Design at scale method for biomanufacturing

Abstract

Biomanufacturing systems include at least one bioreactor and an at least one in silico prediction algorithm that integrates with said bioreactors. The at least one in silico prediction algorithm predicts and optimizes conditions for at least one bioprocess at a large-scale based on data obtained in said bioreactors.

Description

[0001] DESIGN AT SCALE METHOD FOR BIOMANUFACTURING

[0002] Technical Field

[0003] The disclosure relates to systems and methods for designing and conducting bioprocesses.

[0004] Background

[0005] Industrial fermentation generally refers to the use of fermentation in manufacturing processes. Such processes are useful for the production of high-value to commodity-scale products and ingredients, such as fermented foods and drinks as well as chemicals such as acetic acid, citric acid, lactic acid, and ethanol. Moreover, nearly all commercially produced industrial enzymes, such as amylase, lipase, invertase, rennet, and glucose dehydrogenase are made by fermentation with genetically modified organisms. Industrial fermentation is also used for the commodity-scale production of viable cells ("biomass", in this context), such as yeast for baking or brewing and starter cultures for lactic acid bacteria used in cheesemaking. Broadly speaking, fermentation processes produce different categories of products, biomass, medicines, biologies, chemicals, proteins, small molecules (including volatiles), etc.

[0006] Most industrial fermentation processes rely on organisms, particularly bacteria, algae, or fungi such as yeasts and molds. Some industrial fermentation may involve cell cultures from plants and animals, such as Chinese hamster ovary cells or insect cells. Special considerations are required for the specific organisms and conditions used in the fermentation, such as media composition, gases, nutrients, pH, temperature, and mixing. The titer, yield and productivity of producing the fermented material throughout the bioprocess depends on the concentration and fitness of cells, media and cellular components, and enzymes as well as temperature, pH, and gases. Fermentation generally involves or requires product recovery after the fermentation. Product recovery may involve purifying, concentrating and / or drying of fermented material.

[0007] Summary

[0008] The invention provides systems and methods for designing and conducting a bioprocess, such as fermentation, at an industrial production scale by conducting the bioprocess development in a number of small-scale bioreactors (typically smaller than about IL) to inform at least one prediction algorithm. At least one prediction algorithm predicts and optimizes conditions for the bioprocess at a large scale (typically more than about IL and at least about 10 or 100 times larger than the one or more small-scale bioreactors) based on data obtained from the one or more small-scale bioreactors. It will be appreciated that the scope of the term "about" in describing a value or amount includes both an approximation of the value or amount in question, for example taking into account equipment tolerance and / or measurement error, but also includes the exact value or amount in question. For the purposes of the specification, the terms "small-scale" and "large-scale" are to be understood as describing relative sizes of different bioreactors forming part of the invention. Non-limiting examples of small-scale sizes include ImL, 2mL, lOmL, 50mL, lOOmL, 125mL, 250mL, 500mL, IL and smaller. Non-limiting examples of large-scale sizes include IL, 10L, 100L, 500L, lm3, 5m3, 10m3, 20m3, 50m3, 100m3, 150m3and above. It will be appreciated that, in describing the size of bioreactors used in the invention, each of the terms "small-scale" and "large-scale" may represent a single value or a range of values. The large-scale bioreactor(s) may be larger than the small-scale bioreactor(s) by a factor in the range of 2 to 10 or by a larger factor. A group of small- scale bioreactors may have respective different sizes, and a group of large-scale bioreactors may have respective different sizes, where the large-scale bioreactors are larger than the small-scale bioreactors. The at least one prediction algorithm may implement a large-scale digital model of the bioprocess in a large-scale bioreactor (e.g. a reaction vessel). The algorithm models the bioprocess in the large-scale model to predict the results. A design of experiment engine (or also known as a design of experiment module) may make predictions to design additional experiments to be run in the one or more small- scale bioreactors, e.g., varying strains of organisms or nutrients supplies. Optionally, using an iterative loop, the at least one prediction algorithm refines the large- scale predictions until an intended result is obtained. The resultant predictions may be used to implement an industrial scale fermentation system in one or more large-scale bioreactors (and an example could be a fermentation tank) with appropriate downstream processing equipment, e.g., for cell separation, product filtration and / or purification.

[0009] The predictive algorithm allows the bioprocess to be taken from the one or more small-scale bioreactors directly to one or more large-scale bioreactors, without the requirement of scaling up the reaction through a series of incrementally larger bioreactors (e.g. flasks and vessels). Alternatively, the predictive algorithm allows the bioprocess to be taken from the one or more small-scale bioreactors indirectly to one or more large-scale bioreactors by scaling up the reaction through a lower number of incrementally larger bioreactors (e.g. flasks and vessels), as opposed to the higher number of incrementally larger bioreactors (e.g. flasks and vessels) required by conventional methods and techniques. For example, a bioreactor previously classified as a large-scale bioreactor in a first scale-up step might be later classified in a subsequent second scale-up step as a small- scale bioreactor relative to a larger bioreactor classified as a large-scale bioreactor. Under such circumstances, the predictive algorithm is used in each scale-up step. The number of scale-up steps may vary from one to many. Thus, systems and methods of the invention may be used to quickly scale out to industrial-scale production of large and small molecule products, such as a protein, biologic, chemical, medicine, an enzyme, ingredient, or biomass. Beneficially, the systems and methods of the invention permit the design of the bioprocess and resultant product with industrial conditions in mind from the outset, instead of starting from lab scale conditions and slowly scaling up step-by-step to industrial scale conditions which might be very different from the lab scale conditions.

[0010] Designing the process can involve exploring a large number of variables of the experimental space using the one or more small-scale bioreactors. For example, a large number of different mutants or strains of an organism can be tested in parallel under a large number of different reaction conditions, e.g., in different culture media or with different feedstocks, sugars or co-factors and at different temperatures, pH, mixing and aeration conditions. The at least one prediction algorithm can monitor a bioreactor or multiple bioreactors and discover promising strains and reaction condition combinations. A machine learning back-end engine can design additional experiments and conduct those in the one or more small-scale bioreactors. As the optimal strains and conditions are discovered, the at least one prediction algorithm can model those in the large-scale model to predict and confirm optimal conditions for conducting the bioprocess at the industrial, production scale.

[0011] In certain aspects, the invention provides an integrated biomanufacturing system. The system includes one or more small-scale bioreactors and an at least one in silico prediction algorithm that integrates with said one or more bioreactors. By integration, it is intended that the at least one prediction algorithm combines with, receives data from, or otherwise interacts with said one or more bioreactors. The at least one in silico prediction algorithm predicts and optimizes conditions for at least one bioprocess at a large-scale based on data obtained in said one or more bioreactors, e.g., in the one or more small- scale bioreactors (which may each have a volume less than about IL). The at least one in silico prediction algorithm may be configured to predict and optimize conditions for at least one bioprocess at a large scale in accordance with at least one predefined constraint or objective, which may include but are not limited to productivity (for example kg of final product / h of process, or kg of final product / L of reaction volume / h of process), yield (for example kg of feedstock (such as glucose, dextrose, carbon dioxide, nitrogen-source...) I kg of final product), titer (for example kg of final product / L of reaction volume, as supernatant or not), operations efficiency (for example utilities (such as kWh of energy, m3 of process water, m3 of air) consumed during the process I kg of final product), cost (for example production expenses (including manning) / kg of final product, lead time (for example the number of hours or days for a fermentation cycle in batch or fed-batch), or a combination of two or more thereof. The system may include the bioprocess, e.g., being actively conducted in at least one small-scale bioreactor. A design-of-experiment engine may output experimental conditions to be tested in at least one small-scale bioreactor based on the data obtained. A bioreactor control sub-system may receive the experimental conditions from the design-of-experiment engine or an operator and may implement the experimental conditions in at least one small-scale bioreactor. At least some of the bioreactors (e.g. reaction vessels) may be occupied by an organism or combination of organisms conducting the bioprocess. In some embodiments, the bioprocess may be fermentation. In some embodiments, the organism may be a naturally-occurring or transformed bacteria, yeast, fungi, mold, algae, plant, insect or animal cell and the bioprocess comprises expression of a transgene-encoded protein, enzyme, biologic, medicine, a small molecule or a different type of a molecule. In some embodiments, the bioprocess may comprise the production of biomass, enzyme, protein, small molecule, biologic, medicine, or a different type of a molecule. The at least one in silico prediction algorithm may implement an iterative optimization process for the bioprocess in the one or more small-scale bioreactors based on the data obtained. The iterative optimization process may provide a feedback-based cyclical directed and adaptive evolution design process. The algorithm may be cloud-based or may be stored in a local computer or server.

[0012] In some embodiments, the system may further include a test reactor, e.g., in which the system conducts the bioprocess at the large-scale for confirming fermentation conditions and / or improving the at least one prediction algorithm. The at least one prediction algorithm may be provided by a computer control system in operable communication with the one or more small-scale bioreactors via at least one sensor. The at least one sensor is preferably operable to sense, an equipment is preferably operable to analyze, or the computer control system is preferably operable to change: one or more bioprocess / bioreaction conditions in the one or more small-scale bioreactors (e.g., temperature, pH, gas supply, nutrient supply, compound to induce or repress gene expression, foam production, optical density, cell viability, agitation speed, media feed rate, or product withdrawal rate). The system may include a downstream processing subsystem operable under the control of the at least one prediction algorithm, and / or another prediction algorithm, to process product that has been produced in the test reactor. The downstream processing sub-system is preferably operable to perform an operation on a product of the test reactor such as but not limited to centrifugation, filtration, evaporation, distillation, separation, drying, or precipitation. It will be appreciated that, when multiple operations are performed, the operations may be performed in any sequential order unless there is a technical incompatibility.

[0013] In certain embodiments, the system may include a second set of small-scale bioreactors, and optionally third or additional sets of small-scale bioreactors. Each of the set and the second set may be provided with a hardware outfit. The system may include a rack or housing comprising a plurality of bays, such that each bioreactor may be inserted into, or repositioned between, the bays of the rack. The rack may be placed inside a sterile enclosure and coupled to an autosampler and a sensor system and / or bioreactor control systems that are communicatively coupled to the at least one prediction algorithm.

[0014] The at least one prediction algorithm may be provided by a computer control system in operable communication with the small-scale bioreactors via at least one sensor. Preferably, the at least one prediction algorithm operates on inputs such as organism identities (e.g., strain identifiers) and bioprocess conditions in the small-scale bioreactors. The at least one prediction algorithm may be trained on training data that includes the organism identities, the bioprocess conditions, the data obtained in said small-scale and large-scale bioreactors and / or synthetic data that is generated in silico using, for example, simulations, predictions or other algorithms. In some embodiments, the at least one prediction algorithm may predict and optimize the conditions for the bioprocess at the large-scale using a digital model of the bioprocess in a large-scale bioreactor. Content of the large-scale bioreactor may be represented in the model as a collection of submodels obtained by voxel decomposition or finite element analysis. The at least one prediction algorithm may implement, but not limited to, computational fluid dynamics to predict results of performing the bioprocess in the large-scale bioreactor under the optimized conditions.

[0015] The system is preferably implemented using laboratory and industrial hardware. For example, the one or more small-scale bioreactors may be less than about IL and the one or more large-scale bioreactors may be any volume above than about IL and at least 10 times larger than the small scale. Preferably, the system determines optimal conditions for producing a fermentation product in a large-scale bioreactor using data from the one or more small-scale bioreactors without necessarily needing any bioreactors (e.g. vessels) of a volume intermediate between the one or more small-scale bioreactors and the large- scale bioreactor. That is, using data collected only from at least one small-scale bioreactor, the at least one prediction algorithm can output conditions under which to operate the large-scale bioreactor, without the need to perform test fermentations in flasks or bioreactors of any intermediate size (e.g., between about 10 ml and at least about 10 kL). The at least one prediction algorithm may also design, based on (i) the digital model of the bioprocess in the large-scale bioreactor, (ii) data collected from the one or more small- scale bioreactors, or (iii) both (i) and (ii), additional experiments to be run in the small- scale and large-scale bioreactors to refine the at least one prediction algorithm to optimize conditions for the bioprocess at scale in the tank. The at least one prediction algorithm is preferably operable to design experimental processes to run in the one or more small-scale bioreactors to generate data to be used by the at least one prediction algorithm to (i) design additional experimental processes to generate additional data, and / or (ii) refine optimized conditions for the bioprocess at the large scale. Preferably, the set of small-scale bioreactors comprises a plurality of identical or substantially similar reaction vessels (e.g. identical or substantially similar in shape, size and / or volume) and the large-scale bioreactor has a volume at least about 10 times larger than each small-scale bioreactor. It will be understood that the set of small-scale bioreactors may alternatively comprise dissimilar reaction vessels (e.g. dissimilar in shape, size and / or volume). It will be further understood that the set of small-scale bioreactors may comprise a plurality of identical or substantially similar reaction vessels in combination with one or more dissimilar reaction vessels.

[0016] The system may be used to provide a final, industrial-scale production system. For example, the industrial-scale production system may include a plurality of production reactors operably connected to a downstream processing unit, in which the production reactors perform the bioprocess under the conditions optimized by the at least one prediction algorithm. Each of the production reactors may have a volume of about at least IL.

[0017] In preferred embodiments, the at least one prediction algorithm comprises a machine learning system that is trained on data obtained from the one or more small-scale bioreactors and a model of the bioprocess at the large scale, and / or data collected from one or more large-scale bioreactors, such that the machine learning system is operable to output the target parameters for the variables (including bioprocess conditions for growth and production) involved in the biomanufacturing process. The at least one in silico prediction algorithm may create a digital model of the bioprocess operating at the large- scale.

[0018] Aspects of the disclosure provide biomanufacturing methods. The methods include conducting a bioprocess in at least one small-scale bioreactor or a set of small-scale bioreactors, obtaining data from the one or more small-scale bioreactors, training an at least one in silico prediction algorithm using training data that includes the obtained data and a model of the bioprocess and / or data collected from one or more large-scale bioreactors and / or synthetic data that is generated in silico using, for example, simulations or algorithms, and operating the at least one prediction algorithm to provide optimized conditions for the bioprocess at a large scale based on the training step. The method may include iteratively optimizing reaction conditions in the one or more small-scale bioreactors based on readings provided to the at least one prediction algorithm by a sensing system in sensing communication with the one or more small-scale bioreactors. The iterative optimization may be carried out by a design of experiment engine or module or by an operator. The bioprocess may be conducted under the optimized conditions at the large scale in a large-scale bioreactor and readings may be provided to the at least one prediction algorithm by a sensing system in sensing communication with the large-scale bioreactor. Preferably the at least one prediction algorithm is provided by a computer control system in operable communication with the one or more small-scale bioreactors via at least one sensor and the method includes changing, by the computer control system, one or more bioprocess conditions in the one or more small-scale bioreactors. The method may include sensing, analyzing and / or changing a condition such as but not limited to temperature, pH, gas supply, nutrient supply, foam production, optical density, agitation speed, media feed rate, or product withdrawal rate. The method may include conducting the bioprocess under the optimized conditions at the large scale in a large-scale bioreactor or suitable vessel and processing, by a downstream processing sub-system, a product of the bioprocess downstream of the large-scale bioreactor, e.g., by centrifugation, heating, cooling, distillation, filtration, evaporation, drying or precipitation. The small-scale bioreactors may be provided on a hardware outfit and the method may include inserting or repositioning one or more hardware outfit(s) (each with its own bank of small-scale bioreactors) into bays of a bioreactor rack, in which the rack is coupled to sensor systems and / or bioreactor control systems that are communicatively coupled to the at least one prediction algorithm. The method may include processing, by a computer control system having the at least one prediction algorithm resident therein, inputs that include organism identities and bioprocess conditions in the one or more small-scale bioreactors.

[0019] The at least one prediction algorithm may be trained using training data that include, for example, organism identifiers, the bioprocess conditions, data obtained from the small-scale bioreactors, a predetermined product identity, and / or data collected in one or more large-scale bioreactors, and / or synthetic data that is generated in silico using, for example, simulations or algorithms. The design of experiment engine may be operated to design experimental processes for the one or more small-scale bioreactors to refine optimized conditions for the bioprocess at small scale and / or generate additional data used by the at least one prediction algorithm to (i) make predictions that are fed into the design of experiment engine to design additional experimental processes to generate further additional data, and / or (ii) refine optimized conditions for the bioprocess at the large scale, while improving the performance of the algorithms. The method may include modeling, by the at least one prediction algorithm, the bioprocess in a large-scale bioreactor, in which content of the large-scale bioreactor is represented in a large-scale model by a digital 3D model or voxel decomposition. In some embodiments, the at least one prediction algorithm may apply a computational fluid dynamic model to the large-scale model to predict results of performing the bioprocess in the large-scale bioreactor under the optimized conditions.

[0020] In methods of the invention, each of said small-scale bioreactors preferably has a reaction volume smaller than about IL and a large-scale bioreactor preferably has a reaction volume larger than about IL and at least about 10 times larger than the small- scale bioreactors. The at least one prediction algorithm may include a machine learning system that is trained on data obtained from the one or more small-scale bioreactors and a model of the bioprocess at the large scale, and / or data collected from one or more large- scale bioreactors. The method may include operating the machine learning system to output a design plan for a biomanufacturing system (i.e., that operates at the large scale). The design plan for the biomanufacturing system may specify one or more large-scale bioreactors, control hardware operably coupled to the one or more large-scale bioreactors and operable to control reaction conditions in the one or more large-scale bioreactors, and a downstream processing subsystem operable to process a product from the one or more large-scale bioreactors in one or more process steps. Preferably the set of small-scale bioreactors comprises a plurality of substantially similar reaction vessels and the large- scale bioreactors each have a volume at least about 10 times larger than each reaction vessel, and wherein the design plan for the bioprocess includes conditions optimized using the data from the small-scale bioreactors and under which the bioprocess produces a product to a target standard in the large-scale bioreactors.

[0021] The method may be a computer-implemented method. There may be provided a computer program comprising computer code configured to perform the computer- implemented method. The computer control system may comprise at least one processor and at least one memory including computer program code. The at least one memory and computer program code may be configured to, with the at least one processor, cause the computer control system to perform its function(s), non-limiting examples of which are described throughout the specification. It will be appreciated that references to a memory or a processor may encompass a plurality of memories or processors. Further, the computer control system may be, may include, may communicate with or may form part of one or more of an electronic device, a portable electronic device, a portable telecommunications device, a microprocessor, a mobile phone, a personal digital assistant, a tablet, a phablet, a desktop computer, a laptop computer, a server, a cloud computing network, a smartphone, a smartwatch, smart eyewear, and a module for one or more of the same.

[0022] Systems and methods preferably operate without the need for any intermediatesized bioreactors (e.g. flasks or other vessels).

[0023] Aspects of the invention provide a bioreactor system that includes a bank of small- scale bioreactors comprising at least one biosystem generating a product, a sensing system operable to sense a condition of the small-scale bioreactors, and a control system operable to use the sensed condition and a model of product generation to (i) direct a change to inputs to the small-scale bioreactors to optimize generation of an intended product, and (ii) model operation of the biosystem in a large-scale bioreactor. The control system may model operation of the large-scale bioreactor using, but not limited to, voxel decomposition and / or computational fluid dynamics. The biosystem preferably includes a strain or culture of an organism. Each of the small-scale bioreactors may have a separate biosystem. Preferably the product is a fermentation product (e.g., biomass, a chemical, an ingredient, a protein, an enzyme, or a product of an enzyme-catalyzed reaction). The system may sense but not be limited to temperature, pH, gas supply, organism identity, nutrient or media identity or concentration, product quantity produced, cell viability or optical density. Preferably the system includes the large-scale bioreactor and the control system implements the operation of the biosystem in the large-scale bioreactor. In some embodiments, the large-scale bioreactor has a volume at least about 10 times greater than a volume of the small-scale bioreactors, and the system preferably implements the operation of the biosystem in the large-scale bioreactor without, or minimally, using bioreactors (e.g. reaction vessels) of a size intermediate to the small-scale bioreactors and large-scale bioreactor. The control system may implement a machine learning algorithm to direct the changes to the inputs to the small-scale bioreactors. The system may include a reactor rack with a plurality of bays, sections, or slots, wherein banks of small-scale bioreactors are each provided on a hardware outfit that can be removably positioned on the reactor rack.

[0024] Brief Description of the Drawings

[0025] FIG. 1 shows components of a biomanufacturing system.

[0026] FIG. 2 shows a large-scale bioreactor sub-system.

[0027] FIG. 3 shows a design of experiment engine and prediction algorithm.

[0028] FIG. 4 diagrams a biomanufacturing method.

[0029] Detailed Description

[0030] Biomanufacturing systems include at least one (or a plurality of) small-scale bioreactor, a design-of-experiment engine (DOE) [also known as a design-of-experiment module in this specification], and one or more in silico prediction algorithms that integrate with said bioreactors. The one or more in silico prediction algorithms predict and optimize conditions for at least one (or a plurality of) bioprocess at a large-scale based on data obtained in said bioreactors.

[0031] FIG. 1 shows a biomanufacturing system 101. The system 101 includes at least one (or a plurality of) small-scale bioreactor 105 which may have an individual volume smaller than about IL. The system 101 includes a computer control system 119 in operable communication with the small-scale bioreactors 105 via at least one sensor 133 of at least one sensing subsystem 132.

[0032] The computer control system 119 provides one or more prediction algorithms 113 and a DOE 115 that interface with the small-scale bioreactors 105 via at least one sensor 133 of the sensing subsystem 132. Preferably, the computer control system 119 is operable to change conditions in the small-scale bioreactors 105 through the use of a bioreactor control subsystem 135. If needed, an operator 111 (such as a laboratory technician or biologist or fermentation specialist) would be able to supervise and direct the system 101 through the computer control system 119, such as designing specific bioprocess experiments to conduct in the small-scale bioreactors 105.

[0033] The system 101 may further optionally include any of sampling manager 125 and a waste disposal and degassing unit 127. The sampling manager may command operation of an analyzer subsystem 134, routing commands between the individual offline I at-line analyzing equipment to perform such analysis on the samples of the small-scale bioreactors 105, and the prediction algorithm or the DOE. The operator 111 may operate offline I at- line analyzer subsystem 134 and input the data readings of the offline I at-line analyses in the control system 119 or the relevant algorithms.

[0034] FIG. 2 shows a large-scale bioreactor sub-system 201 that may be operably connected to the system 101. The large-scale bioreactor sub-system 201 preferably includes one or more large-scale bioreactors 205. The large-scale bioreactor sub-system 201 may include any of a sampling manager and router unit 225, downstream processing (DSP) sub-system 215 such as a tank, and offline I at-line analyzer subsystem 229. The large-scale bioreactor sub-system 201 preferably includes a waste disposal and degassing unit 227. The sampling manager may command operation of an analyzer subsystem 229, routing commands between the DSP subsystem 215 to the individual offline I at-line analyzing equipment to perform such analysis on the samples of the large-scale bioreactors 205, and the prediction algorithm or the DOE. The operator 111 may operate offline I at- line analyzer subsystem 229 and its individual equipment and input the data readings of the offline I at-line analyses in the control system 119 or the relevant algorithms.

[0035] With reference to the system 101, the sensing subsystem 132 may have one or any number of sensors 133 for sensing, but not limited to e.g., temperature, pH, gas supply, nutrient supply, foam production, optical density, cell viability, dissolved oxygen, agitation speed, medium composition and feed rate, product withdrawal rate, contamination risk, or other conditions. The sensors 133 and the sensing subsystem 132 and the small-scale bioreactors 105 should be designed (in terms of position, type and number of sensors, with respect to the small-scale bioreactors 105 geometry and volume constraints) to collect the relevant data required for the algorithms to predict and / or optimize, and most likely be able to sense directly inside the small-scale bioreactor 105, i.e., traditionally referred to as "inline" or "online" sensing. With this approach, the hardware design fits the algorithm needs and not vice versa. The control subsystem 135 may similarly have controllers or suppliers to optimize one or more of but not limited to temperature, pH, gas supply, nutrient supply, compound to induce or repress gene expression, foam production, agitation speed, media feed rate, product withdrawal rate, or others. For example, for temperature, the sensing subsystem 132 may include a thermometer and the control subsystem 135 may include a Peltier heater or similar. The control subsystem 135 may include one or more supply lines for media and or gas(es), e.g., each under control of valves that may be controlled by the computer control system 119. The sensing subsystem 132 may have one or any number of spectrophotometers and cell viability probes built in, e.g., to read optical density and assess physiological fitness of each small-scale bioreactor 105. The control subsystem 135 may comprise an agitator, e.g., a magnetic stirrer for each small-scale bioreactor 105 (e.g., with a stir magnet in each small-scale bioreactor 105) or impellers.

[0036] For parameters involving sampling for offline or at-line measurements from small- scale bioreactors 105, which may not be managed by the sensing subsystem 132, measurements may be performed by the analyzer subsystem 134 which may include any suitable equipment including, for example, a plate reader, a high-performance liquid chromatography (HPLC) unit, gas chromatography (GC), ion chromatography (IC) a microscope or signal detector (e.g., photodiode, image sensor,, pH sensor, thermometer, mass spectrometer (MS), etc.), a rheometer (to measure viscosity within small-scale bioreactors 105 in response to stress or agitation), others, and optionally, equipment specific to a downstream processing (DSP) subsystem 215 related to the volumes to process when relevant for large-scale bioreactors 205. The DSP subsystem 215 may include a DSP tank with process flow online sensing, and related DSP equipment (e.g. a centrifuge, clarifier, concentrator, filtration units, a lyophilizer, a dryer, etc.).

[0037] In one example, the prediction algorithm 113 may include a machine learning algorithm such as a neural network such that data from the sensing subsystem 132 is transformed into vector space as input to the machine learning algorithm. The in silico prediction algorithm 113 predicts and optimizes conditions for a bioprocess at a large-scale based on data obtained in said small-scale bioreactors. Additionally, the DOE 115 may design a next batch of experiment to be run in small-scale bioreactors to obtain data best suited for the in silico prediction algorithm.

[0038] In preferred embodiments, the prediction algorithm 113 builds or maintains a large- scale model 141 of the bioprocess. The system 101 may optionally include a large-scale bioreactor 205. The large-scale bioreactor 205 may be larger than about IL and at least about 10 times larger than the small-scale bioreactors, for example, a 50 kL stainless steel tank or similar, as is known in the art. See Lee, 2020, Scaling up from the bench : Fermentation tank, Chapter 19 in Boxes: A Field Guide, Bauer et al., Eds., Mattering Press (pp. 289-304), incorporated by reference. The large-scale bioreactors will also have their own sensing subsystem and their own control subsystem. Sensors and sensing subsystems are designed (in terms of position, type and number of sensors, with respect to the large- scale bioreactor 205 geometry and volume constraints) to collect the relevant data required for the algorithm to predict and / or optimize, and are preferably able to sense directly inside the large-scale bioreactor 205, i.e., traditionally referred to as "inline" or "online" sensing. Like with the small-scale bioreactors 105, with this approach, the hardware design fits the algorithm needs and not vice versa. The large scale model 141 of the bioprocess is a digital entity in the computer control system 119. The large scale model 141 preferably provides a digital predictor of conditions inside the large-scale bioreactor 205 when the large-scale bioreactor 205 is included in the system 101 and when the bioprocess is active in the large- scale bioreactor 205.

[0039] For example, the bioprocess may be the production of an antimicrobial peptide or a food ingredient or a nanobody or a cannabinoid or otherwise chosen small or large molecule, and each small-scale bioreactor 105 may be occupied by living cells of Saccharomyces cerevisiae or another yeast or mold or bacteria or fungi or algae or another organism. The predictive algorithm 113 learn the progress of the bioprocess in the small- scale bioreactors 105 in the form of data collected online by sensing subsystem 132 and / or offline I at-line by the analyzer subsystem 134 (as well as optional input parameters provided prior to or during operation, where input parameters may be, for example, data from previous bioprocess experiments conducted or not in the same small-scale bioreactors 105, a change in the target molecule concentration or production organism optical density, or information on various production strains across the various small-scale bioreactors 105 including, optionally, genome or metabolome information for the various strains). Those data may also operate as inputs to the DoE 115. The predictive algorithm 113 creates and operates the large scale model 141 (e.g., by any suitable means such as but not limited to finite element analysis, 3D modeling, decomposition of digital model of a tank into voxels, and / or by computational fluid dynamics) using data from the small- scale bioreactors 105 to show (i) what product is produced in the large-scale model 141 (thereby predicting what would be produced in the large-scale bioreactor 205), (ii) under which optimized conditions it can occur at large scale, and (iii) how bioprocess progressed (e.g., feedstock consumption, product accumulation and related titer, yield and productivity, lapsed process time, energy consumption, etc.).

[0040] The prediction algorithm 113 can test the modelled process against an "expected" (or intended or desired) process target performances or parameters (such as titer, yield, productivity) and, using outputs of an internal calculation, feedback loop, or machine learning algorithm, and cause changes to be made in the small-scale bioreactors (and / or in the large-scale model 141 and / or the large-scale bioreactor 205). The system (e.g., the prediction algorithm 113) may be provided with those specific target parameters, which the prediction algorithm 113 seeks to provide by identifying the relevant growth conditions for, e.g., a given strain. By those means, the prediction algorithm 113 predicts and optimizes the conditions for the bioprocess at the large-scale (e.g., in a 50 kL tank) using a large-scale model 141 of the bioprocess.

[0041] The system may be "live", i.e., the system may actually include or have the active bioprocess or bioprocesses being performed in at least one small-scale bioreactor 105. The DoE 115 may implement an iterative optimization process for the bioprocess in the small- scale bioreactors 105 based on the data obtained. In one example, the system 101 may be used for directed or adaptive evolution design. The system may be used to design or discover an enzyme or cell with properties not known to occur naturally (e.g., a desiccationresistant RNase, or transgenic E. coli with a chemical off-switch, or any other suitable example). The small-scale bioreactors 105 may be populated with a library of variants that are candidate targets. The predictive algorithm may be provided with a test condition (e.g., threshold sugar concentration, enzymatic activity after drying, or ethanol production). The DoE 115 may evaluate results from across the small-scale bioreactors 105 and implement changes in a subsequent round of bioprocesses in a new set of small-scale bioreactors 105. By those means, the system 101 may perform an iterative optimization process embodying a feedback-based cyclical directed and adaptive evolution design process.

[0042] In fact, the predictive algorithm 113 may exploit multiple "banks" of small-scale bioreactors. Additional outfits may be included. In some embodiments, the system includes a second set of small-scale bioreactors, and optionally third or additional sets of small- scale bioreactors. One or more large-scale bioreactors 205 may be included as test bioreactors, such that the system, using data from the small-scale bioreactors 105, conducts the bioprocess at the large-scale in the test reactor (e.g., large-scale bioreactor 205) for confirming fermentation conditions and I or improving the prediction algorithm 113.

[0043] In certain embodiments, each small-scale bioreactor 105 has a milliliter-scale reaction volume (e.g., 10 mL) or microfluidic scale reaction volume. A battery of small- scale and large-scale bioreactors may be remotely controlled. Each may be stirred. The small-scale and large- scale bioreactors are useful for aerobic or anaerobic fermentation. The small-scale and large -scale bioreactors are preferably autoclavable or can otherwise be chemically sterilized. The sensing subsystem 132 or the control subsystem 135 may perform automated sampling. The control subsystem 135 may perform automated inoculation. The small-scale and large-scale bioreactors may be batch-fed, continuously fed, or repeatedly batch-fed. The small-scale and large-scale bioreactors are compatible with a wide variety of natural and synthetic organisms including yeast, fungi, bacteria, archaea, amoeba, protozoa, sponges, corals, biofilms, lichens, algae, synthetic organisms, Chinese hamster ovary cells, insect cells, HeLa cells, chondrocytes, and any other organism. The small-scale and large-scale bioreactors are compatible with acellular bioprocesses including enzymatic reactions, metabolic pathways, chemicals reactions, others, and any combination of any of the foregoing.

[0044] In order to test reactions across all variable space, it may be desirable to vary a large number of conditions, e.g., ingredients, across parallel reactions. It may be desired to test a large number of different strains (e.g., different mutational variants of one organism) for a bioprocess. The system, under direction of the predictions of DOE 115 or the inputs of operator 111, designs experiments that vary growth conditions and / or inocula or strains across small-scale bioreactors 105 and runs the bioprocesses in parallel.

[0045] The system 101 exploits the extrema of scale between volume of one small-scale bioreactor (e.g., 10 ml_) and volume of one large-scale bioreactor 205 (e.g., 50 kL) and runs all informative experiments in parallel and / or sequentially in the small-scale bioreactors 105, and uses the in silico prediction algorithms 113 and the large-scale model 141 to model, predict and optimize how the bioprocess will run at scale.

[0046] FIG. 3 illustrates interoperation of DOE 115 and the prediction algorithm 113. Small- scale bioreactors 105 generate data (collected online by the sensing subsystem 132 or offline I at-line by the analyzer subsystem 134) that may be fed into the prediction algorithm 113. The DOE 115 receives updated models from the prediction algorithm 113 and the data from the small-scale bioreactors and predicts new experiments to be implemented in the small-scale bioreactors 105 to improve the performance of 113 in terms of prediction and optimization. For example, if the at-scale digital model 141 predicts a slow bioprocess, the DOE 115 can define alternative experiments with optimized growth conditions (e.g., different nutrients) for small-scale bioreactors 105 to test if the reaction speed can be increased.

[0047] As discussed, the prediction algorithm 113 is provided by a computer control system 119 in operable communication with the small-scale bioreactors 105 via at least one sensor. The computer control system 119 may operate on inputs that include organism identities and bioprocess conditions in the small-scale bioreactors 105. The prediction algorithm 113 may include or implement machine learning strategies. For example, the prediction algorithm 113 may be trained on training data that includes the organism identifiers, the bioprocess conditions, the data obtained in said bioreactors 105 and / or synthetic data that is generated in silico using, for example, simulations or algorithms. Any suitable bioprocess may be implemented by the system 101. For example, the system 101 may be used to discover or engineer novel enzymes that would expand the feedstock utility of a chosen fermentation organism. In another example, a system 101 of the invention is used to design and implement production of a single-cell oil (e.g., by a filamentous fungus) as a feedstock for biodiesel production. In another example, each small- scale bioreactor 105 is inoculated with a strain of genetically engineered yeast, such as Pichia Pastoris to produce collagen proteins. Optionally using the feedback loops, test data from small-scale bioreactors 105 inform the build of large-scale models 141. The control subsystem 135 uses modules for inoculation and ingredient supply to inoculate small-scale bioreactors 105 with P. Pastoris strains and supply the small-scale bioreactors 105 with glucose as feedstock and relevant optimized media conditions including inducers like methanol and others. The system 101 also designs downstream processing systems for harvesting the final product or ingredient. Thus it can be seen that systems 101, 201 of the invention are useful for methods of biomanufacturing.

[0048] FIG. 4 diagrams a biomanufacturing method 501. The method 501 includes conducting at least one bioprocess 505 in a set of small-scale bioreactors 105 and possibly large-scale bioreactors 205. Data is obtained 509 from the small-scale bioreactors 105 (and in some embodiments from large-scale bioreactors) and used in training 511 an in silico prediction algorithm 113. The training data includes the obtained data and also preferably a model of the bioprocess (which may include a chemical reaction formulae, a list of ingredients, or other data), and / or data collected from large-scale bioreactors, and / or synthetic data that is generated in silico using, for example, simulations or algorithms. The method 501 involves predicting 525, by the prediction algorithm, optimized conditions for the bioprocess at a large scale based on the training step. Preferably the method 501 outputs 531 an at-scale model of the bioprocess (e.g., instructions for implementing the bioprocess in a large-scale bioreactor using data collected by operation of the small-scale bioreactors).

[0049] Importantly, the method 501 can use operations in small-scale bioreactors to plan operation of the bioprocess in a large-scale bioreactor without the requirement to implement the bioprocess at intermediate scales. The small-scale bioreactors 105 may each have a small-scale bioreactor volume less than about IL and the large-scale bioreactor may have a reaction volume above about IL and at least about 10 times larger than the small-scale bioreactor. It is not necessary to conduct a standard step by step scale-up for the bioprocess in flasks or bioreactors such as 250 mL, 1 L, 10L, 100 L, 1.000 L, etc. Using the disclosed systems, the prediction algorithm 113 may implement about 50 kL fermentation after modeling from active bioprocesses conducted in only small-scale (e.g., microfluidic to 10 mL) bioreactors 105. The method may include iteratively optimizing reaction conditions in the small-scale bioreactors based on readings provided to the prediction algorithm by a sensing system in sensing communication with the small- scale bioreactors. The method 501 may include conducting the bioprocess under the optimized conditions at the large scale in a large-scale bioreactor 205 and providing readings to the prediction algorithm 113 by a sensing system 132 in sensing communication with the large-scale bioreactor. The prediction algorithm 113 is provided by a computer control system 119 in operable communication with the small-scale bioreactors 105 via at least one sensor. The method 501 may include changing, by the computer control system 119, one or more bioprocess / bioreaction conditions in the small- scale bioreactors. The DOE engine 115 identifies the relevant growth conditions to be tested to explore the option space of conditions in the most efficient way. The small-scale bioreactors 105 may optionally be provided on a hardware outfit (e.g., a tray or rack) and the method 501 includes inserting or repositioning the hardware outfit into one of a plurality of bays of a bioreactor rack.

[0050] The method 501 may include modeling, by the prediction algorithm 113, the bioprocess in a large-scale model 141 of a large-scale bioreactor, wherein content of the large-scale bioreactor is represented in the large-scale model 141 by voxel decomposition. The DOE engine 115 is preferably operable to design experimental processes to run in the small-scale bioreactors to refine optimized conditions for the bioprocess at the small scale and / or generate additional data to be used by the prediction algorithm 113 to (i) make predictions that are fed into the design of experiment engine to design additional experimental processes to generate additional data, or (ii) refine optimized conditions for the bioprocess at the large scale, while improving the performance of the algorithms. The method 501 may include applying, by the prediction algorithm 113, a computational fluid dynamic model to the large-scale model 141 to predict results of performing the bioprocess in the large-scale bioreactor under the optimized conditions. The method 501 may include modeling, designing, or operating a downstream processing sub-system operable under the control of the prediction algorithm 113 or another algorithm to process and harvest any product produced in a bioreactor (e.g. vessel).

[0051] In certain embodiments of methods and systems of the invention, the prediction algorithm 113 comprises a machine learning system that is trained on data obtained from the small-scale bioreactors and a model of the bioprocess at the large scale, and / or data collected in large-scale bioreactors. The algorithm may learn from different volumes of small-scale bioreactors and different volumes of large-scale bioreactors. A machine learning system may be trained on data obtained from the small-scale bioreactors and a model of the bioprocess at the large scale, such that the machine learning system is operable to output and monitor a control strategy for a biomanufacturing system 101.

[0052] Any suitable machine learning (ML) system or algorithm may be implemented at any suitable stage within the system 101. The system may use reinforcement learning, supervised learning, unsupervised learning, others, or combinations thereof.

[0053] Some embodiments use Reinforcement Learning (RL) to learn how to run bioreactors at scale to generate optimal performance. The RL agent learns by interacting with the environment and deriving inputs and conditions that optimize outcomes. An RL agent may learn from inputs such as the complexity generated by the biological phenomena of interest paired with the fluid dynamics of the small-scale bioreactor. The RL agent learns an internal model of the system and does not require knowledge of the working rules of the system in advance. Whichever ML model(s) are implemented, the training of the ML model may be bootstrapped using computational fluid dynamics (CFD) models and then connected to a bioreactor (e.g. vessel) 205.

[0054] The prediction algorithm 113 preferably provides a probabilistic prediction of the fermentation model at scale. The prediction algorithm 113 may be based on a deep learning model to learn distributions. Deep learning models use multiple layers in a neural network and preferably include at least one input layer and one output layer and optionally any number of hidden layers.

[0055] As shown, the invention provides systems and methods for conducting a bioprocess (e.g., fermentation) at a large, industrial scale using small-scale (e.g., 10 mL) bioreactors to model the bioprocess and generate data with a prediction algorithm that learns from the small-scale bioreactors and I or from the large-scale bioreactors, and creates a model for large-scale implementation. The small-scale bioreactors provide for faster iteration and parallel design of organism and growth conditions than traditional step-by-step scale-up through flasks and bioreactors. Small-scale bioreactors are used to skip scaling up through flasks of intermediate volumes when designing only fermentation conditions, instead providing "Direct to Fermentation" (D2F) techniques that allow for the parallel design of organism (e.g., strain engineering, metabolic engineering...) and fermentation or bioprocess conditions. The small- scale bioreactors may be used to test a wide set of conditions, using the DOE 115 to navigate the option space of design conditions, and use the in silico prediction algorithm 113 to model large scale conditions.

[0056] Preferably, the prediction algorithm 113 operates by continuous learning by leveraging large scale data from biomanufacturing production. Learnings by the algorithm 113 may be used to feed subsequent lab designs and algorithms. The invention provides for scaling up to industrial production by the number of large-scale bioreactors, i.e. scaling- out, (and not the size of large- scale bioreactors or scaling-up).

[0057] The parallelized volumes of the small-scale bioreactors minimize medium loss (and cost) if spoiled, and provide a greater number of data points. For the large-scale model, systems and methods may use finite element analysis, 3D modeling, CFD, or other tools. By the disclosed system and methods, the invention provides Direct to Fermentation (D2F), the ability to conduct and / or predict a bioprocess at the industrial scale from bioprocesses conducted on the small scale.

Claims

What is claimed is:

1. A biomanufacturing system, comprising: at least one small-scale bioreactor; and an at least one in silico prediction algorithm that integrates with said bioreactors, wherein the at least one in silico prediction algorithm predicts and optimizes conditions for at least one bioprocess at a large-scale based on data obtained in said at least one small- scale bioreactor.

2. The system of claim 1, further comprising the bioprocess being performed in the at least one small-scale bioreactor, wherein a design-of-experiment engine outputs experimental conditions to be tested in the at least one small-scale bioreactor based on the data obtained.

3. The system of claim 2, wherein a bioreactor control sub-system receives the experimental conditions from the design-of-experiment engine or an operator and implements the experimental conditions in the at least one small-scale bioreactor.

4. The system of any one of the preceding claims, further comprising a test reactor, wherein the system conducts the bioprocess at the large-scale in the test reactor for confirming fermentation conditions and / or improving the at least one prediction algorithm.

5. The system of claim 4, wherein the at least one prediction algorithm is provided by a computer control system in operable communication with the at least one small-scale bioreactor via at least one sensor.

6. The system of claim 5, wherein the computer control system is operable to change one or more bioprocess / bioreaction conditions in the at least one small-scale bioreactor.

7. The system of claim 6, wherein the at least one sensor senses, an equipment analyzes or the control system changes, a condition selected from the group consisting of but not limited to: temperature, pH, gas supply, nutrient supply, compound to induce or repress gene expression, foam production, optical density, cell viability, agitation speed, media composition, media feed rate, and product withdrawal rate.

8. The system of any one of claims 4 to 7, further comprising a downstream processing sub-system operable under the control of the at least one prediction algorithm to processand harvest product produced in the test reactor, wherein the downstream processing subsystem is operable to perform an operation on a product of the test reactor.

9. The system of any one of the preceding claims wherein the at least one in silico prediction algorithm predicts and optimizes conditions for the at least one bioprocess at a large-scale based on data obtained in said at least one small-scale bioreactor in accordance with at least one predefined constraint or objective.

10. The system of claim 9 wherein the at least one predefined constraint or objective includes yield.

11. The system of claim 9 or claim 10 wherein the at least one predefined constraint or objective includes productivity.

12. The system of any one of claims 9 to 11 wherein the at least one predefined constraint or objective includes titer.

13. The system of any one of claims 9 to 12 wherein the at least one predefined constraint or objective includes cost.

14. The system of any one of claims 9 to 13 wherein the at least one predefined constraint or objective includes operational efficiency.The system of any one of claims 9 to 14 wherein the at least one predefined constraint or objective includes a decrease in lead time and / or production time.

16. The system of any one of the preceding claims, wherein the algorithm is cloudbased.

17. The system of any one of the preceding claims, further comprising a second set of small-scale bioreactors, and optionally third or additional sets of small-scale bioreactors.

18. The system of claim 17, wherein the set and the second set of small-scale bioreactors each comprise a hardware outfit, the system further comprising a rack comprising a plurality of bays, wherein each bioreactor may be inserted into or repositioned between one of the plurality of bays of the rack, wherein the rack is coupled to sensor systems and / or bioreactor control systems that are communicatively coupled to the at least one prediction algorithm.

19. The system of claim 18 wherein the rack is coupled to an autosampler.

20. The system of claim 18 or claim 19, wherein the at least one prediction algorithm is provided by a computer control system in operable communication with the small-scale bioreactors via at least one sensor, wherein the at least one prediction algorithm operates on inputs that include organism identities and bioprocess conditions in the small-scale bioreactors, wherein the at least one prediction algorithm is trained on training data that includes the organism identifiers, the bioprocess conditions, the data obtained in said bioreactors and / or synthetic data.

21. The system of claim 20, wherein at least some of the bioreactors are occupied by an organism or a combination of organisms conducting the bioprocess.

22. The system of claim 21, wherein : the bioprocess comprises fermentation; the organism is a naturally-occurring or transformed bacteria, yeast, fungi, mold, algae, plant, insect or animal cell and the bioprocess comprises expression of a transgene-encoded protein, enzyme, biologic, medicine, a small molecule or a different type of a molecule; or the bioprocess comprises the production of biomass, enzyme, protein, small molecule, food or cosmetics ingredient, biologic, medicine, or a different type of a molecule.

23. The system of any one of the preceding claims, wherein the at least one prediction algorithm predicts and optimizes the conditions for the bioprocess at the large-scale using a model of the bioprocess in a large-scale bioreactor.

24. The system of claim 23 wherein content of the large-scale bioreactor is represented in the model by a digital 3D model or voxel decomposition.

25. The system of claim 23 or claim 24, wherein the at least one prediction algorithm implements computational fluid dynamics to predict results of performing the bioprocess in the large-scale bioreactor under the optimized conditions.

26. The system of any one of claims 23 to 25, wherein the at least one small-scale bioreactor is smaller than about IL and wherein said large-scale bioreactor is larger than about IL and at least about 10 times larger than the at least one small-scale bioreactor.

27. The system of any one of the preceding claims, further comprising a plurality of production reactors operably connected to a downstream processing unit, wherein each ofthe production reactors has a volume of at least about IL, wherein the plurality of production reactors are operable to perform the bioprocess under the conditions optimized by the at least one prediction algorithm.

28. The system of any one of the preceding claims, further comprising a design of experiment engine operable to design experimental processes to run in the at least one small-scale bioreactor to refine optimized conditions for the bioprocess at the small scale and / or generate additional data to be used by the at least one prediction algorithm to (i) make predictions that are fed into the design of experiment engine to design additional experimental processes to generate further additional data, or (ii) refine optimized conditions for the bioprocess at the large scale, while improving the performance of the algorithms.

29. The system of any one of the preceding claims, wherein the at least one prediction algorithm comprises a machine learning system that is trained on training data obtained from small-scale reactions and / or a model of the bioprocess at the large scale, and / or data collected from at least one large-scale bioreactor and / or synthetic data.

30. The system of any one of the preceding claims, wherein the at least one in silico prediction algorithm creates a digital model of the bioprocess operating at the large-scale.

31. A biomanufacturing method, comprising: conducting at least one bioprocess in at least one small-scale bioreactor; obtaining data from the at least one small-scale bioreactor; training an at least one in silico prediction algorithm using training data that includes the obtained data and a model of the bioprocess, and / or data collected from at least one large-scale bioreactor, and / or synthetic data; and operating the at least one prediction algorithm to provide optimized conditions for the bioprocess at a large scale based on the training step.

32. The method of claim 31, further comprising iteratively optimizing reaction conditions in the at least one small-scale bioreactor based on readings provided to the at least one prediction algorithm by a sensing system in sensing communication with the at least one small-scale bioreactor.

33. The method of claim 31 or claim 32 wherein the step of operating the at least one prediction algorithm to provide optimized conditions for the bioprocess at a large scale based on the training step includes operating the at least one prediction algorithm toprovide optimized conditions for the bioprocess at a large scale in accordance with at least one predefined constraint or objective.

34. The method of claim 33 wherein the at least one predefined constraint or objective includes yield.

35. The method of claim 33 or claim 34 wherein the at least one predefined constraint or objective includes productivity.

36. The method of any one of claims 33 to 35 wherein the at least one predefined constraint or objective includes titer.

37. The system of any one of claims 33 to 36 wherein the at least one predefined constraint or objective includes cost.

38. The system of any one of claims 33 to 37 wherein the at least one predefined constraint or objective includes operational efficiency.

39. The system of any one of claims 33 to 38 wherein the at least one predefined constraint or objective includes a decrease in lead time and / or production time.

40. The method of any one of claims 31 to 39, further comprising conducting the bioprocess under the optimized conditions at the large scale in a large-scale bioreactor and providing readings to the at least one prediction algorithm by a sensing system in sensing communication with the large-scale bioreactor.

41. The method of any one of claims 31 to 40, wherein the at least one prediction algorithm is provided by a computer control system in operable communication with the at least one small-scale bioreactor via at least one sensor, the method further comprising changing, by the computer control system, one or more bioprocess / bioreaction conditions in the at least one small-scale bioreactor.

42. The method of claim 41, further comprising sensing, analyzing and / or changing a condition selected from the group consisting of but not limited to: temperature, pH, gas supply, nutrient supply, foam production, optical density, cell viability, agitation speed, medium composition, medium feed rate, and product withdrawal rate.

43. The method of any one of claims 31 to 42, further comprising:conducting the bioprocess under the optimized conditions at the large scale in a large- scale bioreactor; and processing, by a downstream processing sub-system, a product of the bioprocess downstream of the large-scale bioreactor, optionally wherein the operation is selected from the group consisting of but not limited to centrifugation, heating, cooling, filtration, evaporation, distillation, drying and precipitation.

44. The method of any one of claims 31 to 43, wherein bioreactors are provided on a hardware outfit and the method includes inserting or repositioning the hardware outfit into one of a plurality of bays of a bioreactor rack, wherein the rack is coupled to sensor systems and / or bioreactor control systems that are communicatively coupled to the at least one prediction algorithm.

45. The method of any one of claims 31 to 44, further comprising processing, by a computer control system having the at least one prediction algorithm resident therein, inputs that include organism identities and bioprocess conditions in the at least one small- scale bioreactor.

46. The method of claim 45, further comprising: training the at least one prediction algorithm with training data that includes the organism identifiers, the bioprocess conditions, data obtained from the at least one small- scale bioreactor, and / or data collected in one or more large-scale bioreactors, and / or synthetic data, and a predetermined product identity; and operating the design of experiment engine to design experimental processes for the small-scale bioreactors to refine optimized conditions for the bioprocess at the small scale and / or generate additional data used by the at least one prediction algorithm; operating the at least one prediction algorithm to (i) make predictions that are fed into the design of experiment engine to design additional experimental processes to generate further additional data, or (ii) refine optimized conditions for the bioprocess at the large scale, while improving the performance of the algorithms.

47. The method of any one of claims 31 to 46, further comprising modeling, by the at least one prediction algorithm, the bioprocess in a large-scale bioreactor, wherein content of the large-scale bioreactor is represented in a large-scale model by a digital 3D model.

48. The method of claim 47, wherein the at least one small-scale bioreactor has a reaction volume smaller than about 1 liter and wherein said large-scale bioreactor has areaction volume larger than about IL and at least about 10 times larger than the at least one small-scale bioreactor.

49. The method of any one of claims 31 to 48, wherein the at least one prediction algorithm comprises a machine learning system that is trained on data obtained from the at least one small-scale bioreactor and a model of the bioprocess at the large scale, and / or data collected in one or more large-scale bioreactors.

50. The method of any one of claims 31 to 49, wherein the method is a computer- implemented method.

51. A computer program comprising computer code configured to perform the computer-implemented method of claim 50.

Images