Learning device, learning model, screening device, screening program, and method for screening

Abstract

To predict the adsorptivity of an adsorbate on an adsorbent without directly determining the adsorption energy of the adsorbate on the adsorbent's surface.SOLUTION: The learning device according to the present invention has a learning unit for generating a learned model by learning with use of a learning dataset in which a descriptor with the molecular structure of an adsorbate to be adsorbed to an adsorbent and an interaction index Iint of a noted inter-molecule coupling with the adsorbent of the adsorbate are related to each other.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a learning device, a learning model, a screening device, a screening program, and a screening method. [Background technology]

[0002] Attempts have been made to predict predetermined performance, such as the adsorption of an adsorbate (adsorbent material) of a molecule or molecular crystal to an adsorbent made of a solid such as a molecular crystal, metal, or ceramic, by utilizing computational chemistry such as quantum chemical calculations. By using such a method for predicting the performance of an adsorbate, it becomes possible to evaluate the predetermined performance of the adsorbate without actually producing and testing the adsorbate, thereby reducing the cost and time required for discovering, designing, and producing an adsorbate that satisfies the predetermined performance.

[0003] As a method for predicting the performance of an adsorbate using computational chemistry, for example, a computer device has been proposed that models the interaction between the adsorbate and a metal surface and combines this model with quantum chemical calculations for the adsorbate to determine the physical properties of the adsorbate itself, such as its energy (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 4224577 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the computer device of Patent Document 1, when comprehensively screening adsorbates that adsorb to the surface of a specific adsorbent from among a large number of substances using the adsorption energy as an index to select an adsorbate having a predetermined adsorptivity, even calculating the adsorption energy imposes a large calculation load.When using quantum chemical calculations such as first-principles calculations, the process of setting calculation conditions, examining the orientation of the adsorbate relative to the adsorbent, calculating the adsorption energy, and calculating values ​​that serve as an index of the strength of adsorption takes an extremely long time, and therefore there is a problem in that it is not realistic to calculate all adsorbates that are adsorbed to the adsorbent.

[0006] An object of the present invention is to predict the adsorptivity of an adsorbate to an adsorbent without directly determining the adsorption energy of the adsorbate adsorbed on the surface of the adsorbent. [Means for solving the problem]

[0007] One aspect of the present invention is A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of the adsorbate with the adsorbent of interest int The learning device has a learning unit that generates a trained model by learning using a learning dataset in which the and are associated with each other.

[0008] Another aspect of the present invention is A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of the adsorbate with the adsorbent of interest int The learning model is generated using a learning dataset in which the and are associated with each other.

[0009] Another aspect of the present invention is an acquisition unit that acquires a descriptor of a molecular structure of an adsorbent to be predicted and an adsorbate to be adsorbed by the adsorbent to be predicted; A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of the adsorbate with the adsorbent of interest intUsing a trained model generated using a training data set in which the and are associated with each other, the interaction index I of the intermolecular bond of interest between the adsorbate to be predicted and the adsorbent to be predicted is calculated. int A prediction unit for predicting The screening device includes:

[0010] Another aspect of the present invention is On the computer, an acquisition step of acquiring a descriptor of the molecular structure of a prediction target adsorbent and a prediction target adsorbate to be adsorbed by the adsorbent; A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of the adsorbate with the adsorbent of interest int Using a trained model generated using a training data set in which the and are associated with each other, the interaction index I of the intermolecular bond of interest between the adsorbate to be predicted and the adsorbent to be predicted is calculated. int A prediction step of predicting This is a screening program that executes the above.

[0011] Another aspect of the present invention is The computer an acquisition step of acquiring a descriptor of the molecular structure of a prediction target adsorbent and a prediction target adsorbate to be adsorbed by the adsorbent; A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of the adsorbate with the adsorbent of interest int Using a trained model generated using a training data set in which the and are associated with each other, the interaction index I of the intermolecular bond of interest between the adsorbate to be predicted and the adsorbent to be predicted is calculated. int A prediction step of predicting This is a screening method for carrying out the above. Effect of the Invention

[0012] The present invention can predict the adsorptivity of an adsorbate to an adsorbent without directly determining the adsorption energy of the adsorbate adsorbed on the surface of the adsorbent. [Brief description of the drawings]

[0013] [Figure 1] FIG. 2 is an explanatory diagram showing a state in which an adsorbate is adsorbed onto an adsorbent. [Diagram 2] FIG. 2 is an explanatory diagram showing the magnitude of the bond distance between a molecular crystal and a molecule. [Diagram 3] 1 is a block diagram showing a schematic configuration of a learning device according to an embodiment of the present invention. [Figure 4] FIG. 13 is a diagram showing an example of the correlation between the interaction index Iint and the absolute value of the adsorption energy Eads. [Diagram 5] 1 is a diagram showing a configuration of a screening system according to an embodiment of the present invention. [Figure 6] 1 is a block diagram showing a schematic configuration of a screening device according to a first embodiment. [Figure 7] FIG. 2 is a block diagram showing a schematic configuration of a first energy calculation unit. [Figure 8] FIG. 4 is a block diagram showing a schematic configuration of a second energy calculation unit. [Figure 9] FIG. 13 is a diagram illustrating an example of clustering. [Figure 10] FIG. 11 is a block diagram showing a schematic configuration of a screening device according to a second embodiment. [Figure 11] FIG. 2 is a block diagram showing the hardware configuration of a learning device and a screening device. [Figure 12] 1 is a flowchart illustrating a learning method according to an embodiment of the present invention. [Figure 13] 4 is a flowchart showing an example of a screening method according to the first embodiment. [Figure 14] 11 is a flowchart showing an example of a first energy calculation step. [Figure 15] 13 is a flowchart showing an example of a second energy calculation step. [Figure 16] 13 is a flowchart showing an example of a screening method according to a second embodiment. [Figure 17]FIG. 1 is a correlation diagram between the interaction index Iint and the absolute value of the adsorption energy Eads in Example 1. [Figure 18] FIG. 1 is a plot of the average values ​​of the interaction index Iint and the absolute value of the adsorption energy Eads for each cluster in Example 1. [Figure 19] FIG. 1 shows a group of molecules in one cluster. [Figure 20] FIG. 13 is a diagram showing the relationship between the calculated and predicted values ​​of the interaction index Iint in Example 2. [Figure 21] FIG. 1 shows the relationship between the molecules of Example 2 and the predicted value of the interaction index Iint. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] Hereinafter, an embodiment of the present invention will be described in detail. In order to facilitate understanding of the description, the same components in each drawing are given the same reference numerals, and duplicated descriptions will be omitted. In addition, in this specification, unless otherwise specified, "to" indicating a numerical range means that the numerical values ​​before and after it are included as the lower limit and upper limit.

[0015] <Learning device> The learning device according to the present embodiment will be described. The learning device according to the present embodiment uses the descriptor of the molecular structure of the adsorbate to be adsorbed by the adsorbent as an explanatory variable, and calculates the molecular interaction index I of the intermolecular bond of interest between the adsorbate and the adsorbent. int (Hereinafter, simply referred to as the "Interaction Index I int ") as the objective variable, and the interaction index I int Generate a trained model to predict

[0016] In this embodiment, the interaction index I int is an index that represents the intermolecular bond of interest between the adsorbate and the adsorbent, and can be calculated, for example, by the following formula (1).

[0017]

number

[0018] In this embodiment, the adsorbent is a substance to which the adsorbate is adsorbed, and examples of the adsorbent include solid substances such as molecular crystals, metals, and ceramics, as shown in FIG. 1. The adsorbate is a substance (adsorbed substance) that is adsorbed by the adsorbent, and examples of the adsorbate include molecules and molecular crystals, as shown in FIG. 1. One of the adsorbent and the adsorbate may be a molecular crystal. The molecule (adsorbed molecule) that is adsorbed by the adsorbent may be a single molecule, a polymer containing a molecule, or a compound having the polymer. In this embodiment, a case will be described in which the adsorbent is a molecular crystal and the adsorbate is a molecule (hereinafter simply referred to as a molecule) that is adsorbed by the molecular crystal.

[0019] A molecular crystal is generally a substance in which molecules are solidified by agglomeration, forming a periodic crystal structure. Molecular crystals differ from general crystals in the type of interaction. General crystals are mainly formed by interactions between atoms, such as covalent bonds, ionic bonds, and metallic bonds. General crystals include diamond, sodium chloride, and iron. Molecular crystals are similar to general crystals in that the main interactions within each molecule that constitutes a molecular crystal are covalent bonds, but they differ in that they are formed by relatively weak interactions between molecules, such as van der Waals interactions. In addition to being formed by relatively weak interactions, molecular crystals also include hydrogen bonds, which are non-covalent attractive interactions formed between a hydrogen atom covalently bonded to an atom with a high electronegativity and a lone pair of nitrogen, oxygen, sulfur, fluorine, etc. located nearby, or interactions between π electrons of aromatic rings, etc. and other functional groups, etc. Examples of molecular crystals include carbon dioxide (CO2) which constitutes dry ice, molecular iodine (I2), protein crystals, and thickeners which constitute grease used for machine lubrication.

[0020] There are many phenomena in which molecules constituting other substances are adsorbed to the surface of molecular crystals as shown in Figure 1(a), or in which molecular crystals themselves are adsorbed to the surface of other substances as shown in Figure 1(b). In the medical and industrial fields, adsorption phenomena involving molecules and substances such as molecular crystals are important phenomena. In the medical field, for example, understanding the phenomenon in which water contained in a living body, or other biomolecules or molecules of pharmaceutical drugs are adsorbed to the surface of proteins constituting a living body is an important element in biology and pharmacology. The phenomenon in which proteins themselves are adsorbed to the surfaces of materials such as ceramics and polymers is important when designing the material itself of biomaterials that constitute artificial joints, or the lubrication performance of that material. In the industrial field, thickeners that constitute greases used for lubrication of machines may have a crystalline structure, which affects the lubrication state, and is therefore important when designing the lubrication performance of greases and understanding the lubrication state of equipment.

[0021] The surface of such molecular crystals does not have a large surface energy, as, for example, a metal surface, and microscopically, energetically active atoms that can be the center of adsorption are not exposed. As with the explanation of the intermolecular interactions that compose molecular crystals above, the adsorption phenomenon occurs on the surface of molecular crystals as a result of interactions with other substances, mainly through van der Waals interactions and hydrogen bonds.

[0022] In general, the adsorptivity of other molecules to the surface of a molecular crystal, as shown in Figure 1(a), can be determined by estimating the magnitude relationship and tendency from the molecular structure, such as the type of functional group, chain length, and morphology, possessed by the molecules that compose the molecular crystal or the molecules that are adsorbed. The adsorptivity of a molecular crystal to the surface of a substance, as shown in Figure 1(b), can be determined in a similar manner.

[0023] In this embodiment, the trained model M1 generated by the training device 1 calculates an interaction index I int The interaction index I int is correlated with the magnitude of the adsorption energy due to intermolecular bonds (e.g., hydrogen bonds) between a molecule and a molecular crystal, and can be used as an index showing the degree of adsorption of a molecule to the surface of a molecular crystal. int can be obtained directly from the adsorption structure of the molecular crystal and the molecule, which is generated by molecular simulation, etc. int As mentioned above, is an index that correlates with the magnitude of the adsorption energy. Therefore, the interaction index I int This makes it possible to quantitatively and visually predict the type and strength of interaction between molecular crystals and molecules in the phenomenon of adsorption between molecular crystals and molecules, thereby making it possible to predict the adsorptivity of molecules adsorbed onto molecular crystals and easily compare the adsorptivity of molecules adsorbed onto molecular crystals.

[0024] In addition, the interaction index I int is correlated with the molecular bond distance required for adsorption between the molecular crystal and the molecule. intSince can be expressed as a function that becomes smaller as the distance of the molecular bond involved in adsorption becomes longer, it can be regarded as an index showing the strength of the hydrogen bond involved in the adsorption of the molecule of interest. When a molecule is adsorbed onto the surface of a molecular crystal, the distance of the molecular bond can be found by determining one atom on the surface of the molecular crystal and one atom of the molecule as representative coordinates. For example, as shown in Figure 2, when the molecular bond is a hydrogen bond, the hydrogen bond distance R i is calculated as the distance between the hydrogen attached to the more electronegative element in the molecular crystal and the more electronegative element in the adsorbed molecule (see Figure 2(a)). The hydrogen bond distance R i is calculated as the distance between the hydrogen attached to the more electronegative element in the adsorbed molecule and the more electronegative element in the molecular crystal (see Figure 2(b)).

[0025] In this embodiment, the interaction index I output by the trained model M1 int can be used as an index showing the distance of the molecular bond involved in the adsorption between the molecular crystal and the molecule, i.e., the strength of the molecular bond. Therefore, for each molecular crystal, the interaction index I int By outputting this, the degree of adsorption of the molecule to the surface of the molecular crystal can be understood.

[0026] The element having a high electronegativity is not particularly limited, but examples thereof include oxygen, nitrogen, sulfur, fluorine, chlorine, bromine, and iodine.

[0027] In addition, when attention is focused on the interaction between the π electrons of an aromatic ring or the like and other functional groups or the like instead of the distance of the molecular bond involved in the adsorption, the distance between the center of gravity of the aromatic ring as the representative coordinate and the other functional groups or the like may be obtained as the molecular bond distance.

[0028] Fig. 3 is a block diagram showing a schematic configuration of a learning device according to this embodiment. As shown in Fig. 3, the learning device 1 includes a first acquisition unit 11, a second acquisition unit 12, a third acquisition unit 13, a learning data set creation unit 14, a learning unit 15, and an output unit 16. The learning device 1 calculates an interaction index I intGenerate a trained model M1 that predicts

[0029] The first acquisition unit 11 acquires the bulk structure of the molecular crystal.

[0030] Information on the bulk structure of the molecular crystal, etc. may be acquired from the first storage unit 111.

[0031] The first storage unit 111 stores a data table including the bulk structure of a molecular crystal, etc. The data table is not particularly limited as long as it is a database including the bulk structure of a molecular crystal, etc., and may be a general database, for example, the Crystallography Open Database, etc.

[0032] The second acquiring unit 12 acquires a descriptor of the molecular structure of the molecule as an explanatory variable.

[0033] The information of the descriptor of the molecular structure of the molecule may be obtained from the second storage unit 112.

[0034] The second storage unit 112 stores a data table including data related to molecules, such as the molecular structure of the molecule. The data table is not particularly limited as long as it is a database including data related to molecules, and may be a general database, such as Pubchem, PubChemQC, ZINC, ChemSpider, Chembl, GDB, QM7, QM8, or QM9.

[0035] The second storage unit 112 may also store, for example, an RDKit or the like included in a library such as Anaconda (registered trademark), which is software distributed by Anaconda, Inc., USA. When the structural formula is SMILES, the second acquisition unit 12 reads the character string of the SMILES using MolFromSmiles included in the RDKit, and reads the structural formula of the molecule. Note that SMILES is a character string that represents the molecular structure of a molecule. The structural formula of each molecule may be recorded in a table written in data format such as CSV or Excel, which is a spreadsheet software.

[0036] SMILES may be obtained, for example, from a chemical database such as PubChem (a database of chemical substances provided by NCBI in the United States).

[0037] The third acquisition unit 13 acquires an interaction index I int is taken as the objective variable.

[0038] interaction index I int This information may be acquired from the third storage unit 113.

[0039] The third memory unit 113 stores an interaction index I int The data table includes the interaction index I int The third storage unit 113 may store, for example, an interaction index I calculated by an interaction index calculation unit 209 of a screening device 20A (see FIG. 6) and a screening device 20B (see FIG. 10) described later. int A database that records the above may be used as the data table.

[0040] The learning data set creation unit 14 uses the descriptors of the molecular structure of the molecules to be adsorbed as explanatory variables, and calculates the interaction index I int The learning dataset creation unit 14 extracts the input molecular structure descriptors and the interaction index I int Link them together to create a training dataset.

[0041] The learning unit 15 calculates the molecular structure descriptors (explanatory variables) of the molecules and the interaction index I int A trained model M1 is generated by training using a training dataset in which the objective variables are associated with the training data.

[0042] The trained model M1 is a trained model that has been machine-learned in advance using a training data set (training data table) stored in a storage unit (not shown). .... The trained model M1 is a trained model that has been machine-learned in advance using a training data table stored in a storage unit. The trained model M1 is a trained model that has been machine-learned in advance using a training data set stored in a storage unit. The trained model M1 is a trained model that has been machine-learned in advance using a training data table stored in a storage unit. The trained model M1 is a trained model that has int The learned result of the correspondence with (objective variable) is applied. The trained model M1 uses the molecular structure descriptors of the molecules as input data as explanatory variables, and the interaction index I int The output data is the molecular structure descriptor of the molecule and the interaction index I int It is a program for modeling the input / output relationship between the trained model M1 and the trained model M2. The trained model M1 may be expressed as a mathematical formula such as a function.

[0043] Among machine learning techniques, it is preferable to apply a supervised learning algorithm to the trained model M1. Examples of supervised learning include linear regression, regularized regression, partial least squares regression, polynomial regression, kernel regression, logistic regression, random forest, gradient boosting regression tree, support vector machine (SVM), and neural network. The neural network may be deep learning, which is a neural network with more than three layers. As the type of neural network, for example, a convolutional neural network (CNN), a recurrent neural network (RNN), and a general regression neural network may be used. Among these, it is preferable to use a gradient boosting regression tree.

[0044] The output unit 16 outputs, by display or the like, information on the learning dataset used in learning the trained model M1 and information on the trained model M1.

[0045] In this way, since the learning device 1 includes the learning unit 15, the interaction index I int The trained model M1 generated by the learning device 1 can generate an interaction index I int can be predicted.

[0046] interaction index I int As shown in Figure 4, when ads The absolute value of the interaction index I int and the adsorption energy E ads Therefore, when the molecular structure descriptor of the molecule is input to the trained model M1 generated by the learning device 1, the interaction index I int Predicting the adsorption property of a molecule is equivalent to observing the adsorption energy of the molecule, and the magnitude of the adsorption property can be predicted. Therefore, the learning device 1 can be used to predict the adsorption property of a molecule without directly obtaining the adsorption energy of the molecule adsorbed on the surface of the molecular crystal.

[0047] In addition, since the learning device 1 can generate a trained model M1, the learning device 1 can calculate an interaction index I int The predicted interaction index I intBy using this to predict the magnitude of the adsorptivity of a molecule, it is possible to reduce the burden and time required to select a molecule having a predetermined magnitude of adsorptivity. When selecting a molecule effective for molecular crystals, in order to select a molecule with high adsorptivity according to the type of molecular crystal, various molecular crystals and molecules are actually combined in experiments to confirm the adsorption energy or the magnitude of adsorptivity of the molecule. By performing such a process, it is very laborious to select a molecule with appropriate adsorptivity for molecular crystals, and the cost burden is also large due to the preparation of various molecular crystals and molecules. By using the trained model M1 generated by the learning device 1, the interaction index I int Since it is possible to predict the adsorptivity of a molecule to a molecular crystal, the burden involved in predicting this can be reduced, and the time required to select a molecule having the desired adsorptivity can be shortened.

[0048] <Screening System> A screening system according to an embodiment of the present invention will be described. The screening system according to the present embodiment calculates an interaction index I as an index representing the degree of adsorptivity from the descriptors of the molecular structure of a molecule. int is used to screen the molecules.

[0049] 5 is a diagram showing the configuration of the screening system according to the present embodiment. As shown in FIG. 5, the screening system 2 includes a screening device 20, a storage unit 30, and a machine learning potential 40. In the screening system 2, the screening device 20, the storage unit 30, and the machine learning potential 40 are connected via a communication network 50, and input values ​​to the screening device 20, the storage unit 30, and the machine learning potential 40, and output values ​​of the screening device 20, the storage unit 30, and the machine learning potential 40 may be transmitted via the communication network 50. At least one of the storage unit 30 and the machine learning potential 40 may be stored on the cloud.

[0050] In this embodiment, the screening device 20, the storage unit 30, and the machine learning potential 40 are connected via a communication network 50, but may be connected by wire. Also, the screening system 2 may be a standalone device such as a PC (Personal Computer) that includes each component within the device.

[0051] The screening device 20 uses the machine learning potential 40 to calculate the interaction index I when the adsorbate is adsorbed on the adsorbent. int In this embodiment, the adsorbent is a molecular crystal, and the adsorbate is a molecule (molecule) that is adsorbed to the molecular crystal, as in the above-described learning device 1. Details of the screening device 20 will be described later.

[0052] The storage unit 30 stores a data table including information on molecular crystals, molecules, and the like.

[0053] The memory unit 30 may include a first memory unit 301 (see Figure 6) that stores the bulk structure of a molecular crystal, a second memory unit 302 (see Figure 6) that stores information regarding descriptors of the molecular structure of a molecule, and a third memory unit 303 (see Figure 10) that stores descriptor information of the molecular structure of a molecule to be predicted.

[0054] The machine learning potential 40 is an interatomic potential using a machine learning technique that outputs energy from information on the structure of atoms. Examples of the machine learning potential include a neural network potential (NNP), a Gaussian approximation potential (GAP), a spectral neighbor analysis potential (SNAP), and a moment tensor potential (MTP). Among these, the machine learning potential is preferably NNP in terms of the high flexibility of a neural network. For the NNP, an atomic simulator that learns the relationship between atomic coordinates and energy using quantum chemical calculations as teaching data can be used. For the NNP, MATLANTIS (registered trademark) may be used.

[0055] [Screening equipment] The screening device 20 will now be described. The screening device 20 calculates an interaction index I as an index representing the degree of the adsorptivity of a molecule from the descriptor of the molecular structure of the molecule. int The screening device 20 performs screening of molecules using an interaction index I calculated from the descriptors of the molecular structure of the molecule. int Any device capable of screening molecules using the above-mentioned method may be used. In describing the screening device 20, two embodiments of the screening device 20 will be described. One embodiment of the screening device 20 is screening device 20A (see FIG. 6), and the other embodiment is screening device 20B (see FIG. 10).

[0056] (First embodiment) A screening device according to the first embodiment will be described. FIG. 6 is a system configuration diagram showing the configuration of the screening device according to the first embodiment. As shown in FIG. 6, the screening device 20A according to the first embodiment includes a first acquisition unit 201, a second acquisition unit 202, a first energy calculation unit 203, a second energy calculation unit 204, a surface adsorption structure search unit 205, a structure optimization unit 206 for surface adsorption structure, a third energy calculation unit 207, an adsorption energy calculation unit 208, an interaction index calculation unit 209, a correlation diagram creation unit 210, a clustering unit 211, a selection unit 212, and an output unit 213, and calculates an interaction index I as an index representing the magnitude of adsorptivity from the descriptors of the molecular structure of the molecule. int is calculated and a group of candidate molecules is selected.

[0057] The screening device 20A may use the machine learning potential 40 in any of the first energy calculation unit 203, the second energy calculation unit 204, the surface adsorption structure searching unit 205, the structure optimization unit 206 of the surface adsorption structure, the third energy calculation unit 207, the adsorption energy calculation unit 208, the interaction index calculation unit 209, the correlation diagram creation unit 210, the clustering unit 211, and the selection unit 212. It is preferable that the screening device 20A uses the machine learning potential 40 in any one or all of the first energy calculation unit 203, the second energy calculation unit 204, the surface adsorption structure searching unit 205, and the structure optimization unit 206 of the surface adsorption structure, which have a particularly large calculation load.

[0058] The first acquisition unit 201 acquires the bulk structure of the molecular crystal. The bulk structure of the molecular crystal may be acquired from the first memory unit 301 of the memory unit 30. The first memory unit 301 stores the bulk structure of the molecular crystal, similar to the above-mentioned first memory unit 111. The information on the molecular crystal is similar to the information on the molecular crystal acquired from the first memory unit 111 by the above-mentioned learning device 1, and therefore a description thereof will be omitted.

[0059] The second acquisition unit 202 acquires a descriptor of the molecular structure of the molecule as an explanatory variable. The information of the descriptor of the molecular structure of the molecule may be acquired from the second memory unit 302 of the memory unit 30. The second memory unit 302 stores the information of the descriptor of the molecular structure of the molecule, similar to the above-mentioned second memory unit 112. The information of the descriptor of the molecular structure of the molecule is similar to the information of the descriptor of the molecular structure of the molecule acquired from the second memory unit 112 by the above-mentioned learning device 1, and therefore a description thereof will be omitted.

[0060] The first energy calculation unit 203 calculates the energy of the bulk structure of the molecular crystal. As shown in Fig. 7, the first energy calculation unit 203 may include a first molecular crystal structure optimization unit 2031, a surface cutout unit 2032, a second molecular crystal structure optimization unit 2033, and a molecular crystal energy calculation unit 2034.

[0061] The first energy calculation unit 203 is preferably executed using the machine learning potential 40 in the first molecular crystal structure optimization unit 2031 and the second molecular crystal structure optimization unit 2033, which have a particularly large calculation load.

[0062] The first molecular crystal structure optimization unit 2031 performs structure optimization of the bulk structure of the molecular crystal. By performing structure optimization of the bulk structure, calculation accuracy can be improved.

[0063] As a method for optimizing the structure of the bulk structure of a molecular crystal, a commonly used method may be used.

[0064] For example, the first structure optimization unit 2031 for the molecular crystal may take into account the coordinate and energy relationships between the atoms that make up the molecule, place them in appropriate positions, and determine the structure that makes the bulk structure of the molecular crystal the most energetically stable.

[0065] Furthermore, the first molecular crystal structure optimization unit 2031 may perform structure optimization of the bulk structure of the molecular crystal using a general molecular simulation method such as first-principles calculation or molecular mechanics calculation.

[0066] Furthermore, the first molecular crystal structure optimization unit 2031 may create a bulk optimized structure of the molecular crystal optimized by relaxing the bulk structure of the molecular crystal. The relaxation is a general structure optimization method such as the steepest descent method or the conjugate gradient method for obtaining the minimum value of energy in a multidimensional space. For example, the relaxation means that the sum or scalar of force vectors acting on the entire bulk structure of the molecular crystal, or the stress tensor or the main component of the stress tensor, etc., is made to match a predetermined pressure (external pressure) under a certain threshold. The bulk structure of the molecular crystal acquired by the first acquisition unit 201 may not have a correct density that matches the actual state because single molecule information of the bulk structure of the molecular crystal is placed in a simulation cell of an appropriate size. In this state, the energy E of the bulk structure of the molecular crystal is calculated by the molecular crystal energy calculation unit 2034. A Even if you calculate the calculated energy E A The accuracy of the first structure optimization unit 2031 of the molecular crystal can create a bulk optimized structure of the molecular crystal adjusted to a bulk structure of the molecular crystal having a correct density by optimizing the bulk structure of the molecular crystal.

[0067] When the first molecular crystal structure optimization unit 2031 relaxes the bulk structure of the molecular crystal to create a bulk optimized structure of the molecular crystal, the first molecular crystal structure optimization unit 2031 may create the bulk optimized structure under the following two conditions. (1) The volume of the simulation cell is variable. An external force is applied to the simulation cell, and the bulk structure of the molecular crystal may be in a state where a predetermined pressure (external pressure) is applied. (2) The first structure optimization unit 2031 of the molecular crystal may set the above (1) as the structure at absolute zero. A molecular dynamics simulation may be performed to take into account the effect of temperature. In addition to the molecular dynamics simulation, a molecular simulation such as the Monte Carlo method may be performed. The volume of the simulation cell may be variable. The temperature may be set to any appropriate value. An external force may be applied to the simulation cell, and the bulk structure of the molecular crystal may be in a state where a predetermined pressure (external pressure) is applied.

[0068] By creating a bulk optimized structure of the molecular crystal under the above two conditions, the simulation cell is set to reproduce the density of the actual bulk structure of the molecular crystal, i.e., the simulation cell can be set to have the density when the bulk structure of the molecular crystal is relaxed, taking into account the specified pressure and temperature.

[0069] As described above, the first molecular crystal structure optimization unit 2031 has a particularly large calculation load among the first energy calculation unit 203, and therefore is preferably executed using the machine learning potential 40. This allows the structure optimization of the bulk structure of the molecular crystal to be performed in a shorter time.

[0070] The surface cutting unit 2032 cuts out a surface with an arbitrary plane index from the bulk structure of the optimized molecular crystal.

[0071] The arbitrary plane index may be appropriately set to any arbitrary plane index depending on the type of bulk structure of the molecular crystal.

[0072] The second molecular crystal structure optimization unit 2033 performs structure optimization of the bulk structure of the molecular crystal obtained by cutting out a surface of an arbitrary face index. By performing structure optimization of the bulk structure of the molecular crystal obtained by cutting out the surface, the calculation accuracy of the energy of the bulk structure of the molecular crystal obtained by cutting out the surface is improved.

[0073] The second molecular crystal structure optimization unit 2033 may perform structure optimization in the same manner as the first molecular crystal structure optimization unit 2031. Details of the structure optimization method are similar to those of the first molecular crystal structure optimization unit 2031, and therefore will not be repeated here.

[0074] As described above, the second molecular crystal structure optimization unit 2033, like the first molecular crystal structure optimization unit 2031, has a particularly large calculation load among the first energy calculation unit 203, and is therefore preferably executed using the machine learning potential 40. This makes it possible to perform structure optimization of the bulk structure of a molecular crystal cut out of a surface with an arbitrary face index in a shorter time.

[0075] The molecular crystal energy calculation unit 2034 calculates the energy E A is calculated as the first energy.

[0076] Energy E of the bulk structure of a molecular crystal A can be calculated using machine learning potential 40, while the energy E A The calculation method is not particularly limited, and an empirical potential such as OPLS (Optimized Potentials for Liquid Simulations), or the Schrodinger equation and an equation based thereon may be used.

[0077] The second energy calculation unit 204 shown in Fig. 6 calculates the energy of molecular structures of a plurality of (N types (N is an integer of 1 or more)) molecules. As shown in Fig. 8, the second energy calculation unit 204 has a molecular structure optimization unit 2041 and a molecular energy calculation unit 2042.

[0078] The molecular structure optimization unit 2041 optimizes the molecular structures of a plurality of molecules. By optimizing the molecular structures of the molecules, the accuracy of energy calculation can be improved.

[0079] The molecular structure optimization unit 2041 may perform structure optimization in the same manner as the molecular crystal first structure optimization unit 2031. Details of the structure optimization method are the same as those of the molecular crystal first structure optimization unit 2031, and therefore will not be repeated here.

[0080] The molecular structure optimization unit 2041 has a particularly large calculation load among the second energy calculation unit 204, and is therefore preferably executed using the machine learning potential 40. This makes it possible to perform structure optimization of the molecular structure of the molecule in a shorter time.

[0081] The molecular energy calculation unit 2042 calculates the energy E B is calculated as the second energy.

[0082] Energy E of the optimized molecule B The calculation method of is not particularly limited, and a general calculation method may be used. For example, the molecular crystal bulk structure energy E A A method similar to the calculation method of may be used.

[0083] 6, the surface adsorption structure searching unit 205 searches for a surface adsorption structure by combining the molecular structure of an optimized molecule with the bulk structure of the molecular crystal cut out from the surface. The surface adsorption structures may be searched for by the number of molecular structures of a plurality (N types) of molecules optimized in the molecular structure optimization unit 2041.

[0084] The search method is not particularly limited, and a general search method such as Bayesian optimization, sequential optimization, random search, or grid search may be used.

[0085] The surface adsorption structure structure optimization unit 206 optimizes the surface adsorption structure structure.

[0086] The surface adsorption structure structure optimization unit 206 may perform structure optimization in the same manner as the molecular crystal first structure optimization unit 2031. Details of the structure optimization method are similar to those of the molecular crystal first structure optimization unit 2031, and therefore will not be repeated here.

[0087] The third energy calculation unit 207 calculates the energy E A+B is calculated as the third energy.

[0088] Energy E of the optimized surface adsorption structure A+B The calculation method of is not particularly limited, and a general calculation method may be used. For example, the molecular crystal bulk structure energy E A A method similar to the calculation method of may be used.

[0089] The adsorption energy calculation unit 208 calculates the adsorption energy E ads Calculate.

[0090] Adsorption energy E of molecules of optimized surface adsorption structure ads can be calculated from the following formula (I). E ads =E A+B -(E A +E B ) (I)

[0091] The interaction index calculation unit 209 calculates the interaction index I int Calculate.

[0092] interaction index I int can be any value calculated as an index showing the interaction of a molecular bond of interest between a molecule that is an adsorbate and a molecular crystal that is an adsorbent, and for example, a value calculated using the above formula (1) can be used.

[0093] The correlation diagram creation unit 210 calculates the interaction index I int and the adsorption energy E of the molecules of the surface adsorption structure ads Create a relationship diagram showing the correlation with the absolute value of

[0094] The clustering unit 211 clusters the adsorption structures of the structurally optimized surface adsorption structures based on the distance between the molecular crystals and the molecules, and extracts a plurality of groups of candidate molecules. The clustering unit 211 may extract and classify a predetermined number (e.g., 100) of candidate molecules in order of the shortest distance between the molecular crystals and the molecules, based on the distance between the molecular crystals and the molecules in the created surface adsorption structures. The clustering unit 211 also extracts and classifies the candidate molecules in order of the shortest distance between the molecular crystals and the molecules, based on the interaction index I int or the adsorption energy E of the molecules of the surface adsorption structure ads Based on the absolute value of int or the adsorption energy E of the molecule adsAlternatively, a predetermined number (for example, 100) of the highest absolute values ​​may be extracted.

[0095] By clustering candidate molecules into groups, the degree of selectivity of the molecular crystal for the molecule can be confirmed from the distribution of the candidate molecule groups. For example, as shown in Figure 9, molecular groups are extracted for each of three types of molecules, molecule 1, molecule 2, and molecule 3. For molecule 1, the interaction index I int Since the molecules are widely dispersed, the selectivity for a given adsorbent is high, and for molecule 3, the interaction index I int It can be seen that the selectivity for a given adsorbent is low since the solubility is only narrowly distributed.

[0096] A commonly used method can be used for the clustering method, and examples of the clustering method include the k-Means method, the k-Means++ method, and the Gaussian Mixture method. The k-Means method is a method for classifying molecules into k clusters. For example, when the k-Means method is used, a vector x in which the variable values ​​of a predetermined molecule are stored is randomly assigned to k clusters. Next, the center of gravity of the molecules assigned to each cluster is calculated. Next, the distance from the calculated center of gravity of each molecule is calculated, and the vector x is reassigned to the cluster with the closest distance. The above process of calculating the center of gravity of the molecules assigned to each cluster and the process of calculating the distance from the calculated center of gravity of each molecule and reassigning the vector x to the cluster with the closest distance are repeated until the cluster assignment of all molecules converges.

[0097] Furthermore, the clustering unit 211 may visualize the extracted group of candidate molecules, perform nonlinear dimensional reduction of the high-dimensional data to two or three dimensions, and then perform clustering. By calculating the distance between the molecular crystal and the molecule in the surface adsorption structure of the extracted group of candidate molecules, variables with a number of dimensions corresponding to the number of extracted molecules (for example, 100 dimensions when 100 molecules are extracted) are obtained. The clustering unit 211 visualizes the extracted group of candidate molecules, and performs nonlinear dimensional reduction of the high-dimensional data to two or three dimensions, which makes it easier to perform clustering appropriately.

[0098] A method for visualizing the group of extracted candidate molecules includes, for example, principal component analysis (PCA), etc. The clustering unit 211 uses PCA to perform rotational transformation of a coordinate system centered on the sample mean, thereby projecting the data of the group of extracted candidate molecules given to it into a lower dimensional space, so that the data of the group of extracted candidate molecules can be visualized so that the scattering of points appears as large as possible with fewer coordinate axes.

[0099] Methods for nonlinearly reducing high-dimensional data to two or three dimensions include, for example, the t-SNE (t-distributed stochastic neighbor embedding) method, which preserves the relationship between the intermolecular distance between molecular crystals and molecules, and the GTM (generative topographic mapping) method, which preserves the positional relationship between molecular crystals and molecules.

[0100] The selection unit 212 selects a group of candidate molecules from a plurality of molecules. The selection unit 212 may select an arbitrary group of molecules as the group of candidate molecules depending on the level of adsorptive property required for the molecules.

[0101] The method of selecting the group of molecules is not particularly limited, and a general selection method may be used. For example, the selection unit 212 may select the group of molecules by using an interaction index I int Or adsorption energy E adsThe interaction index I int Or adsorption energy E ads The absolute value of is greater than or equal to a specified value, or the interaction index I int Or adsorption energy E ads A selection of groups of molecules may be made based on, for example, having the highest absolute value of

[0102] The output unit 213 outputs the candidate molecule group by display or the like.

[0103] The screening device 20A includes a first energy calculation unit 203, a second energy calculation unit 204, a surface adsorption structure search unit 205, a structure optimization unit 206 for surface adsorption structure, a third energy calculation unit 207, an adsorption energy calculation unit 208, an interaction index calculation unit 209, a correlation diagram creation unit 210, a clustering unit 211, and a selection unit 212. The screening device 20A calculates an interaction index I int and the adsorption energy E ads By clustering based on the absolute value of the interaction index I, a plurality of molecules can be selected into a group of a plurality of molecules according to the strength of adsorption of the molecules adsorbed to the molecular crystal. In this way, the screening device 20A can extract a group of candidate molecules having an arbitrary strength of adsorption from among the plurality of molecules. int and adsorption energy E ads The adsorption property between the molecular crystal and the molecule in the surface adsorption structure can be determined by the magnitude of the absolute value of . Therefore, the screening device 20A can predict the adsorption property of a molecule without directly calculating the adsorption energy of the molecule adsorbed on the surface of the molecular crystal, and can extract a group of candidate molecules having an arbitrary adsorption size.

[0104] Second embodiment A screening device according to the second embodiment will be described. Fig. 10 is a system configuration diagram showing the configuration of the screening device according to the second embodiment. As shown in Fig. 10, the screening device 20B includes a learning model generating unit 221, a third acquiring unit 222, and a predicting unit 223 instead of the correlation diagram creating unit 210 and the clustering unit 211 in the screening device 20A shown in Fig. 6, and calculates an interaction index I as an index representing the degree of adsorptivity from the descriptors of the molecular structure of the molecule to be predicted. int Predict.

[0105] The learning model generation unit 221 uses the descriptor (explanatory variable) of the molecular structure of the molecule acquired by the second acquisition unit 202 and the interaction index I calculated by the interaction index calculation unit 209. int (Objective variable) is used to generate a trained model M2.

[0106] That is, the learning model generation unit 221 uses the descriptor (explanatory variable) of the molecular structure of the molecule acquired by the second acquisition unit 202 and the interaction index I calculated by the interaction index calculation unit 209. int A training dataset is created that matches the (objective variable), and a trained model M2 can be generated by training using the created training dataset.

[0107] The trained model M1 generated by the above-mentioned learning device 1 can be used as the trained model M2.

[0108] The third acquisition unit 222 acquires a descriptor of the molecular structure of the molecule to be predicted as an explanatory variable. The information of the descriptor of the molecular structure of the molecule to be predicted may be acquired from the third memory unit 303 of the memory unit 30. The third memory unit 303 stores the information of the descriptor of the molecular structure of the molecule, similar to the above-mentioned second memory unit 112. The information of the descriptor of the molecular structure of the molecule is similar to the information of the descriptor of the molecular structure of the molecule acquired from the second memory unit 112 by the above-mentioned learning device 1, and therefore a description thereof will be omitted.

[0109] The prediction unit 223 uses the trained model M2 to input the descriptor of the molecular structure of the molecule to be predicted acquired by the third acquisition unit 222, and calculates the interaction index I int Predict.

[0110] That is, the prediction unit 223 inputs the descriptor of the molecular structure of the molecule to be predicted, acquired by the third acquisition unit 222, to the trained model M2, and obtains the interaction index I int is output as the objective variable. The interaction index I int As described above, since it is correlated with the adsorption energy of a molecule, it is an index representing the adsorption energy. Therefore, the prediction unit 223 obtains the interaction index I int By predicting the adsorption energy of the target molecule to the molecular crystal, it is possible to predict the adsorption energy of the target molecule to the molecular crystal.

[0111] The selection unit 212 selects candidate molecules from among a plurality of molecules. The selection unit 212 may select any of a plurality of molecules as candidate molecules depending on the level of adsorptive property required for the molecule, etc.

[0112] The method of selecting the molecules is not particularly limited, and a general selection method may be used. For example, the selection unit 212 may select the molecules by using an interaction index I int Or adsorption energy E ads The interaction index I int Or adsorption energy E ads The absolute value of is greater than or equal to a specified value, or the interaction index I int Or adsorption energy E ads The selection of molecules may be based on having the highest absolute value of

[0113] The output unit 213 outputs the selected molecules as candidate molecules by displaying them or the like.

[0114] As described above, the screening system 2 includes a screening device 20B, which includes a learning model generation unit 221, a prediction unit 223, and a selection unit 212. In the prediction unit 223, the screening device 20B calculates an interaction index I of the molecule to be predicted as an index representing the degree of adsorptivity from the descriptor of the molecular structure of the molecule to be predicted by the learned model M2. int The screening device 20B can predict the interaction index I int By predicting the adsorption property of a molecule, it is possible to predict the adsorption property of the molecule without performing a molecular simulation to obtain an adsorption structure. Then, the screening device 20B selects the interaction index I int Therefore, the screening device 20B can predict the adsorptivity of a molecule without directly calculating the adsorption energy of the molecule adsorbed on the surface of the molecular crystal in the prediction unit 223, and therefore can select a candidate molecule from among a plurality of molecules.

[0115] In the screening system 2, the screening device 20B predicts the interaction index I of the molecule to be predicted using the trained model M2 in the prediction unit 223. int The screening device 20B can predict the interaction index I predicted by the prediction unit 223 using the trained model M2. int By using this method, the degree of adsorptivity of a molecule to a molecular crystal can be predicted more quickly and easily, thereby shortening the time required to screen molecules, such as those with high adsorptivity to molecular crystals.

[0116] For example, the screening system 2 can comprehensively screen and select or design a new substance that will form strong hydrogen bonds on the surface of a specific molecular crystal from among a large number of compound groups. Therefore, the screening system 2 can easily compare the differences in the adsorptivity of different molecules to molecular crystals, allowing for easy screening of molecules in a short time.

[0117] Furthermore, in the screening system 2, the screening device 20B can use the machine learning potential 40 in at least one of the first energy calculation unit 203, the second energy calculation unit 204, the surface adsorption structure search unit 205, and the surface adsorption structure structure optimization unit 206. As a result, the screening device 20B can shorten the process time required to obtain an accurate adsorption structure with a molecular crystal of each molecule, for example, when it is desired to find a molecule that shows a strong adsorption state due to hydrogen bonding from a large number of compound groups. Therefore, the screening device 20B according to this embodiment can shorten the screening time and can be used for a wider range of molecular groups by increasing its versatility.

[0118] (Hardware configuration of learning device 1 and screening device 20 (20A, 20B)) Next, an example of the hardware configuration of the learning device 1 and the screening device 20 will be described. FIG. 11 is a block diagram showing the hardware configuration of the learning device 1 and the screening device 20. As shown in FIG. 11, the learning device 1 and the screening device 20 are configured as an information processing device (computer), and can be physically configured as a computer system including a CPU (Central Processing Unit: processor) 101 which is an arithmetic processing unit, a RAM (Random Access Memory) 102 and a ROM (Read Only Memory) 103 which are main storage devices, an input device 104 which is an input device, an output device 105, a communication module 106, and an auxiliary storage device 107 such as a hard disk. These are connected to each other by a bus 108. The output device 105 and the auxiliary storage device 107 may be provided externally.

[0119] The CPU 101 controls the overall operation of the learning device 1 and the screening device 20, and performs various information processing. The CPU 101 can execute, for example, a learning program or a screening program (described later) stored in the ROM 103 or the auxiliary storage device 107, to generate the trained model M1 or screen molecules.

[0120] The RAM 102 is used as a work area for the CPU 101 and may include a non-volatile RAM for storing main control parameters and information.

[0121] The ROM 103 stores a basic input / output program, etc. A learning program and a screening program may be stored in the ROM 103.

[0122] The input device 104 is an input device such as a keyboard, a mouse, operation buttons, a touch panel, a display screen, etc., and receives information input by a user as an instruction signal and outputs the instruction signal to the CPU 101 .

[0123] The output device 105 is a display device such as a monitor display, an audio device such as a speaker, a printing device such as a printer, etc. In the output device 105, for example, information such as the catalyst selection result is displayed on a display device such as a monitor display, and the displayed screen is updated in response to an input operation via the input device 104 or the communication module 106.

[0124] The communication module 106 is a data transmission / reception device such as a network card, and functions as a communication interface that receives information from an external data recording server or the like, and outputs analysis information to other electronic devices.

[0125] The auxiliary storage device 107 is a storage device such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive), and stores, for example, various data, files, etc. required for the operation of the learning device 1 and the screening devices 20A and 20B.

[0126] Each function of the learning device 1 and the screening device 20 is realized by loading a specified computer software (including a learning program or a screening program) from a main memory device such as RAM 102 or an auxiliary memory device 107, and executing it by the CPU 101, thereby reading and writing data in a main memory device such as RAM 102 or an auxiliary memory device 107, etc., and operating the input device 104, the output device 105 and the communication module 106.

[0127] Therefore, each part of the learning device 1 and the screening device 20 is realized by software and hardware working together in a computer equipped with the learning device 1 and the screening device 20, by the processor executing predetermined computer software (including a learning program or a screening program) that is pre-stored.

[0128] The learning program and screening program can be stored, for example, in a main storage device or auxiliary storage device 107 of a computer. The learning program or screening program may be stored on a computer connected to a communication line such as the Internet, and a part or all of the learning program or screening program may be provided by being downloaded via the communication line. Furthermore, the learning program and screening program may be configured to be provided or distributed via the communication line.

[0129] The learning program and screening program may be recorded (including installed) into a computer from a state in which part or all of them are stored on a portable storage medium, such as an optical disc such as a CD-ROM or DVD-ROM, or a semiconductor memory such as a flash memory.

[0130] <Learning Method> The learning method according to the present embodiment will be described. In the learning method according to the present embodiment, a descriptor of a molecular structure of a molecule and an interaction index I intUsing the learning dataset in which the and are associated, the interaction index I int This is a method to generate a trained model that predicts

[0131] Fig. 12 is a flowchart showing the learning method according to this embodiment. As shown in Fig. 12, in the learning method according to this embodiment, the bulk structure of the molecular crystal is acquired by the first acquisition unit 11 (first acquisition step: step S11).

[0132] Information on the bulk structure of the molecular crystal, etc. may be acquired from the first storage unit 111.

[0133] Next, the second acquisition unit 12 acquires the descriptors of the molecular structure of the molecule as explanatory variables (second acquisition step: step S12).

[0134] The information of the descriptor of the molecular structure of the molecule may be obtained from the second storage unit 112.

[0135] Next, the third acquisition unit 13 calculates the interaction index I int is acquired as a response variable (third acquisition step: step S13).

[0136] interaction index I int This information may be acquired from the third storage unit 113.

[0137] Next, the learning data set creation unit 14 uses the descriptors of the molecular structure of the molecules to be adsorbed as explanatory variables to calculate the interaction index I int are extracted as objective variables and added to a learning dataset (learning dataset creation process: step S14).

[0138] The learning data set creation unit 14 inputs the molecular structure descriptor of the molecule and the interaction index I int Link them together to create a training dataset.

[0139] Next, the learning unit 15 calculates the molecular structure descriptors (explanatory variables) of the molecules and the interaction index I int A trained model M1 is generated by training using a training data set in which the trained variables and the training data are associated with each other (objective variables) (generation process of trained model M1: step S15).

[0140] The learning unit 15 calculates an interaction index I according to the inputted descriptor of the molecular structure of the molecule. int A trained model M1 is generated so that the following is output:

[0141] Next, the output unit 16 outputs information on the learning data set used in learning the trained model M1, information on the trained model M1, and the like by display or the like (output step: step S16).

[0142] The learning method according to the present embodiment includes a learning step (step S15), in which an interaction index I is calculated from the descriptor of the molecular structure of the molecule. int In the learning method according to the present embodiment, by using the learned model M1 generated in the learning process (step S15), it is possible to generate an interaction index I int The interaction index I int is the adsorption energy E ads Since the interaction index I is positively correlated with the absolute value of the interaction index I, it can be used as an index representing the magnitude of the adsorption energy. int By predicting , the adsorption energy of the molecule can be predicted, and the degree of adsorptivity can be predicted. Therefore, the learning method according to this embodiment can be used to predict the adsorptivity of a molecule without directly determining the adsorption energy of the molecule adsorbed on the surface of a molecular crystal.

[0143] In addition, the learning method according to the present embodiment is a method for learning a trained model M1 generated in the learning step (step S15) by calculating an interaction index I intThe predicted interaction index I int By using the generated trained model M1 to predict the magnitude of the adsorptivity of a molecule, it is possible to reduce the burden and time required to predict a molecule having a predetermined magnitude of adsorptivity. Therefore, the learning method according to this embodiment uses the trained model M1 generated to obtain the interaction index I from the descriptor of the molecular structure of the molecule. int Therefore, the predicted interaction index I int By using this method, it is possible to reduce the burden of predicting the adsorptivity of a molecule to a molecular crystal and to shorten the time required to select a molecule having a desired adsorptivity.

[0144] <Screening method> Next, the screening method according to the present embodiment will be described. The screening method according to the present embodiment uses an interaction index I calculated as an index representing the magnitude of the adsorptive property of a molecule from a descriptor of the molecular structure of the molecule. int This is a method for screening molecules using the above method.

[0145] The screening method of this embodiment will be described as a screening method according to the first embodiment when the screening device 20A of the first embodiment described above is used, and as a screening method according to the second embodiment when the screening device 20B of the second embodiment described above is used.

[0146] (First embodiment) A screening method according to the first embodiment will be described. The screening method according to the first embodiment is a method using a screening device 20A having a configuration as shown in FIG. 6. In the screening method according to the first embodiment, the adsorption energy E ads and the interaction index I int and selects a group of candidate molecules. The screening method according to the first embodiment is performed using the screening device 20A described above, and therefore some of the contents already described will be omitted.

[0147] Fig. 13 is a flowchart showing the screening method according to the first embodiment. As shown in Fig. 13, in the screening method, the bulk structure of a molecular crystal is acquired by the first acquisition unit 201 (first acquisition step: step S201).

[0148] The bulk structure of the molecular crystal may be obtained from the first memory section 301 of the memory section 30.

[0149] Next, the second acquisition unit 202 acquires a descriptor of the molecular structure of the molecule as an explanatory variable (second acquisition step: step S202).

[0150] The information of the molecular structure descriptor of the molecule may be obtained from the second memory unit 302 of the memory unit 30.

[0151] Next, the first energy calculation unit 203 calculates the energy of the bulk structure of the molecular crystal (first energy calculation step: step S203).

[0152] In the first energy calculation step (step S203), as shown in Fig. 14, the bulk structure of the molecular crystal is optimized by the first molecular crystal structure optimization unit 2031 (first molecular crystal structure optimization step: step S2031). By optimizing the bulk structure, the calculation accuracy is improved.

[0153] As a method for optimizing the structure of the bulk structure of a molecular crystal, a commonly used method may be used.

[0154] Since the first molecular crystal structure optimization step (step S2031) has a particularly large calculation load among the first energy calculation steps (step S203), it is preferable to execute the first molecular crystal structure optimization unit 2031 using the machine learning potential 40. This allows the structure optimization of the bulk structure of the molecular crystal to be performed in a shorter time.

[0155] Next, a surface with an arbitrary plane index is cut out from the bulk structure of the optimized molecular crystal by the surface cutting unit 2032 (surface cutting process: step S2032).

[0156] The arbitrary plane index may be appropriately set to any arbitrary plane index depending on the type of bulk structure of the molecular crystal.

[0157] Next, the second molecular crystal structure optimization unit 2033 optimizes the structure of the bulk structure of the molecular crystal cut out from a surface with an arbitrary plane index (second molecular crystal structure optimization process: step S2033).

[0158] By optimizing the bulk structure of the molecular crystal having its surface cut out, the accuracy of calculation of the energy of the bulk structure of the molecular crystal having its surface cut out can be improved.

[0159] In the second molecular crystal structure optimization step (step S2033), the second molecular crystal structure optimization unit 2033 may perform structure optimization in the same manner as in the first molecular crystal structure optimization step (step S2031).

[0160] Like the first structural optimization step of the molecular crystal (step S2031), the second structural optimization step of the molecular crystal (step S2033) is one of the first energy calculation steps (step S203) that has a particularly large calculation load, and is therefore preferably executed by the second structural optimization unit 2033 of the molecular crystal using the machine learning potential 40. This makes it possible to perform structural optimization of the bulk structure of the molecular crystal obtained by cutting out a surface with an arbitrary face index in a shorter time.

[0161] Next, the molecular crystal energy calculation unit 2034 calculates the energy E A is calculated as the first energy (molecular crystal energy calculation process: step S2034).

[0162] Energy E of the bulk structure of a molecular crystal Acan be calculated using machine learning potential 40, while the energy E A The calculation method is not particularly limited, and an empirical potential such as OPLS, or the Schrodinger equation and an equation based thereon, etc. may be used.

[0163] Next, as shown in FIG. 13, the second energy calculation unit 204 calculates the energy of the molecular structure of a plurality of (N types (N is an integer of 1 or more)) molecules (second energy calculation step: step S204).

[0164] In the second energy calculation step (step S204), as shown in Fig. 15, the molecular structure optimization unit 2041 performs structure optimization of the molecular structures of a plurality of molecules (molecule structure optimization step: step S2041). By performing structure optimization of the molecular structures of the molecules, the calculation accuracy of the energy is improved.

[0165] In the molecular structure optimization step (step S2041), the molecular structure optimization unit 2041 may perform structure optimization in the same manner as in the first molecular crystal structure optimization step (step S2031).

[0166] Since the molecular structure optimization step (step S2041) has a particularly large calculation load among the second energy calculation steps (step S204), it is preferable to execute the molecular structure optimization step (step S2041) by the molecular structure optimization unit 2041 using the machine learning potential 40. This allows the molecular structure optimization of the molecule to be performed in a shorter time.

[0167] Next, the molecular energy calculation unit 2042 calculates the energy E B is calculated as the second energy (molecule energy calculation process: step S2042).

[0168] Energy E of the optimized molecule BThe calculation method of is not particularly limited, and a general calculation method may be used. For example, the energy E of the bulk structure of the molecular crystal in the energy calculation step of the molecular crystal (step S2034) A A method similar to the calculation method of may be used.

[0169] Next, as shown in FIG. 13, the surface adsorption structure searching unit 205 binds the molecular structure of the optimized molecule to the bulk structure of the molecular crystal cut from the surface to search for a surface adsorption structure (surface adsorption structure searching process: step S205).

[0170] The surface adsorption structures may be searched for as many times as the molecular structures of a plurality of (N types of) molecules optimized in the molecular structure optimization step (step S2041).

[0171] The search method is not particularly limited, and a general search method such as Bayesian optimization, sequential optimization, random search, or grid search may be used.

[0172] Next, the surface adsorption structure structure optimization unit 206 optimizes the surface adsorption structure structure (surface adsorption structure structure optimization step: step S206).

[0173] In the surface adsorption structure structure optimization step (step S206), the surface adsorption structure structure optimization unit 206 may perform structure optimization in a similar manner to the first molecular crystal structure optimization step (step S2031).

[0174] Next, the third energy calculation unit 207 calculates the energy E A+B is calculated as the third energy (third energy calculation step: step S207).

[0175] Energy E of the optimized surface adsorption structure A+B The calculation method of is not particularly limited, and a general calculation method may be used. For example, the energy E of the bulk structure of the molecular crystal in the energy calculation step of the molecular crystal (step S2034) AA method similar to the calculation method of may be used.

[0176] Next, the adsorption energy calculation unit 208 calculates the adsorption energy E ads is calculated (adsorption energy calculation step: step S208).

[0177] Adsorption energy E of molecules of optimized surface adsorption structure ads can be calculated from the following formula (I). E ads =E A+B -(E A +E B ) (I)

[0178] Next, the interaction index calculation unit 209 calculates the interaction index I int is calculated (interaction index calculation process: step S209).

[0179] interaction index I int can be any value calculated as an index showing the interaction of a molecular bond of interest between a molecule that is an adsorbate and a molecular crystal that is an adsorbent, and for example, a value calculated using the above formula (1) can be used.

[0180] Next, the correlation diagram creation unit 210 calculates the interaction index I int and the adsorption energy E of the molecule of the optimized surface adsorption structure ads A relationship diagram showing the correlation with the absolute value of is created (correlation diagram creation step: step S210).

[0181] Next, the clustering unit 211 clusters the structure-optimized surface adsorption structures based on the distance between the molecular crystal and the molecules, and extracts a plurality of groups of candidate molecules (clustering step: step S211).

[0182] In the clustering step (step S211), the clustering unit 211 may extract and classify a predetermined number (e.g., 100) of molecules in order of shortest distance between the molecular crystal and the molecule based on the distance between the molecular crystal and the molecule in the created surface adsorption structure. In addition, the clustering unit 211 may extract and classify the molecular crystals in the order of shortest distance between the molecular crystal and the molecule in the created surface adsorption structure based on the distance between the molecular crystal and the molecule. int or the adsorption energy E of the molecules of the surface adsorption structure ads Based on the absolute value of int or the adsorption energy E of the molecule ads Alternatively, a predetermined number (for example, 100) of the highest absolute values ​​may be extracted.

[0183] By clustering the candidate molecules into groups, the degree of selectivity of the molecular crystal for the molecule can be confirmed from the degree of dispersion of the candidate molecule groups (see Figure 9).

[0184] The clustering method may be a commonly used method, such as the k-Means method, the k-Means++ method, or the Gaussian Mixture method.

[0185] Furthermore, the clustering step (step S211) may involve visualizing a group of extracted candidate molecules by the clustering unit 211, performing nonlinear dimensional reduction of high-dimensional data to two or three dimensions, and then clustering. By calculating the distance between the molecular crystal and the molecule in the surface adsorption structure of the group of extracted candidate molecules, variables with a number of dimensions corresponding to the number of extracted molecules (for example, 100 dimensions when 100 molecules are extracted) are obtained. By visualizing a group of extracted candidate molecules by the clustering unit 211 and performing nonlinear dimensional reduction of high-dimensional data to two or three dimensions, it becomes easier to perform clustering appropriately.

[0186] A method for visualizing the extracted group of candidate molecules includes, for example, PCA, etc. In the clustering step (step S211), the clustering unit 211 can visualize the data of the extracted group of candidate molecules by using PCA so that the scattering of points is as large as possible with fewer coordinate axes.

[0187] Methods for nonlinearly reducing high-dimensional data to two or three dimensions include, for example, the t-SNE method, which preserves the relationship between molecular crystals and intermolecular distances, and GTM, which preserves the positional relationship between molecular crystals and molecules.

[0188] Next, the selection unit 212 selects a group of candidate molecules from among the multiple molecules (selection step: step S212).

[0189] In the selection step (step S212), the selection unit 212 may select an arbitrary group of molecules as a group of candidate molecules depending on the level of adsorptive property required for the molecules.

[0190] The method for selecting the group of molecules is not particularly limited, and a general selection method may be used.

[0191] Next, the candidate molecule group is output by display or the like by the output unit 213 (output step: step S213).

[0192] The screening method according to the first embodiment includes a first energy calculation step (step S203), a second energy calculation step (step S204), a surface adsorption structure search step (step S205), a surface adsorption structure structure optimization step (step S206), a third energy calculation step (step S207), an adsorption energy calculation step (step S208), an interaction index calculation step (step S209), a correlation diagram creation step (step S210), a clustering step (step S211), and a selection step (step S212). The screening method according to the first embodiment includes a clustering step (step S211) in which the interaction index I int and the adsorption energy E adsBy clustering based on the absolute value of the interaction index I, in the selection step (step S212), a plurality of molecules can be selected into a group of a plurality of molecules according to the strength of adsorption of the molecules adsorbed to the molecular crystal. In this way, the screening method according to the first embodiment can extract a group of candidate molecules having an arbitrary strength of adsorption from a plurality of molecules. int and adsorption energy E ads The adsorption property between the molecular crystal and the molecule in the surface adsorption structure can be determined by the magnitude of the absolute value of . Therefore, the screening method according to the first embodiment can predict the adsorption property of a molecule without directly calculating the adsorption energy of the molecule adsorbed on the surface of the molecular crystal, and therefore can extract a group of candidate molecules having any adsorption size.

[0193] In addition, in the screening method according to the first embodiment, the first acquisition step (step S201) may be performed after the second acquisition step (step S202) or simultaneously with the second acquisition step (step S202), and the order is not particularly limited.

[0194] Furthermore, in the screening method according to the first embodiment, the first energy calculation step (step S203) may be performed after the second energy calculation step (step S204) or simultaneously with the second energy calculation step (step S204), and the order is not particularly limited.

[0195] Second embodiment A screening method according to the second embodiment will be described. The screening method according to the second embodiment is a method using a screening device 20B having a configuration as shown in FIG. 10. In the screening method according to the second embodiment, an interaction index I is calculated from the descriptor of the molecular structure of the molecule to be predicted by the screening device 20B having a configuration as shown in FIG. 10, as an index representing the degree of adsorptivity when the molecule is adsorbed to the molecular crystal. int The screening method according to this embodiment is performed using the above-mentioned screening device 20B, and therefore some of the contents already described will be omitted.

[0196] Fig. 16 is a flowchart showing the screening method according to the second embodiment. Each of the steps of the first acquisition step (step S301), the second acquisition step (step S302), the first energy calculation step (step S303), the second energy calculation step (step S304), the surface adsorption structure search step (step S305), the structure optimization step of the surface adsorption structure (step S306), the third energy calculation step (step S307), the adsorption energy calculation step (step S308), and the interaction index calculation step (step S309) of the screening method according to the second embodiment shown in Fig. 16 is similar to each of the steps of the first acquisition step (step S201) to the interaction index calculation step (step S209) of the screening method according to the second embodiment as shown in Fig. 13, and therefore description thereof will be omitted.

[0197] In the screening method according to the second embodiment, the learning model generation unit 221 generates a descriptor (explanatory variable) of the molecular structure of the molecule acquired in the second acquisition step (step S302) and an interaction index I calculated in the interaction index calculation step (step S309). int (objective variable) to generate a trained model M2 (training model generation process: step S311).

[0198] In the learning model generation step (step S310), the learning model generation unit 221 calculates the molecular structure descriptor (explanatory variable) of the molecule acquired in the second acquisition step (step S302) and the interaction index I calculated in the interaction index calculation step (step S209). int A training dataset is created that matches the (objective variable), and a trained model M2 can be generated by training using the created training dataset.

[0199] The trained model M2 can use the trained model M1 generated by the learning method according to this embodiment described above.

[0200] Next, the third acquisition unit 222 acquires, as explanatory variables, descriptors of the molecular structure of the molecule to be predicted (third acquisition step: step S311).

[0201] Descriptor information of the molecular structure of the molecule to be predicted may be acquired from the third storage unit 303 of the storage unit 30. The third storage unit 303 stores the descriptor information of the molecular structure of the molecule, similar to the above-mentioned second storage unit 112. The descriptor information of the molecular structure of the molecule is similar to the descriptor information of the molecular structure of the molecule acquired from the second storage unit 112 by the above-mentioned learning method, and therefore a description thereof will be omitted.

[0202] Next, the prediction unit 223 uses the trained model M2 to input the descriptor of the molecular structure of the molecule to be predicted, which is acquired in the third acquisition step (step S311), to calculate the interaction index I int (prediction step: step S312).

[0203] That is, in the prediction step (step S312), the prediction unit 223 inputs the molecular structure descriptor of the molecule to be predicted, acquired in the third acquisition step (step S311), to the trained model M2, to obtain the interaction index I of the prediction target, which is predicted by the trained model M2. int is output as the objective variable. The interaction index I int As described above, since it is correlated with the adsorption energy of a molecule, it is an index representing the adsorption energy. Therefore, the prediction unit 223 obtains the interaction index I int By predicting the adsorption energy of the target molecule to the molecular crystal, it is possible to predict the adsorption energy of the target molecule to the molecular crystal.

[0204] Next, the selection unit 212 selects candidate molecules from among the multiple molecules (selection step: step S313).

[0205] In the selection step (step S313), the selection unit 212 may select any of a plurality of molecules as candidate molecules depending on the level of adsorptive property required for the molecules, etc.

[0206] The method of selecting the molecule is not particularly limited, and a general selection method may be used. For example, the selection method may be a method using the interaction index I int Or adsorption energy E ads The interaction index I int Or adsorption energy E ads The absolute value of is greater than or equal to a specified value, or the interaction index I int Or adsorption energy E ads The selection of molecules may be based on having the highest absolute value of

[0207] Next, the output unit 213 outputs the selected molecules as candidate molecules by display or the like (output step: step S314).

[0208] The screening method according to the second embodiment includes a learning model generation step (step S310), a prediction step (step S312), and a selection step (step S313). In the prediction step (step S312), the learned model M2 calculates an interaction index I of the molecule to be predicted as an index representing the magnitude of adsorptivity from the descriptor of the molecular structure of the molecule to be predicted. int The screening method according to the second embodiment can predict the interaction index I int By predicting the adsorption property of a molecule, it is possible to predict the adsorption property of the molecule without performing a molecular simulation to obtain an adsorption structure. In the selection step (step S313), the screening method according to the second embodiment selects the interaction index I int Therefore, in the screening method according to the second embodiment, since the adsorption property of a molecule can be predicted without directly calculating the adsorption energy of the molecule adsorbed on the surface of the molecular crystal in the prediction step (step S312), a candidate molecule can be selected from among a plurality of molecules.

[0209] In addition, in the prediction step (step S312) of the screening method according to the second embodiment, the interaction index Iint In the screening method according to the second embodiment, in the prediction step (step S312), the interaction index I predicted using the trained model M2 can be predicted. int By using this method, the degree of adsorptivity of a molecule to a molecular crystal can be predicted more quickly and easily, thereby shortening the time required to screen molecules, such as those with high adsorptivity to molecular crystals.

[0210] For example, the screening method according to the second embodiment can comprehensively screen and select or design a new substance that strengthens hydrogen bonds on the surface of a specific molecular crystal from among a large number of compounds. Therefore, the screening method according to the second embodiment can easily compare the differences in adsorptivity to molecular crystals between different molecules, allowing molecular screening to be performed simply and in a short time.

[0211] Furthermore, the screening method according to the second embodiment can be executed using the machine learning potential 40 in at least one of the first energy calculation step (step S303), the second energy calculation step (step S304), the surface adsorption structure search step (step S305), and the structure optimization step of the surface adsorption structure (step S306). As a result, the screening method according to the second embodiment can shorten the process time required to obtain an accurate adsorption structure with a molecular crystal of each molecule, for example, when it is desired to find a molecule that exhibits a strong adsorption state due to hydrogen bonding from a large number of compound groups. Therefore, the screening method according to the second embodiment can shorten the screening time and can be applied to a wider molecular group by increasing its versatility.

[0212] In addition, in the screening method according to the second embodiment, the learning model generation process (step S310) may be performed after the third acquisition process (step S311) or simultaneously with the third acquisition process (step S311), and the order is not particularly limited.

[0213] In the above embodiment, the adsorbent is a molecular crystal and the adsorbate is a molecule that is adsorbed to the molecular crystal. However, the types of the adsorbent and adsorbate are not limited to this, and may be other types, for example, the adsorbent may be a metal, plastic, etc., and the adsorbate may be a molecular crystal or a molecule.

[0214] As described above, the embodiment has been described, but the above embodiment is presented as an example, and the present invention is not limited to the above embodiment. The above embodiment can be implemented in various other forms, and various combinations, omissions, substitutions, modifications, etc. can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are included in the scope of the invention and its equivalents described in the claims. EXAMPLES

[0215] The embodiments will be described in more detail below with reference to examples and comparative examples, but the embodiments are not limited to these examples and comparative examples.

[0216] <Example 1> (Preparation of screening device) As the screening device, a screening device 20A having the configuration shown in FIG. 6 was produced. (Creating surface adsorption structures) Using a screening device 20A having the configuration shown in FIG. 6, three types of molecular crystals (acetanilide, ε-caprolactam, and imidazole) and 2000 types of molecules to be adsorbed (molecules) were prepared. The 2000 types of molecules were randomly extracted from an existing database (QM9). In order to ensure the consistency of the results, the random number seed was fixed to an arbitrary value. 2000 types of molecules were adsorbed onto each of the three types of molecular crystals, creating a total of 6000 combinations of surface adsorption structures.

[0217] (Adsorption energy E ads (Calculation of Energy E of a molecular crystal A and the molecular energy EB and the energy of the surface adsorption structure E A+B and the adsorption energy E of the molecules of the surface adsorption structure created ads was calculated based on the following formula (I). E ads =E A+B -(E A +E B ) (I)

[0218] (Interaction index I int (Calculation of Interaction index I when focusing on hydrogen bonds of the surface adsorption structure created int was calculated based on the following formula (1).

[0219]

number

[0220] Interaction index I of 6000 surface adsorption structures int and the adsorption energy E ads The correlation diagram plotting the absolute value of the interaction index I int As the adsorption energy E ads It was confirmed that the absolute value of showed a tendency to increase and showed a positive correlation.

[0221] (Clustering) The distance between the molecular crystals and molecules in the created surface adsorption structures was calculated, and the top 100 with the shortest distance were extracted. Calculating the distance between the molecular crystals and molecules in 100 surface adsorption structures means that 100-dimensional variables were obtained. Next, the 100 dimensions were compressed to 2 dimensions using the t-SNE method, and clustering was performed using the k-means method, with molecules with similar properties being color-coded and grouped together. The number of clusters was 10. The clustering results are shown in Figure 18.

[0222] As shown in FIG. 18, it was confirmed that clustering makes it possible to extract candidate molecules as a group including multiple molecules, rather than extracting one specific molecule. Therefore, it can be said that clustering makes it possible to extract a group of molecules with strong hydrogen bonds or a group of molecules with weak hydrogen bonds. For example, it was confirmed that when one wishes to select a molecule with a high hydrogen bond to acetanilide, it is sufficient to select molecule group #3. In addition, as shown in FIG. 19, it was confirmed that multiple molecules belonging to molecule group #3 can be extracted by selecting molecule group #3.

[0223] It was also confirmed that by clustering, the degree of selectivity for each molecular crystal can be understood, as shown in Fig. 18. For example, when the molecular crystal is imidazole, a group of multiple molecules is concentrated in approximately one place, so that imidazole can be said to have low selectivity for the molecule.

[0224] Therefore, since the adsorptivity of each molecule can be confirmed for each molecular crystal, it can be said that it is easy to select molecules according to the type of molecular crystal.

[0225] <Example 2> (Preparation of screening device) As a screening device, a screening device 20B having a configuration shown in FIG. 10 was produced. (Creating surface adsorption structures) Using a screening device 20B having the configuration shown in Figure 10, 2000 types of molecules were adsorbed onto each of three types of molecular crystals in a manner similar to that of Example 1 above, creating a total of 6000 combinations of surface adsorption structures.

[0226] (Generating a trained model) Using the RDKit, we obtained 208 descriptors (number of functional groups, number of branches, charge, etc.) from the SMILES of 2000 molecules, and calculated the molecular structure descriptors and interaction index I int We learned the correlation with (objective variable) and generated a trained model. A gradient boosting regression tree was used for the trained model.

[0227] (Adsorption energy E ads (Calculation of As in Example 1, the energy E of the molecular crystal A and the molecular energy E B and the energy of the surface adsorption structure E A+B and the adsorption energy E of the molecules of the surface adsorption structure created ads was calculated. (Interaction index I int (Prediction of Using the generated trained model, the interaction index I of the descriptor of the adsorption structure of 2000 molecules was calculated. int The interaction index I int Calculated value of and interaction index I int The correlation between the predicted value of and the interaction index I int Calculated value of and interaction index I int A positive correlation was observed between the predicted values ​​of

[0228] The objective variable is the interaction index I int Instead, the adsorption energy E ads If a trained model is generated using the absolute value of int The adsorption energy E ads and the predicted adsorption energy E adsThe coefficient of determination between the absolute values ​​of was large. Therefore, the adsorption energy E ads Rather than using the absolute value of int It was confirmed that the use of the method can predict the adsorption and selectivity due to hydrogen bonds with greater accuracy.

[0229] (Interaction index I using a trained model int (Prediction of Using the generated trained model, the interaction index I of 20,000 molecules that were not included in the training data (2,000 molecules) used to generate the trained model with the surface of a molecular crystal (acetanilide) was calculated. int The SMILES of 20,000 molecules were randomly extracted from a public database (QM9) so as not to overlap with the training data, and the adsorption structure descriptors of 208 molecules were obtained using RDKit. The adsorption structure descriptors of 208 molecules were input into the trained model, and the interaction index I int The predicted value of the molecule number and the molecular interaction index I int The relationship between the predicted value and the actual value is shown in FIG.

[0230] As shown in Figure 21, the adsorption of each molecule to the surface of acetanilide was confirmed. Among them, the 10426th molecule had an interaction index of I int Therefore, it can be said that the 10426th molecule can be selected as a molecule that can increase the adsorption to acetanilide because it can strengthen the hydrogen bond to the surface of acetanilide.

[0231] Therefore, the screening device of this embodiment inputs the descriptor of the adsorption structure of the molecule into the trained model to obtain the interaction index I intSince the screening device of the present embodiment can rapidly screen for candidates for adsorbates having suitable adsorptivity for the surface of a target adsorbent, it can be said that the screening device of the present embodiment can be effectively used in the medical field, industrial field, etc., for screening suitable adsorbents according to the type of adsorbent.

[0232] The aspects of the embodiment of the present invention are as follows, for example. <1> A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of the adsorbate with the adsorbent of interest int A learning device having a learning unit that generates a trained model by learning using a learning dataset in which the and are associated with each other. <2> The interaction index I int is calculated using the following formula (1): <1> The learning device according to claim 1 .

number

number

number

number

number

[0233] 1 Learning device 2. Screening System 11, 201 First Acquisition Section 12, 202 Second Acquisition Section 13, 222 Third Acquisition Section 14. Training Data Set Creation Department 15 Learning Department 16, 213 Output section 20, 20A, 20B Screening device 30 Storage section 40 Machine Learning Potential 203 First Energy Calculation Unit 204 Second Energy Calculation Unit 205 Surface adsorption structure search section 206 Structural optimization of surface adsorption structure 207 Third Energy Calculation Section 208 Adsorption energy calculation section 209 Interaction Index Calculation Section 210 Relationship Diagram Creation Department 211 Clustering Department 212 Selection Department 221 Learning Model Generation Unit 223 Prediction Department 2031 First structural optimization of molecular crystals 2032 Surface cutout 2033 Second structural optimization of molecular crystals 2034 Molecular Crystal Energy Calculation 2041 Molecular Structure Optimization Department 2042 Molecular Energy Calculation

Claims

1. A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of interest between the adsorbate and the adsorbent ｉｎｔ A learning device having a learning unit that generates a trained model by learning using a learning dataset in which the and are associated with each other.

2. The interaction index I ｉｎｔ The learning device according to claim 1 , wherein is calculated by the following formula (1): [0010] (In the formula, I ｉｎｔ is the interaction index, A, R ０ is a constant, N is the number of intermolecular bonds of interest at the interface between the adsorbent and the surface of the adsorbate, and R ｉ is the distance of the intermolecular bond of interest at the interface between the adsorbate and the adsorbent on the surface.)

3. A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of interest between the adsorbate and the adsorbent ｉｎｔ A learning model generated using a learning dataset in which the and are associated.

4. The interaction index I ｉｎｔ The learning model according to claim 3, wherein is calculated by the following formula (1): [0025] (In the formula, I ｉｎｔ is the interaction index, A, R ０ is a constant, N is the number of intermolecular bonds of interest at the interface between the adsorbent and the surface of the adsorbate, and R ｉ is the distance of the intermolecular bond of interest at the interface between the adsorbate and the adsorbent on the surface.)

5. an acquisition unit that acquires a descriptor of a molecular structure of an adsorbent to be predicted and an adsorbate to be adsorbed by the adsorbent to be predicted; A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of interest between the adsorbate and the adsorbent ｉｎｔ Using a trained model generated using a training data set in which the above-mentioned and the above-mentioned are associated with each other, the interaction index I of the intermolecular bond of interest between the adsorbate to be predicted and the adsorbent to be predicted is calculated. ｉｎｔ A prediction unit for predicting A screening device comprising:

6. The interaction index I ｉｎｔ The screening device according to claim 5 , wherein is determined by the following formula (1): [0030] (In the formula, I ｉｎｔ is the interaction index, A, R ０ is a constant, N is the number of intermolecular bonds of interest at the interface between the adsorbent and the surface of the adsorbate, and R ｉ is the distance of the intermolecular bond of interest at the interface between the adsorbate and the adsorbent on the surface.)

7. The energy E of the adsorbent Ａ A first energy calculation unit that calculates The energy E of the adsorbate Ｂ A second energy calculation unit that calculates The energy E of the surface adsorption structure in which the adsorbate is adsorbed on the adsorbent Ａ＋Ｂ A third energy calculation unit that calculates 6. The screening device according to claim 5, further comprising:

8. The energy E of the adsorbent Ａ and the energy E of the adsorbate Ｂ and the energy E of the surface adsorption structure Ａ＋Ｂ From this, the adsorption energy E ａｄｓ The screening device according to claim 7, further comprising an adsorption energy calculation unit that calculates:

9. The interaction index I ｉｎｔ The screening device according to claim 8, further comprising an interaction index calculation unit that calculates:

10. The molecular structure descriptor of the adsorbate used in the second energy calculation unit and the interaction index I calculated by the interaction index calculation unit ｉｎｔ The screening device according to claim 9 , further comprising a learning model generation unit that generates the trained model using the above.

11. The screening device according to claim 7 , wherein at least one of the first energy calculation unit, the second energy calculation unit, and the third energy calculation unit is implemented using a machine learning potential.

12. 7. The screening device according to claim 5, wherein one of the adsorbent and the adsorbate is a molecular crystal.

13. the adsorbent is a molecular crystal; 7. The screening device according to claim 5, wherein the adsorbate is a molecule.

14. On the computer, an acquisition step of acquiring a descriptor of the molecular structure of a prediction target adsorbent and a prediction target adsorbate to be adsorbed by the adsorbent; A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of interest between the adsorbate and the adsorbent ｉｎｔ Using a trained model generated using a training data set in which the above-mentioned and the above-mentioned are associated with each other, the interaction index I of the intermolecular bond of interest between the adsorbate to be predicted and the adsorbent to be predicted is calculated. ｉｎｔ A prediction step of predicting A screening program that executes the following:

15. The computer an acquisition step of acquiring a descriptor of the molecular structure of a prediction target adsorbent and a prediction target adsorbate to be adsorbed by the adsorbent; A descriptor of the molecular structure of the adsorbate to be adsorbed on the adsorbent and an interaction index I of the intermolecular bond of interest between the adsorbate and the adsorbent ｉｎｔ Using a trained model generated using a training data set in which the above-mentioned and the above-mentioned are associated with each other, the interaction index I of the intermolecular bond of interest between the adsorbate to be predicted and the adsorbent to be predicted is calculated. ｉｎｔ A prediction step of predicting A screening method for performing:

Images