Granted Patent B2

US 12,351,791 · App. 16/907,786 · Granted Jul 8, 2025

Inventors: Wolfgang Paul (Penzberg, DE); Arthur Mohr (Penzberg, DE); Sayuri Hortsch (Munich, DE)

Assignee: Hoffman-La Roche Inc.

C12M41/46C12M41/38C12N5/0018G06N3/02G06N20/00G16B5/00

Abstract

Claims (100)

1. A method for controlling a bioreactor which includes cell culture of cells of a specific cell type based upon predicting a metabolic state of the cell culture of cells of the specific cell type, comprising:

providing a metabolic model of a cell of the specific cell type, the metabolic model including a plurality of intracellular metabolites, a plurality of extracellular metabolites and a plurality of intracellular and extracellular fluxes, the metabolic model comprising stoichiometric equations specifying at least one stoichiometric relationship between one among the plurality of intracellular metabolites and one among the plurality of extracellular metabolites;

at each of a first plurality of points in time during cultivation of the cell culture:

receiving a plurality of measurement values measured at said point in time, said measurement values comprising concentrations of a first set of extracellular metabolites among the plurality of extracellular metabolites in a culture medium of the cell culture and a measured cell density of the cells in the cell culture;

inputting, for each among the first set of extracellular metabolites, first measurement values corresponding to a respective second set of extracellular metabolites as input parameter values into a trained machine learning program logic—MLP—, the first measurement values being among the plurality of measurement values, and the respective second set of extracellular metabolites being among the first set of extracellular metabolites;

predicting extracellular fluxes of the first set of extracellular metabolites at a future point in time by the MLP using the received measurement values, the future point in time being a point in time subsequent to the point in time of receiving the measurement values, wherein the extracellular fluxes are uptake rates of the first set of extracellular metabolites into a cell and/or release rates of the first set of extracellular metabolites from a cell into the culture medium; and

performing metabolic flux analysis to calculate the intracellular fluxes at the future point in time using the predicted extracellular fluxes of the first set of extracellular metabolites and the stoichiometric equations of the metabolic model,

wherein

the respective second set of extracellular metabolites used as input parameter values differs for at least two among the first set of extracellular metabolites whose extracellular flux is to be determined; and

the method further includes

measuring the concentrations of all input candidate metabolites over a second plurality of points in time, wherein the input candidate metabolites comprise all extracellular metabolites that are measurably available in a reference bioreactor with a cell culture of the specific type or comprise all extracellular metabolites among the plurality of extracellular metabolites of the metabolic model;

for each respective extracellular metabolite among the first set of extracellular metabolites whose extracellular flux is to be predicted, performing a selection procedure to identify the respective second set of extracellular metabolites to be used as input parameter values for predicting the extracellular flux of the respective extracellular metabolite, the selection procedure comprising, with respect to the respective extracellular metabolite,

(a) defining a third set of extracellular metabolites, the third set comprising all of the candidate input metabolites;

(b) calculating a first relevance score of each respective candidate input metabolite among the extracellular metabolites in the third set as a function of the measured concentrations of the respective candidate input metabolite, the first relevance score indicating a predictive power of the concentration of the respective candidate input metabolite with respect to the extracellular flux of the respective candidate input metabolite;

(c) transferring only a first candidate input metabolite from the third set to a still empty fourth set of extracellular metabolites, removing the first candidate input metabolite from the third set to obtain an updated third set, the first candidate input metabolite having a highest first relevance score among the extracellular metabolites in the third set;

d) calculating a further relevance score of each respective updated candidate input metabolite among the extracellular metabolites in the updated third set as a function of the measured concentrations of the respective updated candidate input metabolite and the measured concentrations of all extracellular metabolites contained in the fourth set, the further relevance score indicating a predictive power of the concentration of the respective updated candidate input metabolite with respect to the extracellular flux of the respective updated candidate input metabolite taking into account one or more extracellular metabolites contained in the fourth set;

(e) transferring only a first updated candidate input metabolite from the updated third set to the fourth set, removing the first updated candidate input metabolite from the updated third set, the transfer taking place only if, by the transfer of the first updated candidate metabolite, the fourth set does not exceed a maximum informative redundancy limit of the metabolites contained therein with respect to prediction of the extracellular flux of the respective extracellular metabolite among the first set of extracellular metabolites, the first updated candidate input metabolite having a highest first relevance score among the extracellular metabolites in the updated third set;

(f) repeating steps d) and e) until no more metabolites can be transferred to the fourth set without the fourth set exceeding the maximum informative redundancy limit; and

(g) using selectively only the metabolites transferred to the fourth set as input parameter values to predict the extracellular flux of the respective extracellular metabolite among the first set of extracellular metabolites;

 determining a deviation between the calculated intracellular fluxes of at least one of the intracellular metabolites from a corresponding reference value or a corresponding reference value range; and

 changing a state of the bioreactor or a medium contained therein to reduce the deviation, the bioreactor including the cell culture.

2. The method according to claim 1 , further comprising a generation of MLP by machine learning, wherein said generation comprises:

generating a training data set, wherein said generating comprises:

at each respective training point in time among a plurality of training points in time during the cultivation of at least one training cell culture of cells of the specific cell type:

receiving a plurality of measurement values measured at the respective training point in time, said measurement values comprising concentrations of a fifth set of extracellular metabolites of the metabolic model in the culture medium of said at least one training cell culture and a measured cell density of the cells in said at least one training cell culture;

receiving a time indication of the respective training point in time; and

calculating extracellular fluxes of the fifth set of extracellular metabolites as a function of the measured values received at the respective training point in time and the measured values received at a respective preceding point in time, wherein the extracellular fluxes are uptake rates of the fifth set of extracellular metabolites into the cell and/or release rates of the fifth set of extracellular metabolites into the culture medium;

training the MLP, wherein the training comprises:

inputting the measured values received at each of the training points in time as training input parameter values to the MLP, and inputting the extracellular fluxes calculated for each of a plurality of following points in time, each respective following point in time among the plurality of following points in time following a corresponding training point in time among the plurality of training points in time, and the extracellular fluxes being calculated as output parameter values associated with the training input parameter values to the MLP; and

performing a learning process by the MLP in such a way that the MLP learns to predict each respective associated output parameter value among the output parameter values based on a corresponding input training parameter value among the training input parameter values; and

storing the trained MLP in a volatile or non-volatile storage medium.

3. The method according to claim 2 , wherein the training data set is generated such that at each of a plurality of training points in time during the cultivation of multiple training cell cultures of cells of the specific cell type, the measured values and time specifications are received and the extracellular fluxes of the fifth set of extracellular metabolites are calculated, wherein the multiple training cell cultures are cultivated in bioreactors of different types, wherein the different types of bioreactors comprise at least two different bioreactor types from the following bioreactor types: a fed-batch bioreactor, a batch bioreactor, a perfusion reactor, a chemostat and a split-batch bioreactor.

4. The method according to claim 1 , wherein the MLP includes at least one neural network.

5. The method according to claim 1 , wherein the measured concentrations of the first set of extracellular metabolites are each:

an indication of volume-related content of a corresponding metabolite among the first set, in particular a mass concentration or a substance concentration, or

a value which correlates in a linear manner or at least approximately linearly with the volume-related content, in particular a corrected or normalised measured metabolite concentration or a measured flux of a corresponding metabolite among the first set.

6. The method according to claim 1 , wherein the MLP includes three or more sub-MLPs, wherein each of the three or more sub-MLPs has been trained to predict the extracellular flux of a single one among the first set of extracellular metabolites and is selectively used to predict the extracellular flux of the single one among the first set of extracellular metabolites at the future point in time.

7. The method according to claim 1 ,

wherein the first relevance score is calculated as a partial mutual information score—PMI score—between a candidate input metabolite among the third set and the respective extracellular metabolite whose extracellular flow is to be predicted; and

wherein the further relevance score is calculated as a PMI score—between a candidate input metabolite among the third set and the respective extracellular metabolite whose extracellular flow is to be predicted, taking into account the one or more extracellular metabolites contained in the fourth set.

8. The method according to claim 1 , wherein the MLP uses measured concentrations of the respective second set of extracellular metabolites as the input parameter values for predicting the extracellular flux of each respective extracellular metabolite among the first set of extracellular metabolites, wherein the respective second set of extracellular metabolites comprises at least one amino acid.

9. The method according to claim 1 , wherein the plurality of first points in time are separated by time intervals of 10 minutes to 48 hours.

10. The method according to claim 1 , wherein the cell type is a genetically modified cell type which is maintained and/or grown in the bioreactor for the purpose of obtaining a biomolecule.

11. The method according to claim 1 , wherein at each of the future points in time a calculation of a plurality of the intracellular fluxes of the metabolic model is performed.

12. The method according to claim 1 , further comprising:

identifying one or more intracellular metabolites of the metabolic model whose calculated intracellular fluxes deviate from a respective reference value or reference value range by more than a threshold value, the one or more intracellular metabolites including the at least one of the intracellular metabolites, and the respective reference value or reference value range including the corresponding reference value; and

automatically identifying that intracellular flux which acts as a limiting factor for cell growth or production of a desired biomolecule.

13. The method according to claim 1 , further comprising:

comparing the calculated intracellular fluxes with reference values or reference value ranges for intracellular fluxes of respective one or more intracellular metabolites to determine the deviation, the respective one or more intracellular metabolites including the at least one of the intracellular metabolites, and the reference values or reference value ranges including the corresponding reference value;

issuing a warning if the deviation of the calculated intracellular flux of at the least one of the intracellular metabolites from its respective reference value or reference value range exceeds a limit value; and

sending a control command to the bioreactor to automatically initiate steps which change the state of the bioreactor or the medium contained therein in such a way as to reduce the deviation, wherein the automatically initiated steps include in particular a change in a quantity and/or a change in a composition of a culture medium.

14. The method according to claim 13 , further comprising:

identifying a reaction within the metabolic model of the cells which acts as a limiting factor for cell growth or production of a desired biomolecule; and

automatically adding of selectively those substances which modify the intracellular flux acting as a limiting factor in such a way as to promote cell growth or production of the desired biomolecule or a quality of the biomolecule.

15. The method according to claim 1 , further comprising:

calculating current extracellular fluxes of two or more among the first set of extracellular metabolites from the concentrations of the first set of extracellular metabolites measured at a current point in time and at a previous point in time;

performing a further metabolic flux analysis to calculate current intracellular fluxes at the current point in time using the calculated current extracellular fluxes of the two or more among the first set of extracellular metabolites and the stoichiometric equations of the metabolic model; and

using the calculated current intracellular fluxes as a characterization of a current metabolic state of the cells of the cell culture.

16. The method according to claim 1 ,

wherein a concentration of lactate dehydrogenase—LDH—measured in the culture medium of the cell culture is further received at each of the first points in time during the cultivation of the cell culture; and

wherein the prediction of the extracellular fluxes of the first set of extracellular metabolites at each of the future points in time by the MLP is made using a corrected rather than the measured cell density, wherein the calculation of the corrected cell density comprises for each of the first points in time:

calculating the density of lysed cells in the culture medium of the cell culture as a function of the measured LDH concentration, said function being an empirically determined, heuristic and linear function representing the relationship of the LDH concentration in the culture medium to a number of the lysed cells of that specific cell type; and

calculating the corrected cell density as a sum of the measured cell density in the culture medium and the calculated density of the lysed cells.

17. A system, comprising:

one or more processors;

a bioreactor, the bioreactor including a cell culture of cells of a specific cell type;

a first interface for receiving measurements from a bioreactor containing the cell culture; and

a volatile or non-volatile storage medium comprising:

a metabolic model of a cell of the specific cell type, the metabolic model including a plurality of intracellular metabolites, a plurality of extracellular metabolites and a plurality of intracellular and extracellular fluxes, the metabolic model including stoichiometric equations specifying at least one stoichiometric relationship between one among the plurality of intracellular metabolites and one among the plurality of extracellular metabolites;

a trained machine learning program logic—MLP—; and

a program logic that, when executed by the one or more processors, causes the one or more processors to perform a method at each of a first plurality of points in time during cultivation of the cell culture, the method comprising:

receiving a plurality of measurement values measured at said point in time via the first interface, the measurement values comprising concentrations of a first set of extracellular metabolites among the plurality of extracellular metabolites in a culture medium of the cell culture and a measured cell density of the cells in the cell culture;

inputting, for each among the first set of extracellular metabolites, first measurement values corresponding to a respective second set of extracellular metabolites as input parameter values to the MLP, the first measurement values being among the plurality of measurement values, and the respective second set of extracellular metabolites being among the first set of extracellular metabolites;

predicting extracellular fluxes of the first set of extracellular metabolites at a future point in time by the MLP using the received measurement values, the future point in time being a point in time subsequent to the time of receipt of the measurement values, wherein the extracellular fluxes are uptake rates of the first set of extracellular metabolites into a cell and/or release rates of the first set of extracellular metabolites from a cell into the culture medium; and

performing metabolic flux analysis to calculate the intracellular fluxes at the future point in time using the predicted extracellular fluxes and the stoichiometric equations of the metabolic model,

wherein

the respective second set of extracellular metabolites used as input parameter values differs for at least two among the first set of extracellular metabolites whose extracellular flux is to be determined; and

the method further includes

measuring the concentrations of all input candidate metabolites over a second plurality of points in time, wherein the input candidate metabolites comprise all extracellular metabolites that are measurably available in a reference bioreactor with a cell culture of the specific type or comprise all extracellular metabolites among the plurality of extracellular metabolites of the metabolic model;

for each respective extracellular metabolite among the first set of extracellular metabolites whose extracellular flux is to be predicted, performing a selection procedure to identify the respective second set of extracellular metabolites to be used as input parameter values for predicting the extracellular flux of the respective extracellular metabolite, the selection procedure comprising, with respect to the respective extracellular metabolite,

(a) defining a third set of extracellular metabolites, the third set comprising all of the candidate input metabolites;

(b) calculating a first relevance score of each respective candidate input metabolite among the extracellular metabolites in the third set as a function of the measured concentrations of the respective candidate input metabolite, the first relevance score indicating a predictive power of the concentration of the respective candidate input metabolite with respect to the extracellular flux of the respective candidate input metabolite;

(c) transferring only a first candidate input metabolite from the third set to a still empty fourth set of extracellular metabolites, removing the first candidate input metabolite from the third set to obtain an updated third set, the first candidate input metabolite having a highest first relevance score among the extracellular metabolites in the third set;

d) calculating a further relevance score of each respective updated candidate input metabolite among the extracellular metabolites in the updated third set as a function of the measured concentrations of the respective updated candidate input metabolite and the measured concentrations of all extracellular metabolites contained in the fourth set, the further relevance score indicating a predictive power of the concentration of the respective updated candidate input metabolite with respect to the extracellular flux of the respective updated candidate input metabolite taking into account one or more extracellular metabolites contained in the fourth set;

(e) transferring only a first updated candidate input metabolite from the updated third set to the fourth set, removing the first updated candidate input metabolite from the updated third set, the transfer taking place only if, by the transfer of the first updated candidate metabolite, the fourth set does not exceed a maximum informative redundancy limit of the metabolites contained therein with respect to prediction of the extracellular flux of respective extracellular metabolite among the first set of extracellular metabolites, the first updated candidate input metabolite having a highest first relevance score among the extracellular metabolites in the updated third set;

(f) repeating steps d) and e) until no more metabolites can be transferred to the fourth set without the fourth set exceeding the maximum informative redundancy limit; and

(g) using selectively only the metabolites transferred to the fourth set as input parameter values to predict the extracellular flux of the respective extracellular metabolite among the first set of extracellular metabolites;

 determining a deviation between the calculated intracellular fluxes of at least one of the intracellular metabolites from a corresponding reference value or a corresponding reference value range; and

 changing a state of the bioreactor or a medium contained therein to reduce the deviation, the bioreactor including the cell culture.

18. The system according to claim 17 , wherein the method further comprises a generation of MLP by machine learning, wherein said generation comprises:

generating a training data set, wherein said generating comprises:

at each respective training point in time among a plurality of training points in time during cultivation of at least one training cell culture of cells of the specific cell type:

receiving a plurality of measurement values measured at the respective training point in time, said measurement values comprising concentrations of a fifth set of extracellular metabolites of the metabolic model in the culture medium of said at least one training cell culture and a measured cell density of the cells in said at least one training cell culture;

receiving a time indication of the respective training point in time; and

calculating extracellular fluxes of the fifth set of extracellular metabolites as a function of the measured values received at the respective training point in time and the measured values received at a respective preceding point in time, wherein the extracellular fluxes are uptake rates of the fifth set of extracellular metabolites into the cell and/or release rates of the fifth set of extracellular metabolites into the culture medium;

training the MLP, wherein the training comprises:

inputting the measured values received at each of the training points in time as training input parameter values to the MLP, and inputting the extracellular fluxes calculated for each of a plurality of following points in time, each respective following point in time among the plurality of following points in time following a corresponding training point in time among the plurality of training points in time, and the extracellular fluxes being calculated as output parameter values associated with the training input parameter values to the MLP; and

performing a learning process by the MLP in such a way that the MLP learns to predict each respective associated output parameter value among the output parameter values based on a corresponding input training parameter value among the training input parameter values; and

storing the trained MLP in the volatile or non-volatile storage medium.

19. The system according to claim 18 , wherein the training data set is generated such that at each of a plurality of training points in time during cultivation of multiple training cell cultures of cells of the specific cell type, the measured values and time specifications are received and the extracellular fluxes of the fifth set of extracellular metabolites are calculated, wherein the multiple training cell cultures are cultivated in bioreactors of different types, wherein the different types of bioreactors comprise at least two different bioreactor types from the following bioreactor types: a fed-batch bioreactor, a batch bioreactor, a perfusion reactor, a chemostat and a split-batch bioreactor.

Assignments (5)

Priority Claims (1)

Continuity (2)

References Cited (69)