Finding process protocols for generating organoid structures

Abstract

Various techniques described herein allow process protocols to be found for processes for growing organoid structures, wherein the growth of the organoid structures in a candidate process can be monitored, and one or more additional process protocols can be generated on the basis of the corresponding result of such a monitoring process.

Description

[0001] Description

[0002] Identification of process protocols for the generation of organoid structures

[0003] TECHNICAL AREA

[0004] The present invention relates to the field of the production of organoid structures, in particular methods for generating process protocols for the production of organoid structures.

[0005] BACKGROUND

[0006] Organoids are three-dimensional cell cultures. They can be derived from stem cells, such as pluripotent or adult stem cells. Organoids can mimic the structural and functional properties of organs and offer, for example, the possibility of modeling and studying human or animal tissues in the laboratory without relying on living organisms.

[0007] Organoid production involves isolating stem cells from various sources, such as embryonic stem cells, induced pluripotent stem cells, or tumor samples. These cells can then be differentiated under appropriate conditions to develop into different cell types typically found in a particular organ. The differentiated cells are cultured into three-dimensional structures that resemble the cellular organization and function of the organ in question. In organoids, the 3D structure can arise through self-assembly, growth, and development. Differentiation occurs concurrently. Structure and function are interdependent. The organoid cultures are then cultivated and matured under optimal conditions to promote their growth and development. An organoid culture can consist of one or more organoids.Spheroids and tumoroids are further types of three-dimensional cell cultures. In spheroids, a mixture of different cell types is brought together, i.e., aggregated, which can then partially (especially in the case of tumors) develop and grow further.

[0008] Organoid cultures, spheroid cultures, and tumoroid cultures have applications in various fields. They are frequently used for disease modeling to investigate the development and progression of diseases such as cancer, genetic disorders, and infectious diseases. Furthermore, they serve as a platform for drug development by enabling testing in a human tissue environment, leading to more accurate predictions of drug efficacy and toxicity. Organoid cultures can also be used therapeutically. They can contribute to personalized medicine by being generated from a specific patient's cells to develop personalized treatment approaches. Finally, they contribute to basic research by serving as model systems for investigating organ development, cell-cell interactions, and the underlying mechanisms of various diseases.

[0009] The production of organoids can require processes with long cultivation times and numerous steps. Typical cultivation times can range from several days to several months. The necessary steps can be defined in a process protocol, which includes, for example, information on the timing of media changes, media compositions, media temperatures, and environmental conditions such as light and atmospheric conditions. Further possible process steps include, for example, the addition of extracellular matrix, determining cell or...Object density, setting of cultivation volumes, adaptation / selection of a type of culture vessel and mechanical work steps, such as cutting or otherwise dividing or sorting the organoids, detaching objects from each other and / or from the bottom of the culture vessel (biochemically or mechanically), mechanical or biochemical singulation into cell clumps or individual cells.

[0010] It has been shown that the success of organoid cultivation depends significantly on the protocol. Only a suitable protocol allows for cultivation on an industrial scale, meaning with reproducible results even after multiple iterations of the same protocol. In conventional approaches, finding a suitable protocol requires considerable time and experience. Often, trial-and-error techniques are employed. This is because the impact of varying a protocol or changing a protocol step (sensitivity) on cultivation success is difficult or impossible to predict. The number of possible protocol variations is large, making systematic testing of all protocols often impractical. Furthermore, the resource, time, and labor costs of a single protocol run are high, so simply testing all different protocols is not feasible.

[0011] SUMMARY

[0012] The task, therefore, is to improve process protocols for the production of organoid structures, taking into account the problems mentioned above.

[0013] This problem is solved by a method and a system defined in the independent claims. The dependent claims define embodiments.

[0014] One aspect concerns a procedure that comprises one or more iterations of the following steps. One or more process protocols, each defining a process for growing organoid structures in an automated growth device, are provided. The automated growth device is controlled based on the one or more process protocols to perform each of the one or more processes, at least partially. The growth of the organoids or organoid precursors (or organoid structures in general) is monitored by means of one or more monitoring modalities of the automated growth device, and based on a result of the monitoring, one or more further process protocols are generated.

[0015] Another aspect concerns a system with an automated propagator for growing organoid structures, comprising one or more monitoring modalities and a control device. The control device is designed to execute one or more iterations of the following steps. The control device provides one or more process protocols, each defining a process for growing organoid structures in the automated propagator. Based on these one or more process protocols, the control device directs the automated propagator to at least partially execute each of the one or more processes. Using the one or more monitoring modalities, the control device monitors the propagation of the organoid structures and, based on a monitoring result, generates one or more additional process protocols.

[0016] The features set out above and those described below can be used not only in the corresponding explicitly set out combinations, but also in further combinations or in isolation, without leaving the scope of protection of the present invention.

[0017] BRIEF DESCRIPTION OF THE FIGURES

[0018] FIG. 1 shows process steps of a method for optimizing process protocols for the production of organoid structures.

[0019] FIG. 2 schematically shows a system for optimizing process protocols for the production of organoid structures. FIG. 3 schematically shows details of the execution and evaluation of a process protocol.

[0020] DETAILED DESCRIPTION

[0021] Some examples in this disclosure generally provide for a variety of circuits or other electrical devices. All references to the circuits and other electrical devices and the functionality they provide are not intended to be limited to only what is shown and described herein. Even if certain designations may be assigned to the various circuits or other electrical devices, these are not intended to limit the scope of function of the circuits and other electrical devices.

[0022] It is understood that the following description of embodiments is not to be understood in a limiting sense. The scope of the invention is not to be limited by the embodiments described below or by the drawings, which serve only for illustration.

[0023] The drawings are to be regarded as schematic representations, and the elements depicted in the drawings are not necessarily drawn to scale. Rather, the various elements are shown in such a way that their function and general purpose are recognizable to a person skilled in the art. Any connection or coupling between functional blocks, devices, components, or other physical or functional units shown in or described herein may also be realized by an indirect connection or coupling. Coupling between components may also be established via a wireless connection. Functional blocks may be implemented in hardware, firmware, software, or a combination thereof. Some groups of elements, e.g., the multiple plates, are identified by reference numerals consisting of a number and, optionally, a trailing letter.Depending on the context, the same reference numeral can denote a single element or all elements of the group. For example, each plate 206A, 206B, 206C, 206D, 206E, 206F, 206G, 206H, 206J is one of several plates 206. The same reference numerals in different drawings refer to similar or identical components.

[0024] While many cell cultures, especially two-dimensional cell cultures grown on an agar layer, for example, show essentially homogeneous growth under the same growth conditions, it has been found that the growth of organoid structures is less deterministic and variations can occur even when using a nominally identical protocol.

[0025] However, the large-scale, and especially industrial, use of organoids requires manufacturing processes that deliver consistently good results with reasonable effort, particularly in terms of time, resource commitment, and material usage. Finding an optimized process protocol for generating organoid structures (i.e., finished organoids or precursor structures) can therefore be crucial. Improving process protocols can also be very costly, particularly in terms of time, resource commitment, and material usage. Different types of organoids, such as models of the heart, stomach, intestines, liver, retina, brain, lungs, or kidneys, may require different process protocols, also depending on the precise composition of the culture media used.

[0026] The following describes embodiments of methods and systems that enable the efficient optimization of process protocols for the generation of organoid structures. Organoids are three-dimensional cell structures grown from stem cells, such as embryonic stem cells, induced pluripotent stem cells, or cells from tumor tissue, and can mimic specific tissues or organ parts. They are characterized by a complex, self-organized architecture that reflects important functional and structural properties of the target tissue.In contrast to traditional cell cultures, which are often set up as two-dimensional monolayers of cells (also called cell lawns) and exhibit limited cell differentiation and interaction, organoids offer an improved model for the study of organ development, disease mechanisms, and drug testing because they better mimic the microenvironment and cellular interactions of a real organ.

[0027] The following describes techniques related to the generation and production of organoid structures. The cultivation of a finished organoid generally proceeds through several phases. The division into different phases is not strictly defined, and individual phases (regardless of the specific definition used) can overlap or not be strictly distinguished relative to one another. This is also due to the partially instance-specific variation in organoid growth. Nevertheless, a possible sequence of phases in organoid production is described below for better illustration. However, this sequence is not limiting for the reasons mentioned above. First, the stem cells are isolated and cultured: The production of an organoid begins with the isolation of pluripotent or tissue-specific stem cells or cells from tumor tissue.

[0028] Pluripotent stem cells are cultivated under controlled conditions to ensure their proliferation and survival. These stem cells are not yet functionally differentiated. Therefore, depending on the specific process employed, a single stem cell can give rise to various types of organoids that mimic different organs. The 3D structure of these organoids can emerge through self-assembly, growth, and development. Differentiation can occur concurrently. During this process, the stem cells are exposed to specific growth factors and signaling molecules that direct differentiation into the desired cell types. These biochemical signals mimic those that occur in the cells' natural environment during organ development.Three-dimensional aggregation and organization then occur: The differentiated cells are induced to arrange themselves (self-organize) within a three-dimensional matrix, often using hydrogels as a scaffold. This matrix supports the spatial arrangement of the cells and promotes the cellular interactions typical of the organ structure. Finally, the organoid matures and becomes functionalized: In this phase, the cells within the aggregates develop complex organoid structures that mimic specific functions of the target organ. Maturation can take several weeks and requires a continuous supply of nutrients and growth factors, which may be tailored to the specific growth stage. During this time, specialized microenvironments also form, which are crucial for the organoid's final function. Additional steps are possible.For example, organoids can be sorted or cut manually or automatically. Specific organoid types may require additional stimuli; for instance, respiratory or lung organoids may require stimuli via a liquid-air interface, or muscle organoids may require mechanical stimuli.

[0029] Using the techniques described herein, process protocols can be generated that enable the cultivation of an organoid in individual growth phases or from the first phase to the last. All such processes are fundamentally aimed at the final cultivation of finished organoids, even if a process only covers one, for example, an early growth phase of the organoids. Organoid structures, in this context, refer not only to finished organoids (i.e., with developed functionality that mimics an organ) but also to 3D cell structures in an earlier growth phase. Several examples of this disclosure are based on the realization that, in such a production of organoid structures, a multitude of parameters can be adjusted to achieve the desired result. The range of freely adjustable parameters, whose values ​​can be varied when generating a process protocol, is very large.

[0030] Several examples of revelations are further based on the finding that the manufacturing process often varies from organoid structure to organoid structure. This means that strictly deterministic execution of one and the same process several times in succession will typically lead to different results. A dispersion of process outcomes is observed even with a nominally identical process.

[0031] Several examples are based on the experience that the conventional development of process protocols for cultivating organoid structures requires a high degree of specific expertise to achieve good results. However, the sensitivity of a parameter to the process outcome is often not readily apparent, even to experts in the respective field. This means that even a slight variation in the protocol can, for example, cause a significant increase in rejects or even prevent the desired cultivation result from being achieved at all. The variability in cultivation results can increase so dramatically that scaling up production becomes impractical. The sensitivity of a parameter to the process outcome quantifies the magnitude of the change in the process result when the value of the parameter in question is incrementally altered.The effects of changing a process step are therefore difficult to predict in advance. Examples include a slightly earlier or later change of the medium, a slightly varied target pH value, a different temperature setting, earlier selection of poorly developing organoid structures, a varied timing of growth factor addition, etc. However, knowledge of a parameter's sensitivity is helpful for developing an optimized process for cultivating organoid structures. A particularly important optimization outcome is that the respective process enables the cultivation of a specific organoid structure. Optimization criteria can include, for example, the yield of generated organoid structures, the development of organ-specific cell types and structures, and the onset of organ-specific functions.

[0032] The complexity of developing an optimized process protocol increases even further when—as is optionally possible—it is considered that, in addition to the desired outcome of cultivating the organoid structures, the objective of a specific organoid cultivation process includes not only the successful cultivation of a finished organoid of a particular species, but also other target parameters. For example, particularly in industrial settings, it is desirable to reduce or even minimize the time between the start and end of a specific organoid cultivation process. The overall duration of a process is therefore often a quality criterion for evaluating the quality of the corresponding process protocol. Another quality criterion involves the time required by users for manual interaction with the cultivation environment. The number of manual steps should often be reduced or minimized.The use of materials, such as chemical reagents, should often be reduced or minimized. Available systems for cultivating organoid structures may be limited, so the throughput or cultivation time should be minimized or at least reduced in certain systems. Another quality criterion can relate to the variability or dispersion of the organoid structures produced. The variability of organoid structures quantifies the dispersion of process results (for example, regarding functional and / or structural properties of the organoid structures produced by a specific process) when the respective process is repeated multiple times. Variability thus represents, in a sense, a standard deviation of a frequency distribution that describes a specific property of the organoid structure produced by the process.From the above considerations, it is evident that, at least in some examples, a multidimensional quality function can be formulated for the generation of a process protocol for the cultivation of organoid structures, which evaluates the quality of such a generated process protocol with regard to several target specifications (for example, execution time for the process, i.e., the total duration of the process; a number of steps to be carried out manually, material input, variability of the generated organoid structures and / or occupancy time in certain systems, etc.).

[0033] Based on these findings, techniques are described below that enable the creation of a process protocol with improved properties. These techniques are used to generate process protocols for cultivating organoid structures that meet one or more quality criteria—for example, the aforementioned quality criteria. In particular, techniques are described for generating process protocols in a targeted manner, meaning that instead of erratically traversing the space of free parameters until a protocol meeting the target specifications is found, the free parameter space is systematically explored. The goals of this process protocol optimization can include, for example, better results and higher yields in the cultivation of the desired organoid structures, reduced resource consumption, and faster execution.

[0034] Using the techniques described herein, it is possible to quickly and reliably find a customized process protocol for the cultivation of organoid structures, i.e., specifically finished organoids of a particular species. This customized process protocol exhibits optimized properties with respect to one or more quality criteria. These quality criteria can be flexibly defined by a user. Alternatively or additionally, the customized process protocol may have an optimized property for a specific automated cultivation device. For example, different types of automated cultivation devices may execute certain process protocols better than other types.This can be due, for example, to the fact that certain manipulations are only possible within certain limits or not at all in a particular automated rearing device, while corresponding limitations do not exist in another type of automated rearing device. Accordingly, it may be possible to identify particularly effective process protocols for different automated rearing devices using the techniques described herein, taking into account the specific limitations and restrictions of each type of automated rearing device or even the specific instance.

[0035] Optimization techniques can be employed; that is, starting from a result for one or more process protocols, a targeted variation of parameter values ​​of free parameters can be carried out using an optimization algorithm. In other words, this allows for a quality assessment of a process defined by a process protocol.

[0036] The techniques described herein for generating an organoid structure can, for example, proceed in three or more stages. An exemplary multi-stage procedure for generating process protocols for cultivation processes would be: (1) Automated initial generation of promising process protocols to test culture conditions. (2) Execution of the proposed process protocols and evaluation of success based on a quality assessment. (3) Automated generation of further promising process protocols based on the protocols already executed and their results, by varying the initial process protocols from (1). Then, steps (2)-(3) can be repeated, for example, until a quality criterion is met. Such a three-stage procedure as explained above can, for example, be implemented according to the procedure shown in FIG. 1, which is described below.Figure 1 shows process steps of a method 100 according to one embodiment. The method 100 from Figure 1 can be computer-implemented. For example, at least some steps of the method from Figure 1 can be performed by an electronic data processing device. For example, a processor can load and execute program code from memory. When the processor executes the program code from memory, the processor implements the method from Figure 1 or at least some steps of this method.

[0037] The process is an iterative process.

[0038] In step 102, several initial process protocols are typically provided (although it is also possible for only a single initial process protocol to be provided). Each process protocol defines a process, i.e., a sequence of steps for growing organoid structures in an automated growth device. Each process protocol can be used to test the culture conditions defined within it. The process protocols are different to allow for testing the generation of organoid structures under different culture conditions.

[0039] Further boundary conditions for the process protocols can be specified by a user. For example, it can be specified which process steps (events) are adaptable (e.g., media change in the differentiation phase) and which possible actions are provided for (e.g., time of media change, concentration of the medium). Furthermore, for each action, the limits within which variations can occur can be specified (e.g., no less than 6 hours between two media changes), as well as the expected effect of each action. It can also be specified how many protocol variants can be implemented. Implementing fewer protocol variants is fast and cost-effective but results in less precise optimization. Implementing many protocol variants takes longer and is more expensive but provides more precise optimization.The generation of organoid structures may require a sterile culture environment. Therefore, performing the process protocols provided in step 102 in an automated growth device may be advantageous. Organoid structures may require a liquid nutrient medium that allows them to grow spatially in a three-dimensional structure. Different process protocols may vary in their process parameters, such as the timing of medium changes or the temperatures maintained by the nutrient media. Other process parameters may include the type of nutrient medium, ambient pressure, mechanical influences such as stirring or shaking the medium, the addition of substances (e.g., growth factors, hormones, or inhibitors), or irradiation, for example, with visible light, UV light, or infrared light, and the corresponding timing of their application.For example, process parameters may include the type, composition, concentration and strength of a suitable matrix, cell density, type and number of objects per process vat, time and method of sorting / separating "good" and "bad" organoid structures.

[0040] A process protocol can therefore include control instructions for one or more system components of an automated growth device. For example, an automated growth device could have one or more so-called multiwell plates or culture plates, which have several cavities, hereinafter also referred to as (process) trays, in which the growth of organoid structures can take place under specific, adjustable process conditions (temperature, humidity, UV illumination, etc.). An automated growth device could, for example, have a dispenser module by which liquids can be automatically filled into a tray. The trays could have a controllable drain so that liquid contained in the tray can be automatically drained. The automated growth device could, for example, have a pipetting module (e.g., a pipetting robot) to dispense a medium (e.g.,to aspirate a liquid and / or transport objects, e.g., liquids, cells, cell clumps, cysts, organoid structures, etc. It would be conceivable for the automated growth device to include a robotic device to move individual plates with trays from a plate or tray storage area (where the trays can be stored for the duration required to grow the organoid structures) to individual modules, for example, to the dispenser module or to a drainage module where the liquid in a tray can be drained by opening the tray's drain. An automated growth device could also include a manipulator.Using such a manipulator, individual organoid structures from a group—for example, an aggregate or simply a common quantity in a specific vat—can be grasped, manipulated, and separated, for instance, by physical grasping, suction, adhesion, or cutting. Organoid structures can then be transferred between different vats. The automated growing device can also provide one or more modules for process monitoring. For example, such modules could include various sensors or data acquisition devices. For instance, temperature or humidity sensors could be provided in each vat. A module could also be included that enables optical, particularly microscopic, imaging of a vat or a portion thereof.For example, the imaging modality could be varied in such a microscopy module. Phase contrast could be switched on or off. Fluorescence imaging could be switched on or off. Light sheet imaging could be selectively activated. Coherence tomography imaging could be used. Non-magnifying optical imaging could be used, for example, for sufficiently large organoid structures. Typically, this might require a robotic module to remove the appropriate tray from a tray storage area and place it on a suitable sample stage for image acquisition. After image acquisition, the robotic module can then pick up the tray again and return it to its corresponding position in the tray storage area.In other examples, imaging modalities can be used, such as wide-field imaging, optical contrast imaging (e.g., digital phase contrast (DPC), quantum phase contrast (QPC), digital imaging contrast (DIC), transport of intensity (TIE), phase contrast), fluorescence imaging, confocal imaging (lightsheet microscopy, lattice lightsheet microscopy), optical coherence tomography (OCT), or dynamic OCT. The acquired images can then be processed and enhanced using image processing techniques or machine-learned algorithms, for example, regarding noise level, contrast level, focus level, etc.

[0041] The cultivation of organoids requires clean, ideally sterile, conditions. Individual elements and / or the entire cultivation apparatus can be equipped accordingly, e.g., with suitable enclosures, filtration of the ambient air, and / or the technical capability to sterilize surfaces or individual units, e.g., by UV, steam, or manual cleaning.

[0042] Using such an automated cultivation device, certain process protocols can be carried out fully or semi-automatically. In semi-automated execution of process protocols, individual process steps may be performed by a user. For example, a robot module could remove a specific tray from the tray hotel and place it in a sample sluice, allowing the user to manipulate the corresponding organoid structures within the tray. Such manually performed process steps can be displayed in the process protocol. For instance, the process protocol could include control instructions for a human-machine interface.It would then be conceivable that, upon reaching the relevant stage in the process, control instructions could be issued to the human-machine interface, prompting the user to perform the corresponding manipulation step. For example, a corresponding message could be displayed via a graphical user interface or sent to the user's portable device.

[0043] The process protocol can be defined in an event-oriented manner. For example, the process protocol can define specific process landmarks that describe the occurrence of a particular event in the cultivation of organoid structures. The precise time of occurrence of this event is then not specified in the process protocol, but can be identified through process monitoring, for example, using an image-based monitoring modality. Thus, an event could be defined as an organoid structure exceeding a certain volume, and this event could trigger the singulation of that organoid structure into its own container.

[0044] The process protocol can contain, at an abstract level, corresponding control instructions for the various modules, which can then be translated by driver software into concrete control signals for the different modules.

[0045] For example, the process logs provided in step 102 can be loaded from a database of process logs. Such a database could, for instance, be populated based on historically executed logs. Various selection criteria can be considered when making selections from the database. For example, a user could specify a particular type of finished organoid to be cultivated. Then, a selection could be made from the database based on a similarity measure between these organoids to be cultivated on the one hand and organoids cultivated using process logs from the database on the other. The various process logs in the database can thus be generally linked to metadata, enabling targeted selection from the database based on such metadata.The metadata can, for example, also include information on a quality assessment based on various quality criteria, as described above. For instance, a user could define a specific target in step 102. The user could then evaluate different quality criteria according to their importance (for example, particularly low material consumption, while the duration of the process is less important). Based on a similarity measure between the corresponding quality target and its fulfillment, the metadata could then be used to select promising initial process protocols from the database.In this way, promising process protocols can be selected as a starting point for the subsequent process steps, which can promote convergence towards a "good" process protocol that meets all required quality criteria.

[0046] Another technique for providing the multiple process protocols in step 102 involves evaluating one or more predefined initial process protocols, for example, from a database or via a user interface (where the user can define them), and then varying these initial process protocols. The result of this variation is then the process protocols that are subsequently executed in step 104. This means that instead of using existing initial process protocols unchanged, a variation of these process protocols can be introduced. Such a variation can be achieved by statistically varying the parameter values ​​of one or more parameters of the process protocols. Such a variation could also take into account one or more physical-technical models.For example, it would be possible to determine a technically achievable maximum bandwidth for parameter values ​​of a specific parameter in a process protocol and then, based on this technically achievable maximum bandwidth, to vary the corresponding parameter value in the initial process protocol. Using such a variation ensures that the parameter space of free parameters is sampled more reliably across its entire width by subsequent steps. This avoids an initial bias on a specific sub-range of the parameter space. In this way, a globally improved optimization result can be achieved.

[0047] Furthermore, the user can evaluate the variation in terms of its expected effect on the quality criteria. This allows for a more effective selection of the process parameters with the highest expected effect to be varied first.

[0048] Furthermore, the user can define maximum deviations for the variation, which can be taken into account when selecting variations of process parameters.

[0049] In step 104, the one or more process protocols provided in step 102 are executed. For example, the automated rearing device can be controlled based on the process protocols from step 102. If a process protocol requires manual intervention from a user, the execution of this process protocol includes automatically issuing a corresponding instruction to the user via a suitable human-machine interface.

[0050] Each process protocol provided in step 102 can be executed multiple times, for example, to obtain a statistical distribution of the results from these process protocols. The number of process instances can be defined in the respective process protocol. It would also be conceivable to determine the number of instances based on one or more properties of the process protocol or a metric for evaluating the quality of the process protocol; details of a corresponding statistical design of experiments will be explained later. Multiple executions of a process protocol can be achieved, for example, by applying this process protocol to multiple samples simultaneously (i.e., in parallel) in several process tanks of the propagation apparatus, or by applying this process protocol multiple times consecutively in the propagation apparatus or in a different propagation apparatus to multiple samples.

[0051] In step 105, the growth of the organoid structures, which is effected by carrying out the process protocols in step 104, is monitored.

[0052] For example, it would be conceivable to perform steps 104 and 105 concurrently. Alternatively, the results of the organoid structure cultivation could be evaluated after the process has been completed.

[0053] In step 105, one or more monitoring modalities of the automated growth device are used. In particular, image-based monitoring can be performed, e.g., using optical, especially microscopic, imaging. This means that during the execution of a process, images of the respective organoid structure at the various stages of growth can be acquired, for example, at a predefined acquisition frequency. These images can then be automatically evaluated by applying suitable image processing algorithms. In particular, machine-learned models can be used to perform localization or, in particular, segmentation of the organoid structures or their components. Specifically, individual cells, different cell types, or different parts of an organoid structure, such as outer cell layers and cell-free lumens, can be localized and segmented.Based on the segmentation, quality metrics can then be derived, such as area ratios, achievement of minimum sizes or volumes, appearance of specific cell types, the quantity or volume ratio of different cell types to each other, etc. Determining localization information, for example, instance-specific segmentation of organoid structures, can be achieved using a machine-learned model. Such a machine-learned model can receive as input a corresponding image, for example, an optical image such as a microscopic image, showing one or more organoid structures. As output, a segmentation mask can be provided, which can be superimposed on the image. The basis for such a machine-learned model can be, for example, a vision transformer architecture; see, for example, Chen, Wuyang, et al."A simple single-scale vision transformer for object localization and instance segmentation." arXiv preprint arXiv:2112.09747 (2021) oder Enze, et al. "SegFormer: Simple and efficient design for semantic segmentation with transformers." arXiv:2105.15203 oder Cheng, Bowen, et al. "Masked-attention mask transformer for universal image segmentation." proceedings of the IEEE / CVF conference on computer vision and pattern recognition. 2022.

[0054] Alternatively or additionally, anomalies could be detected using a machine-learned model. Anomaly detection offers the advantage of identifying previously unseen deviations from the desired process outcome. This means that classification is not limited to predefined classes; it also allows for the detection of deviations from the norm that are not specified a priori. Alternatively or additionally, specific properties of organoid structures could be classified. Furthermore, for example, the size or shape of an organoid structure could be evaluated. Such and other geometric parameters of an organoid structure could be determined and evaluated in two-dimensional projection images of a microscopic photograph. However, the use of an image-based monitoring modality is just one example, and other modalities can be employed.For example, one or more sensors can be used to measure environmental conditions. Examples would be temperature or humidity. The quantity of a medium, the pH value of the medium, or of an organoid structure could be measured. Monitoring can specifically track the variation in values ​​of one or more properties of organoid structures grown using the same process protocol in multiple instances of the respective process. This can include, for example, determining instance-specific values ​​of one or more specific properties of organoid structures grown in multiple instances of the respective process.For example, a specific process could be executed multiple times in parallel, and for each organoid structure generated, its size or shape could be determined based on a microscopic image. Then, the distribution of the size or shape of the organoid structure across the multiple instances of this process could be determined.

[0055] It is evident from the above that monitoring the growth of organoid structures can include, in particular, determining instance-specific values ​​of one or more properties of multiple organoid structures. For example, it would be conceivable that an image-based evaluation of organoid structure properties could use images showing several organoid structures arranged side by side in a group or cluster. Such organoid structures are grown using the same process, that is, the same process protocol. In such an example, it would be conceivable to first localize individual organoid structures in the images, for example, using instance segmentation techniques or bounding boxes. More generally, it would be conceivable to determine localization information for individual organoid structures.Then, for each localized organoid structure, a corresponding value can be determined for the respective property, such as its appearance in the image, morphology, size, shape, color, or the distance to one or more neighboring organoid structures (i.e., the inter-organoid distance). Properties can also relate to substructures of the organoid structure, such as cell-free lumens, specific layering of cell strata, or the presence of organ-specific cell types. These properties can be captured from images, for example, from various microscopic modalities, such as fluorescence microscopy. The result is a list of organoid structures, and for each of these structures, one or more values ​​are obtained for one or more properties. This allows the quality of each individual organoid structure to be evaluated.For example, it would be possible to determine the statistical distribution of such instance-specific values. This would allow, for instance, monitoring the dispersion of corresponding values ​​for a particular property.

[0056] Monitoring the variance of values ​​for one or more properties of organoid structures is particularly helpful in the context of organoid growth. This stems from the understanding that organoid growth in its various stages is not strictly deterministic, but rather subject to statistical variations from instance to instance of a given process (even with nominally identical process protocols). Therefore, knowledge of such statistical variations can be especially useful for enabling scalable and automated organoid growth.Since organoids are macroscopic objects, the properties of individual organoid structures are what matters (which differs from a cell culture, where the microscopic cells are present in large numbers anyway, so that individual deviations from the norm are less important due to averaging over the cells in the culture).

[0057] Techniques for segmenting or determining localization information for organoid structures are generally known from the prior art. Such known algorithms can also be used in connection with the present invention. See, for example, MacDonald, Michael, et al. "Improved automated segmentation of human kidney organoids using deep convolutional neural networks." Medical Imaging 2020: Image Processing. Vol. 11313. SPIE, 2020; or Haja, Asmaa, et al. "Towards automation of organoid analysis: A deep learning approach to localize and quantify organoid images." Computer Methods and Programs in Biomedicine Update 3 (2023): 100101; or Hradeckä, Lucia, et al. "Segmentation and tracking of mammary epithelial organoids in brightfield microscopy." IEEE Transactions on Medical Imaging 42.1 (2022): 281-290; or Bremer, Jan P., et al. "GOAT: Deep learning-enhanced generalized organoid annotation tool." bioRxiv (2022): 2022-09.

[0058] In step 106, each completed process protocol is evaluated. This evaluation can be automated based on the properties of the generated organoid structure. These properties can be obtained from the monitoring data from step 105.

[0059] Several criteria that can be considered within a quality assessment have already been explained above. These criteria can include, in particular, the difficulty of carrying out the process, the amount of material consumed, the total process duration, the occupancy time of a manufacturing system for carrying out the process, the number of manual steps required for cultivation using the process, and / or the variability of the organoid structures produced. The assessment can be performed during the execution of the process protocol, immediately after its execution, or at a later time. For example, the assessment can be performed later and on a different device after the execution and monitoring data have been transferred to that device.

[0060] In particular, the statistical distribution can be factored into the evaluation if a specific process protocol has been performed multiple times. For example, if a particular process protocol is executed multiple times, low variation in the quality or in certain properties of the generated organoid structure when a single process protocol is repeated multiple times (i.e., intra-process variation) can contribute to a positive evaluation of that process protocol. Typically, low process variation is desirable in industrial-scale cultivation. If intra-process variation is needed as a measure of the quality of a particular process protocol, it would be conceivable to determine the number of repetitions of the corresponding process protocol in step 104 based on a confidence level required for this statistical evaluation.In other words, this means that the number of times a particular process needs to be repeated is determined by a statistical design of experiments or can be defined by the user. For example, if a certain level of confidence is required for the intra-process variation of organoid structure size, a suitable statistical design of experiments can determine how many repetitions of the corresponding process are needed to determine this variation or standard deviation of the organoid structure size with a certain level of confidence.

[0061] For example, the number of repetitions can be determined as follows:

[0062] • The quality metric is the diameter of the organoid structures in mm.

[0063] • It is known that the results can, in principle, vary between 0.5 mm and 4 mm.

[0064] • It can be estimated that there is a standard deviation of s = 0.5 mm (i.e., a variance of v = s). 2 = 0.5 2 mm).

[0065] • It is assumed that when testing for a statistically significant difference between two process protocols, the significance level a should be 5% and the discrimination index 1 - β should be 80%.

[0066] • Then it can be assumed that the quality metric diameter varies normally when a process protocol is carried out multiple times and varies equally for all process protocols.

[0067] • For these reasons, a two-sided two-sample t-test can be used to determine a significant difference between two process protocols. • Now it can be determined how many repetitions of two process protocols are necessary to prove the null hypothesis that the mean quality metrics of both process protocols under the chosen a and β do not exceed d. 2 to refute the claim that they are 0.5 mm apart. m in From these assumptions, N « 16 • 2 0 5

[0068] • Unt 2 = 16 • 16.

[0069] • If the required detectable difference is only allowed to be 0.1 mm, a significantly larger minimum number of s results. 2 0 5

[0070] N min « 16 • = 16 2 ln d 2 • 0.1 2 = 400.

[0071] Alternatively, the number of required process repetitions can be specified by the user to accommodate external limitations in time and resources.

[0072] The quality assessment in step 106 can be generally formulated using a heuristic quality metric. This means, for example, that a quality assessment function can be manually parameterized based on user specifications. However, it would also be conceivable to use the quality assessment function as a machine-learned model.

[0073] In this way, even higher-dimensional spaces can be appropriately represented for quality expectations. In some examples, the pre-trained machine-learned model could be fine-tuned for quality assessment for specific rearing tasks, e.g., for particular users or processes, to enable improved quality assessment even in individual situations. This might be the case, for example, if certain monitoring modalities for determining the properties of the organoid structures differ from one rearing device to another. Variations—e.g., in the contrast or noise level of captured images—can be taken into account through such fine-tuning.

[0074] In the flowchart of FIG. 1, step 106 is shown following step 105, and step 105 following step 104. However, it would be possible, in principle, to evaluate a process protocol even while the corresponding process is being executed. This means that step 104 and step 106 can be executed at least partially simultaneously. If, during the evaluation of a process protocol, it is determined before the end of the protocol that predefined intermediate goals or quality levels will not be met, the execution of this process protocol can be aborted; that is, a specific process protocol can also be executed only partially in step 104. This reduces the time required to sample the parameter space of the free parameters of the process protocols.

[0075] In step 108, one or more additional process protocols are automatically generated based on the previously executed process protocols. Step 108 ensures that the large parameter space is efficiently sampled to identify the most suitable process protocol. The appropriate selection of process protocols in step 108 is crucial for efficiently sampling the parameter space of the available parameters within the process protocols.

[0076] Several implementations are conceivable for step 108, and in particular for the logic that generates further process logs. An optimization procedure can be used, parameterized based on the monitoring from step 105. This optimization procedure can, for example, receive the quality assessment of the processes, performed in step 106, as input.

[0077] For example, the process protocols can be treated as black-box objects, and further process protocols can be generated using sequential optimization methods, such as Bayesian optimization. The dependence of at least one quality measure of the organoid structure growth on the selected process protocol parameters can be approximated using a surrogate model. This model can be initially defined and then updated based on the previously executed and evaluated process protocols. In particular, an anisotropic heteroscedastic Gaussian process regressor can be trained as a surrogate model for Bayesian optimization. Other possibilities for implementing the sequential optimization methods include gradient-free optimization such as the Nelder-Mead method, greedy algorithms, evolutionary algorithms, biology-inspired optimization methods such as particle swarm optimization, and FireFly optimization.

[0078] The dispersion of quality measures across multiple iterations of a process protocol can be used to model the noise of the process protocol, for example, based on the variance or standard deviation of the quality measures. The predicted mean for a process protocol can serve as the predicted quality measure, and the predicted standard deviation as the expected dispersion of the quality measure. The quality measure can also be a derived value from several quality measures. Furthermore, multiple quality measures can be predicted together, a process also known as structured output regression. Multiple quality measures can also be modeled independently.

[0079] The fundamental idea of ​​Bayesian optimization is based on the principle of "exploration and exploitation." This means that when proposing new points, a compromise is found between improving already known, good process protocols (exploitation) and searching for new, potentially even better process protocols (exploration). Based on the surrogate model, an expected improvement can be estimated. A new process protocol can then be determined that maximizes the expected improvement. The principle of Bayesian optimization, therefore, consists of training a probabilistic model (surrogate model) whose inputs are protocol configurations, and which predicts the estimated performance of these configurations.The next best protocol configuration can then be found by having the model evaluate the quality of different options and using an acquisition function for selection. This can be particularly advantageous if the model can be evaluated much faster than the entire process, if the model can be retrained quickly (for example, with fewer than 10,000 training data points), and if the dimensionality of the configuration options is not very large (fewer than 1,000 parameters, preferably fewer than 100 parameters, or even just 10 parameters). Examples of surrogate models that can be used include Gaussian processes, (artificial) neural networks, Parzen-Tree Estimators (TPE), Random Forests, or other bootstrap-aggregated models.A Gaussian process regression, as mentioned above, with anisotropic kernel functions (i.e., individual, learnable weighting / scaling of the input parameters) and / or with heteroscedastic (i.e., not independent and identically distributed random variables) noise assumptions per observation can be particularly suitable, so that the dispersion of the results after repeated execution of a process protocol can be considered the uncertainty of the parameter point and modeled accordingly. Examples of acquisition functions are the probability of improvement and the expected improvement.Furthermore, in addition to the expected result from the surrogate model (for example, sufficient average quality of the generated organoid structures with sufficiently low variance), other aspects can be included in the generation of further process protocols, such as the difficulty of carrying out individual process steps, the amount of material consumed, and the overall duration. In this way, a process protocol can ultimately be provided that leads to the best result under the given conditions.

[0080] The generation of additional process protocols can also be performed in batch mode, meaning that not just one (better) additional process protocol is generated, but several. Preferably, the two or more next-best process protocol suggestions according to the optimization (e.g., surrogate model / acquisition function) are not used, because these are likely to be relatively similar. Instead, diversity among the process protocol suggestions is rewarded. This can be implemented, for example, as "greedy batch selection," where the acquisition function iteratively determines the currently best process protocol for a total set k, and then locally reduces the weighting of the protocols around this configuration. The best process protocol is then determined, and this process is iterated until k protocols are reached.

[0081] Another example of implementing step 108 would be the use of a reinforcement learning (RL) model: An agent observes the environment (here, step 105) and selects an action (here, a new process log in step 108) that it deems appropriate based on its observations of the given situation. The environment then sends a feedback signal, which serves as the basis for determining a reward using a reward function and assigning a value to this action (next iteration of step 106, for evaluating the generated process log). The reward function can be predefined or can be trained. The reward function replicates the quality assessment.The reward helps the agent recognize whether the decision was good (based on the quality assessment from step 106) and worth repeating in similar situations, or whether it was a bad decision with an undesirable outcome that should be avoided. The next action (here, generating the next process protocol) is determined within the framework of the so-called "policy." A "greedy strategy" or an "exploratory strategy" can be pursued. Combinations are also conceivable, such as the "greedy" strategy, in which the best known action is usually chosen, but there is a small probability of a random action being performed. During the training process, the reinforcement learning model automatically adjusts weights to improve the accuracy of predicting expected rewards.This is typically achieved through a method called "Temporal Difference Learning" (TD learning) or through methods based on Monte Carlo simulation. Both are unsupervised techniques that allow for model parameterization based on the insights gained through monitoring. In Temporal Difference Learning, values ​​are adjusted based on the difference between the previously estimated values ​​of a state and the rewards actually received, plus the estimated values ​​of the next state after an action is executed. In such a case, unlike Bayesian optimization, for example, no initial search strategy is required to determine one or more subsequent process protocols based on the result of an iteration. This search strategy is learned and optimized by the reinforcement learning model itself.

[0082] Another example of implementing step 108 involves examining various process logs to derive underlying causal dependencies between specific process steps and the result, dependencies that go beyond mere correlations. Methods of so-called "causal discovery" can be used for this purpose.

[0083] In other examples, a correlation-based analysis can be performed on the results of the various process logs to determine linear and non-linear dependencies, for example, based on so-called "gradient-boosted trees". This allows, for example, the determination of which process steps with which settings have a significant influence on the overall result, such as quality or process stability.

[0084] The further process logs can be carried out and evaluated as previously described in steps 104 and 106 and used in step 108 as a basis for generating further process logs.

[0085] Procedure steps 104 to 108 are performed iteratively until a termination criterion is met in step 110. A termination criterion may be met, for example, when the available resources, particularly the available time or media or stem cells, are exhausted; when a process protocol has been found that achieves sufficient organoid structure quality; and / or when a process protocol with a sufficiently low variance in organoid structure quality has been found. Another termination criterion may be the finding of a process protocol that generates organoid structures within a specified time. Yet another termination criterion concerns the variation in quality assessment. For example, the variation in quality assessment from iteration to iteration of step 106 may be taken into account.If this variation in the quality assessment is below a certain threshold, the termination criterion may be met.

[0086] In this way, the best possible process protocol can be found within the given parameters and made available in step 112 for, for example, large-scale and industrial production of organoid structures. The process protocol is specifically configured to carry out the corresponding process for growing organoid structures using a suitable automated growing device. The process protocol can, for example, include corresponding control instructions.

[0087] In some examples, an "optimal" process protocol can be determined in step 112. This could be the protocol that shows the best quality rating in a corresponding quality metric.

[0088] FIG. 2 shows an exemplary system 200 configured to carry out the previously described method 100. The system 200 comprises a control device 202 and a rearing device 204. The elements of the rearing device 204 can be physically contained within a single housing or be physically separate. The control device 202 can, for example, be an electronic control system, such as a computer system. The rearing device 204 can be at least partially automated and coupled to the control device 202.

[0089] The control device 202 can, for example, be one or more computer systems with main memory, mass storage, a processing unit, a user interface, and input / output interfaces for coupling with the rearing device 204. For example, one computer system can be provided for automatically executing the protocols, another computer system for monitoring and evaluating the process protocols, and yet another computer system for generating further protocols. The control device 202 can be located near the rearing device 204. In other examples, the control device 202 can be located remotely from the rearing device 204 (remote computing) and coupled to the rearing device 204 via a remote data link. The control device 202 can be configured to perform the procedures described herein.In particular, program code (software) can be provided in a working memory of the control device 202 which, when executed by the processing unit of the control device 202, performs the procedures described herein.

[0090] The growing device 204 can have a sterile interior in which several plates 206 are arranged, for example, in a rack (so-called "tub hotel" or "plate hotel"). In addition to sterility, the growing device 204 can also ensure an optimal temperature, humidity, and gas atmosphere for growth (for example, regulated CO2, O2, or N2). In the example shown in FIG. 2, nine plates 206A-206J are depicted. However, this number of plates is exemplary, and any other number of plates 206 can be provided in the growing device 204. The plates 206 can also be arranged differently, for example, on a conveyor belt or in a carousel. Each plate 206 has several process tubs 208. In the example shown in FIG. 2, each plate 206 has four tubs 208A-208D. This number is only exemplary.In other examples, each plate 206 can have more than four wells, for example, 6, 12, 24, 48, 96, 384, 1536, 3456, 9600, or each plate 206 can have only one well 208. The plates 206 can have different numbers of wells 208. Each well 208 can contain a nutrient solution in which an organoid structure or a composite of several organoid structures can be grown. Alternatively or additionally, each well 208 can contain other substances, for example, gel-like substances, to support the growth of the organoid structures. The plates can have lids (not shown) to ensure closure and / or sterility, which are removed before treatment. The wells can be shaped differently, for example, depending on the organoid structures to be produced or to create different cultivation environments.One or more lighting devices in the growth apparatus 204 can irradiate one or more trays with light, for example, infrared light, visible light, and / or ultraviolet light. The trays can be coupled to mechanical actuators, for example, a shaking or pivoting actuator (e.g., an orbital shaker), to allow slight movement of the organoid structures in their nutrient solution.

[0091] The rearing device 204 may further comprise one or more monitoring modalities 210, one or more actuator or robot devices 212, and one or more treatment devices 214. Modules for implementing a respective monitoring modality may, for example, be provided in each shelf location or at a central location, so that a robot device grasps the respective tray and moves it to the central location to perform the corresponding measurement.

[0092] A transport device (not shown) can move one of the plates 206 such that one of the trays 208 of the corresponding plate 206 is positioned in a working area 216 of the monitoring modalities 210, the robot devices 212, and the treatment devices 214. In other embodiments, several working areas 216 with corresponding monitoring modalities, robot devices, and treatment devices can be provided, so that several trays 208 can be processed simultaneously. In the example shown in FIG. 2, for instance, tray 208B of plate 206E is located in the working area 216 of the monitoring modality 210, the robot device 212, and the treatment device 214.

[0093] The monitoring modalities 210 can include, for example, a camera, a temperature sensor, a pressure sensor, a pH sensor, and the like. The monitoring modalities 210 can be used to record properties of the organoid structure culture and / or properties of the culture medium in the vat 208B, such as the temperature and pH of the culture medium, the ambient pressure in the area of ​​the vat 208B, and an image of the organoid structure culture in the vat 208B.

[0094] The robotic device 212 can, for example, perform mechanical tasks, such as stirring the culture medium in the vat 208B, disassembling the organoid structure culture into different parts, or removing parts of the organoid structure culture from the vat 208B. The robotic device 212 is shown schematically in Fig. 2. The robotic device 212 can comprise several robotic devices, for example, a robotic direction for moving the plate 206 and opening and closing the vats 208, and another robotic device or manipulator for handling the organoid structures and the medium in the vats 208.

[0095] The treatment device 214 can, for example, introduce substances into the vat 208B, such as adding liquid or solid substances, or remove parts or all of the solution from the vat 208B, for example by suction, and replace it with a new solution. The treatment device 214 can, for example, take samples of the solution in the vat 208B for analysis, for example, using appropriate monitoring modalities 210. For example, the composition of the solution in the vat 208B can be determined, such as its nutrient content. Individual or groups of organoid structures can also be removed from their plate by the treatment device and transferred to a new culture vessel. The aim may be to keep the individual objects intact. However, it may also be to deliberately break down the objects into cell aggregates of various sizes, down to individual cells.

[0096] The treatment device 214 can also be used to treat substances in

[0097] To convert plates without cells, for example to prepare plates for use with cells / organoid structures by applying a surface coating.

[0098] Another embodiment of treatment device 214 can, for example, separate parts of the organoid structures, cut them, crush them or otherwise mechanically process them.

[0099] By coupling the growing device 204 with the control device 202, the control device 202 can move any tub 208 of any plate 206 into the working area 216 at a desired time and control and monitor the growing of the organoid structures in the tub 208 according to a process protocol.

[0100] In particular, the method 100 described in FIG. 1 can be carried out with the system 200 to perform efficient optimization of process protocols for the production of organoid structures.

[0101] To do this, an initial process protocol can first be created. See also step 102 of the previously described method 100 in FIG. 1. For example, a database provided in or accessible to the control device 202 may contain a number of historical protocols that were used in the past for the generation of similar and / or identical organoid structures. The initial process protocol can be selected from this set.

[0102] A process protocol can, for example, describe a series of process steps performed for the cultivation of organoid structures and, in particular, finished organoids. Each process step can optionally be associated with one or more events that begin or end the respective process step. A process protocol includes, for example, information on a media change, a media composition, a temperature or temperature profile, and the like. A user can specify, via the user interface of the control device 202, which process steps are adaptable. For example, it can be specified whether a medium change is performed during a differentiation phase of the organoid structure and what type of medium is to be used.Furthermore, the user can specify which actions are possible, such as a time for changing the medium or adjusting the specific composition of the medium. Additionally, for each action, the limits within which variations are permitted can be defined. For example, time windows for changing the medium or a maximum interval between two media changes can be set. Other parameters may need to be considered, such as a maximum dosage of substances, for example, growth factors added to a nutrient solution, or a maximum number of media changes. Limit values ​​for temperature, pressure, or pH can also be specified, within which these parameters can be varied and adjusted. Cell density or the number of cysts / organoid structures per tank relative to the available surface area (A / volume) may also need to be taken into account.

[0103] Finally, the user can specify the expected effect for each action. Relevant sensitivity information can be obtained from historical process logs. A suggested expected effect can also be determined by the control device 202, for example, based on historical or machine-learned information. If no information on an expected effect is available, this can be explicitly marked as "unknown".

[0104] For the action with the strongest expected effect (i.e., greatest sensitivity), an alternative process protocol can be created based on the initial process protocol(s). For example, the alternative process protocol can be created by modifying the action with the strongest expected effect as much as possible within the given limits. The user can also specify how many process protocols (hereinafter also referred to as process protocol variants) should be considered and executed in total. A small number of process protocol variants can be executed quickly and cost-effectively, but may result in a less optimized process protocol. Executing many process protocol variants takes longer and is more expensive, but can lead to a better optimized protocol.

[0105] Furthermore, statistical design of experiments can be used to estimate how many trials of each process protocol variant are needed to conclude, for a given significance level, that the alternative process protocol variant is significantly better than the initial process protocol variant.

[0106] Depending on the available budget in terms of, for example, time and resources, and taking into account the specified number of process protocol variants to be considered, further alternative process protocol variants can be created based on the initial process protocol variant by changing actions with a strong expected effect within the possible limits.

[0107] The process protocol variants thus determined are carried out and repeated as frequently as necessary to ensure statistical significance of the results (Step 104). During or after the execution of the process protocol variants, the growth of the organoid structures is monitored using monitoring modalities 210. For example, the execution of each process protocol variant can be evaluated by imaging at specific time points or process steps. The evaluation can be based, for example, on the formation of the desired organoid structures, their size, their morphological properties, their color, their shape, or on the spacing between organoid structures in an organoid structure culture (Step 106). The evaluation can also include, for example, the formation of organ-specific cell types (which, for example,This may be characterized by the onset of a specific fluorescence or a change in morphology, density / refraction), the formation of an organ-specific structure (e.g. rosettes or multiple lumens), the metabolism or production of organ-specific substances or proteins (which can be measured in the culture medium, for example), the onset of further organ-specific functions (e.g. contraction), the occurrence of signaling (e.g. electrical impulses or coordinated calcium spikes), or the occurrence of substance transport.

[0108] The results of all runs of a process protocol variant are merged by determining, for example, the mean and standard deviations of the evaluations. The mean can be used, for instance, as a criterion for the quality of the organoid structures that can be produced with this process protocol variant, and the standard deviation can indicate the applicability of this process protocol variant for the large-scale production of organoid structures, such as on an industrial scale. Low variance indicates consistent quality of the organoid structures, which is generally desirable for large-scale production. Process protocol variants with high variance are generally less suitable for industrial-scale use, as they can result in high scrap rates and significant quality variations.

[0109] Figure 3 shows exemplary details of the execution and evaluation of a specific process protocol (process protocol variant). Reference numeral 302 denotes a time axis t over the course of the growth of organoid structures. The process protocol variant can specify process steps E1 to E5, which are to be carried out at specific times. The execution of these process steps can be varied within a predefined time frame, as indicated by the dashed arrows in relation to process step E1. For example, process step E1 can be carried out slightly earlier or slightly later, but always before process step E2. Process step E1 can, for example, indicate a 50% change of the medium in which the organoid structure is grown. Process step E2 can, for example, indicate a 100% change of the medium.Process step E3, for example, could indicate the addition of a first growth factor to the medium, and process step E4, for example, could indicate the addition of a second growth factor to the medium. Process step E5, for example, could be performed when the organoid is complete and indicate another 100% change of the medium.

[0110] During the execution of the process protocol variant (see step 104 in FIG. 1), the state of the organoid structure culture can be continuously or at specific time points using, for example, an imaging device (see step 105 in FIG. 1). Reference numeral 306 denotes corresponding images over time t. Some or all of the acquired images of the organoid structure culture can be evaluated by a user, particularly an expert, with regard to the quality of the generated organoid structures and annotated accordingly (see reference numeral 308). Physical parameters can also be recorded during the execution of the process protocol variant (see reference numeral 310).

[0111] Multiple execution instances of a process protocol variant can be carried out sequentially or simultaneously in several vats 208. Likewise, the different process protocol variants can be carried out sequentially in the rearing device 204 or simultaneously in several vats 208. Mixed configurations are also possible to make the best possible use of the available capacity of the rearing device 204.

[0112] For good scalability, it can be advantageous to automate as many process steps as possible for each process protocol variant. This can be achieved using an automated propagation device, as described in connection with FIG. 2, and continuous image acquisition to monitor the progress of the current process steps. In particular, the processing of the image data can be carried out using machine learning methods, for example, based on the annotated images (reference numbers 306 and 308).

[0113] Based on the merged values ​​from the execution of the process protocol vanants, further process protocol variants can be automatically generated (step 108 in FIG. 1). For example, a Bayesian optimization can be performed based on the merged values.

[0114] For this additional process protocol variant found in this way, it can also be determined how often this process protocol variant needs to be carried out in order to expect statistically significant results, for example using a statistical design of experiments or based on previously used values.

[0115] The next process protocol variant can be executed multiple times (see FIG. 1: Step 104) and evaluated (see FIG. 1: Step 106). Further process protocol variants can be automatically generated until a termination criterion is reached (see FIG. 1: Step 110). The termination criterion can be reached when a predefined time budget is exhausted, a predefined resource budget is exhausted, or a process protocol variant has been generated that produces the organoid structure with statistical certainty above a predefined threshold.

[0116] It is possible that with some process protocol variants, it will be determined relatively early on, or at least predicted with sufficient probability (before all growth phases of the finished organoid are complete), that the protocol will not lead to the desired result, for example, if no meaningful formation of an organoid structure from the stem cells is visible during the differentiation phase. In these cases, the process can be terminated prematurely, and the process protocol variant can be assigned a poor result. Such early termination can be advantageously exploited by optimization methods with dynamic search budgets and can also save valuable culture media.

[0117] Naturally, the features of the embodiments and aspects of the invention described above can be combined with one another. In particular, the features can be used not only in the combinations described, but also in other combinations or individually, without leaving the scope of the invention.

[0118] As an example, in some cases the optimization of process protocols described above, which involves the physical implementation of process protocols in an automated growth device, is at least partially replaced or supplemented by appropriate simulations. For instance, if a digital twin of the growth process exists, the effects of different process protocols can be simulated instead of being investigated physically. The digital twin could, for example, be a physical-chemical-biological model and / or model the effect of individual actions on the organoid in a data-driven manner. This reduces the time required for implementation.For example, it would be possible to use the digital twin to generate hypotheses for the validity assessment of a specific process protocol and, if a hypothesis appears promising, subsequently initiate the execution of the corresponding process in the automated rearing device (i.e., not in the digital twin). Such an implementation using a digital twin can offer particular advantages, especially in connection with the reinforcement learning model, as described above in connection with step 108 from FIG. 1. The automatic generation of further process protocol variants (step 108) can, for example, be carried out using reinforcement learning techniques. In this case, no assumptions about good search strategies, i.e., about expected effects and improvements, are required.This typically requires a large number of executions of the process protocol vanants. Therefore, using the previously described digital twin to simulate the growth process of organoid structures can be advantageous in conjunction with reinforcement learning techniques. Furthermore, the reward function of reinforcement learning can also be automatically trained using the digital twin.

[0119] Furthermore, the present invention is not limited to the process sequences described above, e.g., from FIG. 3. For example, the process protocols can consist of several phases. These phases can include, for example, seeding, differentiation, expansion, and maturation. In many cases, the transition to the next phase only occurs once the preceding phase has been completed with a satisfactory result. In these cases, the selection and design of the process steps in a later phase are unaffected by when and how the process steps in a previous phase were carried out. A kind of Markov assumption applies. This can be exploited during optimization (step 108), so that initially only the process steps of the first phase are optimized, and then all phases are optimized successively.

[0120] The term organoid structure, as used herein, is not limited to the final product, i.e., the finished organoid, but also encompasses precursors of an organoid during cultivation, in particular 3D cell cultures that can arise during the development of a finished organoid, so-called organoid precursors. The term organoid structure therefore also includes 3D cell structures, such as cysts or embryonic bodies.

[0121] Furthermore, techniques for growing organoid structures have been described above. The methods disclosed herein are also generally applicable to other 3D cell cultures, such as spheroids and tumoroids. As with the growth of organoid structures, it has been observed that even with nominally identical growth protocols, instance-specific variations in growth results or growth rate can occur with spheroids and tumoroids. Therefore, in such cases, it can also be helpful to apply the techniques described herein for finding optimized process protocols.

Claims

Patent claims 1. A procedure comprising one or more iterations of the following steps: - Providing one or more process protocols, each defining a process for growing organoid structures in an automated growth device, - Controlling the automated rearing device based on the one or more process protocols in order to carry out each of the one or more processes at least partially, - Monitoring the growth of the organoid structures using one or more monitoring modalities of the automated growth device, and - based on a result of monitoring, generating one or more further process logs.

2. The method according to claim 1, wherein the one or more further process protocols are generated based on an optimization method.

3. The method of claim 2, wherein the optimization method is parameterized based on monitoring.

4. Method according to claim 2 or claim 3, wherein the optimization method is parameterized based on predetermined parameter limits and / or an expected effect.

5. A method according to one of claims 2 to 3, wherein the optimization method determines the one or more process protocols taking into account a quality assessment of the corresponding processes, wherein the quality assessment takes into account one or more of the following criteria: - a level of difficulty in carrying out the corresponding process, - a quantity of material consumed in the corresponding process, - the total duration of the corresponding process, - the occupancy time of a manufacturing system; - a number of manual steps involved in raising the animals; - a dispersion of the generated organoid structures; - a yield of generated organoid structures, - the development of organ-specific cell types; - the development of organ-specific structures; and / or - the onset of organ-specific functions.

6. Method according to one of the preceding claims, wherein the generation of one or more further process protocols comprises Bayesian optimization.

7. Method according to any of the preceding claims, wherein monitoring the growth of the organoid structures comprises determining instance-specific values ​​of one or more properties of several organoid structures.

8. Method according to claim 7, wherein the monitoring comprises determining a statistical distribution of the instance-specific values.

9. The method of claim 7 or 8, wherein one or more of the properties are selected from the following group: appearance of the organoid structures or components of the organoid structures in an optical image; morphology; size; shape; Color; Inter-organoid structure distance.

10. Method according to any one of claims 7 to 9, wherein the instance-specific values ​​are determined by means of a machine-learned Models are determined.

11. Method according to any of the preceding claims, comprising providing the multiple process protocols: - Selections from one or more predefined initial process protocols, - Varying one or more predefined initial process protocols to obtain the multiple process protocols.

12. Method according to claim 11, wherein one or more predefined initial process protocols are varied based on predetermined parameter limits and / or an expected effect.

13. Method according to any of the preceding claims, wherein the automated rearing device is controlled to perform each of the processes multiple times.

14. Method according to claim 13, wherein a number of multiple executions for each process is determined by means of statistical design of experiments for the respective process protocol.

15. Method according to claim 13 or claim 14, wherein for each repeatedly performed process, monitoring the growth of the organoid structures comprises determining instance-specific values ​​of one or more properties of several organoid structures and determining a statistical distribution of the instance-specific values.

16. Method according to any of the preceding claims, wherein one or more monitoring modalities provide an optical image of the organoid structures.

17. Method according to one of the preceding claims, wherein the process protocols provided in a subsequent iteration of the multiple iterations replace those provided in a preceding iteration of the Several iterations generated one or more additional process logs.

18. Method according to any of the preceding claims, wherein one or more iterations are performed until a termination criterion is met, the termination criterion being selected from the following group: a predetermined time budget is exhausted, a variation in a quality rating occurs between the multiple iterations, a predetermined resource budget is exhausted, and / or a process log has been generated which has a quality rating that meets one or more predetermined quality criteria.

19. Method according to any of the preceding claims, wherein the one or more iterations further comprise the following step: - optionally terminating the execution of a process based on monitoring the growth of the organoid structures.

20. System comprising: an automated growing device for growing organoid structures with one or more monitoring modalities, and a control device configured to perform one or more iterations of the following steps: - Providing one or more process protocols, each defining a process for growing organoid structures in the automated growing device, - Controlling the automated rearing device based on the one or more process protocols in order to carry out each of the one or more processes at least partially, - Monitoring the growth of the organoid structures using one or more monitoring modalities, and - based on a result of monitoring, generating one or more further process logs.

Images