Information processing device and method

Abstract

To support a search for an aroma component indicating an olfactory property identical or similar to a predetermined olfactory property.SOLUTION: A method is provided in which a computer executes the steps of: specifying a property parameter group of one or more olfactory property values; obtaining a feature amount of the property parameter group; applying a prediction model having learned a correlation between the olfactory property value of an aroma component and a configuration parameter of the aroma component, to the feature amount, thereby calculating a configuration parameter of an aroma component that is predicted to indicate an olfactory property corresponding to the property parameter group.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present disclosure relates to an information processing device, a method, a program, and a system. [Background technology]

[0002] When an aroma component with known olfactory properties (e.g., aroma descriptors, olfactory receptor (OR) activity properties, or a combination thereof) has problems such as ease of handling, availability, or price, it may be considered to substitute another aroma component with equivalent olfactory properties. However, since there are a huge number of different aroma components, it is not realistic from the standpoint of time and cost to blindly conduct sensory evaluations of aromas by experts or OR assays until the desired aroma component is discovered.

[0003] Patent Document 1 discloses a technical idea for determining a combination of multiple candidate substances that reproduces the aroma of a target substance. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2021 / 235544 Summary of the Invention [Problem to be solved by the invention]

[0005] The technical idea of ​​Patent Document 1 requires information on the maximum response intensity and EC50 of each candidate substance for each olfactory receptor. In other words, it is not envisaged to handle candidate substances for which this information is unknown (unmeasured).

[0006] The purpose of the present disclosure is to aid in the search for aroma components that exhibit the same or similar olfactory characteristics as desired. [Means for solving the problem]

[0007] A method according to one aspect of the present disclosure includes a step in which a computer acquires features relating to a characteristic parameter group, which is one or more olfactory characteristic values, and a step in which a predictive model that has learned the correlation between the olfactory characteristic values ​​of an aroma component and the constituent parameters of the aroma component is applied to the features to calculate the constituent parameters of the aroma component that are expected to exhibit olfactory characteristics corresponding to the characteristic parameter group. [Brief description of the drawings]

[0008] [Figure 1] 1 is a block diagram showing a configuration of an information processing system according to an embodiment of the present invention; [Diagram 2] FIG. 2 is a block diagram showing a configuration of a client device according to the present embodiment. [Diagram 3] FIG. 2 is a block diagram showing a configuration of a server according to the present embodiment. [Figure 4] FIG. 1 is an explanatory diagram of one aspect of the present embodiment. [Diagram 5] FIG. 4 is a diagram showing a data structure of a molecular property value table according to the present embodiment. [Figure 6] FIG. 2 is a diagram showing an overall flow of information processing according to the present embodiment. [Figure 7] FIG. 2 is an explanatory diagram of a prediction model according to the present embodiment. [Figure 8] 11A to 11C are diagrams illustrating examples of screens displayed in the information processing of the present embodiment. [Figure 9] FIG. 13 is an explanatory diagram of a prediction model according to the first modified example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In the drawings for explaining the embodiment, the same components are generally designated by the same reference numerals, and the repeated description will be omitted.

[0010] (1) Information Processing System Configuration The configuration of the information processing system will be described below. Fig. 1 is a block diagram showing the configuration of the information processing system according to the present embodiment.

[0011] As shown in FIG. 1, the information processing system 1 includes a client device 10 and a server 30. The client device 10 and the server 30 are connected via a network (eg, the Internet or an intranet) NW.

[0012] The client device 10 is an example of an information processing device that transmits a request to the server 30. The client device 10 is, for example, a smartphone, a tablet terminal, or a personal computer.

[0013] The server 30 is an example of an information processing device that provides the client device 10 with a response in response to a request transmitted from the client device 10. The server 30 is, for example, a server computer.

[0014] (1-1) Client device configuration The configuration of the client device will now be described with reference to Fig. 2, which is a block diagram showing the configuration of the client device of this embodiment.

[0015] 2, the client device 10 includes a storage device 11, a processor 12, an input / output interface 13, and a communication interface 14. The client device 10 is connected to a display 21.

[0016] The storage device 11 is configured to store programs and data. The storage device 11 is, for example, a combination of a Read Only Memory (ROM), a Random Access Memory (RAM), and a storage (for example, a flash memory or a hard disk).

[0017] The programs include, for example, the following programs: ·OS (Operating System) programs Applications that process information (e.g. web browsers)

[0018] The data includes, for example, the following data: Databases referenced in information processing Data obtained by performing information processing (i.e., the results of performing information processing)

[0019] The processor 12 is a computer that realizes the functions of the client device 10 by running a program stored in the storage device 11. The processor 12 is, for example, at least one of the following: ·CPU(Central Processing Unit) ·GPU(Graphic Processing Unit) ·ASIC(Application Specific Integrated Circuit) ·FPGA(Field Programmable Gate Array)

[0020] The input / output interface 13 is configured to obtain information (eg, a user's instruction) from an input device connected to the client device 10, and to output information (eg, an image) to an output device connected to the client device 10.

[0021] The input device is, for example, a keyboard, a pointing device, a touch panel, or a combination thereof. The output device is, for example, a display 21, a speaker, or a combination thereof.

[0022] The communication interface 14 is configured to control communications between the client device 10 and an external device (eg, a server 30).

[0023] The display 21 is configured to display an image (still image or moving image). The display 21 is, for example, a liquid crystal display or an organic EL display.

[0024] (1-2) Server configuration The configuration of the server will now be described with reference to Fig. 3, which is a block diagram showing the configuration of the server according to this embodiment.

[0025] As shown in FIG. 3, the server 30 includes a storage device 31, a processor 32, an input / output interface 33, and a communication interface .

[0026] The storage device 31 is configured to store programs and data, and is, for example, a combination of a ROM, a RAM, and a storage (for example, a flash memory or a hard disk).

[0027] The programs include, for example, the following programs: -OS programs Application programs that perform information processing

[0028] The data includes, for example, the following data: Databases referenced in information processing Results of information processing

[0029] The processor 32 is a computer that realizes the functions of the server 30 by starting a program stored in the storage device 31. The processor 32 is, for example, at least one of the following. ·CPU GPU ·ASIC FPGA

[0030] The input / output interface 33 is configured to obtain information (eg, a user's instruction) from an input device connected to the server 30, and to output information (eg, an image) to an output device connected to the server 30.

[0031] The input device is, for example, a keyboard, a pointing device, a touch panel, or a combination thereof. The output device is, for example, a display.

[0032] The communication interface 34 is configured to control communications between the server 30 and external devices (eg, the client device 10).

[0033] (2) One aspect of the embodiment An embodiment of the present invention will now be described with reference to Fig. 4, which is an explanatory diagram of an embodiment of the present invention.

[0034] As shown in Fig. 4, in response to an instruction from a user US1, the client device 10 acquires a group of characteristic parameters that are the values ​​of one or more scent descriptors desired by the user US1 (i.e., the presence or absence or degree of relevance to one or more scent descriptors). A scent descriptor is an example of an olfactory characteristic. As an example, the user US1 specifies to the client device 10 the scent descriptor values ​​desired for the alternative target scent component.

[0035] The client device 10 transmits a search request for aroma components (hereinafter, referred to as "candidate aroma components") that are estimated to indicate aroma descriptors corresponding to the characteristic parameter group to the server 30. The search request includes the characteristic parameter group.

[0036] The server 30 acquires a plurality of feature amounts related to the group of characteristic parameters based on the search request received from the client device 10. Note that instead of the server 30, the client device 10 may extract a plurality of feature amounts from the group of characteristic parameters (which may include preprocessing such as standardization or normalization) and transmit the plurality of feature amounts to the server 30.

[0037] The server 30 applies the prediction model LM11 to the acquired multiple feature quantities. As a result, the server 30 estimates molecular property values ​​of aroma components indicating aroma descriptors corresponding to the characteristic parameter group (one example of "constituent parameters"). The prediction model LM11 corresponds to a trained model (classifier) ​​that has learned the correlation between the aroma descriptor values ​​of aroma components and the molecular property values ​​of the aroma components. The characteristic parameter group corresponds to the input parameters of the prediction model LM11, and the constituent parameters (molecular property values ​​in this embodiment) correspond to the output parameters of the prediction model LM11.

[0038] The server 30 searches a molecular property table TB12, which will be described later, for one or more aroma components that have molecular property values ​​that are the same as or similar to the estimated molecular property value.

[0039] The server 30 sets the aroma components found by the search as candidate aroma components and transmits information about the candidate aroma components to the client device 10. The client device 10 presents the received information about the candidate aroma components to the user US1. This allows the user US1 to quickly and at low cost obtain information about aroma components that are expected to exhibit aroma descriptors corresponding to the characteristic parameter group (i.e., a desired) such as aroma components that have the potential to replace existing aroma components.

[0040] (3) Database The database of this embodiment will now be described. Tables constituting the database of this embodiment are stored in the storage device 31.

[0041] (3-1) Molecular property table The molecular property value table of this embodiment will be described below with reference to Fig. 5, which is a diagram showing the data structure of the molecular property value table of this embodiment.

[0042] In the molecular property value table, information on a plurality of molecular property values ​​of each of a plurality of aroma components that may be included in the search range (hereinafter, referred to as "molecular property value information") is registered.

[0043] 5, the molecule physical property table includes an "aroma component name" field and N (N is an integer equal to or greater than 2) "physical property" fields. Each field is associated with the others.

[0044] The "aroma component name" field stores information about the name of the aroma component corresponding to the record (hereinafter referred to as "aroma component name information").

[0045] The N "physical property" fields store N types of physical property information of the aroma components corresponding to the record. Each piece of physical property information can be identified, for example, by executing software (e.g., PaDEL-descriptor) that calculates the molecular physical property values ​​of a compound.

[0046] (4) Information processing The information processing of this embodiment will be described. Fig. 6 is a diagram showing an overall flow of the information processing of this embodiment. Fig. 7 is an explanatory diagram of a prediction model of this embodiment. Fig. 8 is a diagram showing an example of a screen displayed in the information processing of this embodiment.

[0047] The information processing of this embodiment may begin, for example, when the client device 10 launches an application for searching for aroma components that are estimated to exhibit a desired fragrance descriptor, or when the client device 10 accesses a web page that provides SaaS (Software as a Service) for such a search.

[0048] As shown in FIG. 6, the client device 10 acquires a group of characteristic parameters (S110). Specifically, the client device 10 acquires a set of characteristic parameters, which are values ​​of one or more scent descriptors. The set of characteristic parameters may include the presence or absence or the degree of one or more scent descriptors.

[0049] Here, the aroma descriptors correspond to evaluation items in the sensory evaluation of aroma components, and specific examples of aroma descriptors include floral, animal, sweet, fresh, roasted, celery, green, etc. There is no particular limit to the number of types of aroma descriptors, but 138 types of aroma descriptors were used in the examples described below.

[0050] As a first example of acquiring a characteristic parameter group (S110), the client device 10 accepts, from a user, designation of, for example, the values ​​of descriptors of fragrances constituting the characteristic parameter group. For example, the client device 10 accepts input of the characteristic parameter group through an object (e.g., an input form) arranged on a UI (User Interface) screen displayed on the display 21. The client device 10 then acquires the characteristic parameter group by analyzing the input contents of this object. Alternatively, the client device 10 acquires the characteristic parameter group by selecting a file in which the characteristic parameter group is recorded in response to a user instruction and analyzing the contents of the file.

[0051] As a second example of acquiring a group of characteristic parameters (S110), the client device 10 accepts a designation of an aroma component (hereinafter referred to as a "target aroma component") from the user. Then, the client device 10 identifies the value of an aroma descriptor indicated by the target aroma component designated by the user. As an example, the client device 10 identifies the value of an aroma descriptor indicated by the target aroma component by referring to a database in which information describing the relationship between a plurality of aroma components and the values ​​of their aroma descriptors is registered. Alternatively, the client device 10 may identify the value of an aroma descriptor indicated by the target aroma component using a trained model. The client device 10 acquires the value of the identified aroma descriptor as a group of characteristic parameters.

[0052] The second example of acquiring the characteristic parameter group (S110) may be executed by the client device 10 and the server 30 in cooperation with each other. That is, the client device 10 may receive a target aroma component designated by the user and notify the server 30 of the target aroma component, and the server 30 may acquire the characteristic parameter group based on the target aroma component. In this case, the server 30 extracts feature amounts from the acquired characteristic parameter group in step S111 described below.

[0053] The term "fragrance component" as used herein refers to a volatile compound whose smell is sensorily detected by humans. Specific examples of the fragrance component include limonene and ethyl acetate.

[0054] After step S110, the client device 10 executes a search request (S111). Specifically, the client device 10 generates a search request for the candidate aroma components based on the characteristic parameters acquired in step S110. The search request may include the characteristic parameters acquired in step S110, or may include other information related to the characteristic parameters (e.g., feature quantities or information on aroma components specified by the user). The client device 10 transmits the generated search request to the server 30.

[0055] After step S111, the server 30 executes acquisition of the feature amount (S130). Specifically, the server 30 receives the search request transmitted in step S111. The server 30 acquires a feature amount based on the search request.

[0056] As a first example of acquiring the feature amount (S130), the server 30 extracts a group of characteristic parameters included in the search request, and extracts the feature amount from the group of characteristic parameters using a predetermined algorithm.

[0057] As a second example of acquiring the feature amount (S130), the server 30 extracts the feature amount included in the search request. That is, in this example, the client device 10 extracts the feature amount.

[0058] After step S130, the server 30 executes estimation of molecular property values ​​(S131). Specifically, the server 30 estimates the molecular property value by applying the prediction model LM11 to the feature amount acquired in step S130. The estimation result may be, for example, a vector consisting of numerical values ​​each representing one or more molecular property values.

[0059] The predictive model LM11 can be constructed by machine learning about the correlation between the values ​​of the scent descriptors (an example of "olfactory characteristics") of aroma components and the molecular property values ​​of the aroma components.

[0060] As an example, as shown in FIG. 7, the prediction model LM11 is constructed by performing supervised learning on the untrained model LM21 using a plurality of sets of learning data each including a learning feature related to an aroma descriptor of each of a plurality of aroma components and a correct answer data corresponding to the learning feature. The machine learning algorithm is not particularly limited, but a random forest is used in the embodiment described later. In this learning data, the correct answer data corresponding to the learning feature represents the molecular property value of the aroma component from which the feature is obtained. The aroma descriptor of the aroma component from which the learning feature is obtained may be identified, for example, by performing a sensory evaluation of the aroma of each aroma component, or may be set based on known information (e.g., academic papers, past experimental results, aroma descriptor database, etc.). The molecular property value of the aroma component from which the learning feature is obtained may be identified, for example, by executing software for calculating the molecular property value of a compound for each aroma component, or may be set based on known information (e.g., academic papers, past experimental results, molecular property value database, etc.). There is no particular limit to the number of types of aroma components used in learning, but in the embodiment described below, 1271 types of aroma components were used.

[0061] Moreover, the molecular property value is information on the physical properties estimated from the molecules of the aroma components. In particular, the molecular property value refers to a parameter specific to each molecule that can be understood or calculated from structural information of the molecule, such as the number of constituent elements, partial structure, or LogKow (octanol / water partition coefficient). In addition, for example, the number of atoms, fingerprint, logarithmic value of hydrophobicity (Log(hydrophobicity)), solubility, etc. are also specific examples of molecular property values. In addition, the number of types of molecular property values ​​is not particularly limited, but 1503 types of molecular property values ​​were used in the examples described later.

[0062] After step S131, the server 30 executes a search for candidate aroma components (S132). Specifically, the server 30 specifies a search range. Here, the search range may be all aroma components registered in the molecular property table (FIG. 5), or may be some of the aroma components. As an example, the user may specify, to the client device 10, the conditions for aroma components to be included / not included in the search range (e.g., the presence or absence of a specific functional group, molecular weight, etc.) and aroma components to be included / excluded in the search range, and the server 30 may specify the search range according to the user's instructions.

[0063] The server 30 compares the molecular property values ​​of each of the aroma components included in the specified search range with the molecular property value estimated in step S131. As an example, the server 30 calculates the similarity (e.g., the distance between two vectors) between the molecular property value estimated in step S131 and the molecular property values ​​of each aroma component included in the search range, or the dispersion of the difference or ratio (e.g., the variance of the difference or ratio between corresponding components of two vectors). Specifically, the server 30 may calculate the similarity between the molecular property value estimated in step S131 and the molecular property value of each aroma component included in the search range. The similarity can be expressed based on, for example, the distance between the molecular property values, the difference or ratio of the values ​​for each molecular property, or a combination thereof. Alternatively, the server 30 may calculate the difference or ratio of the values ​​for each molecular property between the molecular property value estimated in step S131 and the molecular property value of each aroma component included in the search range, and calculate the dispersion of the difference or ratio. The dispersion may be, for example, the variance, the standard deviation, or the like.

[0064] The server 30 treats aroma components that satisfy a predetermined standard for at least one of the similarity and dispersion as candidate aroma components.

[0065] After step S132, the server 30 outputs the candidate aroma components (S133). Specifically, the server 30 transmits information on the candidate aroma components obtained in step S131 to the client device 10.

[0066] After step S133, the client device 10 executes screen display (S112). Specifically, the client device 10 receives the information on the candidate aroma components (e.g., component names and aroma descriptors) transmitted in step S133. The client device 10 displays the information on the candidate aroma components on the display 21. For example, the client device 10 may display the screen of FIG. 8 on the display 21.

[0067] The screen of FIG. 8 includes an object J20. Object J20 displays the search results for candidate aroma components. In the example of Fig. 8, object J20 displays information (e.g., component name and aroma descriptor) of an aroma component (quinoline) specified by the user, and information (e.g., component name and aroma descriptor) of candidate aroma components (cyclohexanone and 3-methylcyclohexanone) estimated to have aroma descriptors corresponding to the characteristic parameters identified from the aroma component.

[0068] The notification of the search result may be made on an app screen or a website screen, or may be made by email or other messages. The notification may include, for example, a notice by posting a message on an electronic bulletin board, a chat tool, or an SNS in a format that can be viewed by specific or unspecified people. Alternatively, the estimation result may be stored in a file, for example in CSV format, and provided to the user.

[0069] (5) Examples An example of this embodiment will be described. The inventors selected 1271 types of aroma components that can be used in food as training aroma components. For each aroma component, the inventors specified an aroma descriptor as a feature. This aroma descriptor (feature) indicates whether or not the component corresponds to 138 aroma descriptors, including sweet, fresh, roasted, celery, etc. The inventors selected the 138 aroma descriptors from a publicly known database in which information on the relationship between aroma components and aroma descriptors is registered, from the perspective of the number of corresponding aroma components.

[0070] The inventors also used PaDEL-Descriptor to calculate approximately 2000 molecular property values, including the number of constituent elements, partial structure, LogKow (octanol / water partition coefficient), number of atoms, fingerprints, etc., for each training aroma component. The inventors selected 1503 molecular property values ​​by excluding those with missing values ​​in any of the aroma components and those with values ​​of 0 across all aroma components. The inventors used all molecular property values ​​for which scent descriptors were obtained as feature quantities as ground truth data corresponding to the feature quantities. The inventors used random forests to perform supervised learning using training data based on the above 1271 aroma components, and constructed a prediction model.

[0071] The inventors provided the constructed prediction model with inputs of features based on the quinoline aroma descriptors. As a result, cyclohexanone was found to be the aroma component with the highest similarity, and 3-methylcyclohexanone was found to be the aroma component with the lowest dispersion. The inventors carried out an aroma sensory evaluation of the found aroma components (cyclohexanone and 3-methylcyclohexanone) and confirmed that both of these aroma components have an aroma similar to quinoline. Furthermore, according to the results of this aroma sensory evaluation, 3-methylcyclohexanone has the same aroma descriptor "medicinal" as quinoline.

[0072] (6) Summary As described above, the server 30 of the present embodiment identifies a characteristic parameter group, which is the value of one or more aroma descriptors, and acquires feature quantities related to the characteristic parameter group. The server 30 calculates molecular property values ​​of aroma components expected to show aroma descriptors corresponding to the characteristic parameter group by applying a prediction model that has learned the correlation between the aroma descriptor values ​​of aroma components and the molecular property values ​​of the aroma components to the feature quantities. This makes it possible to estimate molecular property values ​​of aroma components expected to show aroma descriptors identical or similar to a desired aroma descriptor in a short time and at low cost. In other words, it is possible to assist in the search for aroma components that show aroma descriptors identical or similar to a desired aroma descriptor.

[0073] The server 30 may search for aroma components based on the calculated molecular property values ​​and output information on the searched aroma components, allowing the user to obtain information on aroma components that are expected to show the same or similar aroma descriptor as the desired aroma descriptor.

[0074] The server 30 may calculate at least one of the similarity or the dispersion of the difference or ratio between the molecular property values ​​of multiple aroma components included in the search range and the calculated molecular property values, and search for aroma components for which at least one of the calculated similarity or dispersion satisfies a predetermined criterion, thereby obtaining aroma components that are expected to show aroma descriptors that are closer to the desired aroma descriptor.

[0075] In response to the designation of an aroma component whose value of the aroma descriptor is specified, the server 30 may specify the value of the aroma descriptor indicated by the aroma component as a characteristic parameter group. In this way, even if the user does not fully understand the aroma descriptors indicated by the aroma component, the characteristic parameter group can be specified by simply designating the aroma component, and information on the aroma component expected to show the aroma descriptor corresponding to the characteristic parameter group can be obtained.

[0076] The prediction model may be constructed by supervised learning using a plurality of sets of training data each including training features related to training aroma descriptor values ​​and ground truth data corresponding to the training features. The ground truth data corresponding to the training features may represent molecular property values ​​of the aroma components from which the features are obtained. This enables learning-based estimation of the correlation between the aroma descriptors indicated by the aroma components and the molecular property values ​​of the aroma components.

[0077] The training features may represent at least one of the results of an aroma sensory evaluation of one or more aroma descriptors of the aroma components from which the features are obtained, thereby enabling a prediction model to be built using high-quality training data based on aroma sensory evaluations by experts.

[0078] (7) Variations A modification of this embodiment will now be described.

[0079] (7-1) Variation 1 Modification 1 will be described. Modification 1 is an example in which human olfactory receptor (hereinafter simply referred to as "olfactory receptor" or "OR") activity characteristics are used as olfactory characteristics instead of scent descriptors. Note that this embodiment and modification 1 can be combined. In other words, a combination of scent descriptors and olfactory receptor activity characteristics can also be used as olfactory characteristics.

[0080] A part of the explanation regarding Modification 1 can be understood by replacing "descriptor of aroma" in the explanation of this embodiment with "olfactory receptor activity characteristic." Other matters will be described in the following explanation.

[0081] (7-1-1) Information processing A description will be given of information processing according to Modification 1. Fig. 9 is an explanatory diagram of a prediction model according to Modification 1.

[0082] The information processing of variant 1 may begin, for example, when client device 10 launches an application for searching for aroma components that are suspected to exhibit desired olfactory characteristics (i.e., olfactory receptor activity characteristics), or when client device 10 accesses a web page that provides SaaS for such searching.

[0083] As shown in FIG. 6, the client device 10 acquires a group of characteristic parameters (S110). Specifically, the client device 10 acquires a group of characteristic parameters that are one or more olfactory receptor activity characteristics. The group of characteristic parameters may include the presence or absence or the degree of one or more OR activity characteristics.

[0084] Here, for example, OR1C1, OR10G7, OR1A1, OR2J3, OR4Q3, etc. are specific examples of olfactory receptors. The third character (number) of the string representing the olfactory receptor represents the olfactory receptor family, the fourth character (alphabet) represents the subfamily, and the fifth character (number) represents the identification value of each receptor within the subfamily. The number of types of olfactory receptors is not particularly limited, but it is expected that, for example, about 103 types of olfactory receptors will be used.

[0085] After step S110, the client device 10 executes a search request (S111) in the same manner as in the present embodiment.

[0086] After step S111, the server 30 executes acquisition of feature amounts (S130) in the same manner as in the present embodiment.

[0087] After step S130, the server 30 executes estimation of molecular property values ​​(S131). Specifically, the server 30 estimates the molecular property value by applying the prediction model LM11 to the feature amount acquired in step S130. The estimation result may be, for example, a vector consisting of numerical values ​​each representing one or more molecular property values.

[0088] As in this embodiment, the prediction model LM11 can be constructed by machine learning about the correlation between the olfactory receptor activity characteristics of an aroma component (an example of an "olfactory characteristic") and the molecular property values ​​of the aroma component (an example of a "constituent parameter").

[0089] As an example, as shown in FIG. 9, a prediction model LM11 is constructed by performing supervised learning on an untrained model LM21 using a plurality of sets of learning data each including a learning feature related to the olfactory receptor activity characteristic of each of a plurality of aroma components and ground truth data corresponding to the learning feature. The machine learning algorithm is not particularly limited, but it is assumed that a random forest is used, for example. In this learning data, the ground truth data corresponding to the learning feature represents the molecular property values ​​of the aroma component from which the feature is obtained. The olfactory receptor activity characteristic of the aroma component from which the learning feature is obtained may be identified, for example, by performing an OR assay on each aroma component, or may be set based on known information (e.g., academic papers, past experimental results, an olfactory receptor activity characteristic database, etc.).

[0090] After step S131, the server 30 performs a search for candidate aroma components (S132) in the same manner as in this embodiment. Specifically, the server 30 specifies a search range. Here, the search range may be all aroma components registered in the molecular property table (FIG. 5), or may be some of the aroma components. As an example, the user may specify, to the client device 10, the conditions for aroma components to be included / not included in the search range (e.g., the presence or absence of a specific functional group, molecular weight, etc.) and aroma components to be included / excluded in the search range, and the server 30 may specify the search range according to the user's instructions.

[0091] After step S132, the server 30 outputs the candidate aroma components (S133) in the same manner as in this embodiment.

[0092] After step S133, the client device 10 executes screen display (S112). Specifically, the client device 10 receives the information on the candidate aroma components (for example, component names and olfactory receptor activity characteristics) transmitted in step S133. The client device 10 displays the information on the candidate aroma components on the display 21.

[0093] (7-1-2) Summary As described above, the server 30 of the first modification specifies a characteristic parameter group that is one or more olfactory receptor activity characteristics, and acquires feature quantities related to the characteristic parameter group. The server 30 calculates molecular property values ​​of aroma components that are expected to exhibit olfactory characteristics corresponding to the characteristic parameter group by applying a prediction model that has learned the correlation between the olfactory receptor activity characteristics of aroma components and the molecular property values ​​of the aroma components to the feature quantities. This makes it possible to estimate molecular property values ​​of aroma components that are expected to exhibit olfactory characteristics identical or similar to desired olfactory characteristics in a short time and at low cost. In other words, it is possible to assist in the search for aroma components that exhibit olfactory characteristics identical or similar to desired olfactory characteristics.

[0094] The training features may represent at least one of the results of one or more olfactory receptor assays of the aroma components from which the features were obtained, making it possible to build a prediction model using high-quality training data based on the olfactory receptor assays.

[0095] (7-2) Variation 2 Modification 2 is an example in which, in this embodiment, olfactory receptor activity characteristics are used as configuration parameters instead of molecular property values.

[0096] Specifically, the server 30 of the second modification specifies a characteristic parameter group, which is the value of one or more aroma descriptors, and acquires feature quantities related to the characteristic parameter group. The server 30 calculates the olfactory receptor activity characteristic of an aroma component that is expected to show an aroma descriptor corresponding to the characteristic parameter group by applying a prediction model that has learned the correlation between the aroma descriptor value of an aroma component and the olfactory receptor activity characteristic of the aroma component to the feature quantities. This makes it possible to estimate the olfactory receptor activity characteristic of an aroma component that is expected to show an aroma descriptor identical or similar to a desired aroma descriptor in a short time and at low cost. In other words, it is possible to assist in the search for aroma components that show an aroma descriptor identical or similar to a desired aroma descriptor.

[0097] The server 30 may also search for aroma components based on the calculated olfactory receptor activity characteristics and output information on the searched aroma components. This allows the user to obtain information on aroma components that are expected to show an aroma descriptor identical or similar to a desired aroma descriptor. In order to perform such processing, in the second modification, a table can be constructed and used in which information on a plurality of olfactory receptor activity characteristics of each of a plurality of aroma components that may be included in the search range is registered.

[0098] Furthermore, the prediction model used in Modification 2 may be constructed using supervised learning using multiple sets of training data each including training features related to the values ​​of training aroma descriptors and ground truth data corresponding to the training features. The ground truth data corresponding to the training features may represent the olfactory receptor activity profile of the aroma component from which the features were obtained. This enables learning-based estimation of the correlation between the aroma descriptors indicated by the aroma components and the olfactory receptor activity profile of the aroma components.

[0099] (7-3) Variation 3 Variation 3 is an example in which, in variation 1, scent descriptors are used as constituent parameters instead of molecular property values.

[0100] Specifically, the server 30 of the third modification specifies a group of characteristic parameters that are one or more olfactory receptor activity characteristics, and acquires feature quantities related to the group of characteristic parameters. The server 30 calculates the scent descriptors of the aroma components that are expected to exhibit olfactory receptor activity characteristics corresponding to the group of characteristic parameters by applying a prediction model that has learned the correlation between the olfactory receptor activity characteristics of the aroma components and the scent descriptors of the aroma components to the feature quantities. This makes it possible to estimate the scent descriptors of the aroma components that are expected to exhibit the same or similar olfactory receptor activity characteristics as the desired olfactory receptor activity characteristics in a short time and at low cost. In other words, it is possible to assist in the search for aroma components that exhibit the same or similar olfactory receptor activity characteristics as the desired olfactory receptor activity characteristics.

[0101] The server 30 may also search for aroma components based on the calculated aroma descriptors and output information on the searched aroma components. This allows the user to obtain information on aroma components that are expected to exhibit the same or similar olfactory receptor activity characteristics as the desired olfactory receptor activity characteristics. In order to perform such processing, in the third modification, a table can be constructed and used in which information on a plurality of aroma descriptors for each of a plurality of aroma components that may be included in the search range is registered.

[0102] Furthermore, the prediction model used in Modification 3 may be constructed using supervised learning using multiple sets of training data each including a training feature related to a training olfactory receptor activity characteristic and ground truth data corresponding to the training feature. The ground truth data corresponding to the training feature may represent an aroma descriptor of the aroma component from which the feature was obtained. This enables learning-based estimation of the correlation between the olfactory receptor activity characteristic exhibited by an aroma component and the aroma descriptor of the aroma component.

[0103] (8) Other modifications The storage device 11 may be connected to the client device 10 via a network NW. The display 21 may be integrated into the client device 10. The storage device 31 may be connected to the server 30 via the network NW.

[0104] In the above description, an example has been shown in which the server 30 searches for candidate aroma components based on molecular property values ​​estimated by the prediction model LM11 and outputs the search results to the client device 10. However, the server 30 may be configured to output the molecular property values ​​estimated by the prediction model LM11 to the client device 10. In this case, the client device 10 may search for candidate aroma components using a rule base or a trained model, or a user may search for candidate aroma components.

[0105] An example in which the information processing system of the embodiment is implemented by a client / server type system has been shown. However, the information processing system of the embodiment can also be implemented by a stand-alone type computer or a peer-to-peer type system. For example, the client device 10 can execute all or part of the processes of steps S130 to S133. In this case, the prediction model LM11 can be stored in the storage device 11.

[0106] Each step of the above information processing can be executed by either the client device 10 or the server 30. For example, a process described as being performed by one device may be executed by another device, and a process described as being performed by the interaction of multiple devices may be executed by a single device. In addition, although the above description shows an example in which each step is executed in a specific order in each process, the execution order of each step is not limited to the described example as long as there is no dependency between the steps.

[0107] The configuration parameters may be determined in two stages. As an example, by performing learning by reversing the roles of the learning feature quantities of the prediction model of Modification 1 and the corresponding correct answer data, it is possible to construct a model that predicts the olfactory receptor activity characteristics of an aroma component that is expected to have a specific molecular property value. By using such a prediction model in combination with the prediction model of this embodiment, it is possible to predict the molecular property values ​​of an aroma component that is expected to show a desired aroma descriptor, and further to predict the olfactory receptor activity characteristics of the aroma component that is expected to have the molecular property value. Then, the aroma component may be searched for based on the olfactory receptor activity characteristics, and information on the searched aroma component may be output. As a result, in this embodiment and Modification 2, information on an aroma component that is expected to show an aroma descriptor identical or similar to a desired aroma descriptor can be obtained by a different approach.

[0108] As another example, by performing learning by reversing the roles of the learning feature quantities of the prediction model of this embodiment and the corresponding ground truth data, it is possible to construct a model that predicts the odor descriptors of aroma components that are expected to have specific molecular property values. By using such a prediction model in combination with the prediction model of Modification 1, it is possible to predict the molecular property values ​​of aroma components that are expected to exhibit desired olfactory receptor activity characteristics, and further to predict the odor descriptors of the aroma components that are expected to have the molecular property values. Then, aroma components may be searched for based on the odor descriptors, and information on the searched aroma components may be output. In this way, information on aroma components that are expected to exhibit the same or similar olfactory receptor activity characteristics as desired olfactory receptor activity characteristics can be obtained by an approach different from Modifications 1 and 3.

[0109] Although the embodiment of the present invention has been described in detail above, the scope of the present invention is not limited to the above embodiment. Furthermore, the above embodiment can be improved or modified in various ways without departing from the spirit of the present invention. Furthermore, the above embodiment and the modified examples can be combined. [Explanation of symbols]

[0110] 1: Information processing system 10: Client device 11:Storage device 12: Processor 13: Input / Output Interface 14: Communication interface 21: Display 30: Server 31:Storage device 32 : Processor 33: Input / Output Interface 34: Communication interface

Claims

1. The computer Identifying a set of characteristic parameters, which are one or more olfactory characteristic values; acquiring a feature value relating to the group of characteristic parameters; calculating constituent parameters of aroma components that are expected to exhibit olfactory characteristics corresponding to the group of characteristic parameters by applying a prediction model that has learned correlations between olfactory characteristic values ​​of aroma components and constituent parameters of the aroma components to the feature quantities; A method for performing.

2. The computer, searching for aroma components based on the calculated constituent parameters; outputting information on the searched aroma components; The method of claim 1 , further comprising:

3. the searching step calculates at least one of a degree of similarity or a degree of dispersion of a difference or a ratio between constituent parameters of a plurality of aroma components included in a search range and the calculated constituent parameters, and searches for aroma components for which at least one of the calculated degree of similarity or degree of dispersion satisfies a predetermined criterion. The method of claim 2.

4. In the step of specifying, in response to designation of an aroma component whose olfactory characteristic value has been specified, the olfactory characteristic value exhibited by the aroma component is specified as the characteristic parameter group. The method of claim 1.

5. the prediction model is constructed by supervised learning using a plurality of sets of learning data each including a learning feature related to a learning olfactory characteristic value and ground truth data corresponding to the learning feature, the correct answer data corresponding to the learning feature represents constituent parameters of the aroma components from which the learning feature is obtained, The method of claim 1.

6. the training feature represents at least one of the result of an aroma sensory evaluation of one or more aroma descriptors of the aroma component from which the feature is obtained, or the result of one or more olfactory receptor assays of the aroma component from which the feature is obtained. The method according to claim 5.

7. The olfactory characteristic value is at least one of an aroma descriptor or an olfactory receptor activity characteristic; The method of claim 1.

8. The configuration parameters are molecular physical properties. The method of claim 1.

9. The olfactory characteristic value is a scent descriptor, The configuration parameter is an olfactory receptor activity profile; The method of claim 1.

10. The olfactory characteristic value is an olfactory receptor activity characteristic, The configuration parameters are scent descriptors. The method of claim 1.

11. A means for identifying a group of characteristic parameters which are one or more olfactory characteristic values; A means for acquiring a feature amount relating to the group of characteristic parameters; a means for calculating constituent parameters of an aroma component that is expected to exhibit an olfactory characteristic corresponding to the group of characteristic parameters by applying a prediction model that has learned the correlation between the olfactory characteristic value of the aroma component and the constituent parameters of the aroma component to the feature amount; An information processing device comprising:

12. Computer, A means for identifying a group of characteristic parameters which are one or more olfactory characteristic values; means for acquiring a feature quantity relating to the group of characteristic parameters; a means for calculating constituent parameters of aroma components that are expected to exhibit olfactory characteristics corresponding to the group of characteristic parameters by applying a prediction model that has learned correlations between olfactory characteristic values ​​of aroma components and constituent parameters of the aroma components to the feature quantities; A program that functions as a

13. A system including a first information processing device and a second information processing device, The first information processing device is A means for identifying a group of characteristic parameters which are one or more olfactory characteristic values; A means for acquiring a feature amount relating to the group of characteristic parameters; a means for calculating constituent parameters of an aroma component that is expected to exhibit an olfactory characteristic corresponding to the group of characteristic parameters by applying a prediction model that has learned the correlation between the olfactory characteristic value of the aroma component and the constituent parameters of the aroma component to the feature amount; a means for outputting information based on a calculation result of constituent parameters of aroma components expected to exhibit olfactory characteristics corresponding to the group of characteristic parameters to the second information processing device; Equipped with system.

Images