Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F30/00—Computer-aided design [CAD]
  • G06F30/20—Design optimisation, verification or simulation
• • C—CHEMISTRY; METALLURGY
  • C12—BIOCHEMISTRY; BEER; SPIRITS; WINE; VINEGAR; MICROBIOLOGY; ENZYMOLOGY; MUTATION OR GENETIC ENGINEERING
  • C12M—APPARATUS FOR ENZYMOLOGY OR MICROBIOLOGY; APPARATUS FOR CULTURING MICROORGANISMS FOR PRODUCING BIOMASS, FOR GROWING CELLS OR FOR OBTAINING FERMENTATION OR METABOLIC PRODUCTS, i.e. BIOREACTORS OR FERMENTERS
  • C12M41/00—Means for regulation, monitoring, measurement or control, e.g. flow regulation
  • C12M41/48—Automatic or computerized control
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
  • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
  • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B17/00—Systems involving the use of models or simulators of said systems
  • G05B17/02—Systems involving the use of models or simulators of said systems electric

Definitions

• the present invention relates to the field of predicting outcome of and modelling of a process in a bioreactor, especially a method for predicting outcome of and modelling of a process used for manufacturing a sample intended to be used in another system.
• a digital twin may be used to modelling a process, e.g. a bioreactor, to predict the outcome from a process run ahead of time, provided the digital twin has access to a process model that accurately describe the process run. Users spend huge effort on developing protocols for the process runs to ensure maximal growth of cells, and continue to invest in optimizing the process. This is currently through trial and error, vanilla statistical techniques, and experience. Thus, there is a need to develop a procedure for creating process models that can serve as a digital representation of a process used for manufacturing a sample in a biological system, such as a bioreactor.
• Cell growth is a highly non-linear process, following 4 phases of growth - linear, exponential, stationary and death phase. It is also highly variable, and can be influenced in complex fashion by several known and unknown environmental and genetic factors, with the result that two successive cell culture batches can follow very different growth patterns.
• Evaluation of data from bioprocesses is normally performed off-line and post-run. In addition, the evaluation is not done as a comparison to the "expected outcome".
• One reason is the lack of connectivity for different sources of data, e.g. log data, in-process-controls, product quality data, etc.
• Another reason is the inability to adequately model the performance of the bioprocess.
• An object of the present disclosure is to provide methods and devices configured to execute methods and computer programs which seek to mitigate, alleviate, or eliminate one or more of the above-identified deficiencies in the art and disadvantages singly or in any combination.
• the object is achieved by a method for predicting outcome of a process used for manufacturing a sample in a bioreactor, the process belonging to a category.
• the method comprises selecting a process model based on the category; accessing historic data related to past process runs for manufacturing the sample; and accessing current data obtained from a current process run of the process.
• the obtained current data which is based on the selected process model, comprises: process strategy data, bioreactor instrument data, data from online sensors and/or data from offline sensors.
• the method further comprises predicting an outcome of at least one selected parameter of the current process run for manufacturing the sample based on the accessed historic data and current data.
• An advantage is that an undesired behaviour of a selected parameter may detected ahead of time and measures may be instituted that will affect the outcome.
• the object is also achieved by a method for modelling of a process used for manufacturing a sample in a bioreactor, the process belonging to a category.
• the method comprises selecting a process model based on the category; accessing historic data related to past process runs for manufacturing the sample; and accessing current data obtained from a current process run of the process.
• the current data which is based on the selected process model, comprises: process strategy data, bioreactor instrument data, data from online sensors and/or data from offline sensors.
• the method further comprises predicting outcome at least one parameter of the current process run for manufacturing the sample; and updating the process model based on historic data and the monitored at least one parameter when the current process run is completed.
• An advantage is that a process model used to model a process is automatically updated based on the results from the previous process run.
• the object is also achieved by a control system for controlling a process used for manufacturing a sample in a bioreactor, the process belonging to a category.
• the control system is configured to simulate the process and is further configured to select a process model based on the category; access historic data related to past process runs for manufacturing the sample; and access current data obtained from a current process run of the process.
• the obtained data which is based on the selected process model, comprises: process strategy data, bioreactor instrument data, data from online sensors and/or data from offline sensors.
• the control unit is further configured to predict an outcome of at least one selected parameter of the current process run for manufacturing the sample; and to control the process used for manufacturing the sample in a bioreactor based on the predicted outcome of the at least one selected parameter of the current process run.
• Fig. 1 illustrates a control arrangement for modelling of a process used for manufacturing a sample in a bioreactor.
• Fig. 2 illustrates extracting current data from a current process run into a database.
• Fig. 3 shows an example of how different processes are categorized in order to assign a suitable process model to a process.
• Fig. 4 illustrates a system for producing a product based on a sample created in a cell culture process.
• Fig. 5 illustrates a flowchart for adaptive modelling of a process used for manufacturing a sample in a bioreactor.
• Fig. 6 illustrates a flowchart for predicting outcome of a process used for manufacturing a sample in a bioreactor.
• Fig. 7 illustrates a flowchart for predicting features in advance when manufacturing a sample in a bioreactor.
• Figs. 8a-8c illustrate different steps in a process to predict features in advance using the process described in connection with figure 7.
• Figs. 9a-9b illustrate a traditional, and a model based, bioprocess control and data evaluation, respectively.
• process model refers to the proprietary model of the cell culture process which can forecast outcomes of interest and enables "what if analysis for process optimization.
• feed refers to the solution that is added to the culture to prevent nutrient depletion.
• media refers to the base liquid or gel designed to support the growth of cells.
• a typical media comprises of amino acids, vitamins, inorganic salts, glucose, serum etc.
• cell-line refers to a cell culture developed from a single cell and therefore consisting of cells with a uniform genetic make-up.
• the term "clone” refers to an organism or cell, or group of organisms or cells, produced asexually from one ancestor or stock, to which they are genetically identical.
• the term “outcome” refers to the measurable output/product of a cell culture. This can be cells, proteins, by-products like lactate, ammonium etc.
• the term “strategy” refers to the protocol for process parameters like feed regime, instrument set points (e.g. pH, DO, C02) etc.
• supply refers to additional nutrients added apart from feed and base media.
• capture means in the context of a chromatography method the first chromatography step, wherein a large amount of target compound is captured or, for a flow- through process, a large amount of impurities is captured.
• Digital representation of a process used for manufacturing a sample in a biological system is desired in order to be able to evaluate and improve the process before it is used to manufacture the sample.
• a ramification of this premise is that the associated process model, used as a digital representation, may require self-learning capabilities and novel analytics to faithfully represent the biological system.
• a result of the process may be predicted, such as an outcome (e.g. cell viability, cell count, product titre, product quality, etc.).
• an outcome e.g. cell viability, cell count, product titre, product quality, etc.
• instrument parameters e.g. pH, rocking rate, rocking angle, temperature, oxygen/C(3 ⁇ 4 control, etc.
• user control factors e.g. feed, feed strategy, media, clone, glucose Stock Solution, etc.
• outcomes has to be modelled as a function of instrument parameters and measured parameters using offline/online sensors of the biological system, e.g. the cell culture process, during a current process run.
• measured parameters are: pH, dissolved (3 ⁇ 4, C(3 ⁇ 4, Glucose, Glutamine, Glutamate, Lactate, Ammonium, Sodium Ion, Potassium Ion, Osmolality, etc.
• instrument parameters depend on the reactor type used (e.g. shake flask or stir tank) and comprises: pH, rocking rate, rocking angle, stirring rate, impeller speed, temperature, oxygen/C0 2 control, aeration rate, etc. Studies regarding which instrument parameters influence cell growth have been performed, but so far there is no comprehensive model to find optimal values for the instrument parameters and other related parameters, such as calculated amount of supplement provided, feed rates, feed strategies, etc. In the prior art, optimal settings are typically obtained after significant process development and process optimization efforts. However, if an analytic approach is implemented involving domain knowledge, process data and historic data, i.e. incorporating domain knowledge into statistical understanding of the process data, the efforts required to arrive at optimal instrument parameters and perform "what-if ' analysis will be significantly reduced.
• the disclosed process models are analytic in nature as they self-learn from past process runs, or historic process runs, to predict outcomes from the current process run.
• the model can also incorporate information from soft sensors, in line sensors, and commercial asset performance management solutions to refine predictions of outcomes.
• These analytic models automatically fine tune predictions of the current process for manufacturing a sample, e.g. using Kalman Filters.
• anomalies e.g. contaminants, metabolites out of boundary values
• the disclosed process models are able to detect patterns from offline measurements made and thereby detect anomalies early. This information may be used by the operator to take necessary corrective action or actions to improve yield from the process run.
• Figure 1 illustrates a control arrangement for adaptive modelling of a process used for manufacturing a sample in a biological process, such as a bioreactor used in a cell culture process 11.
• the control arrangement comprises a control system 10 which is configured to model the cell culture process and is further configured to: select a process model, access historic data, access current data from a current process run and predict an outcome of at least one selected parameter of the current process run for manufacturing the sample.
• the process model is selected based on the type of process used to manufacture a sample in the cell culture process 11.
• a system to categorize the different processes is disclosed in connection with figure 3, in which each process is assigned to a category and is stored in a database for process models 12.
• Historic data related to past process runs for manufacturing the sample is accessed from a database with historic data 13, and current data is accessed from a current process run of the process, as described in connection with figure 2.
• Historic data comprises data of completed process runs or experiments and the typical data is similar to the data obtained from a current process run.
• the control unit is also configured to predict an outcome of at least one selected parameter. This includes cell viability, cell counts, product titre, product quality, etc.
• a database 14 is used for consolidating historic data, current data and data related to the process model. All data related to the current and past processes, in addition to the data related to the process model are consolidated and stored in a place for easy access.
• the control unit 10 may also be configured to control the process used for manufacturing the sample in a bioreactor based on the predicted outcome of the at least one selected parameter of the current process run, as indicated by the dashed arrow 15.
• Figure 2 illustrates extracting current data from a current process run into a database 14.
• the current data is obtained from the Cell culture process 11 and is based on the selected process model from the process model database 12 and the current data comprises: process strategy data 21, bioreactor data 22 (including instrument data and data from online sensors), and/or data from offline measurements 23.
• Process strategy data 21 comprises strategy information regarding the process as such, e.g. media, feed type, feed regime, supplements (type and concentration) & supplements regime, etc.
• Bioreactor data 22 comprises data related to the process from the Bioreactor instrument (e.g. agitation, aeration, etc.) and any available online sensors attached to the Bioreactor (e.g. pH, dissolved (3 ⁇ 4, etc.).
• Offline measurement data 23 comprises data related to process samples measured on offline sensors (e.g. data for cell count, product titre, metabolites concentration, partial pressure of gasses, etc.).
• Figure 3 shows an example of how different processes are categorized in order to assign a suitable process model to a process.
• the system comprises a number of levels representing the detail of categorization made, such as a default level 30, first level 31, second level 32, third level 33, etc. This information is stored in a database for process models 12. It should be noted that the description below is only an example of how the categorization may be applied, and other features for category selection may be used, e.g. scale-up/scale-down.
• the default level 30, a default process model "D" is a basic process model used and assigned to any process that has not been previously categorized.
• the first level 31 represents in this example process models for based on cell lines, Ci,...,C n , wherein different cell lines may require an adapted process model in order to predict an outcome correctly.
• Each process model may in turn be further adapted based on e.g. reactor type Ri,...,R k , as illustrated in the second level 32.
• some of the process models for cell line Ci and reactor type R3 ⁇ 4 has further been adapted based on media, Mi,...,M j , as illustrated in the third level 33.
• a corresponding process model is selected by the control system 10 and used for a current process run.
• the process to categorize a process into a suitable process model is described in more detail in connection with figures 5 and 6.
• Figure 4 illustrates an example embodiment of a system 40 for producing a product based on a sample created in a cell culture process 11.
• the control system has been described in connection with figure 1 and has access to a consolidated database 14 containing all data related to the current and historic process runs.
• sample refers to a liquid which contains two or more compounds to be separated.
• compound or “product”
• target compound or “target product” means herein any compound which it is desired to separate from a liquid comprising one or more additional compounds.
• a target compound me be a compound desired e.g. as a drug, diagnostic or vaccine; or, alternatively, a contaminating or undesired compound which should be removed from one or more desired compounds.
• the system 40 further comprises a capture step, in this example illustrated by a continuous chromatography system 41, into which the sample from the cell culture process 11 is fed.
• the sample comprising a target product.
• the continuous chromatography system 41 capture the target product to be delivered 42.
• the continuous chromatography system 41 measures a number of parameters in order to obtain an efficient high quality manufacturing process and information 43 regarding impurities, product quality, etc. may be transferred to the control system 10. This information may be used to further adapt the process model and to control the cell culture process 11 in order to increase the performance of the complete system and provide an improved yield.
• Figure 5 illustrates a flowchart for adaptive modelling of a process used for manufacturing a sample in a bioreactor, the process belonging to a category.
• Each process used to manufacture a sample is assigned to a default category, unless a category has been assigned to the current process.
• the process starts at 50 and in step 51, a process model is selected based on the category, as described in connection with figure 3.
• the selected process model defines relevant data to be obtained from the cell culture process 11.
• Step 52 is an optional step, which comprises consolidating historic data, current data and data related to the process model in a database, as described in connection with figure 1.
• historic data is accessed either from a separate database or in a consolidated database.
• the historic data is related to past process runs for manufacturing the sample and comprises data of completed process runs or experiments.
• step 53 current data is accessed either directly from the cell culture process or from a consolidated database.
• the current data comprises: process strategy data, bioreactor instrument data, data from online sensors and/or data from offline sensors, as described in connection with figure 2.
• the accessed data is based on the selected process model, and in step 54 the current data is obtained from a current process run of the process in the cell culture process. This step enables accommodation of additional or less process data of the parameters in comparison to data required by the process model.
• the input parameters are selected, step 54a. Either automatically based on available data and the process model, or user specified input parameters is used, wherein a user of the system selects parameters in addition to the mandatory parameters required by the process model. According to some aspects, data of selected parameters is read to be available for further processing in step 54b.
• Step 55 is an optional step in which missing data in the current data obtained from the current process run is handled and enables the process model to perform data imputation when encountered with missing data.
• a missing data value is replaced with an imputed value based on historical trends, interpolation and predictions based on other available data of the parameters in step 55 a.
• data with missing data values are removed, i.e. clean data with missing values, in step 55b.
• Data is preferably removed only when determined not to materially affect the predictions.
• step 56 at least one parameter of the current process run for manufacturing the sample is monitored and the process model is adapted in real-time based on historic data and/or the monitored at least one parameter when the current process run is completed.
• This step provides the process model a capability to adapt the process model for different cell lines, clones, reactor types, media, etc. This reduces the need to manually build process models for each variation. Self-learning helps to adjust the prediction errors of the process model for improved accuracy.
• This step is to train and update the process models, either by updating a process model or create a new process model, based data from completed process runs or experiments.
• Techniques used for self-learning is Kalman Filters, Fuzzy logic etc.
• the step of adapting the process model further comprises updating the process model for the category, step 56a, using historic data for better predictions or forecasts.
• the process used in the current process run is determined to belong to a new category and the step of adapting the process model further comprises creating a new process model (step 56b) by: assigning the process to the new category; and storing the process model as a new process model.
• the process ends in step 57.
• the process described in connection with figure 5 may be implemented in a computer program for modelling of a process used for manufacturing a sample in a bioreactor.
• the computer program comprises instructions which, when executed on at least one processor, cause the at least one processor to carry out the method described in figure 5.
• the computer program may be stored on a computer-readable storage medium.
• Figure 6 illustrates a flowchart for predicting outcome of a process used for manufacturing a sample in a bioreactor, the process belonging to a category. Each process used to manufacture a sample is assigned to a default category, unless a category has been assigned to the current process.
• the process starts at 60 and continuous to steps 51 , 52, 53, 54 and 55, which have been described in connection with figure 5. These steps include:
• Step 51 selecting a process model based on the category.
• Optional step 52 consolidating historic data, current data and data related to the process model in a database.
• Step 53 accessing historic data related to past process runs for manufacturing the sample in step and accessing current data obtained from a current process run of the process.
• the current data comprises: process strategy data, bioreactor instrument data, data from online sensors and/or data from offline sensors.
• the input parameters are selected, step 54a. Either automatically based on available data and the process model, or user specified input parameters is used. According to some aspects, data of selected parameters is read to be available for further processing in step 54b.
• Optional step 55 handling missing data in the current data obtained from the current process run and enabling the process model to perform data imputation when encountered with missing data.
• a missing data value is replaced with an imputed value based on historical trends, interpolation and predictions based on other available data of the parameters in step 55a.
• data with missing data values are removed, i.e. clean data with missing values, in step 55b.
• step 61 at least one parameter of the current process run for manufacturing the sample is monitored and the process model is adapted based on the monitored at least one parameter during the current process run.
• This step provides the process model a capability to adapt the process model for different cell lines, clones, reactor types, media, etc. This reduces the need to manually build process models for each variation.
• Self-learning helps to adjust the prediction errors of the process model for improved accuracy. The purpose of this step is to train and update the process models, either by temporarily or permanently, based on obtained data from process runs or experiments. Techniques used for self-learning is Kalman Filters, Fuzzy logic etc.
• the step of adapting the process model further comprises updating the process model for the category, step 61a, using new data from the current process run and temporarily updating the process model as "new" process model for the category, i.e. applying the updated process model when predicting the outcome from the current process run.
• the "new" process model is permanently stored for the category in step 61b, i.e. the updated process model is applied when predicting the outcome in future process runs using a process belonging to this category.
• the process used in the current process run is determined to belong to a new category and the step of adapting the process run further comprises assigning the process to the new category; and storing the process model as a new process model.
• the actual predictions during the process run may also be updated based on measured values vs. predicted values.
• the process continuous with the final step 62 in which an outcome of at least one selected parameter of the current process run for manufacturing the sample is predicted based on the accessed historic data and current data.
• the step of predicting the outcome of at least one selected parameter further comprises at least one of the following: predicting a forecast of the at least one parameter, step 63.
• the outcomes of interest are predicted ahead of time, until the end of the current process run.
• predicting a forecast of anomalies step 64.
• Anomalies are predicted ahead of time, e.g. metabolite concentration out of expected ranges in 24 hours if no adjustments is made to the cell culture process.
• the process described in connection with figure 6 may be implemented in a computer program for predicting outcome a current process run for manufacturing a sample in a bioreactor.
• the computer program comprises instructions which, when executed on at least one processor, cause the at least one processor to carry out the method described in figure 6.
• the computer program may be stored on a computer-readable storage medium.
• the disclosed process is a lean approach. It can work with a minimal set of parameters, but can use more parameters if available.
• a method, and a system which has a self-evolving data based approach to learning patterns of cell growth. This can be used for a) Prediction of various outcome parameters/metabolites of the current process runs such as viable cell concentration and product concentration b) Identifying anomalous growth patterns and c) Learn relationships between cell growth outcome parameters and various levers of experimental control.
• a system for predicting features such as viable cell concentration, total cell concentration, product or metabolites days in advance. This is a step wise approach, where coarser predictions are made in the earlier steps, and refined at later steps. It is required that the system has access to a database containing historical data of output and input features from past process runs stored in the database, e.g. indexed by experiment ID, either in raw format or potentially as a knowledge tree based on some distance/nearness criteria such as Euclidean distance, etc.
• Figure 7 illustrates a flowchart for an alternative process for predicting features in advance when manufacturing a sample in a bioreactor.
• the process is a curve evolution approach, which is based on a series of steps to improve the possibility to predict the outcome, or forecasts for a feature, of a process for manufacturing a sample in a bioreactor during a process run.
• step 1) Fitting an overall base model to historic data - illustrated in Fig. 8a.
• This is a base model which takes a simple mean/median/polynomial fit of the historic data in the database versus culture time.
• all the process runs in the database are used.
• a subset of only those process runs which satisfy certain conditions, i.e. pre-existing knowledge e.g. choose only experiment in database which use same media as current process run
• step 2) Improving base model in real time based on curve evolution - illustrated in Fig.
• this step comprises running a pattern match of the current process run's time series until the current time, against all process runs in the database till time T, to identify the closest K matches which satisfy a certain distance threshold.
• This step is repeated at every T. step 3) Building model from fed back errors - Illustrated in Fig. 8c.
• the prediction made for time T+forecast horizon after applying step 2 is further refined by building a model on historical process runs in the database f (or a subset as chosen from step 2), to predict how errors made in all the forecasts till the current time of experiment are able to predict errors made in the future, and update current prediction made after step 2 by this error.
• step 4) Adding covariate information to correct predictions further (optional step). Explicitly using measured metabolites like glucose, lactate, etc. to build a model of errors g after applying step 3 against metabolite information, to further refine forecasts.
• Prediction of how cell growth/product production by cell is going to evolve a few days in advance is a hard problem, as cells are complex and affected by several factors.
• An associated challenge is identifying anomalous growth patterns in advance, and identifying if recovery from anomalous growth is possible by changing the cell environment operationally, and if so, how.
• the system described herein is a continually learning prediction system, which learns from other historical process runs in the past, and also from previous data points of the current experiment in a feedback mode, to issue accurate predictions.
• Model 2 The learning from other process runs step (Model 2) can be used for other associated functions as well, such as identifying bad or unrecoverable process runs, and emerging anomalies by comparing to other flagged anomalous process runs in the past, or deviations from normal process runs.
• this system can also be used to predict interesting experimental outcomes, such as the time when cell growth peaks, the time when cell viability reaches a certain threshold, etc. which can be of use to the experimentalist for logistical purposes.
• the system described herein can also be potentially extended to understand how cell growth is influenced by measured parameters, and in experimental design, by looking at multidimensional cluster patterns.
• Step 71 is an optional step in which conditions for the Base Model, BM, for the process run is set. If this step is omitted, the base model is created based on the historic data from all previous process runs. In optional step 72, historic data matching the conditions set in step 71 is obtained from previous runs (which normally are stored in a database accessible to the system.
• BM Base Model
• step 73 a If not, the process continues to step 73b, where the conditions for the base model is updated before new historic data is obtained in step 72.
• step 74 to create a model, in this disclosure called "Base Model", based on historic data.
• the historic data may be selected based on conditions set in step 71.
• step 74 builds a coarse model based on mean/median/any curve fit, indicated by 80 in Fig. 8a, of all the data which is similar to the current experiment for which prediction has to be made. Similarity can be similar conditions (similar reactor, media, cell line), or a more sophisticated measure of similarity.
• a prediction for the course of the batch for a specific cell line e.g. a CHO cell line
• a specific cell line e.g. a CHO cell line
• All previous process runs in the historical database which have a similar cell line, and have been grown in a similar bioreactor, with similar media are considered and are selected.
• This subselected set of historic process runs is called El.
• the mean/median/polynomial fit of all these process runs El is calculated, and is indicated with reference numeral 80 in Figs 8a and 8b, to get an estimate of prediction for the current process run.
• this "similarity" can be defined any way, it may be a more looser similarity threshold if no sets of previous process runs from the historic data match the conditions of the current process run, or have a tighter threshold if there are exactly run batches with the same conditions in the past.
• This similarity threshold can also depend on the stage of the pharma workflow, a tighter similarity threshold might be desire in manufacturing, and a looser threshold in process development.
• step 75 as the current process run progresses through time, the set of similar process runs used to build the model keep changing. This helps in continuously refining predictions.
• step 75 while a set of process runs El which are similar are selected to build the base model, in step 75, amongst the process runs in El, all previous process runs which are closest to the time series of the current process run up to that particular day are chosen, this model is called Base Model with Learning, BMWL.
• BMWL Base Model with Learning
• the subset of El whose time series from day 0 to day 5 (in a univariate or multivariate sense) is closest to the current process run is considered. Let this subset be E2, and then calculate mean/median/any curve fit, as indicated by reference numeral 81 in Fig.
• step 76 the forecasts issued by step 75 is revised to create finer forecasts.
• the errors made by the forecasts are fed back to improve future forecasts at the next time instant. In other words, we build a model of form
• BMWL(+EC) stands for Base Model with Learning and Error Correction.
• step 77 Residual error correction using metabolite information is applied.
• This step revises forecasts from step 76, by regressing remaining error between step 76 forecasts and actual data against measured metabolite information such as glucose, lactate, ammonia , etc.
• the metabolite information is preferably measured in advance and stored in a database accessible to the system.
• step 78 the data from the current process run is stored as historic data for future process runs.
• the historic data is normally stored in a database accessible to the system.
• the technical advantage with method described above is that it allows for a system which continuously learns from new data, to ensure it can predict output features as accurately as possible. Since this is entirely automated, it also obviates the need to manually build, and update, custom solutions/models for each type of bioreactor or each experimental setup. As it is continuously fed more data, it also keeps getting more powerful with time as its learning repository gets bigger.
• the process described in connection with figure 7 may be implemented in a computer program for predicting outcome a current process run for manufacturing a sample in a bioreactor.
• the computer program comprises instructions which, when executed on at least one processor, cause the at least one processor to carry out the method described in figure 7.
• the computer program may be stored on a computer-readable storage medium.
• the method for predicting forecasts for a feature in a process used for manufacturing a sample in a bioreactor during a process run may be expressed as follow. Values related to the feature is continuously measured during a current process run, and the method comprises:
• the method further comprising performing residual error correction 77 on the revised forecasts using metabolite information.
• the method further comprising setting 71 a set of conditions for the model, and obtaining 72 an amount of historic data from previous process runs to form the selection of historic data used to create the model.
• the method further comprises controlling 73a if the amount of historic data obtained from previous runs is in a predetermined interval, and updating 73b conditions for the model if the amount of historic data is outside the predetermined interval and repeat the step of obtaining 72 an amount of historic data, or proceed to the step of creating 74 a model if the amount of historic data is within the predetermined interval.
• the predetermined interval is selected to be historic data from at least ten previous process runs.
• the predetermined interval is selected to be historic data from not more than one hundred previous process runs.
• the method further comprises storing data from the current process run as historic data for future runs.
• the evaluation of data from bioprocesses is normally performed off-line and post-run, and not done as a comparison to the expected outcome.
• the following is implemented: connect all data-generating sources in the bioprocess (as explained in more detail below).
• Figure 9a illustrates a traditional system setup 90 for bioprocess control and data evaluation.
• This may be used for a parameter related to the cell, e.g. pH, that is produced in a process 91.
• the controller 94 will cause a change of the parameter measured in the process by an addition 95 (in this example CO2 or Base addition) to affect the measured parameter to achieve a desired difference between the set point and the measured value.
• an addition 95 in this example CO2 or Base addition
• the output 96 from the process is a sample for further processing in downstream processes and features, e.g. biomass concentration is measured on the sample.
• Figure 9b illustrates an improved system setup 100 for model based control and data evaluation. This may be used to further improved the prediction of the outcome from the process.
• the improved system setup 100 further comprises a unit 97 for creating a process model.
• Inputs from the addition 95 as well as model parameter estimations 98, are used to estimate at least one selected parameter 99 (Viable Cell Density (VCD), viability, titer (product concentration), concentration of metabolites, partial pressure of carbon dioxide (pC(3 ⁇ 4), buffer capacity, PID parameters, oxygen mass transfer coefficient (k L a), cell specific rates, concentration of lactic acid, etc.)
• VCD Variable Cell Density
• viability viability
• titer product concentration
• concentration of metabolites concentration of metabolites
• pC(3 ⁇ 4) partial pressure of carbon dioxide
• buffer capacity PID parameters
• oxygen mass transfer coefficient (k L a) oxygen mass transfer coefficient
• concentration of lactic acid etc.
• the value of the at least one selected parameter is fed back to the controller 94 and used to control the process 91.
• Each parameter is indicative of a benefit for the user of the system, and provide a soft sensing.
• VCD or viability When VCD or viability is estimated by the model, the harvest time may be estimated. When titer, [metabolites] or pC(3 ⁇ 4 is estimated by the model, sampling frequency may be decreased. When pC(3 ⁇ 4, buffer capacity, PID parameters or 3 ⁇ 4a is estimated, control of the process may be improved and when cell specific rates are estimated On-line data evaluation may be performed.
• bioreactors are used to produce samples for downstream capture processes in a chromatography system.
• the amount of sample produced from the bioreactors has to be adapted to match the capacity of the downstream capture processes. This means that the downstream process may influence the input to the controller 94 and/or when building the model in the unit 97.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• Chemical & Material Sciences (AREA)
• General Physics & Mathematics (AREA)
• Health & Medical Sciences (AREA)
• Theoretical Computer Science (AREA)
• Evolutionary Computation (AREA)
• Computer Hardware Design (AREA)
• General Engineering & Computer Science (AREA)
• Zoology (AREA)
• Automation & Control Theory (AREA)
• Wood Science & Technology (AREA)
• Organic Chemistry (AREA)
• Bioinformatics & Cheminformatics (AREA)
• Life Sciences & Earth Sciences (AREA)
• Analytical Chemistry (AREA)
• Geometry (AREA)
• Genetics & Genomics (AREA)
• General Health & Medical Sciences (AREA)
• Biochemistry (AREA)
• Sustainable Development (AREA)
• Artificial Intelligence (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Microbiology (AREA)
• Medical Informatics (AREA)
• Software Systems (AREA)
• Biotechnology (AREA)
• Biomedical Technology (AREA)
• Apparatus Associated With Microorganisms And Enzymes (AREA)
• Management, Administration, Business Operations System, And Electronic Commerce (AREA)
• Testing And Monitoring For Control Systems (AREA)
• Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)

Priority Applications (5)

Applications Claiming Priority (2)

Related Child Applications (2)

Publications (1)

Family

ID=64659011

Family Applications (1)

Country Status (5)

Cited By (6)

Families Citing this family (6)

Citations (5)

Family Cites Families (27)

• 2018
  • 2018-06-18 WO PCT/IN2018/050398 patent/WO2018229802A1/en not_active Ceased
  • 2018-06-18 CN CN201880052944.XA patent/CN110945593A/zh active Pending
  • 2018-06-18 EP EP18817323.1A patent/EP3639171A4/en active Pending
  • 2018-06-18 JP JP2019569403A patent/JP2020523030A/ja active Pending
  • 2018-06-18 US US16/620,940 patent/US12182482B2/en active Active
• 2024
  • 2024-11-14 US US18/948,020 patent/US20250068794A1/en active Pending

Patent Citations (5)

Non-Patent Citations (3)

Also Published As

Similar Documents

Legal Events