Prediction device and prediction method

Abstract

A computer acquires a parameter value set including values of a plurality of parameters, the values of the plurality of parameters being respectively classifiable into at least one factor of a plurality of factors that affect generation of a crystal. The computer performs a prediction related to generation of a crystal form available to a substance by inputting the parameter value set into a crystal generation model. The crystal generation model represents a process of generation of a crystal, and is a model including time as an element. The computer outputs information related to a result of the prediction, that is, prediction result information.

Description

Technical Field

[0001] This invention generally relates to predictive techniques related to the available crystal forms of materials. Background Technology

[0002] As a predictive technique related to the available crystal form of a substance, it is known to use first-principles calculations to calculate energy information related to the crystal structure (structure of the crystal form), and to predict the crystal structure based on the calculated energy information (e.g., Patent Document 1).

[0003] Existing technical documents Patent documents Patent Document 1: Japanese Patent Application Publication No. 2020-166706 Summary of the Invention The technical problem that the invention aims to solve Typically, first-principles calculations are based on quantum mechanics (first principles), are computationally demanding, and are used to deal with molecular behavior and intermolecular interactions at the microscopic level of time (typically a few picoseconds to tens of picoseconds). Furthermore, first-principles calculations are essentially calculations that predict the final morphology of precipitated crystals (the process of which is not explicitly defined).

[0004] On the other hand, the process of crystal formation from a substance (liquid phase) is a macroscopic process in terms of time.

[0005] Therefore, if first-principles calculations are used to process macroscopic processes over time, it is necessary to construct the macroscopic process by accumulating the calculated microscopic processes over time. This results in a high computational load. Analyzing processes lasting from minutes to hours is practically impossible.

[0006] Technical solutions for solving technical problems The computer obtains a set of parameter values, which can be categorized into at least one of several factors influencing crystal formation. By inputting this set of parameter values ​​into a crystal formation model, the computer makes predictions related to the formation of the achievable crystal form of the material. The crystal formation model represents the crystal formation process and includes time as a factor. The computer outputs information related to the prediction results, i.e., the prediction result information.

[0007] Invention Effects According to the present invention, predictions related to the macroscopic formation of crystals over time can be made with a smaller computational load than calculations using first-principles methods. Attached Figure Description

[0008] Figure 1 This is a diagram showing the configuration of the prediction device according to the first embodiment.

[0009] Figure 2 This is a diagram illustrating a summary of the process performed in the first embodiment.

[0010] Figure 3 This is a diagram showing an overview of the extraction prediction.

[0011] Figure 4A This is a diagram showing the effect of infrared radiation on the precipitation of crystal form 1.

[0012] Figure 4B This is a diagram showing the effect of infrared radiation on the precipitation of crystal form 2.

[0013] Figure 5 This is a schematic diagram of the processing during the runtime phase.

[0014] Figure 6 This is an example of a screen showing the prediction results based on the output information.

[0015] Figure 7 This is a diagram illustrating a first example of the application of the prediction device.

[0016] Figure 8 This is a diagram illustrating a second example of the application of the prediction device.

[0017] Figure 9 This is a diagram showing the configuration of the prediction device according to the second embodiment.

[0018] Figure 10 This is a schematic diagram of the precipitation prediction involved in the second embodiment. Detailed Implementation

[0019] In the following description, "interface device" can be one or more interface devices. These one or more interface devices can be at least one of the following: • One or more I / O (Input / Output) interface devices. An I / O (Input / Output) interface device is an interface device between at least one of an I / O device and a remote display computer. The I / O interface device for the display computer can be a communication interface device. At least one I / O device can be any of a user interface device, such as an input device like a keyboard and point-and-click devices, and an output device like a display device.

[0020] • More than one communication interface device. More than one communication interface device can be more than one of the same type of communication interface device (e.g., more than one NIC (Network Interface Card)) or more than two different types of communication interface devices (e.g., NIC and HBA (Host Bus Adapter))).

[0021] Additionally, in the following description, "memory" refers to one or more storage devices as an example, typically a main storage device. At least one storage device in the memory can be a volatile storage device or a non-volatile storage device.

[0022] Additionally, in the following description, "permanent storage device" can refer to one or more permanent storage devices as an example of more than one storage device. Permanent storage devices can typically be non-volatile storage devices (such as secondary storage devices), specifically, such as HDD (Hard Disk Drive), SSD (Solid State Drive), NVME (Non-Volatile Memory Express) drives, or SCM (Storage Class Memory).

[0023] Additionally, in the following description, "storage device" can be at least a memory or a permanent storage device.

[0024] Additionally, in the following description, "processor" can refer to more than one processor device. At least one processor device can typically be a microprocessor device such as a CPU (Central Processing Unit), but can also be other types of processor devices such as a GPU (Graphics Processing Unit). At least one processor device can be single-core or multi-core. The at least one processor device can be a processor core. At least one processor device can also be a processor device in a broader sense, such as a circuit that is described in a hardware description language as a "collection of gate arrays" (e.g., FPGA (Field-Programmable Gate Array), CPLD (Complex Programmable Logic Device), or ASIC (Application Specific Integrated Circuit)) that performs some or all of the processing.

[0025] Furthermore, in the following description, functions are sometimes described using the term "yyy part," but functions can also be implemented by a processor executing more than one computer program, by more than one hardware circuit (e.g., FPGA or ASIC), or by a combination thereof. When a function is implemented by a processor executing a program, appropriate use of storage devices and / or interface devices, etc., is made for the determined processing; therefore, the function can also be considered at least a part of the processor. The processing described with the function as the subject can also be processing performed by the processor or a device having the processor. Programs can also be installed from a program source. A program source can be, for example, a program-distributed computer or a computer-readable recording medium (e.g., a non-transitory recording medium). The description of each function is only one example; multiple functions can also be combined into one function, or one function can be divided into multiple functions.

[0026] Hereinafter, several embodiments of the present invention will be described with reference to the accompanying drawings. It should be noted that in the following description of the embodiments, the substance is dissolved in a solvent as a solute, and crystals are formed by drying the solvent in which the solute is dissolved, etc. Therefore, the term "precipitation" is used as an example of "formation" of crystals.

[0027] [First Implementation Method] Figure 1 This is a diagram showing the configuration of the prediction device according to the first embodiment.

[0028] In this embodiment, the prediction device 100 is a physical computer system, but it can also be a logical computer system based on a physical computer system, or a combination of at least a portion of a physical computer system and at least a portion of a logical computer system. The physical computer system can consist of one or more physical computers, including an interface device 51, a storage device 52, and a processor 53 connected to them. The logical computer system can include virtual machines or systems that provide cloud computing services.

[0029] An input / output console 80 is connected to the interface device 51. The input / output console 80 is a human-machine interface device operated by the user 10. The input / output console 80 may be, for example, the input and display devices of the prediction device 100, or a remote information processing terminal (e.g., a client) communicatively connected to the prediction device 100 (e.g., a server). The information processing terminal may be, for example, a personal computer or a smartphone.

[0030] Data and programs are stored in storage device 52. For example, the data includes a known substance database 190 containing information on each known substance and a processing result database 110 that is built or updated in the processing described later. For example, the programs include prediction software 191.

[0031] The processor 53 executes the prediction software 191, which realizes the functions of the input unit 120, the prediction unit 133, the fitting parameter optimization unit 134, and the output unit 140.

[0032] The input unit 120 receives information input from the user 10 via the input / output console 80. Based on at least a portion of the information input from the user 10, information from the known material database 190, information from the processing result database 110, and information as a result of processing performed by the fitting parameter optimization unit 134, the input unit 120 obtains a parameter value set. The "parameter value set" includes the values ​​of multiple parameters, each categorized as at least one of multiple factors affecting crystal precipitation. The parameter value set includes the fitting parameter values ​​described later.

[0033] The precipitation prediction unit 133 predicts the achievable crystal form of the substance by inputting the set of parameter values ​​obtained by the input unit 120 into the crystal precipitation model 132. The crystal precipitation model 132 is a model representing the crystal precipitation process and including time as a factor. The crystal precipitation model 132 can be a machine learning model or other types of models. In this embodiment, at least a portion of the crystal precipitation model 132 can be a model used for computer simulation (e.g., an approximate model), particularly a continuous simulation model. The precipitation prediction unit 133 constructs or updates the processing result database 110 based on the processing results.

[0034] The fitting parameter optimization unit 134 optimizes one or more fitting parameter values ​​contained in the parameter value set input to the crystal precipitation model 132. Optimization is performed, for example, using a fitting model 136. The fitting model 136 can be a machine learning model or other types of models, such as linear regression, logistic regression, SVM (Support Vector Machine), decision tree model, neural network (e.g., CNN (Convolutional Neural Network), RNN (Recurrent Neural Network)) or Bayesian optimization.

[0035] The output unit 140 generates information based on the prediction results made by the extraction prediction unit 133, and outputs this information to the input / output console 80 (typically for display).

[0036] Figure 2 This is a diagram illustrating a summary of the process performed in the first embodiment.

[0037] In this embodiment, there are a preparation phase and an operation phase.

[0038] In the preparation phase, known substance information related to known substances is processed (S210). During this processing, a processing result database 110 is constructed and updated. The known substance information can be information input by user 10 or information obtained from a known substance database 190. "Known substances" are substances with known crystal forms. As described later, the processing result database 110 includes functional group patterns and sets of one or more fitted parameter values.

[0039] During the operation phase, the processing results database 110 and information about unknown substances related to the unknown substances are used for processing (S220). "Unknown substance" refers to a substance whose available crystal form is unknown. The prediction device 100 can assist in screening for the available crystal forms of the unknown substance. As described later, the unknown substance information includes functional group pattern information representing one or more functional group patterns for the unknown substance. These one or more functional group patterns can be determined by the chemical formula of the unknown substance or inferred from its absorption spectrum (infrared absorption spectrum).

[0040] The execution phase can begin once the given start conditions are met. The execution phase can begin after the preparation phase has ended, or it can begin midway through the preparation phase (i.e., there can be a period where the preparation phase and the execution phase run concurrently).

[0041] Figure 3 This is a diagram showing an overview of the extraction prediction.

[0042] The crystal precipitation model 132 comprises a network of equations. At least a portion of this network can be an approximate model. The network has one or more equations for each of the multiple factors (multiple analytical indices) affecting crystal precipitation. These equations are either time-evolution equations (e.g., differential equations) that take time as a factor, or non-time-evolution equations that do not take time as a factor. Figure 3 In the diagram, equations 301A to 301H, represented by thick straight lines with arrows, are time-evolution equations, while equation 301J, represented by thick straight lines without arrows, is a non-time-evolution equation. The equation network can be a directed graph where equations are nodes and the relationships between equations are directed edges. The direction of the directed edge (from the starting point to the ending point) means that the value output from the equation at the starting point is input into the equation at the ending point. The ratio of starting point to ending point can be 1:1, many:1, 1:many, and many:many. For example, input and output of values ​​obtained over a certain time period can be performed between equations 301.

[0043] Crystal precipitation model 132 represents the crystal precipitation process. The crystal precipitation process is a macroscopic process over time, equivalent to a change in thermodynamic stability. For each crystal form, crystal precipitation depends on the relationship between temperature and saturation concentration. An example of this relationship is the solubility curve. Therefore, multiple factors (multiple analytical indicators) include factors related to temperature or concentration, such as product temperature, drying rate, solvent concentration, solute concentration, supersaturation, and solubility (it should be noted that "product" refers to the solvent in which the substance (solute) is dissolved). Regarding solubility, for each crystal form, there exists an equation related to the solubility curve, namely, solubility equation 301J. For example, if there are crystal form 1, crystal form 2, ..., then there are solubility equations 301J1, 301J2, ... The solubility curve is represented using a two-dimensional orthogonal coordinate system of temperature and concentration, and does not include the factor of time; therefore, solubility equation 301J, as described above, is a time-independent evolution equation.

[0044] Furthermore, in this embodiment, crystal precipitation can be achieved under infrared irradiation at a selected wavelength. That is, the precipitated crystal form can be controlled through the interaction of infrared light at a specific wavelength with the solute (solvent). The wavelength region where infrared absorption is significant (including the wavelength region with the peak wavelength) varies depending on the functional group mode of the substance. Therefore, by observing the infrared absorption at wavelengths that depend on the functional group mode, easily precipitated and / or difficult-to-precipitate crystal forms can be determined. For example, this can be illustrated by using febuxostat as a case study. Figure 4A As shown, a certain wavelength W absorbed by the carboxyl group COOH Infrared irradiation selectively excites the stretching vibrations of the OH groups within the carboxyl groups, causing the hydrogen bonds between functional groups to dissociate, making it difficult to form crystal form 1 (a crystal form in which the carboxyl groups of two molecules are strongly bonded together by hydrogen bonds). On the other hand, as... Figure 4B As shown, through the same wavelength W COOH Infrared irradiation causes the carboxyl groups to rotate and position at an angle different from the usual position, resulting in a higher likelihood of crystal form 2. Thus, crystal precipitation that is difficult to achieve with crystal form 1 and easy to achieve with crystal form 2 can be achieved by controlling the infrared wavelength. Since crystal precipitation can be achieved through such infrared wavelength control, there are infrared-related factors, such as infrared radiation energy and infrared absorption energy, as analytical indicators. In other words, crystal precipitation model 132 represents the crystal precipitation process under infrared irradiation at a selected wavelength. It should be noted that equation 301A corresponding to infrared radiation energy and equation 301B corresponding to infrared absorption energy are both time-evolution equations, but when the energy is constant regardless of time, at least one of equations 301A and 301B can also be a time-independent evolution equation.

[0045] In addition, as mentioned above, solvent concentration and solute concentration are factors. There is a time evolution equation 301E for solvent concentration. Furthermore, for solute concentration, there is a time evolution equation 301F for the liquid phase and a time evolution equation 301G for each crystal form. The changes in the depth of the solid arrows (straight lines) representing these time evolution equations 301 indicate the changes in concentration over time. That is, since the solvent concentration and liquid phase concentration decrease over time, the solid arrows representing time evolution equations 301E and 301F gradually become lighter. On the other hand, as time passes, the solute concentration associated with the crystal form increases (the amount of precipitated crystals increases), therefore, the solid arrows representing the time evolution equations 301G1 to 301Gn for crystal forms 1 to n gradually become darker.

[0046] Input unit 120 acquires input information 150. Input information 150 may include solute chemical formula 151 (an example of information representing the structure of a substance). Input information 150 includes a set of parameter values ​​152. The parameters corresponding to the parameter values ​​included in the set of parameter values ​​152 depend on multiple factors. Specifically, for example, the parameters corresponding to the parameter values ​​included in the set of parameter values ​​152 include: parameters belonging to the experimental conditions for crystal precipitation (e.g., heater temperature, irradiation wavelength (wavelength of infrared light), initial product temperature, initial concentration), i.e., experimental condition parameters; parameters belonging to physical properties (solvent and / or solute), i.e., physical property parameters; and parameters defined as theoretically difficult to determine, i.e., fitting parameters.

[0047] The parameter values ​​in parameter value set 152 are classified as any factor and input into equation 301 corresponding to the factor of the classification target. The parameter value:factor (analysis index) ratio can be 1:1, many:1, 1:many, and many:many. The relationship between the number of factors m and the number of parameter values ​​n can be arbitrary, but in this embodiment, m < n (i.e., fewer factors than parameter values).

[0048] As fitting parameters, for example, there are solubility parameters, interfacial energy, and the number of crystal forms. The solubility parameter value is input into the solubility equation 301J. The interfacial energy value is input into the time evolution equation 301G associated with the crystal form. For each crystal form in the number of crystal forms corresponding to the value of the number of crystal forms, there are solubility parameter values ​​and interfacial energy values. The value of the number of crystal forms affects the number of time evolution equations 301G associated with the crystal form (for example, the value of the number of crystal forms is the same as the number of time evolution equations 301G used for prediction). In cases where the value of the number of crystal forms has been determined, such as when the crystal precipitation model 132 is applied to a known substance (a known compound), it may not be a fitting parameter. The value of the number of crystal forms can be inferred by analogy based on the functional group pattern of the substance, other information, etc. Hereinafter, for the sake of simplicity, the fitting parameters are set as the solubility parameter and the interfacial energy.

[0049] Interfacial energy refers to the new energy generated at the interface when, for example, a phase transition occurs from a liquid phase to a solid phase (crystal). In this embodiment, the value of the interfacial energy is optimized as one of the fitting parameters.

[0050] Regarding solubility parameters, for example, it can be described as follows. That is, for each crystal form, the precipitation of crystals depends on temperature and saturation concentration, and an example of this relationship is the solubility curve. The solubility curve can ideally be uniquely determined by thermodynamic formulas (such as the van der Hoff equation) if there is only one crystal form and the heat of solution can be rigorously measured; however, in reality, when multiple crystal forms exist, in most cases, even experimentally, it is impossible to accurately determine the solubility curve for each crystal form. Furthermore, there are sometimes examples where the van der Hoff equation cannot be approximated. The solubility curve is determined based on more than one solubility parameter value.

[0051] The following describes the processing for the preparation and operation phases respectively.

[0052] The output obtained from the crystal precipitation model 132 by inputting the parameter value set 152 can contain the values ​​calculated (predicted) for each equation 301 at each elapsed time point, including at least one or more experimental condition parameter values ​​at the end of crystal precipitation. Hereinafter, the experimental condition parameter values ​​contained in the parameter value set 152 can be referred to as "first experimental condition parameter values," and the experimental condition parameter values ​​at the end of crystal precipitation contained in the output of the crystal precipitation model 132 can be referred to as "second experimental condition parameter values." One or more first experimental condition parameter values ​​can be referred to as "first experimental condition parameter value set," and one or more second experimental condition parameter values ​​can be referred to as "second experimental condition parameter value set." For each given experimental condition item (parameter), there can be first experimental condition parameter values ​​and second experimental condition parameter values.

[0053] The treatment in the preparation phase is an example of the first treatment. The treatment in the preparation phase is carried out for every known substance that is so numerous that it can be assumed to cover most (ideally all) functional group patterns.

[0054] For various known substances, one or more parameter value sets 152 can be prepared. Parameter value set 152 includes one or more first experimental condition parameter values ​​and one or more fitted parameter values. In this embodiment, for known substances, the value of the number of crystal forms may not be a fitted parameter value. The precipitation prediction unit 133 obtains a prediction result containing one or more predicted second experimental condition parameter values ​​by inputting the parameter value set into the crystal precipitation model 132. The fitted parameter optimization unit 134 optimizes multiple fitted parameter values ​​(in this embodiment, the solubility parameter value and interfacial energy value of each crystal form among the number of crystal forms corresponding to the value of the number of crystal forms) based on the predicted one or more second experimental condition parameter values. Specifically, for each parameter value set 152, the fitted parameter optimization unit 134 learns one or more fitted parameter values ​​that make the predicted one or more second experimental condition parameter values ​​close to or overlap with one or more second experimental condition parameter values ​​obtained through experiments (one or more second experimental condition parameter values ​​at the end of crystal precipitation in the experiment). More specifically, for example, the fitting parameter optimization unit 134 learns a fitting model 136 for each parameter value set 152 based on data including one or more predicted second experimental condition parameter values, one or more second experimental condition parameter values ​​obtained through experiments, and one or more fitting parameter values. The input to the fitting model 136 can be one or more predicted second experimental condition parameter values, and the output of the fitting model 136 can be one or more fitting parameter values.

[0055] The precipitation prediction unit 133 registers in the processing result database 110 data containing functional group patterns representing a substance (e.g., a known substance) (including at least one of the number of functional groups, the type of functional groups, the position of the functional groups, and the bonds between functional groups) and sets of values ​​or ranges of each fitting parameter. Depending on the functional group pattern, the aforementioned sets may sometimes also include information about infrared irradiation conditions (e.g., one or more wavelengths or wavelength regions of the irradiated infrared radiation, the irradiation start time and irradiation duration of each of the one or more wavelengths or wavelength regions) or other information. For example, the processing result database 110 may contain a set of fitting parameter values ​​(one or more fitting parameter values, such as solubility parameter values ​​and interfacial energy values) for each functional group pattern or each set of functional group patterns (one or more functional group patterns possessed by a known substance) for each crystal form of that functional group pattern or set of functional group patterns.

[0056] In the preparation phase, the above processing is performed on every known substance among a wide variety sufficient to be assumed to cover most functional group patterns. Therefore, it is expected that in the processing result database 110, for each functional group pattern among most (ideally all) functional group patterns, a set containing the functional group pattern and the value or range of each fitting parameter is registered. Thus, in the preparation phase, the sets containing the functional group pattern and the value or range of each fitting parameter are accumulated in the processing result database 110 for learning the fitting parameter values. Therefore, the processing in the preparation phase can also be referred to as the "first learning processing".

[0057] Figure 5 This is a schematic diagram of the processing during the runtime phase.

[0058] The processing during the operational phase is an example of the second processing. The processing during the operational phase is performed on unknown substances. Specifically, the following operations are conducted.

[0059] (S701) Input unit 120 acquires input information 150 related to the unknown substance and obtains unknown substance information from the input information 150. If the input information 150 contains the chemical formula of the unknown substance, input unit 120 can determine the functional group pattern based on the chemical formula and obtain unknown substance information containing information representing the determined functional group pattern. If the input information 150 contains information representing the absorption spectrum, input unit 120 can estimate the functional group pattern of the unknown substance based on the absorption spectrum and the known substance database 190 (or another database representing the relationship between functional group patterns and absorption spectra), and obtain unknown substance information containing information representing the estimated functional group pattern. The unknown substance information may contain information on one or more functional group patterns. Furthermore, the unknown substance information may contain a set of parameter values ​​contained in the input information 150. The set of parameter values ​​may contain a set of first experimental condition parameter values. In the set of parameter values, the parameter values ​​of physical properties can be determined based on the basic data of the unknown substance. "Basic data" can be data obtained through case studies of unknown substances (e.g., comparisons with similar substances) or analysis by any method (e.g., DTA analysis (Differential Thermal Analysis), DSC analysis (Differential Scanning Calorimeter)).

[0060] (S702) The input unit 120 uses one or more functional group patterns represented by unknown substance information as key quantities and refers to the processing result database 110.

[0061] (S703) The input unit 120 estimates the number of crystal forms obtainable from the unknown substance, and for each crystal form among the number of crystal forms corresponding to the estimated number of crystal forms, estimates the fitting parameter value for each fitting parameter. The number of crystal forms can be estimated based on the number and content of functional group patterns represented by the unknown substance information and the processing result database 110. In addition, the fitting parameter value can be estimated by either of the following methods (X) and (Y).

[0062] (X) Input unit 120 estimates the value range of each fitting parameter (the value range of each solubility parameter and the value range of interfacial energy among more than one solubility parameter). For each fitting parameter, the upper and lower limits of the value range can be the maximum and minimum values ​​estimated for that fitting parameter using one or more functional group modes as key quantities. For each fitting parameter, the value range is the range of values ​​to be explored. For example, in a two-dimensional orthogonal coordinate system of temperature and concentration, the range of exploration of the solubility curve (specifically, the range of exploration for each solubility parameter value) is determined. It should be noted that the lower limit of the solubility curve can be a reference solubility curve determined as a solubility curve in the crystal precipitation of an unknown substance. The "reference solubility curve" can be a curve corresponding to the most precipitated crystal form, or it can be an experimentally based solubility curve. Input unit 120 sets the exploration range (value range) of each fitting parameter for each crystal form among the number of crystal forms corresponding to the estimated number of crystal forms (e.g., included in the parameter value set of unknown substance information). The fitting parameter optimization unit 134 refers to the parameter value set and, for each crystal form, explores the fitting parameter value from the exploration range through machine learning (using fitting model 136) for each fitting parameter. Thus, for each crystal form in the estimated number of crystal forms, a solubility parameter value and an interfacial energy value are determined.

[0063] (Y) In cases where the structure of the unknown substance being studied is similar to the structure of multiple known substances in the known substance database 190 (e.g., the distance between the structural features is less than a given value), it may not be necessary to predetermine the exploration range of each fitting parameter. The input unit 120 uses the number and content of functional group patterns represented by the unknown substance information as key quantities, and estimates (determines) the number of crystal forms and fitting parameter values ​​corresponding to structural features that are the same as or similar to these structural features from the processing result database 110.

[0064] Furthermore, the method described above (X), i.e., the method of setting the exploration range, can also be used for prediction of known substances (e.g., processing in the preparation stage). For example, when a parameter value set containing a new first experimental condition parameter value set of a certain known substance is obtained, the input unit 120 can set the exploration range of each fitting parameter for each crystal form among the number of crystal forms corresponding to the value of the number of crystal forms based on the parameter value set, and the fitting parameter optimization unit 134 can search again for the fitting parameter value corresponding to the known substance within the exploration range of the fitting parameter value.

[0065] (S704) The precipitation prediction unit 133 performs a prediction by inputting a set of parameter values ​​(the value of the number of crystal forms and one or more fitted parameter values ​​for each crystal form among the number of crystal forms corresponding to the value of the number of crystal forms) estimated in S703 into the crystal precipitation model 132. In this prediction, one or more fitted parameter values ​​(solubility parameter value and interfacial energy value) estimated in S703 for each crystal form among the number of crystal forms corresponding to the value of the estimated number of crystal forms are used as reference values.

[0066] (S705) The prediction unit 133 derives the experimental conditions based on at least a portion of the prediction results in S704 and the estimated fitting parameter values ​​in S703. It should be noted that the experimental conditions can also be arbitrarily derived by other devices or by the user 10 instead of the prediction device 100.

[0067] (S706) Under the experimental conditions obtained in S705, an experiment was conducted to precipitate crystals from an unknown substance.

[0068] (S707) Evaluate the crystal sample of the crystals precipitated in S706 (e.g., XRD analysis or DSC analysis). It should be noted that evaluation items may include thermal analysis (e.g., thermodynamic stability), crystal profile estimation by XRD analysis, etc.

[0069] (S708) In the evaluation in S707, if the termination conditions of the process are met (e.g., if the evaluation result is consistent with the prediction for all crystal forms), information related to the unknown substance is provided to the utilization entity that provided the unknown substance (described later).

[0070] In the evaluation in S707, if the termination condition of the process is not met (e.g., if the evaluation result does not conform to the prediction for at least one crystal form), the process flow returns to S703. In this S703, the fitting parameter optimization unit 134 changes (e.g., makes a minor correction) the fitting parameter values ​​estimated in the previous S703. This change can be made using the method (X) described above. Alternatively, this change can be made using the fitting model 136. For example, for a crystal form, the values ​​of the evaluation results (e.g., for each first set of experimental condition parameter values, the second set of experimental condition parameter values ​​corresponding to that first set of experimental condition parameter values) and the values ​​of the prediction results (e.g., for each first set of experimental condition parameter values, the second set of experimental condition parameter values ​​predicted for that first set of experimental condition parameter values) are input into the fitting model 136. For this crystal form, the amount of change in the parameter values ​​of each fitting parameter can depend on the magnitude of the difference between the values ​​of the evaluation results and the predicted values.

[0071] Based on the processing during the operation phase, by repeatedly performing the processing including S703 and S704, for unknown substances, for each crystal form in the number of crystal forms corresponding to the estimated number of crystal forms, the fitting parameter values ​​can be optimized so that the predicted set of second experimental condition parameter values ​​is close to or overlaps with the set of second experimental condition parameter values ​​obtained through experiments (learning the fitting parameter values).

[0072] It should be noted that upon completion of the processing during the operational phase, the precipitation prediction unit 133 can register a set of one or more functional group patterns containing the unknown substance and the estimated fitting parameter values ​​into the processing result database 110. Additionally, at least a portion of the information related to the unknown substance can also be registered into the known substance database 190. During the operational phase, for the unknown substance, the estimation (also referred to as optimization or learning) of fitting parameter values ​​can be performed, and a set of functional group patterns and values ​​or value ranges of each fitting parameter is accumulated in the processing result database 110. Therefore, the processing during the operational phase can also be referred to as the "second learning processing."

[0073] Figure 6 This is an example of a screen showing the prediction results based on the output information.

[0074] In at least one of the preparation and operation phases of the process, the output unit 140 outputs information 170 (see reference 133) based on the results predicted by the extraction prediction unit 133. Figure 3 The prediction results screen 800 is displayed. The prediction results screen 800, for example, contains crystal form information 803 and solubility information 804.

[0075] Crystal form information 803 indicates the available crystal forms of the substance and the precipitation rate of each crystal form. Crystal form information 803 can include information such as precipitation amount and average critical radius for each crystal form.

[0076] Solubility information 804 includes solubility curves depicted in a two-dimensional orthogonal coordinate system of temperature and concentration (the solubility curves for each crystal form represented by crystal form information 803). The solubility curve for each crystal form is based on a set of solubility parameter values ​​(more than one solubility parameter value) estimated for that crystal form.

[0077] Figure 7 This is a diagram illustrating a first example of the application of the prediction device 100.

[0078] There exists an entity 1200 that provides the utilization of the unknown substance as the object of prediction and a service-providing entity 1250 that provides the service. In this embodiment, an "entity" can be a natural person or a legal person, such as a for-profit or non-profit enterprise, organization, research institution or other group.

[0079] In the first usage example, the service providing entity 1250 has a prediction device 100 and a crystal fabrication auxiliary system 1260. The crystal fabrication auxiliary system 1260 includes at least one of a basic data acquisition device 1210, a crystal precipitation device 1220, and an evaluation device 1230.

[0080] The basic data acquisition device 1210 is a device for acquiring basic data of a substance.

[0081] The crystal precipitation apparatus 1220 is an example of an experimental setup for precipitating crystals. The crystal precipitation apparatus 1220 includes an infrared irradiation device (e.g., an infrared heater) that irradiates infrared light of a selected wavelength. The infrared irradiation device causes the solvent containing the solute to evaporate and crystallize by irradiating it with a specific wavelength.

[0082] Evaluation device 1230 is an apparatus for evaluating the crystal precipitated by crystal precipitation apparatus 1220 (e.g., an apparatus used for evaluation in S708).

[0083] For example, the following loop can be performed: For each crystal form, the crystal precipitation device 1220 precipitates a crystal of that crystal form according to the same set of experimental condition parameter values ​​as input to the prediction device 100, and the evaluation device 1230 evaluates the crystal. If the evaluation results for all crystal forms (e.g., the second set of experimental condition parameter values ​​obtained through experiments) differ from the prediction results of the prediction device 100 (e.g., the predicted second set of experimental condition parameter values), then the prediction device 100 performs another prediction. The prediction device 100 can input the first set of experimental condition parameter values ​​from the parameter value set into the crystal precipitation device 1220.

[0084] In the first utilization example, the utilizing entity 1200 provides an unknown substance to the service providing entity 1250 and entrusts the crystal form screening of the unknown substance. The service providing entity 1250 accepts the crystal form screening entrustment, uses the crystal making auxiliary system 1260 and the prediction device 100 to perform crystal form screening of the unknown substance, and provides the information as the result to the utilizing entity 1200.

[0085] Figure 8 This is a diagram illustrating a second example of the application of the prediction device.

[0086] Entity 1200 has a crystal fabrication auxiliary system 1260. Entity 1200 inputs information (e.g., a set of parameter values) to a remote prediction device 100 via a communication network such as the Internet, for example, using an input / output console. In response to the input information, it obtains output information (prediction results) from the prediction device 100. Based on the output information, crystal precipitation and evaluation are performed using the crystal fabrication auxiliary system 1260.

[0087] As described above, all the constituent elements of the crystal manufacturing assistance system 1260 may be present in one of the utilizing entity 1200 and the service providing entity 1250, or a portion of the elements of the crystal manufacturing assistance system 1260 may be present in the utilizing entity 1200, and another portion of the elements of the crystal manufacturing assistance system 1260 (e.g., the remaining elements) may be present in the service providing entity 1250.

[0088] In addition, the aforementioned user 10 may belong to entity 1200, service provider 1250, or other entities besides entity 1200 and entity 1250.

[0089] [Second Implementation] The second embodiment will be described below. The main focus will be on the differences from the first embodiment; commonalities with the first embodiment may be omitted or simplified in the description.

[0090] Figure 9 This is a diagram showing the configuration of the prediction device according to the second embodiment. Figure 10 This is a schematic diagram of the precipitation prediction involved in the second embodiment.

[0091] The precipitation prediction unit 133 includes an evaluation unit 135. The evaluation unit 135 records one or more contribution rate sets calculated using the crystal precipitation model 132 into the processing result database 1110. The crystal precipitation model 132 may also not include the equation network as described in the first embodiment. Alternatively, a contribution rate set may be prepared for each elapsed time.

[0092] For example, Figure 10As shown, the contribution rate set consists of the contribution rates of multiple factors affecting crystal precipitation (e.g., concentration factor, temperature factor, infrared factor, others). For each factor, the contribution rate is the degree to which that factor contributes to the overall contribution of the multiple factors. For example, the sum of the multiple contribution rates in the contribution rate set is a given value (e.g., "1"). The contribution rates of each factor may differ depending on the elapsed time. Figure 10 In the example shown, a set of contribution rates consists of a contribution rate of "0.90" for the concentration factor, a contribution rate of "0.00" for the temperature factor, a contribution rate of "0.00" for the infrared factor, and a contribution rate of "0.10" for other factors. The contribution rate can be the input-output value between factors (or between equations) (or a value obtained based on that value).

[0093] For each experimental condition (set of experimental condition parameter values), for each available crystal form, the degree of influence of each factor on that crystal form is determined for each of the multiple factors. The combination of the contribution rate set and the factor degree for each experimental condition and each factor for each crystal form is, for example, recorded in the processing results database 1110. For each crystal form, the influence degree of each factor can be referred to as the weighted average of that factor for that crystal form. For example, for experimental condition 1, the influence degree of the concentration factor for crystal form 1 is "1.00", and the influence degree of the infrared factor for crystal form 1 is "0.20".

[0094] Evaluation unit 135 calculates a score for each crystal form under each experimental condition, based on the influence and contribution rate of each factor. Precipitation prediction unit 133 predicts the crystal form to be precipitated under each experimental condition based on the scores of each crystal form. For example, evaluation unit 135 calculates a value (e.g., product) based on the influence and contribution rate for each factor for each crystal form, and calculates a score for that crystal form (e.g., the sum of multiple products) based on multiple values ​​calculated for multiple factors. Evaluation unit 135 predicts that a crystal form with a high precipitation score will precipitate. The crystal form predicted to precipitate can be the crystal form with the highest score, or a crystal form that has obtained a score higher than a predetermined score. Evaluation unit 135 makes this prediction for each experimental condition, i.e., predicts the crystal form to precipitate.

[0095] Evaluation unit 135 investigates the consistency (to what extent they match) between the predicted results (predicted crystal form) and the actual experimental results (crystal form precipitated under those experimental conditions) for each experimental condition, i.e., the appropriateness of the contribution rate set. Evaluation unit 135 calculates the consistency between the predicted results and the experimental results for each contribution rate set.

[0096] The output unit 140 can also display the output information of the contribution rate set representing the optimal consistency and the crystal form predicted using the contribution rate set for each experimental condition on the input / output console 80.

[0097] The above description describes several embodiments, but these are merely illustrative examples of the invention and are not intended to limit the scope of the invention to these embodiments. The invention can also be implemented in various other ways. For example, in the above description, the database is an example of data that yields output from input (e.g., key quantities), which can be arbitrarily constructed data (e.g., structured or unstructured data).

[0098] For example, the above explanation can be summarized as follows. The following summary may include supplementary explanations and descriptions of variations of the above explanation.

[0099] As an example of a computer, the prediction device performs the following (A) to (C). (A) can be performed by the input unit 120, (B) can be performed by the prediction extraction unit 133, and (C) can be performed by the output unit 140.

[0100] (A) Obtain a set of parameter values ​​containing the values ​​of at least one of a plurality of factors (e.g., a plurality of analytical indicators) that are respectively classified as factors affecting crystal formation.

[0101] (B) By inputting the obtained set of parameter values ​​into a crystal formation model, which represents the crystal formation process and includes time as an element, predictions are made related to the formation of the crystal form that can be obtained from the material.

[0102] (C) Output information related to the prediction results, i.e., prediction result information.

[0103] Therefore, predictions related to the macroscopic formation of crystals over time can be made with a smaller computational load than calculations using first-principles methods.

[0104] The parameter set may include one or more first experimental condition parameter values ​​and one or more fitted parameter values. The experimental condition parameter values ​​may be values ​​related to the experimental conditions. The one or more fitted parameter values ​​may include at least one parameter value selected from parameters related to the solubility curve, i.e., solubility parameter values ​​(an example of fitted parameter values ​​related to the temperature / saturation concentration relationship); interfacial energy values, i.e., interfacial energy values; and the number of crystal forms. The predicted result may include one or more second experimental condition parameter values ​​(one or more experimental condition parameter values ​​at the end of crystal formation) obtained from the crystal formation model by inputting the parameter set into the crystal formation model. The prediction device may perform a first processing. In the first processing, the prediction device (e.g., the fitting parameter optimization unit 134) may learn one or more fitted parameter values ​​that make the predicted one or more second experimental condition parameter values ​​for the one or more first experimental condition parameter values ​​included in the parameter set close to or overlap with one or more second experimental condition parameter values ​​obtained experimentally for the one or more first experimental condition parameter values. Therefore, an improvement in prediction accuracy is expected.

[0105] Specifically, for example, the prediction device can perform processing including the following (a) to (d), namely object processing.

[0106] (a) The prediction device (e.g., a precipitation prediction unit) outputs a prediction result by inputting the acquired set of parameter values ​​into a crystal formation model. The set of parameter values ​​includes a first set of experimental condition parameter values ​​and a set of fitted parameter values. The first set of experimental condition parameter values ​​is one or more first experimental condition parameter values. The set of fitted parameter values ​​is one or more fitted parameter values. The prediction result includes a predicted set of second experimental condition parameter values. The second set of experimental condition parameter values ​​is one or more second experimental condition parameter values.

[0107] (b) The prediction device (e.g., a precipitation prediction unit) determines, for the first set of experimental condition parameter values, whether the relationship between the predicted second set of experimental condition parameter values ​​in (a) and the actual (typically obtained through experiment) second set of experimental condition parameter values ​​satisfies a given condition. "The relationship satisfies the given condition" may mean, for example, that the difference between the predicted second set of experimental condition parameter values ​​(e.g., its characteristic quantities) and the actual second set of experimental condition parameter values ​​(e.g., its characteristic quantities) is less than a certain amount.

[0108] (c) If the determination result in (b) is false (i.e., the difference exceeds a certain amount), the prediction device (e.g., the fitting parameter optimization unit) changes at least a portion of the fitting parameter values ​​in the first experimental condition parameter value set. The prediction device (e.g., the fitting parameter optimization unit) can determine the amount of change in the fitting parameter values ​​based on the magnitude of the difference obtained in (b), and change the fitting parameter values ​​by the determined amount of change.

[0109] (d) The prediction device obtains a set of parameter values ​​that includes the first set of experimental condition parameter values ​​and the modified set of fitted parameter values ​​in (c), and performs the processing in (a) (i.e., the processing flow returns to (a)).

[0110] It should be noted that the prediction device (e.g., the precipitation prediction unit) can terminate the object processing if the determination result in (b) is true (e.g., the difference between the sets of second experimental condition parameter values ​​is less than a certain amount (e.g., the difference is zero)). Furthermore, the prediction device (e.g., the precipitation prediction unit) can accumulate data as a set of data containing data representing the structural features of the substance (e.g., functional group patterns) and the fitted parameter value set at the end of the object processing for that substance into a storage device as a learning result. The object processing can be included in either the first processing or the second processing. In the object processing of the first processing, the substance can be a first substance (e.g., a known substance), and in the object processing of the second processing, the substance can be a second substance (e.g., an unknown substance).

[0111] It should be noted that when one or more fitted parameter values ​​include a solubility parameter value, it is expected that an appropriate value can be estimated as the solubility parameter value, which is one of the parameter values ​​for which an appropriate value is difficult to determine experimentally or theoretically. Therefore, improved prediction accuracy can be expected. Furthermore, when one or more fitted parameter values ​​include an interfacial energy value, it is expected that an appropriate value can be estimated as the interfacial energy value, which is one of the parameter values ​​for which an appropriate value is difficult to determine experimentally or theoretically. Therefore, improved prediction accuracy can be expected. Additionally, when one or more fitted parameter values ​​include a value for the number of crystal forms, improved prediction accuracy for the number of crystal forms available for the substance can be expected.

[0112] In addition to the first processing, the prediction device can also perform a second processing. The first processing may be processing one or more fitted parameter values ​​contained in a set of learning parameter values ​​for multiple first substances. The second processing may be processing estimating one or more fitted parameter values ​​for a second substance based on the learning results of one or more fitted parameter values. An example of the first substance may be a known substance, and an example of the second substance may be an unknown substance. An example of the learning results may be a processing result database 110. At least one of the first and second processing may include (A) to (C). For each experiment, there may be a pair of first experimental condition parameter value sets and second experimental condition parameter value sets, that is, the first experimental condition parameter value set and the second experimental condition parameter value set may correspond 1:1. Since the learning results of one or more fitted parameter values ​​are used to determine one or more fitted parameter values ​​for the second substance, it is expected that appropriate one or more fitted parameter values ​​will be obtained for the second substance, and therefore, an improvement in the prediction accuracy for the second substance is expected.

[0113] An example of a crystal formation model is crystal precipitation model 132. A crystal formation model can include multiple model components. These multiple model components can consist of multiple time-evolution components and one or more non-time-evolution components. The time-evolution components correspond to a subset of factors and each incorporates elapsed time as a factor. The one or more non-time-evolution components correspond to another subset of factors and each does not incorporate elapsed time as a factor. The one or more non-time-evolution components can include a temperature / saturation concentration relationship component, which is a model component used to define the relationship between temperature and saturation concentration, i.e., the temperature / saturation concentration relationship, and includes fitting parameters related to the temperature / saturation concentration relationship. The multiple time-evolution components can include a model component that takes as input the output value from the temperature / saturation concentration relationship component, which has inputted fitting parameter values ​​related to the temperature / saturation concentration relationship. Specifically, for example, a crystal formation model can include an equation network, where the multiple model components can be multiple equations constituting the equation network, and the multiple time-evolution components can be multiple time-evolution equations. Thus, by setting the multiple model components included in the crystal generation model as one or more non-time evolution components including a temperature / saturation concentration relationship component and multiple time evolution components including a reference temperature / saturation concentration relationship component, an improvement in prediction accuracy can be expected.

[0114] The learning results of one or more fitted parameter values ​​can include a relationship between functional group patterns and the values ​​or ranges of each of the one or more fitted parameters. Functional group patterns can include at least one of the following: the number of functional groups, the types of functional groups, the positions of the functional groups, and the bonds between the functional groups. Since the functional group patterns of a substance influence the crystal forms that can be obtained from that substance, it is expected that appropriate fitted parameter values ​​can be obtained based on the above learning results, thus improving prediction accuracy. It should be noted that the "relationship" between functional group patterns and the values ​​or ranges of each of the one or more fitted parameters can be implemented as a dataset such as a database, or in other forms.

[0115] Specifically, for example, the learning results may include a set of fitted parameter values ​​for each functional group pattern or set of functional group patterns (one or more functional group patterns possessed by the first substance) regarding that functional group pattern or set of functional group patterns (e.g., solubility parameter values ​​and interfacial energy values ​​for each crystal form among the number of crystal forms corresponding to the value of the number of crystal forms). It is expected that, based on one or more functional group patterns of the second substance and the learning results, the value of the number of crystal forms, the solubility parameter values ​​and interfacial energy values ​​for each crystal form among the number of crystal forms corresponding to the value of the number of crystal forms can be estimated.

[0116] Crystal formation models can incorporate associations with functional group influencing factors (factors that affect functional groups). These factors are those that cause hydrogen bonding or dissociation between functional groups, specifically infrared radiation, as mentioned above. Depending on the functional group pattern, the wavelength regions where infrared radiation is significantly absorbed differ; therefore, it is hoped that controlling the wavelength of the irradiated infrared radiation can lead to the formation of more crystal forms. That is, it may be possible to obtain crystal forms that have not been discovered previously. Furthermore, since the crystal formation model represents the crystal formation process using functional group influencing factors such as infrared radiation, it is expected to reduce the possibility of omissions in the predicted crystal forms.

[0117] In the second process, the prediction device can estimate the number of crystalline forms obtainable by the second substance based on learning results (e.g., accumulated learning results) and one or more functional group patterns of the second substance. For each of the estimated number of crystalline forms for the second substance, the prediction device can learn one or more fitted parameter values ​​that make the predicted values ​​of one or more second experimental condition parameters for the second substance close to or overlap with one or more second experimental condition parameter values ​​obtained through experiments on the second substance. Thus, optimization of the fitted parameter values ​​for the second substance can be expected.

[0118] In the second processing step, the prediction device can, for the second substance, incorporate the relationship between one or more functional group patterns and the values ​​or ranges of values ​​explored for each of the one or more fitting parameters into the learning results. This further enriches the learning results, and improved accuracy of the prediction results in subsequent second processing steps can be expected.

[0119] In the first and / or second processing, the prediction device may, based on the learning results and one or more functional group patterns of the substance, estimate not only the number of crystalline forms that can be obtained for the substance, but also an exploration range for each of the one or more fitting parameters. For each of the estimated number of crystalline forms for the substance, a fitting parameter value is explored from the estimated exploration range for each of the one or more fitting parameters. Because this narrows the exploration range, a reduction in the computational load associated with exploring the fitting parameter values ​​can be expected.

[0120] The prediction results information may include at least one of information representing the ratio of each crystalline form of the substance that is available and information representing the temperature / saturation concentration relationship for each crystalline form of the substance that is available. Therefore, the prediction results can be easily understood based on the displayed prediction results information.

[0121] In (B), the prediction device can perform the following operations. As a result, an improvement in prediction accuracy can be expected.

[0122] • Define one or more sets of contribution rates consisting of the contribution rate of each factor (e.g., for each factor specified for the crystal formation model).

[0123] • Based on the influence of each factor determined for each crystal form for each experimental condition (set of experimental condition parameter values) and the contribution rate of each factor in the contribution rate set for each elapsed time, the score of each crystal form is calculated for each experimental condition.

[0124] • For each contribution rate set, based on the fraction of each crystal form under each experimental condition, predict the crystal form that will precipitate under each experimental condition.

[0125] • For each contribution rate set, calculate the consistency between the prediction results obtained using that contribution rate set for each experimental condition and the actual experimental results for each experimental condition.

[0126] For example, for each experimental condition parameter (e.g., each factor (analytical indicator)), there can exist experimental condition parameter values ​​for each of multiple time points. For instance, a first experimental condition parameter value could be an example of an experimental condition parameter value at a first time point. A second experimental condition parameter value could be an example of an experimental condition parameter value at a second time point. The term "multiple time points" can include two or more time points in the time elapsed during calculations (e.g., simulations) using a crystal formation model, or two or more time points in actual elapsed time (e.g., elapsed time in an actual experiment). An example of a second time point could be the time point of crystal precipitation. An example of an experimental condition parameter value at a first time point could be an experimental condition parameter value input into the crystal formation model, such as a first experimental condition parameter value. An example of an experimental condition parameter value at a second time point could be an experimental condition parameter value output from the crystal formation model, such as a second experimental condition parameter value. Fitting parameter values ​​can be learned so that one or more predicted experimental condition parameter values ​​at a second time point are close to or overlap with one or more actual (e.g., experimentally obtained) experimental condition parameter values ​​at the second time point. For example, the fitting parameter values ​​can be learned so that the feature values ​​of one or more experimental condition parameter values ​​predicted at the second time point are close to or overlap with the feature values ​​of one or more actual experimental condition parameter values ​​at the second time point.

[0127] Furthermore, based on the above explanation, a method for optimizing one or more fitted parameter values ​​can be described as follows. It should be noted that, in the following description, "one or more fitted parameter values" can, as described above, include at least one of the values ​​for solubility parameter, interfacial energy, and the number of crystal forms. The one or more fitted parameter values ​​can include both a solubility parameter value and an interfacial energy value for each of the number of crystal forms corresponding to the value of the number of crystal forms. Additionally, an example of the "prediction model" described below can be a crystal formation model. The prediction model can be a model that includes time as a factor. At least a portion of the prediction model can also be an approximate model. Furthermore, in learning the fitted parameter values, a model for fitting the parameter values ​​(e.g., a model similar to fitted model 136) can be used.

[0128] <Example of a method for optimizing the fitted parameter values> An optimization method for fitting parameter values ​​involves optimizing one or more fitting parameter values ​​from the set of parameter values ​​input into a prediction model that influence the prediction results obtained using the prediction model. For a given set of one or more parameter items, learn one or more fitted parameter values ​​that make the predicted values ​​of one or more parameters in the predicted results obtained using the prediction model close to or overlap with the actual values ​​of one or more parameters.

[0129] Explanation of reference numerals in the attached figures 100…Prediction device.

Claims

1. A prediction method, characterized in that, The computer performs the following operations: (A) Obtain a set of parameter values ​​containing the values ​​of multiple parameters, wherein the values ​​of the multiple parameters are respectively classified as at least one of multiple factors affecting the formation of crystal; (B) By inputting the set of parameter values ​​into a model representing the crystal formation process and including time as a factor, i.e., a crystal formation model, predictions related to the formation of the crystal forms that can be obtained from the material are made. as well as (C) Output information related to the prediction result, i.e., prediction result information.

2. The prediction method according to claim 1, wherein, The parameter value set includes one or more first experimental condition parameter values ​​and one or more fitting parameter values. Experimental condition parameter values ​​are values ​​related to experimental conditions. The one or more fitting parameter values ​​include at least one parameter value selected from the following: a parameter value related to the solubility curve, i.e., a solubility parameter value; an interfacial energy value, i.e., an interfacial energy value; and a value indicating the number of crystal forms. The predicted result includes one or more second experimental condition parameter values, which are one or more experimental condition parameter values ​​obtained from the crystal generation model at the end of crystal generation by inputting the parameter value set into the crystal generation model. The computer performs a first process, which is a process of learning to make one or more second experimental condition parameter values ​​predicted for one or more first experimental condition parameter values ​​contained in the parameter value set close to or overlap with one or more second experimental condition parameter values ​​obtained experimentally for the one or more first experimental condition parameter values.

3. The prediction method according to claim 2, wherein, The first process involves learning one or more fitted parameter values ​​from the parameter value set for multiple first substances. The computer performs a second process, which involves estimating one or more fitted parameter values ​​for the second substance based on the learning results of the one or more fitted parameter values. At least one of the first process and the second process includes (A) to (C).

4. The prediction method according to claim 2, wherein, The crystal generation model includes multiple model components. The multiple model components consist of multiple time-evolutionary components and one or more non-time-evolutionary components. The plurality of time evolution components correspond to a subset of the plurality of factors and each uses elapsed time as its element. The one or more non-temporal evolution components correspond to another subset of the multiple factors and do not use elapsed time as a factor. The one or more non-time-evolutionary components include a temperature / saturation concentration relationship component, which is a model component used to define the relationship between temperature and saturation concentration, i.e., the temperature / saturation concentration relationship, and includes fitting parameters related to the temperature / saturation concentration relationship. The plurality of time evolution components include: a model component that takes as input the output of a temperature / saturation concentration relationship component that takes as input the fitted parameter values ​​related to the temperature / saturation concentration relationship.

5. The prediction method according to claim 4, wherein, The crystal formation model includes a network of equations. The multiple model components are multiple equations that constitute the equation network. The multiple time evolution components are multiple time evolution equations.

6. The prediction method according to claim 3, wherein, The learning results of the one or more fitted parameter values ​​include functional group patterns and relationships between values ​​or ranges of values ​​associated with each of the one or more fitted parameters. The functional group pattern includes at least one of the following: the number of functional groups, the type of functional groups, the position of functional groups, and the bonds between functional groups.

7. The prediction method according to claim 6, wherein, The crystal formation model includes the correlation with factors that influence functional groups, i.e., the factors that influence functional groups.

8. The prediction method according to claim 7, wherein, The functional group influencing factor is infrared radiation.

9. The prediction method according to claim 6, wherein, In the second process, Based on the learning results and one or more functional group patterns of the second substance, the estimated number of achievable crystal forms of the second substance is determined. For each crystal form in the estimated quantity of the second substance, learn one or more fitted parameter values ​​that make one or more predicted values ​​of second experimental condition parameters for the second substance close to or overlap with one or more values ​​of second experimental condition parameters obtained in experiments for the second substance.

10. The prediction method according to claim 9, wherein, In the second processing, for the second substance, the relationship between one or more functional group patterns and the values ​​or ranges of values ​​explored for each of the one or more fitting parameters is reflected in the learning results.

11. The prediction method according to claim 6, wherein, In the first process and / or the second process Based on the learning results and one or more functional group patterns of the substance, in addition to estimating the number of achievable crystal forms of the substance, the exploration range for each of the one or more fitted parameter values ​​is also estimated. For each crystal form of the substance, for each of the one or more fitted parameter values, explore fitted parameter values ​​from the estimated range of exploration.

12. The prediction method according to claim 1, wherein, The prediction result information includes at least one of information representing the ratio of each crystal form that can be obtained from the substance and information representing the relationship between temperature and saturation concentration for each crystal form that can be obtained from the substance, i.e., the temperature / saturation concentration relationship.

13. The prediction method according to claim 1, wherein, In (B), Define one or more sets of contribution rates consisting of the contribution rate of each factor specified for the crystal formation model. Based on the influence of each factor determined for each crystal form and the contribution rate of each factor in each time-lapsed contribution rate set, which can be obtained for each set of experimental condition parameter values ​​used for crystal formation (more than one set of experimental condition parameter values), a score for each crystal form is calculated for each set of experimental condition parameter values. For each contribution rate set, based on the score of each crystal form in each set of experimental condition parameter values, the crystal form to be precipitated is predicted for each set of experimental condition parameter values. For each contribution rate set, the consistency between the prediction results obtained using that contribution rate set for each set of experimental condition parameter values ​​and the actual experimental results for each set of experimental condition parameter values ​​is calculated.

14. A prediction device, characterized in that, have: The input unit acquires a set of parameter values, which includes the values ​​of multiple parameters, the values ​​of which are respectively classified into at least one of multiple factors that affect the formation of the crystal; The prediction unit, wherein the set of parameter values ​​is input into a crystal formation model, representing the crystal formation process and including time as a factor, performs predictions related to the formation of the available crystal forms of the material; and The output section outputs information related to the prediction result, i.e., prediction result information.

15. A recording medium, characterized in that, The record contains computer programs that cause the computer to execute the following (A) to (C): (A) Obtain a set of parameter values ​​containing the values ​​of multiple parameters, wherein the values ​​of the multiple parameters are respectively classified as at least one of multiple factors affecting crystal formation; (B) By inputting the set of parameter values ​​into a model representing the crystal formation process and including time as a factor, i.e., a crystal formation model, predictions related to the formation of the crystal forms that can be obtained from the material are made. as well as (C) Output information related to the prediction result, i.e., prediction result information.

16. A computer program, characterized in that, Make the computer perform the following (A) to (C): (A) Obtain a set of parameter values ​​containing the values ​​of multiple parameters, wherein the values ​​of the multiple parameters are respectively classified as at least one of multiple factors affecting crystal formation; (B) By inputting the set of parameter values ​​into a model representing the crystal formation process and including time as a factor, i.e., a crystal formation model, predictions related to the formation of the crystal forms that can be obtained from the material are made. as well as (C) Output information related to the prediction result, i.e., prediction result information.

Images