Function estimation device

Abstract

To provide a function estimation device that can estimate a functionality component and an activity value from research data.SOLUTION: A function estimation device has: a first calculation unit that calculates a first correlation value indicating the correlation between a component name and functionality and a first description frequency of the component name, in a first document group having a description related to functionality from a document including a predetermined article name; a first component selection unit that selects a first component on the basis of the first correlation value and the first description frequency; a second calculation unit that calculates a second correlation value indicating the first component and the chemical structure of the first component and a second description frequency of the chemical structure, in a second document group related to the first component extracted from the first document group; a characteristic chemical structure selection unit that selects a characteristic chemical structure on the basis of the second correlation value and the second description frequency; an activity value extraction unit that extracts the activity value of a second component that is one classification of the first component from the second document group; and a prediction unit that predicts the activity value of a third component that is one classification of the first component by using a learned model trained by using the activity value of the second component and the characteristic chemical structure.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a function estimation device. [Background technology]

[0002] In recent years, research into the functions of food has been attracting attention. Examples of the functions of food include nutritional functions, taste functions, and bioregulatory functions. Bioregulatory functions are functions that regulate various bodily functions, such as lowering blood pressure and cholesterol.

[0003] The bioregulatory function of food varies in activity (a numerical value indicating the degree of effectiveness) depending on the type and amount of functional ingredients contained in the food.

[0004] Technologies relating to functional ingredients in foods are disclosed below. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] WO2016 / 181945 Summary of the Invention [Problem to be solved by the invention]

[0006] However, functional ingredients in food are currently in the research stage, and the functionality and activity values ​​of all ingredients have not yet been clarified. In order to calculate the functionality of a certain food, it is necessary to know the types of functional ingredients contained in the food and data on the activity values ​​of those functional ingredients. However, as mentioned above, since not all functional ingredients and activity values ​​have been clarified, it is difficult to calculate the functionality of a food.

[0007] Identifying functional ingredients requires long-term and advanced research, such as extracting functional ingredients and activity values ​​from a huge number of research papers, understanding the characteristics and structure of the ingredients, and estimating the activity values ​​of other ingredients.

[0008] Therefore, one disclosure provides a function estimation device that can estimate functional components and activity values ​​from research data. [Means for solving the problem]

[0009] The system includes a first calculation unit that calculates a first correlation value indicating the correlation between a component name and functionality and a first mention frequency of the component name in a first group of documents having descriptions of functionality from documents including a predetermined product name; a first component selection unit that selects a first component based on the first correlation value and the first mention frequency; a second calculation unit that calculates a second correlation value between the first component and a chemical structure of the first component and a second mention frequency of the chemical structure in a second group of documents related to the first component extracted from the first group of documents; a characteristic chemical structure selection unit that selects a characteristic chemical structure based on the second correlation value and the second mention frequency; an activity value extraction unit that extracts an activity value of a second component, which is a type of the first component, from the second group of documents; and a prediction unit that predicts the activity value of a third component, which is a type of the first component, using a trained model trained using the activity value of the second component and the characteristic chemical structure. [Effects of the Invention]

[0010] One disclosure is that functional ingredients and activity values ​​can be estimated from research data. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a diagram illustrating an example of the configuration of a function estimation system. [Figure 2] FIG. 2 is a diagram illustrating an example of the configuration of the function estimation device 100. As shown in FIG. [Figure 3] FIG. 3 is a diagram showing an example of a processing flowchart of the function estimation process S100. [Figure 4] FIG. 4 is a diagram showing an example of the output of the analysis tool. [Figure 5] FIG. 5 shows an example of the correlation value and the percentage of articles that describe functionality. [Figure 6] FIG. 6 is a diagram showing an example of a list of class keys and associated function information. [Figure 7] FIG. 7 is a diagram showing an example of the relationship between the frequency of appearance of a word in a paper and the correlation value. [Figure 8] FIG. 8 is a diagram illustrating an example of an output of the second analysis tool. [Figure 9] FIG. 9 is a diagram showing an example of a summary of characteristic parameters of components. [Figure 10] FIG. 10 is a diagram showing an example of learning of the activity value estimation tool. [Figure 11] FIG. 11 is a diagram showing an example of a list of estimated activity values ​​predicted collectively. DETAILED DESCRIPTION OF THE INVENTION

[0012] [First embodiment] A first embodiment will be described. In the following example, food is used as an example, but the subject of estimation of functional components and activity values ​​is not limited to food, and may include, for example, medicines, cosmetics, etc. In the following example, a biological regulation function is used as an example of functionality, but nutritional function and preference function may also be included.

[0013] <System configuration example> Fig. 1 is a diagram showing an example of the configuration of a function estimation system. The function estimation system 10 is a system that estimates functional components and activity values ​​using document data of papers and research results. The function estimation system 10 is composed of a function estimation device 100 and a network N1.

[0014] The function estimation device 100 is a device that estimates functional components and activity values, and is, for example, a computer machine. The function estimation device 100 is connected to a network N1 and acquires literature data D1 from the network N1. The network N1 is, for example, the Internet or an intranet, and literature such as various papers and research results is digitized and stored. The literature data D1 is, for example, data on papers and research results related to food. The network N1 may also be a large-scale storage medium.

[0015] The function estimation device 100 estimates what functional components are contained in a target food and what the activity values ​​of those functional components are.

[0016] <Configuration example of a function estimation device> 2 is a diagram showing an example of the configuration of the function estimation device 100. The function estimation device 100 is, for example, a computer machine or a server machine. The function estimation device 100 has a CPU 110, a storage 120, a memory 130, and a communication circuit 140.

[0017] The storage 120 is an auxiliary storage device such as a flash memory, HDD, or SSD that stores programs and data. The storage 120 stores a function estimation program 121, a document extraction program 122, an analysis tool program 123, a second analysis tool program 124, and an activity value estimation tool program 125.

[0018] The memory 130 is an area into which the programs stored in the storage 120 are loaded. The memory 130 may also be used as an area in which the programs store data.

[0019] The communication circuit 140 is a circuit that connects to the network N1 and performs communication with the network N1, and acquires document data D1 from the network N1.

[0020] The CPU 110 is a processor that loads a program stored in the storage 120 into the memory 130, executes the loaded program, configures each unit, and realizes each process.

[0021] By executing the function estimation program 121, the CPU 110 constructs a first literature extraction unit, a first calculation unit, a first component selection unit, a second literature extraction unit, a second calculation unit, a characteristic chemical structure selection unit, an activity value extraction unit, and a prediction unit, and performs function estimation processing. The function estimation processing is a process for estimating the activity value of a certain function of a component contained in a target object (such as food). In the function estimation processing, the function estimation device 100 extracts components and activity values ​​having a certain function from literature data, estimates the activity value of the component contained in the target object according to the degree of similarity between the chemical structure of the extracted component and the chemical structure of the component contained in the target object, and calculates the activity value of the certain function of the target object as a whole.

[0022] By executing the document extraction program 122, the CPU 110 constructs a first document extraction unit and a second document extraction unit and performs document extraction processing. The document extraction processing is processing for extracting documents that match conditions from multiple documents. The document extraction processing extracts documents that include a predetermined character string, for example. Extraction may involve, for example, outputting a list of document names or outputting a document summarizing the contents of the documents.

[0023] The CPU 110 executes the analysis tool program 123 to construct an analysis tool. When information about a document is input, the analysis tool outputs a correlation value indicating the correlation between character strings included in the document, the frequency of occurrence of the character string, etc. When paper data is input, the analysis tool outputs the frequency of occurrence of a certain character string, a correlation value with a target character string, etc., and is, for example, IBM Watson.

[0024] The CPU 110 executes the second analysis tool program 124 to construct a second analysis tool. The second analysis tool outputs patent documents having a specific character string as a cluster. When patent document data is input, the second analysis tool outputs the number of patent documents in which an extracted characteristic word appears as the number of clusters among patent documents containing a certain character string, for example, VALUENEX Radar.

[0025] The CPU 110 executes the activity value estimation tool program 125 to construct an activity value estimation tool. The activity value estimation tool is an AI (Artificial Intelligence), such as a neural network. The activity value estimation tool is a tool, such as Data Robot, that learns the association between component names, chemical structures, functions, and activity values ​​and estimates the activity value of a function from the chemical structure of a target component.

[0026] <Function estimation processing> 3 is a diagram showing an example of a processing flowchart of the function estimation process S100. The function estimation device 100 acquires document data and executes the function estimation process S100.

[0027] The function estimation device 100 collects literature data (S101). The literature data is collected, for example, from the Internet or a storage medium storing literature data. The function estimation device 100 collects literature data including the names of foods whose functionality is to be estimated.

[0028] For example, if a tester wants to estimate what functional components are contained in cacao, he or she will collect literature data containing the word "cacao."

[0029] The function estimation device 100 centralizes the collected literature data and imports it into an analysis tool (S102). The function estimation device 100 outputs, for example, a list of the centralized data. The centralized data includes, for example, the author's name, title, summary, and content (including URLs and shortcuts where the content is written).

[0030] The analysis tool outputs the frequency of appearance of words included in the imported literature data, correlation values ​​with specific words, etc. (S103). In order to detect words that have a high correlation value with functionality, the function estimation device 100 inputs the word "function" into the analysis tool and performs analysis. Note that the word to be input is in the language used in the literature data; for example, "function" is input in the case of English literature.

[0031] FIG. 4 is a diagram showing an example of the output of the analysis tool. The words included in the literature data (values ​​in FIG. 4), their frequency of occurrence (frequency in FIG. 4; number of documents in which they appear), and their correlation value with "function" (correlation in FIG. 4) are output. The correlation value is the correlation with "function," and for example, the closer (fewer) the position of "function" and the word in question in the sentence (such as the number of words between the two words) the higher the value. Furthermore, the correlation value only needs to indicate the relationship between "function" and other words, and may be a value that takes into account, for example, whether they appear in the same chapter, paragraph, or phrase, or the context of their appearance positions.

[0032] Returning to the processing flowchart of FIG. 3, the function estimation device 100 extracts feature words contained in the key components (S104). These are components selected from the components of the class keys. The class keys are component names (major category names) in academic classifications, such as "AEA (acylethanolamine)" and "CmE (campesterol ester)." The function estimation device 100 extracts component names with high correlation values ​​and frequency of appearance from, for example, the output results of the analysis tool.

[0033] 5 is a diagram showing an example of correlation values ​​and the proportion of articles that describe functionality. For example, when the correlation value is 2.0, functionality-related content is included in 94.2% of articles, so the function estimation device 100 extracts component names of class keys from words with a correlation value of 2.0 or greater. The function estimation device 100 excludes and does not extract words that are clearly not component names.

[0034] FIG. 6 is a diagram showing an example of a list of function information associated with a class key. The function information is obtained by extracting words related to functionality from among words that have a high correlation value with the component name of the class key. This extraction process is performed in step S104. The representative component is the component name of a medium classification of the component of the class key, and may be output by an analysis tool or generated based on other academic data (for example, a document listing the component name of the class key and a detailed list of those components).

[0035] The function estimating device 100 selects, for example, "ceramide" as the key component. Thereafter, the function estimating device 100 estimates the activity value related to the "skin-beautifying effect" for each type (subcategory) of "ceramide" contained in "cacao."

[0036] The function estimation device 100 extracts words related to chemical structure as feature words of the key component from among words having a high correlation value with the key component. FIG. 7 is a diagram showing an example of the relationship between the frequency of appearance of a word in papers (the proportion of papers in which the word appears) and the correlation value. In the table of FIG. 7, the vertical axis indicates the frequency of appearance (number of papers in which the word appears / total number of papers), and the horizontal axis indicates the correlation value with "ceramide." The numerical values ​​in the table also indicate the appearance rate of the feature word (ceramide) (number of papers in which the word was appropriate as a feature word / number of papers in which the word appeared). The function estimation device 100 selects words with a high appearance rate. For example, the function estimation device 100 extracts words related to chemical structure that appear in papers with an appearance frequency of 1.1% to 50% and a correlation value of 2.0 to 9.9 as feature words. For example, the function estimation device 100 extracts "sphingosine length," "fatty acid length," "hydroxyl group," etc. as feature words corresponding to "ceramide."

[0037] Here, the function estimation device 100 may use another analysis tool (second analysis tool) to extract more characteristic words. FIG. 8 is a diagram showing an example of the output of the second analysis tool. For example, the second analysis tool outputs the frequency of occurrence of a certain characteristic word in literature containing the word "ceramide" (the number of documents in which it appears) as the number of clusters. The function estimation device 100 may extract, as characteristic words, those clusters of the second analysis tool that are equal to or greater than a predetermined value and related to chemical structure. Characteristic words may be extracted from the results output by both or either the analysis tool and the second analysis tool.

[0038] Returning to the processing flowchart of FIG. 3, the function estimation device 100 extracts the activity values ​​of the components from the paper (S105). The function estimation device 100 extracts the type of ceramide described in the paper and the activity value of the function of the ceramide from the paper. For example, if the function is "water retention," the activity value is extracted from the paper that describes the activity value related to water retention.

[0039] The function estimation device 100 compiles the characteristic parameters of the components and imports them into the activity value estimation tool (S106).

[0040] 9 is a diagram showing an example of a summary of the characteristic parameters of components. The vertical axis shows the type of ceramide (subcategory name), and the horizontal axis shows the activity value, function, and characteristic term. The type of ceramide is, for example, the type of ceramide contained in "cacao."

[0041] Furthermore, in FIG. 9, "water retention" is entered as a function. "Water retention" is a specific function that realizes the "skin-beautifying effect" in FIG. 6, and is, for example, a function name extracted from a paper. Since "skin-beautifying effect" is a general and unclear function name, and "water retention" is a more specific function name, "water retention" is selected as the function for estimating activity values, as it is likely to be described in papers. The function estimation device 100 investigates specific numerical values ​​of the chemical structures related to the characteristic words for each type of ceramide and enters them in a list. Furthermore, the function estimation device 100 enters the types of ceramides described in the paper and the water retention activity values ​​of those ceramides extracted in step S105 in a list. For example, in FIG. 9, the water retention activity value of Cer(d18:0_12:0) is described in the paper, so "36" is extracted from the paper and entered. Ceramide types not described in the paper are left blank. Note that the blanks will be filled in with estimated values ​​in later processing.

[0042] The activity value estimation tool is, for example, an AI. By importing a summary (teaching data: chemical structure or component name and the activity value of the target function of the component), the activity value estimation tool learns the relationship between the component with the activity value, chemical structure, function, and activity value. In particular, the activity value estimation tool learns the relationship between chemical structure and activity value. In the case of Figure 9, by importing the summary, the activity value estimation tool learns the relationship between the type and chemical structure of ceramide and the activity value of water retention. Note that the activity value estimation tool may separately learn the component name and chemical structure.

[0043] FIG. 10 is a diagram showing an example of learning of the activity value estimation tool. The vertical axis represents chemical structure, and the horizontal axis represents effect. The effect is, for example, the effect that differences in chemical structure have on differences in activity value. The higher the effect value, the greater the influence on the activity value. For example, it can be seen that the sphingosine length has a large influence on the activity value, but the number of double bonds has little influence on the activity value (has a small influence).

[0044] Returning to the processing flowchart of Figure 3, the function estimation device 100 collectively predicts the activity values ​​of multiple components, identifies the functional components and the required intake amount, etc. (S107), and ends the processing. Using an activity value estimation tool, the function estimation device 100 outputs estimated function activity values ​​(estimated activity values) for components not described in papers. The activity value estimation tool estimates the activity values ​​of components whose activity values ​​are not described from their chemical structures, and outputs the estimated activity values.

[0045] Figure 11 shows an example of a list of estimated activity values ​​predicted collectively. The estimated activity values ​​for each ceramide type that were blank in Figure 9 are output. The estimated activity values ​​listed in this list are used to fill in the blanks in Figure 9.

[0046] For example, the function estimation device 100 calculates the total activity value contained in a food from the activity value of each component related to a certain function of the food (including activity values ​​described in papers and estimated activity values). By comparing the total activity values ​​for each food, the effect of each food related to a certain function can be evaluated. In the above example, the water-retaining activity value of the ceramide component contained in cacao was output. In addition, by outputting the same information for other foods, foods can be compared and relatively evaluated. Furthermore, if there is a component other than ceramide that is related to water-retaining properties, the activity value of this component can also be calculated and added to the activity value of ceramide, allowing for a more accurate calculation of the total activity value. [Explanation of symbols]

[0047] 10: Function estimation system 100: Function estimation device 110:CPU 120: Storage 121: Function estimation program 122: Document extraction program 123: Analysis tool program 124: Second analysis tool program 125: Activity value estimation tool program 130: Memory 140: Communication circuit N1: Network

Claims

1. a first calculation unit that calculates a first correlation value indicating a correlation between an ingredient name and functionality and a first description frequency of the ingredient name in a first group of documents having a description of functionality from documents including a predetermined product name; a first component selection unit that selects a first component based on the first correlation value and the first description frequency; a second calculation unit that calculates a second correlation value between the first component and a chemical structure of the first component and a second description frequency of the chemical structure in a second document group related to the first component extracted from the first document group; a characteristic chemical structure selection unit that selects a characteristic chemical structure based on the second correlation value and the second description frequency; an activity value extracting unit that extracts activity values ​​of a second component, which is a type of the first component, from the second document group; a prediction unit that predicts the activity value of a third component, which is a type of the first component, using a trained model trained using the activity value of the second component and the characteristic chemical structure; A function estimation device having the above.

2. The correlation indicates the degree of proximity between the character string related to the ingredient name and the character string related to the functionality. The function estimation device according to claim 1 .

3. the first frequency of mention indicates the number of documents in which the component name is mentioned; The first component selection unit selects, as the first component, a component name whose first correlation value is equal to or greater than a first threshold and whose first description frequency is equal to or greater than a second threshold. The function estimation device according to claim 2.

4. The characteristic chemical structure selection unit selects, as the characteristic chemical structure, a chemical structure whose second correlation value is equal to or greater than a third threshold and whose second description frequency is equal to or greater than a fourth threshold. The function estimation device according to claim 3.

5. The trained model predicts the activity value of the third component using the similarity of the characteristic chemical structure. The function estimation device according to claim 1 .

Images