use in tissue engineering or biochemical engineering. Based on their mode of operation, a bioreactor may be classified as batch, fed batch or continuous. Organisms growing in bioreactors may be submerged in liquid medium or may be attached to the surface of a solid medium. Submerged cultures may be suspended or immobilized.

[0025] In a continuously operated process the organisms will remain in the reactor and byproducts will be removed from the reactor with the effluent. A typical bioreactor comprises the following parts: an agitator or impeller - used for the mixing of the contents of the reactor which keeps the “cells” in the perfect homogenous condition for better transport of nutrients and oxygen to the cells, a baffle - used to break the vortex formation in the vessel, a sparger - to supply adequate gasses (such as oxygen or CO2) to the growing cells, and a jacket - the jacket provides the annular area for circulation of certain temperature of water which keeps the temperature of the bioreactor at a desired value, although the use of Peltier heating or heating blankets is also conceivable. Bioreactors generally are classified as having a volume of more than 20 liters or a volume of less than 20 liters.

[0026] Machine learning models concern data-driven methods aiming to automatically learn the underlying process mechanisms from experimental data. Examples in the present field include the so-called reinforcement learning approach for batch bioprocess optimization, the Artificial Neural Network (ANN) to predict e.g. the growth of microalgae, and the ANN used to e.g. model the rate of change of the dynamic biosystem. Machine learning (ML) can be seen as a set of tools that contribute to the automation of the whole experimental cycle, including model building and practical planning, thus allowing human experts to focus on the more demanding and overarching cognitive tasks.

[0027] The method may further comprise: d) acquiring measurement data of one or more bioprocesses; and e) training the machine learning model based on the measurement data acquired during the one or more bioprocesses in the one or more bioreactors.

[0028] Thus, the user may obtain or acquire measurement data of one or more bioprocesses in order to train the model, instead of obtaining a pre-trained machine learning model.

[0029] The method may further comprise: f) performing one or more bioprocesses in one or more bioreactors for acquiring measurement data of one or more bioprocesses. Thus, the user may actually perform one or more bioprocesses, e.g. in his or her “own” (local) bioreactor, to obtain or acquire measurement data of one or more bioprocesses in order to train the model, instead of obtaining pre-acquired measurement data and / or a pre-trained machine learning model.

[0030] The user may also manually provide additional information, such as bioreactor size, type, impeller type, bioreactor configuration or media composition, as input for the modelling.

[0031] The machine learning model may employ a Gaussian process. Gaussian processes are highly advantageous to e.g. model cell growth, fluorescence or any other time series measurements based on certain bioreactor parameters such as setpoints, cell qualities or any other quantifiable quality effecting the bioreactor parameters. The model can be plotted so the user can analyze and make decisions. The advantage of using Gaussian processes is such processes are suited for situations where data is “expensive”. Fewer datapoints are necessary and the insecurity / uncertainty of the model is indicated allowing the user to make decisions based on this. The model combined with an acquisition function, as mentioned hereafter, may allow for reaching optimal bioprocess parameters with fewer experiments compared to traditional statistical approaches. The strategy can also be tailored to the user’s wishes. Data-driven methods are simple and easy to develop but typically require a large amount of high-quality data to train. In addition, predictions from data-driven models may be unreliable due to their black-box nature, and they may have limited use for conditions outside the training dataset. Gaussian processes are well-suited for such situations, i.e. situations where only limited or low-quality measurement data is available. The skilled person will, however, understand that various machine learning or Al algorithms, and combinations thereof can be used, such as neural nets, genetic algorithms, et cetera, to model a bioprocess (apart from a Gaussian process).

[0032] The method may further comprise: g) outputting a result of the modelling suitable for interpretation by a user.

[0033] The outputting a result of the modelling may comprises displaying a plot or graph of the one or more other time series parameters. The outputting the result of the modelling may for instance also comprise outputting a table or spreadsheet. The plot or graph may be a 2D-plot or graph, with an X-axis and a Y- axis. Thus, an intuitive presentation of the modelling / prediction result is provided to the user / biologist.

[0034] Preferably, one of the X-axis or the Y-axis displays time and the other one of the X-axis or Y-axis displays one of the one or more other time series parameters, such that the user / biologist can easily see what happens with the respective other time series parameter over time.

[0035] The result of the modelling may comprise an uncertainty of one or more of the one or more other parameters of the model. The uncertainty may be indicated by means of a reliability score, i.e. a number or percentage, or visually in a plot, e.g. by displaying a confidence interval.

[0036] The method may further comprise: h) repeating steps d) - e) for acquiring further measurement data to reduce an uncertainty of one or more of the one or more other parameters of the model.

[0037] The steps d) - e) may be repeated until an uncertainty of one or more of the one or more other parameters of the model is reduced below a threshold set by a user. Preferably, outputting the result of the modelling may comprise presenting one or more preferably automated suggestions for obtaining further measurement data (for the one or more time series parameters) to improve the reliability of the “virtual bioreactor”.

[0038] Preferably, the one or more bioprocesses comprise one or more cell cultivation processes. Cell cultivation therein generally refers to the removal of cells from an animal or plant and their subsequent growth in a favorable artificial environment, such as a bioreactor.

[0039] The measurement data and / or the one or more time series parameters and / or the one or more other time series parameters comprise time, pH, DO (dissolved oxygen), feed, infection status, temperature, cell concentration, cell growth, cell quality, cell viability, protein expression and / or fluorescence. The one or more other time series parameters (outputs) may for instance comprise cell concentration, cell growth, cell quality, cell viability, protein expression and / or fluorescence. The one or more time series parameters (inputs) may for instance comprise time, pH, DO, feed, infection status and / or temperature.

[0040] The step of b) defining the one or more time series parameters (inputs) may comprise creating a setpoint profile for the one or more time series parameters. Such setpoint profiles may e.g. be shown to the user in a plot or graph, e.g. a 2D plot or graph, wherein e.g. an X-axis indicates time and a Y-axis shows one or more time series parameters (inputs), basically showing how the one or more time series parameters develop / change over time. The user may then create and move points to define the setpoint profile(s).

[0041] The modelling / predicting may employ one or more acquisition functions for suggesting further measurement data to be acquired when repeating steps d) - e). Thus, optimal bioprocess parameters may be achieved with fewer experiments compared to traditional statistical approaches. Acquisition functions are basically mathematical techniques that guide how a parameter space should be explored during optimization. They may e.g. use the predicted mean and predicted variance generated by a Gaussian process model.

[0042] The one or more acquisition functions may favor exploitation or exploration based on input from the user. Exploitation essentially means “trying slightly different things that have already been proven to be good solutions”. Exploration means “trying totally different things from areas that have not yet been probed”. Exploitation or exploration normally concerns a tradeoff.

[0043] The one or more acquisition functions may comprise an Upper Confidence Bound (UCB) acquisition function. With UCB, the exploitation versus exploration tradeoff is relatively straightforward and easy to tune via e.g. a parameter A. Concretely, UCB is a weighted sum of the expected performance captured by the Gaussian process, and of the uncertainty captured by the standard deviation of the Gaussian process.

[0044] As mentioned before, in embodiments of the aforementioned method the results of the modelling may be used for performing or controlling one or more bioreactor processes.

[0045] One or more steps of the method may be performed on or by a biocontroller during use communicatively connected to a bioreactor.

[0046] The results of the modelling are preferably used by a biocontroller during use communicatively connected to a bioreactor for controlling one or more bioprocesses in a / the bioreactor.

[0047] Bioreactor controllers (or, in short, “biocontrollers”) in general are provided / supplied with one or more bioreactors. Bioreactor controllers contain electronics and software to read from the sensors in the bioreactor. Such bioreactor controllers also contain software to interpret the sensor signals and, based on human preset control parameters, control the bioreactor actuators, such as the impeller motor, the liquid addition pumps, gas flow additions, et cetera. Bioreactor controllers provide an interface, typically in the form of a touch screen or connection to a PC, for the human operator to monitor the bioprocess in the one or more bioreactors connected to the (one or more) bioreactor controller(s). Bioreactor controllers come in many different shapes and sizes, to match the requirements of the bioreactor. A bioreactor controller may be a single cabinet, or a combination of interconnected modular cabinets. One or more bioreactors, controlled by the one or more bioreactor controllers, are together referred to as a bioreactor system.

[0048] The bioreactor controller may be configured as a single or dual bioreactor control system to match bioprocess requirements.

[0049] Another aspect of the disclosure concerns a computer software program, comprising instructions for performing one or more steps, e.g. combinations thereof, of the method.

[0050] Another aspect of the disclosure relates to a computer system, comprising: one or more computer system processors for executing the computer software program; and memory storing the computer software program.

[0051] Another aspect of the disclosure concerns a biocontroller during use communicatively connected to a bioreactor, comprising an aforementioned computer system.

[0052] Another aspect of the disclosure concerns a biocontroller network comprising one or more aforementioned communicatively connected biocontrollers. Such biocontrollers may be communicatively connected, e.g. over a LAN or similar connection.

[0053] Another aspect of the disclosure concerns an aforementioned biocontroller network, further comprising one or more computer servers, wherein the one or more computer servers comprise an aforementioned computer system.

[0054] Another aspect of the disclosure relates concerns a non-transitory computer-readable medium comprising the computer software program. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The disclosure will be explained in more detail below, with reference to illustrative embodiments shown in the drawings. Therein:

[0056] Figure 1 shows an example embodiment of a method according to the present disclosure;

[0057] Figures 2 - 7 shows an example embodiment of a method according to the present disclosure, wherein the method employes an iterative process to reduce uncertainty;

[0058] Figures 8 - 13 show a graphical user interface (GUI) with visual presentations related to example embodiments of the method according to the present disclosure;

[0059] Figure 14 shows an example embodiment of a of a computer system for carrying out the method according to the present disclosure; and

[0060] Figure 15 shows an example embodiment of a bioreactor.

[0061] DETAILED DESCRIPTION

[0062] Figure 1 shows an example embodiment of an at least partly computer-implemented method 100 according to the present disclosure, for modelling one or more bioprocesses 2 in a bioreactor 3.

[0063] Purely for illustration purposes, Figure 15 shows an example embodiment of a bioreactor 3. According to the present disclosure, a bioprocess 2 normally occurs in such a bioreactor 3, in particular a bioprocess 2 comprising the cultivation of cells. The skilled person will understand that the method 100 according to the present disclosure can be used with all sorts of bioreactors 3, not just the one shown in Figure 15, such as single-use or multi-use bioreactors 3.

[0064] The method 100 comprises the steps of: a) obtaining 101 a trained machine learning model 4, wherein the machine learning model 4 is trained based on measurement data 6 acquired during one or more bioprocesses 2 in one or more bioreactors 3; b) defining 102 one or more time series parameters 1 , 7; and c) modelling 103 one or more other time series parameters 1 , 8 of one or more bioprocesses 2 in a bioreactor 3 using the trained machine learning model 4 and the one or more time series parameters 1 , 7 defined in step b).

[0065] The iterative method 100 as shown in Figure 1 may further comprise determining initial inputs at step 108. The initial inputs, such as temperature settings, may be fed to step 106, where an actual experiment 2 is run in a bioreactor 3, or a bioprocess 2 is performed. Subsequently, measurement data 6 is acquired and fed to step 105, wherein a machine learning model 4 is trained based on the acquired measurement data 6 to obtain a trained model 4 at step 101. Preferably, the machine learning model 4 employs a Gaussian process 9, as illustrated in Figures 2 - 7. The trained machine learning model 4 may also be obtained 101 in other ways, such as by obtaining a pre-trained machine learning model 4, as mentioned previously. At step 102 time series parameters 1 , 8 are then defined, such as by the user, as input for the modelling at step 103.

[0066] Step 107 the comprises outputting 107 a result 5 of the modelling 103 suitable for interpretation by a user. The outputting 107 a result 5 of the modelling may comprise displaying a plot or graph 10 of the one or more other time series parameters 1 , 8. As shown in Figures 2 - 7, the plot or graph 10 may be a 2D-plot or graph, with an X-axis and a Y-axis. One of the X-axis or the Y-axis may display time and the other one of the X-axis or Y-axis displays one of the one or more other time series parameters 1 , 8. The result 5 of the modelling 103, such as the plot or graph 10, may comprise an uncertainty 14 of one or more of the one or more other parameters 1 , 8 of the model 4. The computer software program behind the method basically assumes there is a smooth line between datapoints (see e.g. the datapoints 19 in e.g. Figure 3) to form the basis of a prediction function 18, which will more and more approximate the “true” function 15 once more datapoints 19 are added.

[0067] As shown in Figure 1 , the method may further comprise: h) repeating 110 a.o. steps d) - e) for acquiring further measurement data 6 to reduce an uncertainty 14 of one or more of the one or more other parameters 1 , 8 of the model 4. The steps d) - e) may be repeated 110 until an uncertainty 14 of one or more of the one or more other parameters 1 , 8 of the model 4 is reduced below a threshold set by a user. Preferably, the model 4 is tested for many combinations of time series parameters 1 , 7 (inputs) and other time series parameters 1 , 8 (outputs) to determine proper suggestions for next experiments / bioprocess 2 for obtaining further measurement data 6.

[0068] As mentioned before, the one or more bioprocesses 2 may comprise one or more cell cultivation processes.

[0069] The measurement data 6 and / or the one or more time series parameters 1 , 7 and / or the one or more other time series parameters 1 , 8 may comprise time, pH, DO, feed, infection status, temperature, cell concentration, cell growth, cell quality, cell viability, protein expression and / or fluorescence.

[0070] As shown in Figure 11 , b) defining 102 the one or more time series parameters 1 , 7 may comprise creating a setpoint profile 12 for the one or more time series parameters 1 , 7.

[0071] The modelling 103 may employ one or more acquisition functions 11 for suggesting further measurement data 6 to be acquired when repeating steps d) - e). The one or more acquisition functions 11 may favor exploitation or exploration based on input from the user. The one or more acquisition functions 11 may comprise an Upper Confidence Bound (UCB) acquisition function, as shown in Figures 2 - 7.

[0072] Figures 2 - 7 illustrate a “true” function 15 that is to be approximated with predictions 18. A confidence interval is indicated by reference numeral 14. Lower and upper UCB bounds are indicated with reference numerals 16 and 17, respectively. Actual data / measurement points are indicated with reference numeral 19.

[0073] Figure 3 shows a first run or experiment (‘Run T). Figure 7 shows a last run (‘Run 5’). From Figure 3, onwards to Figure 7, measurement data 6 / observation points 19 are consecutively added at the UCB bounds 16, 17, leading to increased machine learning model 4 accuracy for each subsequent run / experiment 2.

[0074] Figures 8 - 13 show a graphical user interface (GUI) with example visual presentations of (steps of) exemplary embodiments of the method 100 according to the present disclosure.

[0075] Figure 8 shows a project management screen, where projects can be set up, in particular by selecting time series parameters / input parameters 1 , 7 (bioprocess parameters), such as time, temperature, pH and dissolved oxygen (DO) and other time series parameters / output parameters 1 , 8 (bioprocess parameters), such as cell viability and cell density. Figure 9 shows example cultivation (experiment / bioprocess 2) data / measurement data 6 to be used as input for the machine learning model 4. The screen shown in Figure 9 shows that data from e.g. 9 cultivations / experiments may be used to “teach” the machine learning model 4.

[0076] Figure 10 shows an example of a “virtual bioreactor” dashboard, where input parameters 7 (DO, pH, temperature and feed), output parameters 8 (cell density, viability and protein expression), as well as a reliability score are shown as a result 5 from the modelling 103. Lower and / or upper ranges may be set for the values of the input parameters 1 , 7, as shown in Figure 10.

[0077] Figure 11 shows an example of a GUI, showing that the step of b) defining 102 the one or more time series parameters (inputs) 1 , 7 may comprise creating one or more setpoint profiles 12 for the one or more time series parameters 1 , 7, such as pH, DO and feed. Such setpoint profiles 12 may e.g. be shown to the user in a plot or graph, e.g. a 2D plot or graph, wherein e.g. an X-axis indicates time and a Y-axis shows one or more time series parameters (inputs) 1 , 7, basically showing how the one or more time series parameters 1 , 7 are to develop / change over time (to provide the user with insight into how the one or more time series parameters 1 , 7 influence the one or more other time series parameters 1 , 8 in the “virtual bioreactor”). The user may create and move points to define the setpoint profile(s) 12. The user may thus “play around” with different setpoint profiles 12.

[0078] Figure 12 shows a GUI, wherein a reliability score is shown, including an explanation for the reliability score, aw well as (automated) suggestions to improve the reliability of the “virtual bioreactor”.

[0079] Figure 13 shows a plurality of 2D-plots or graphs 10, respectively showing the development of cell density and cell viability over time in the “virtual bioreactor”.

[0080] Figure 14 illustrates an example of a computer system 30, which may be an individual computer system 30 or an arrangement of distributed computer systems 30. The computer system 30 includes a computer software program or one or more analysis module(s) 36 configured to perform various tasks according to some embodiments, such as one or more methods 100 disclosed herein. To perform these various tasks, the analysis module 36 executes independently, or in coordination with, one or more processors 31 , which is (or are) connected to one or more storage media 34. The processor(s) 31 is (or are) also connected to a network interface 37 to allow the computer system 30 to communicate over a data network 35 with one or more additional computer systems 30 (note that computer systems 30 may or may not share the same architecture, and may be located in different physical locations, e.g., some computer systems 30 may be located in a processing facility, while in communication with one or more other computer systems 30 that are located in one or more data centers, and / or located in varying countries on different continents).

[0081] A processor 31 can include a microprocessor, microcontroller, processor module or subsystem, programmable integrated circuit, programmable gate array, or another control or computing device.

[0082] The storage media 34 can be implemented as one or more computer- readable or machine-readable storage media 34. Note that while in the example embodiment of Figure 14 storage media 34 is depicted as within computer system 30, in some embodiments, storage media 34 may be distributed within and / or across multiple internal and / or external enclosures of computer system 30 and / or additional computer systems 30. Storage media 34 may include one or more different forms of memory including semiconductor memory devices such as dynamic or static random access memories (DRAMs or SRAMs), erasable and programmable read-only memories (EPROMs), electrically erasable and programmable read-only memories (EEPROMs) and flash memories, magnetic disks such as fixed, floppy and removable disks, other magnetic media including tape, optical media such as compact disks (CDs) or digital versatile disks (DVDs), BLU-RAY® disks, or other types of optical storage, or other types of storage devices. Note that the instructions 33 discussed above can be provided on one computer-readable or machine-readable storage medium 34, or in another embodiment, can be provided on multiple computer-readable or machine readable storage media distributed in a large system having possibly plural nodes. Such computer-readable or machine-readable storage medium 34 or media 34 is (are) considered to be part of an article (or article of manufacture).

[0083] An article or article of manufacture can refer to any manufactured single component or multiple components. The storage medium 34 or media 34 can be located either in the machine running the machine-readable instructions, or located at a remote site from which machine-readable instructions can be downloaded over a network for execution.

[0084] It should be appreciated that computer system 30 may have more or fewer components than shown, may combine additional components not depicted in the example embodiment of Figure 14, and / or computer system 30 may have a different configuration or arrangement of the components depicted in Figure 14. The various components shown in Figure 14 may be implemented in hardware, software, or a combination of both hardware and software, including one or more signal processing and / or application specific integrated circuits.

[0085] Further, the processing methods described herein may be implemented by running one or more functional modules in information processing apparatus such as general purpose processors or application specific chips, such as ASICs, FPGAs, PLDs, or other appropriate devices. These modules, combinations of these modules, and / or their combination with general hardware are all included within the scope of protection of the disclosure, as defined by the appended claims.

[0086] As mentioned before, Figure 15 shows an example embodiment of a bioreactor 3. The one or more steps of the method may be performed on a biocontroller 38 during use communicatively connected to a / the bioreactor 3. The results of the modelling may be used by the biocontroller 38 for controlling, measuring or configuring one or more bioprocesses in the bioreactor 3. The biocontroller 38 may comprise an aforementioned computer system 30. The computer system 30 may be implemented in, or form part of, the hardware of the biocontroller 38.

[0087] Another aspect of the disclosure concerns a biocontroller network 39 comprising one or more communicatively connected biocontrollers 38. The biocontroller network 39 may further comprise one or more computer servers 40, which may comprise an aforementioned computer system 30. A computer server 40 may be broadly interpreted as a smartphone, laptop or desktop, or similar devices.

[0088] The computer server 40 may also store and or process, and or format and / or analyze data from one or more bioreactors 3 or biocontrollers 38 connected to the server.

[0089] The computer server 40 may contain one or more databases, such as one for the experiment data and / or message data and / or account data. The computer server 40 may employ several applications for the different functionalities. The computer server 40 may also manage and control the licenses for the biocontroller 38 and / or server functionalities, possibly through a connection to external servers. The computer server 40 may contain redundant hardware, such as multiple devices for storage and / or processors and / or power supply. Two or more computer server 40 instances may be set up in the biocontroller network 39 for additional redundancy. The one or more servers 40 may have their own interfaces and / or may be reachable through the other interfaces in the biocontroller network 39. The server (functionality) may be provided on external hardware through an internet connection (i.e. a Cloud). The computer server 40 may provide connection to a market place, such as one for bioreactor 3 and / or biocontroller 38 and / or bioreactor system and / or bioreactor network 39 functionalities.

[0090] As mentioned before, bioreactor controllers 38 are provided / supplied with one or more bioreactors 3. Bioreactor controllers 38 contain electronics and software to read from the sensors in the bioreactor 3. Such bioreactor controllers 38 also contain software to interpret the sensor signals and, based on human preset control parameters, control the bioreactor actuators, such as the impeller motor, the liquid addition pumps, gas flow additions, et cetera. Bioreactor controllers 38 provide an interface, typically in the form of a touch screen or connection to a PC, for the human operator to monitor the bioprocess in the one or more bioreactors 3 connected to the (one or more) bioreactor controller(s) 38.

[0091] Bioreactor controllers 38 come in many different shapes and sizes, to match the requirements of the bioreactor 3. A bioreactor controller 38 may be a single cabinet, or a combination of interconnected modular cabinets. The one or more bioreactors 3, controlled by the one or more bioreactor controllers 38, are together referred to as a bioreactor system.

[0092] The bioreactor controller 38 may be configured as a single or dual bioreactor control system to match bioprocess requirements.

[0093] The bioreactor controllers 38 and / or computer servers 40 may communicate via appropriate communication devices (not shown), e.g. over the data network 35 as shown in Figure 14, a LAN, WAN (e.g. cloud) network, and may be implemented by means of Ethernet cabling, Wi-Fi, 5G, et cetera.

[0094] Although the disclosure has been described above with reference to example embodiments, variants within the scope of the present disclosure will readily occur to those skilled in the art after reading the above description. Such variants are within the scope of the independent claims and the dependent claims. In addition, it is to be understood that express rights are requested for variants as described in the dependent claims. It should also be noted that the example embodiments shown in the Figures, or features thereof, may be combined to yield embodiments not explicitly shown in the Figures. LIST OF REFERENCE NUMERALS

[0095] 1. Bioprocess parameters

[0096] 2. Bioprocess

[0097] 3. Bioreactor

[0098] 4. Machine learning model

[0099] 5. Result of the modelling

[0100] 6. Measurement data

[0101] 7. Time series / input parameters

[0102] 8. Other time series / output parameters

[0103] 9. Gaussian process

[0104] 10. Plot / graph of one or more other time series parameters

[0105] 11. Acquisition function

[0106] 12. Setpoint profile

[0107] 13.

[0108] 14. Confidence

[0109] 15. True function

[0110] 16. Lower UCB bound

[0111] 17. Upper UCB bound

[0112] 18. Prediction

[0113] 19. Observation (measured data)

[0114] 30. Computer system

[0115] 31. Processor

[0116] 32. Memory

[0117] 33. Instructions

[0118] 34. Non-transitory computer-readable medium

[0119] 35. Data network

[0120] 36. Analysis module / computer software program

[0121] 37. Network interface

[0122] 38. Biocontroller

[0123] 39. Biocontroller network

[0124] 40. Computer server 100. Method

[0125] 101. Obtain trained model

[0126] 102. Define time series parameters

[0127] 103. Run model / predict 104. Acquire measurement data

[0128] 105. Train model

[0129] 106. Perform bioprocess

[0130] 107. Output results

[0131] 108. Determine initial inputs 109. -

[0132] 110. Output / feed inputs to run next experiment

Claims

CLAIMS1. Method (100) for modelling one or more bioprocesses (2) in a bioreactor (3), comprising the steps of: a) obtaining (101) a trained machine learning model (4), wherein the machine learning model is trained based on measurement data (6) acquired during one or more bioprocesses in one or more bioreactors (3); b) defining (102) one or more time series parameters (1 , 7); c) modelling (103) one or more other time series parameters (1 , 8) of one or more bioprocesses in a bioreactor (3) using the trained machine learning model (4) and the one or more time series parameters (1 , 7) defined in step b).

2. Method (100) according to claim 1 , further comprising: d) acquiring (104) measurement data (6) of one or more bioprocesses (2); and e) training (105) the machine learning model (4) based on the measurement data (6) acquired during the one or more bioprocesses in the one or more bioreactors (3).

3. Method according to claim 2, further comprising: f) performing (106) one or more bioprocesses (2) in one or more bioreactors (3) for acquiring (104) measurement data (6) of one or more bioprocesses.

4. Method (100) according to any one of the preceding claims, wherein the machine learning model (4) employs a Gaussian process (9).

5. Method (100) according to any one of the preceding claims, further comprising: g) outputting (107) a result (5) of the modelling suitable for interpretation by a user.

6. Method (100) according to claim 5, wherein the outputting (107) a result (5) of the modelling comprises displaying a plot or graph (10) of the one or more other time series parameters (1 , 8).

7. Method (100) according to claim 6, wherein the plot or graph (10) is a 2D-plot or graph, with an X-axis and a Y-axis.

8. Method (100) according to claim 7, wherein one of the X-axis or the Y-axis displays time and the other one of the X-axis or Y-axis displays one of the one or more other time series parameters (1 , 8).

9. Method (100) according to any one of the claims 5 - 8, wherein the result (5) of the modelling comprises an uncertainty (14) of one or more of the one or more other parameters (1 , 8) of the model (4).

10. Method (100) according to any one of the preceding claims 2 - 9, further comprising: h) repeating (110) steps d) - e) for acquiring further measurement data(6) to reduce an uncertainty of one or more of the one or more other parameters (1 , 8) of the model (4).

11. Method (100) according to claim 10, wherein steps d) - e) are repeated (110) until an uncertainty of one or more of the one or more other parameters (1 , 8) of the model (4) is reduced below a threshold set by a user.

12. Method (100) according to any one of the preceding claims, wherein the one or more bioprocesses (2) comprise one or more cell cultivation processes.

13. Method (100) according to any one of the preceding claims, wherein the measurement data (6) and / or the one or more time series parameters (1 , 7) and / or the one or more other time series parameters (1 , 8) comprise time, pH, DO, feed, infection status, temperature, cell concentration, cell growth, cell quality, cell viability, protein expression and / or fluorescence.

14. Method (100) according to any one of the preceding claims, wherein b) defining (102) the one or more time series parameters (1 , 7) comprises creating a setpoint profile (12) for the one or more time series parameters (1 , 7).

15. Method (100) according to any one of the preceding claims, wherein the modelling (103) employs one or more acquisition functions (11) for suggesting further measurement data (6) to be acquired when repeating steps d) - e) as found under claims 10 and 11.

16. Method (100) according to claim 15, wherein the one or more acquisition functions (11) favor exploitation or exploration based on input from the user.

17. Method (100) according to claim 15 or 16, wherein the one or more acquisition functions (11) comprise an Upper Confidence Bound (UCB) acquisition function.

18. Method (100) according to any one of the preceding claims, further comprising: using the results of the modelling for performing one or more bioreactor processes.

19. Method (100) according to any one of the preceding claims, wherein one or more steps of the method are performed on a biocontroller (38) during use communicatively connected to a bioreactor (3).

20. Method (100) according to claim 18 or 19, wherein the results of the modelling are used by a biocontroller (38) during use communicatively connected to a bioreactor (3) for controlling one or more bioprocesses in the bioreactor.

21. Computer software program, comprising instructions (23) for performing one or more steps of the method (100) according to any one of the preceding claims.

22. Computer system (30), comprising: one or more computer system processors (31) for executing the computer software program according to claim 21 ; andmemory (32) storing the computer software program according to claim 21 .

23. Biocontroller (38) during use communicatively connected to a bioreactor (3), comprising a computer system (30) according to claim 22.

24. Biocontroller network (39) comprising one or more communicatively connected biocontrollers (38) according to claim 23.

25. Biocontroller network (39) according to claim 24, further comprising one or more computer servers (40), wherein the one or more computer servers comprise a computer system (30) according to claim 22.

26. Non-transitory computer-readable medium (34) comprising the computer software program according to claim 21.

Images