JP7281261B2 - Analysis method for sensory irritation of substances - Google Patents

Description

TECHNICAL FIELD The present invention relates to a method for analyzing the sensory irritation of a substance to the skin.

Safety is an important factor that cosmetics should have. On the other hand, the number of people who feel that they have sensitive skin is on the rise, and not a few consumers have experienced some discomfort (feeling of irritation) when using cosmetics. In order to provide cosmetics that can be used more safely, it is desired to reduce the organoleptic properties of cosmetics. So far, preservatives, moisturizing agents, perfume ingredients such as menthol, etc. have been reported as components of cosmetics having sensory irritation to the skin. However, no comprehensive or systematic knowledge has been established regarding the sensory irritation to the skin of ingredients used in cosmetics. On the other hand, there is no clear guideline so far as to what kind of skin is sensitive to what kind of substance.

In recent years, in various fields, methods for predicting or controlling events using models and systems constructed based on machine learning have been developed. In the field of skin analysis, for example, in Patent Document 1, using a program constructed by a machine learning method, blemishes are classified based on the morphological feature amount and color feature amount of blemishes extracted from a photographed image of the skin. method is described.

JP 2013-90752 A

The present invention provides a method for analyzing the organoleptic properties of substances on the skin.

In one embodiment, the present invention is a method for analyzing the sensory irritation of a substance to the skin, comprising:
obtaining the physicochemical properties of the target substance;
Acquiring the skin characteristics of the skin, and analyzing the sensory irritation of the target substance to the skin by a machine learning model based on the physicochemical characteristics of the target substance and the skin characteristics of the skin;
including
the skin properties include stratum corneum water content, TEWL and pH of the skin;
the physico-chemical properties include melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient;
provide a way.
In another embodiment, the present invention provides a database for machine learning the organoleptic properties of substances on the skin, comprising:
Measurements of stratum corneum water content, TEWL and pH of skin for multiple subjects;
Melting points, molecular weights, octanol-water partition coefficients, and skin permeation coefficients for a plurality of substances; and results of measurements of the organoleptic properties of the substances on the skin of the plurality of subjects;
I will provide a.
In another embodiment, the present invention provides a method for constructing a model for analyzing the organoleptic properties of substances on the skin, comprising:
Measurements of stratum corneum water content, TEWL and pH of skin for multiple subjects;
Melting points, molecular weights, octanol-water partition coefficients, and skin permeation coefficients for a plurality of substances; and results of measurements of the organoleptic properties of the substances on the skin of the plurality of subjects;
to create a classifier for identifying the presence or absence of sensory irritation to the skin of substances by machine learning with descriptors,
including,
provide a way.
In another embodiment, the present invention provides a program or apparatus for performing the method for analyzing the sensory irritation of the substance to the skin.

The present invention provides a machine learning model for analyzing the organoleptic properties of substances on the skin. Using this model, it becomes possible to predict the sensory irritation of various substances to the skin of a subject with higher probability. In particular, the model can be used to analyze the presence or absence of a subject's sensitivity to sensory stimulation by a given substance or cosmetic product. Furthermore, the model enables exhaustive or systematic analysis of the sensory irritation of various substances to the skin. INDUSTRIAL APPLICABILITY According to the present invention, it is possible to evaluate the suitability of a subject's skin for cosmetics or the like, or to select cosmetics suitable for the subject without performing a stinging test.

A conceptual diagram of an apparatus for analyzing the sensory irritation of a substance to the skin using a machine learning model. Mean temperature of the skin of Japanese subjects with various skin types. Impurity reduction amount by each feature amount in the random forest discriminator for the sensory irritation of the substance to the skin.

All patents, non-patents, and other publications cited herein are hereby incorporated by reference in their entirety.

The present invention provides a machine learning model for analyzing the organoleptic properties of substances on the skin. The present invention also provides a method for analyzing the sensory irritation of a substance to the skin without actually applying the substance to living skin or cultured skin cells based on the machine learning model, and implements the method. We provide a device and a program for

As used herein, "sensory irritation" caused by a substance refers to discomfort (eg, pain, hot flashes, itching, itching, etc.) that occurs on the skin to which the substance is applied. In addition, the ``sensory irritation to the skin'' of a substance (hereinafter also simply referred to as the ``sensory irritation'' of a substance) means that the skin to which it is applied causes discomfort (e.g., pain, hot flashes, itching, itching, etc.). It refers to the property of the substance to give.

One aspect of the present invention is the analysis of the organoleptic properties of substances using machine learning models. The machine learning model uses skin characteristics collected from a human population and physicochemical characteristics of a plurality of test substances as explanatory variables, and the sensory irritation of the plurality of test substances to the skin of a person belonging to the human population. It is constructed by machine learning with the measurement results of Based on the machine learning model, the physicochemical properties of the target substance are analyzed to determine whether the target substance has organoleptic properties on the skin of a given subject, or the skin characteristics of the subject are analyzed. is analyzed whether the subject's skin is sensitive to sensory stimulation by a given substance of interest.

Examples of skin characteristics used as explanatory variables in the machine learning include stratum corneum water content, transepidermal water loss (TEWL), skin pH, electrical skin threshold (Current Perception Threshold; CPT), sebum content, and skin color. , skin temperature, and more detailed examples include stratum corneum water content, TEWL, skin pH, skin CPT at 5 Hz (CPT5), skin CPT at 250 Hz (CPT250), sebum content, skin color L* value, a * and b* values.

In one embodiment, the skin properties used as explanatory variables are preferably stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value At least one selected from the group consisting of, more preferably at least stratum corneum water content, TEWL and pH, still more preferably stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and Skin color L*, a* and b* values.

The stratum corneum moisture content can be measured by a skin surface stratum corneum moisture content measuring device such as Corneometer (Courage+Khazaka electronic GmbH). TEWL can be measured by a transepidermal water loss measurement device such as the Tewameter (Courage+Khazaka electronic GmbH). Skin pH can be measured by a pH meter or the like. Skin CPT can be measured by a sensory neuroautomation device such as the Neurometer (Neurotron, Inc.). The amount of sebum can be measured by an oil meter such as Sebumeter (Courage+Khazaka electronic GmbH). Skin color (L* value, a* value and b* value) can be measured by a spectrophotometer such as Chromameter (KONICA MINOLTA JAPAN, INC.).

The skin characteristics are measured from a group of test subjects for model construction of a plurality of people, preferably 80 or more, more preferably 100 or more. When the number of people in the group for collecting skin characteristics is small, the accuracy of machine learning decreases, and when the number of people in the group is large, it takes time to learn, resulting in a decrease in learning efficiency.

Examples of physicochemical properties of substances used as explanatory variables in the machine learning include boiling point, melting point, molecular weight, octanol-water partition coefficient (logKow), skin permeability coefficient, solubility parameter, IOB (Inorganic Organic Balance). , and molecular descriptors such as Chi Connectivity Indices, Kappa Shape Indices, Electrotopological State Indices, and Topological descriptors. As the skin permeation coefficient (logP), Potts and Guy's skin permeation coefficient prediction formula (Pharm Res., 9, 663-669, 1992) is used. Solubility parameters include Hildebrand solubility parameter (SP), Hansen solubility parameter (HSP) δd (dispersion force term), δp (polar term), and δh (hydrogen bond term), Abraham solvation parameter (ASP A, ASP B, ASP E, ASP L, ASP S, and ASP V). Chi Connectivity Indices include x0, x1, x2, xp3-xp10, xc3, xc4, xpc4, and knotp. The above molecular descriptors can be determined using The Toxicity Estimation Software Tool (EPA TEST; [www.epa.gov/chemical-research/toxicity-estimation-software-tool-test]).
In one embodiment, the physicochemical properties used as the explanatory variables are preferably boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeability coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A , ASP B, ASP E, ASP L, ASP S, ASP V, and IOB, more preferably at least one selected from the group consisting of melting point, molecular weight, octanol-water partition coefficient, and skin permeability coefficient more preferably melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient, still more preferably boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeation coefficient, SP, HSP δd, HSP δp, HSP δh , ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and IOB.
Less variation in the physico-chemical properties reduces the accuracy of machine learning.

The physico-chemical properties are values for several different types of test substances selected for model building. Preferably, the model-building test substances comprise at least a pH adjuster, a preservative, a polar oil and a humectant. A small variation in the types of test substances reduces the accuracy of machine learning. Examples of the pH adjuster include cosmetically acceptable pH adjusters such as organic acids such as lactic acid and citric acid, and examples of the preservative include cosmetically acceptable pH adjusters such as methylparaben, chlorphenesin, and phenoxyethanol. Preservatives include cosmetically acceptable polar oils such as isostearyl isostearate, isononyl isononanoate, isopropyl isostearate, and ethylhexyl hydroxystearate. Cosmetically acceptable humectants such as propylene glycol, propylene glycol, glycerin, diglycerin, 1,3-butylene glycol and the like.
In one embodiment, the model-building test substance used in the machine learning is preferably at least one selected from the group consisting of lactic acid and citric acid, and the group consisting of methylparaben, chlorphenesin, and phenoxyethanol. and at least one selected from the group consisting of isostearyl isostearate, isononyl isononanoate, isopropyl isostearate, and ethylhexyl hydroxystearate, and dipropylene glycol, propylene glycol, glycerin, and diglycerin. , and 1,3-butylene glycol, more preferably lactic acid, citric acid, methylparaben, chlorphenesin, phenoxyethanol, isostearyl isostearate, isononyl isononanoate, isopropyl isostearate , ethylhexyl hydroxystearate, dipropylene glycol, propylene glycol, glycerin, diglycerin, and 1,3-butylene glycol.

The sensory irritation of the test substance for model construction to the skin of a group of test subjects for model construction can be examined by a stinging test or the like. For example, one of the test substances is applied to the skin of a subject in the population of subjects for model construction, and the subject's sensations (pain, hot flashes, itching, itching, etc.) are investigated, The organoleptic properties of the test substances on the subject's skin are evaluated. For example, the subject's skin discomfort due to the test substance may be scored stepwise. In a preferred example, the rank of skin discomfort due to the test substance (e.g. 3-10 scale, preferably 5-7 scale from "no discomfort" to "very uncomfortable" for overall discomfort). and score the discomfort. A similar test is conducted for each subject with all types of the model-building test substances. Furthermore, the procedure of scoring the discomfort for one test substance may be performed multiple times in one subject, and their statistical values may be obtained as the score of the subject's skin discomfort due to the test substance. good. Based on the resulting discomfort scores, an organoleptic score for each subject's skin for each test substance can be determined. Furthermore, the organoleptic score may be binarized to determine the presence or absence of organoleptic irritation of the test substance to the subject's skin. In one embodiment, the organoleptic score of a test substance for a subject's skin can be the sum or average of the scores obtained from the subject for the test substance. Alternatively, if the total or average value of the obtained scores is a predetermined threshold or more (or above the threshold), 1 (irritant), and below the threshold (or below the threshold), 0 (irritant) None), the organoleptic score may be binarized. In another embodiment, the skin organoleptic score of a test substance for a subject can be expressed as the percentage of scores obtained from the subject for the test substance that include a score value greater than 0. . Alternatively, if the obtained ratio is a predetermined threshold or more (or above the threshold), set it to 1 (irritant), and if it is below the threshold (or below the threshold), set it to 0 (no irritation), The organoleptic score may be binarized. The procedure for obtaining the organoleptic score listed above is exemplary, and in the present invention, various other means capable of assessing the organoleptic potential of a test substance on the skin of a subject are used to evaluate each test substance. Sensory irritation to the skin of a subject can be evaluated. Preferably, the organoleptic potential of each test substance on each subject's skin is expressed as a binary value (no or no organoleptic potential), such as irritating (1) or non-irritating (0).

said skin properties for all of said plurality of modeling subjects, said physicochemical properties for all of said plurality of modeling test substances, and said model building test substances for all said modeling test substances; A database containing measurement results of sensory irritation to the skin (sensory irritation score) is useful as a database for machine learning of the sensory irritation of substances. The database may consist of the data of the skin properties, the data of the physicochemical properties, and the data of the organoleptic score, but may also contain other supplementary information. good.

Using the skin properties and physicochemical properties described above as explanatory variables and the sensory irritation score as an objective variable, the sensory irritation properties of substances are machine-learned. The machine learning can be performed by random forests, support vector machines, neural networks, naive Bayes classifiers, and the like. The model constructed by machine learning, for example, when inputting the subject's skin characteristics and the physicochemical characteristics of the target substance, outputs whether the target substance is sensory irritant to the subject's skin. It is a discriminator. Alternatively, the machine learning model is a model that, when inputting the subject's skin properties and the physicochemical properties of the target substance, outputs the sensory irritation score of the target substance when applied to the subject. Furthermore, by evaluating the obtained machine learning model using impurity, OOB (Out-Of-Bag), etc., the explanatory variables that are important for the decision by the model are extracted, and the extracted explanation Additional machine learning models may be constructed by further machine learning using only the variables.

The machine learning model constructed above can be used to analyze the sensory irritation properties of substances. For example, if the subject's skin characteristics and the physicochemical characteristics of the target substance, which are variables, are input to the machine learning model, whether or not the subject's skin receives sensory stimulation from the target substance is output. Based on the output results, it is possible to analyze the sensory irritation of the target substance to the subject's skin.

Accordingly, one preferred embodiment of the present invention is a method for analyzing the organoleptic properties of substances on the skin. The method comprises
obtaining the physicochemical properties of the target substance;
Acquiring the skin characteristics of the skin, and analyzing the sensory irritation of the target substance to the skin by the above-described machine learning model based on the physicochemical characteristics of the target substance and the skin characteristics of the skin,
including.

The subject's skin characteristics measured in the analysis method may be at least one selected from the group consisting of skin characteristics used as machine learning variables when constructing the machine learning model used in the method. Therefore, the skin characteristics used as variables are at least one selected from the group consisting of stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value. If it is a species, the skin characteristics measured in the method are preferably the stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and At least one selected from the group consisting of b* values, more preferably at least stratum corneum moisture content, TEWL and pH, still more preferably stratum corneum moisture content, TEWL, skin pH, CPT5, CPT250, sebum content and skin color L*, a* and b* values.

The physicochemical properties of the target substance acquired in the analysis method may be at least one selected from the group consisting of physicochemical properties used as variables in the machine learning model used in the method. Therefore, the physicochemical properties of the target substance obtained in the method are preferably boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeability coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, At least one selected from the group consisting of ASP B, ASP E, ASP L, ASP S, ASP V, and IOB, more preferably at least melting point, molecular weight, octanol-water partition coefficient, and skin permeability coefficient. , more preferably melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient, still more preferably boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeation coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and IOB. The type of target substance used in the method is not particularly limited as long as the above physicochemical properties can be obtained.

Analysis of the organoleptic property of the target substance to the skin may include, for example, determining whether the target substance is organoleptic to the subject's skin. Furthermore, it is possible to determine whether or not the cosmetic product is sensory stimulating to the skin of the subject based on the determination result and the pre-confirmed prescription of the given cosmetic product.

In addition, the machine learning model constructed above can be used to analyze the sensitivity of a subject or skin to sensory stimulation by a substance. As used herein, the term "sensitivity to sensory stimulation by a substance" (sometimes simply referred to as "sensitivity") of a subject or skin refers to the property of the subject or the skin that the application of the substance causes sensory irritation to the skin. . For example, the variables of the subject's skin characteristics and the physicochemical characteristics of the target substance are input into the machine learning model, and based on the output results obtained, the sensitivity of the subject or the skin to the target substance ( For example, it is possible to analyze whether or not a sensory stimulus is received from the target substance.

Accordingly, another embodiment of the present invention is a method of analyzing skin sensitivity to sensory stimulation by substances. The method comprises
measuring skin characteristics of a subject;
Obtaining the physicochemical properties of the target substance, and Sensitivity of the subject's skin to sensory stimulation by the target substance by the aforementioned machine learning model based on the subject's skin properties and the physicochemical properties of the target substance to parse
including.
Yet another embodiment of the present invention is a method of analyzing a subject's sensitivity to sensory stimulation by a substance. The method comprises
measuring skin characteristics of a subject;
Acquiring the physicochemical properties of the target substance, and analyzing the sensitivity of the subject to sensory stimulation by the target substance using the aforementioned machine learning model based on the subject's skin properties and the physicochemical properties of the target substance. to do
including.
The subject's skin properties measured in this method, the physicochemical properties of the target substance obtained, and the type of target substance used in this method are the same as in the method for analyzing the sensory irritation of the substance to the skin described above. is. Analysis of sensitivity of a subject or skin by the method can include, for example, determining whether the subject or skin is sensitive to sensory stimulation by the substance of interest.

The above-described method for analyzing the sensory irritation of a substance according to the present invention and the method for analyzing the sensitivity of a subject or skin according to the present invention can be applied to the skin of a subject without actually applying the substance to the skin of a living body or cultured skin cells. It makes it possible to investigate the organoleptic properties of substances, or the sensitivity of a subject's skin to a given substance of interest. Therefore, according to the method, it is possible to evaluate the suitability of the subject's skin to cosmetics or the like, or to select cosmetics suitable for the subject, without performing a stinging test. For example, if a customer's skin characteristics are measured at a store, appropriate cosmetics can be selected and provided to the customer. Moreover, according to the method, it is possible to examine the sensory irritation of the candidate substance for cosmetics to the skin and evaluate the suitability of the candidate substance as a cosmetic product without performing a stinging test. Moreover, the method for analyzing the sensory irritation of a substance according to the present invention makes it possible to examine the sensory irritation of a substance to the skin without actually applying the substance to the skin of a living body or cultured skin cells. Therefore, the method makes it possible to examine the sensory irritation properties of various substances more simply and quickly, and enables comprehensive or systematic analysis of the sensory irritation properties of substances. The method is also useful in determining appropriate cosmetic formulations for individuals with particular skin characteristics.

A further aspect of the present invention is an apparatus for analyzing the sensory irritancy of a substance or the sensitivity of a subject or his skin by means of the machine learning model described above. The device utilizes the machine learning model of the present invention from the data of the subject's skin properties and the physicochemical properties of the target substance to calculate the organoleptic properties of the target substance or the sensitivity of the subject or the skin thereof. and an output unit for outputting the result of the calculation. The device may comprise an input for inputting skin properties of the subject and/or physico-chemical properties of the substance of interest. Data input to the input unit is sent to the calculation unit and can be used for calculation using the machine learning model. The device may also be connected to a skin property measuring device. In this case, the measured skin property data of the subject is automatically sent to the device and used for calculation in the calculation unit. good too. In addition, the device may include a data storage unit, and the data storage unit stores the calculation result in the calculation unit, the physicochemical property data of the target substance, the prescription of the existing cosmetic product or its candidate composition. data, measured skin characteristics data of the subject, etc. can be saved. For example, physicochemical property data of a predetermined target substance may be stored in advance in the data storage unit. The data stored in the data storage unit can be called up as needed and used for computation in the computation unit. Also, the device may simultaneously comprise two or more of the input section, the connection with the measuring device, and the data storage section.

A further aspect of the present invention is a program for analyzing the sensory irritation of a substance or analyzing the sensitivity of a subject or skin using the machine learning model described above. The program uses the machine learning model of the present invention from the data of the subject's skin properties and the physicochemical properties of the target substance to calculate the sensory irritation of the target substance or the sensitivity of the subject or the skin. is. For example, the program may be a program for performing calculations in the calculation unit of the device of the present invention described above.

FIG. 1 shows a conceptual diagram of one embodiment of an apparatus for analyzing the sensory irritancy of a substance or the sensitivity of a subject or skin using a machine learning model. In this device, the subject's skin characteristic data is measured by the measuring device 1, and the obtained measured values are automatically or manually sent to the input unit 2 of the device main body. In this apparatus, the physicochemical property data of substances are stored in the data storage unit 3 and sent to the calculation unit as needed. The calculation unit 4 analyzes the subject's sensitivity to the substance using a machine learning model from the subject's skin property data and the physicochemical property data of the substance, and outputs the result to the output unit 5 . In one embodiment, the computing unit 4 executes a program for analyzing skin sensitivity to substances or sensory irritation of substances using a machine learning model based on skin property data and physicochemical property data. In another embodiment, the calculation unit 4 executes a program for analyzing sensitivity of a subject or skin by machine learning model based on skin property data and physicochemical property data. In one embodiment, computing unit 4 outputs whether the subject is sensitive to a particular target substance. In one embodiment, the computing unit 4 outputs substances having sensory stimulation properties to the subject from among the substances registered in the data storage unit 3 . In one embodiment, the calculation unit 4 outputs whether or not a target substance selected from the substances registered in the data storage unit 3 has sensory stimulation to the subject. In a further embodiment, the data storage unit 3 of the present device stores a database of prescriptions of existing cosmetics or their candidate compositions, and the computing unit 4 determines whether the target substance causes sensory irritation to the subject. Cosmetics and candidate substances that are sensory stimulating (or non-sensory stimulating) for the subject are output based on the result of calculation as to whether or not they are present and the prescriptions of the cosmetics and candidate substances referenced from the data storage unit 3. do.

A further embodiment of the present invention is the assessment or analysis of a subject's skin type or skin characteristics using facial skin temperature. As shown in the examples below, when skin types were classified based on the above skin characteristics of subjects, the skin temperature distribution of the face tended to have different characteristics for each skin type. More specifically, the average temperature of the whole face, the high or low skin temperature of each part of the forehead, nose or cheeks with respect to other parts are related to the skin type classified based on the skin characteristics of the subject as described above tended to have Therefore, the subject's skin type or skin characteristics can be evaluated or analyzed based on the facial skin temperature parameter. The parameters related to the skin temperature of the face include the average skin temperature of the entire face, the average skin temperature of facial parts (e.g. forehead, nose and cheeks), the skin temperature distribution between the parts (e.g. forehead, nose and cheeks). (High or low skin temperature of any part of the body relative to other parts). These parameters can be obtained by measuring the whole facial skin temperature. The whole face skin temperature can be measured by a skin surface temperature measuring device such as Thermotracer (NEC Sanei Co., Ltd.). For example, a subject whose cheek temperature is higher than other sites can be evaluated as having an atopic dry skin-like skin type with abnormal barrier function and skin dryness, or It can be evaluated to tend to have skin characteristics of low, high TEWL, low sebum content, high skin pH, reddish and dull complexion, and low threshold to electrical stimulation.

Further disclosed herein are the following materials, methods of manufacture, uses, methods, etc., as exemplary embodiments of the present invention. However, the invention is not limited to these embodiments.

[1] A method for analyzing the sensory irritation of a substance to the skin, comprising:
obtaining the physicochemical properties of the target substance;
Acquiring the skin characteristics of the skin, and analyzing the sensory irritation of the target substance to the skin by a machine learning model based on the physicochemical characteristics of the target substance and the skin characteristics of the skin;
including,
Method.
[2] A method for analyzing skin sensitivity to sensory stimulation by a substance, comprising:
measuring skin characteristics of a subject;
Acquiring the physicochemical properties of the target substance, and analyzing the sensitivity of the subject's skin to sensory stimulation by the target substance using a machine learning model based on the subject's skin properties and the physicochemical properties of the target substance. to do
including,
Method.
[3] A method for analyzing a subject's sensitivity to sensory stimulation by a substance, comprising:
measuring skin characteristics of a subject;
Acquiring the physicochemical properties of the target substance, and analyzing the subject's sensitivity to sensory stimuli by the target substance using a machine learning model based on the subject's skin properties and the physicochemical properties of the target substance. ,
including,
Method.
[4] The skin characteristics are
Preferably, at least one selected from the group consisting of stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value,
More preferably, at least the stratum corneum water content, TEWL and pH,
More preferably, stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value.
The method according to any one of [1] to [3].
[5] The physicochemical properties are
Preferably, boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeation coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and IOB At least one selected from the group consisting of
More preferably, at least the melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient,
More preferably, the melting point, molecular weight, octanol water partition coefficient and skin permeability coefficient,
More preferably, boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeability coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and is an IOB;
The method according to any one of [1] to [4].
[6] Preferably, the machine learning model comprises:
subject skin characteristics for model building;
Physico-chemical properties of the model-building test material, wherein the model-building test material comprises at least a pH adjuster, a preservative, a polar oil and a humectant; and
Measurement results of the sensory irritation of the model construction test substance on the skin of the model construction subject,
The method according to any one of [1] to [3], which is a model constructed by machine learning using
[7] The skin characteristics of the subject for model construction are
Preferably, at least one selected from the group consisting of stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value,
More preferably, at least the stratum corneum water content, TEWL and pH,
More preferably, stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value.
[6] The method described.
[8] The physicochemical properties of the model-building test substance are
Preferably, boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeation coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and IOB At least one selected from the group consisting of
More preferably, at least the melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient,
More preferably, the melting point, molecular weight, octanol water partition coefficient and skin permeability coefficient,
More preferably, boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeability coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and is an IOB;
The method according to [6] or [7].
[9] The model-building test substance is preferably at least one selected from the group consisting of lactic acid and citric acid, and at least one selected from the group consisting of methylparaben, chlorphenesin, and phenoxyethanol; at least one selected from the group consisting of isostearyl isostearate, isononyl isononanoate, isopropyl isostearate, and ethylhexyl hydroxystearate, and from dipropylene glycol, propylene glycol, glycerin, diglycerin, and 1,3-butylene glycol is at least one selected from the group consisting of
More preferably, lactic acid, citric acid, methylparaben, chlorphenesin, phenoxyethanol, isostearyl isostearate, isononyl isononanoate, isopropyl isostearate, ethylhexyl hydroxystearate, dipropylene glycol, propylene glycol, glycerin, diglycerin, and 1, is 3-butylene glycol,
[6] The method according to any one of [8].
[10] Preferably, the measurement result of the sensory irritation of the model-building test substance on the skin of the model-building subject is obtained by the following procedure:
obtaining, for each of the modeling test materials, an organoleptic score for the skin of each of the modeling subjects;
calculating the presence or absence of sensory irritation to the skin of each subject for each of the test substances for model construction based on the sensory irritation score;
The method according to any one of [6] to [9], which is data on the presence or absence of sensory irritation to the skin of each subject of each test substance obtained by the method.

[11] A database for machine learning the sensory irritation of substances to the skin, including:
skin characteristics for multiple subjects;
physico-chemical properties of the substances; and results of measurements of the organoleptic properties of the substances on the skin of the subjects.
[12] The skin characteristics are
Preferably, at least one selected from the group consisting of stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value,
More preferably, at least the stratum corneum water content, TEWL and pH,
More preferably, stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value.
[11] The database described.
[13] The physicochemical properties are
Preferably, boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeation coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and IOB At least one selected from the group consisting of
More preferably, at least the melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient,
More preferably, the melting point, molecular weight, octanol water partition coefficient and skin permeability coefficient,
More preferably, boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeability coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and is an IOB;
The database according to [11] or [12].
[14] Preferably, the measurement results of the sensory irritation of the skin of the plurality of subjects of the plurality of substances are obtained by the following procedure:
obtaining an organoleptic score for each of the plurality of subjects for each of the plurality of substances;
calculating the presence or absence of sensory irritation to each skin of the subject for each substance based on the sensory irritation score;
The database according to any one of [11] to [13], which is data on the presence or absence of sensory irritation to the skin of each subject of each of the substances obtained by the database.

[15] A method for constructing a model for analyzing the sensory irritation of a substance to the skin, comprising the following:
skin characteristics for multiple subjects;
physicochemical properties of a plurality of substances; and measurement results of the organoleptic properties of the plurality of substances on the skin of the plurality of subjects;
to create a classifier for identifying the presence or absence of sensory irritation to the skin of substances by machine learning with descriptors,
A method, including
[16] The skin characteristics are
Preferably, at least one selected from the group consisting of stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value,
More preferably, at least the stratum corneum water content, TEWL and pH,
More preferably, stratum corneum water content, TEWL, skin pH, CPT5, CPT250, sebum content, and skin color L* value, a* value and b* value.
[15] The method described.
[17] The physicochemical properties are
Preferably, boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeation coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and IOB At least one selected from the group consisting of
More preferably, at least the melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient,
More preferably, the melting point, molecular weight, octanol water partition coefficient and skin permeability coefficient,
More preferably, boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeability coefficient, SP, HSP δd, HSP δp, HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, and is an IOB;
The method according to [15] or [16].
[18] Preferably, the measurement results of the sensory irritation of the skin of the plurality of subjects of the plurality of substances are obtained by the following procedure:
obtaining an organoleptic score for each of the plurality of subjects for each of the plurality of substances;
calculating the presence or absence of sensory irritation to each skin of the subject for each substance based on the sensory irritation score;
The method according to any one of [15] to [17], which is data on the presence or absence of sensory irritation to the skin of each subject of each of the substances obtained by the method.

[19] A program for performing the analysis method according to any one of [1] to [10].
[20] An apparatus for performing the analysis method according to any one of [1] to [10].
[21] The device according to [20], which preferably comprises an input unit, a computing unit, and an output unit, further comprises a data storage unit as necessary, and is further optionally connected to a skin characteristic measuring device. Device.
[22] Preferably, the device according to [21], wherein the computing unit performs computation using the program according to [19].

EXAMPLES The present invention will be described in more detail below based on examples, but the present invention is not limited to these.

Example 1 Building a machine learning model 1. Method 1) Subjects Skin characteristic data were collected from 200 women (120 Japanese, 80 Germans) who were aware of their sensitive skin. On another day, a stinging test with the target substance was performed on the same subject.

2) Target Substances The following 14 compounds were used as target substances.
pH adjusters: lactic acid, citric acid Preservatives: methylparaben, chlorphenesin, phenoxyethanol Oil agents: isostearyl isostearate, isononyl isononanoate,
Isopropyl Isostearate, Ethylhexyl Hydroxystearate Humectants: Dipropylene Glycol, Propylene Glycol, Glycerin,
Diglycerin, 1,3-butylene glycol Boiling point, melting point, molecular weight, octanol-water partition coefficient (log Kow), skin permeability coefficient, Hildebrand solubility parameter (SP), Hansen solubility parameter (HSP) ) δd, δp, and δh, Abraham solvation parameters (ASP A, ASP B, ASP E, ASP L, ASP S, and ASP V), and IOB (Inorganic Organic Balance) were obtained. These physicochemical properties were calculated by ChemDraw, EpiSuite and HSPiP. The skin permeation coefficient (log P) was calculated using Potts and Guy's skin permeation coefficient prediction formula (Pharm Res., 9, 663-669, 1992). In addition, as the physicochemical properties of the target substance, Chi Connectivity Indices (x0, x1, x2, xp3 to xp10, xc3, xc4, xpc4, knotp), Kappa Shape Indices, Electrotopological State Indices, Topological description 798 molecules such as tors Descriptors were determined using The Toxicity Estimation Software Tool (EPA TEST; [www.epa.gov/chemical-research/toxicity-estimation-software-tool-test]).

3) Collection of skin characteristic data Skin characteristics include stratum corneum water content, transepidermal water loss (TEWL), skin pH, electrodermal threshold (CPT5 and CPT250) at 5 Hz and 250 Hz, sebum content, skin color (L*, a* and b*) and whole face skin temperature distribution were measured. After washing the face, the subject was acclimatized in a constant temperature and humidity room (temperature 22±2° C., humidity 50±5%) for 15 minutes, and then various skin characteristics were measured. In the skin characteristic measurement, the subject took a full face photograph using VISIA-CR (Canfield Scientific Inc.), whole face skin temperature, stratum corneum water content, TEWL, skin pH, skin color, CPT5 and CPT250, The amount of sebum was measured and the stratum corneum was collected. The whole face skin temperature was measured once using a thermotracer (Thermotracer TH3100; NEC Sanei Co., Ltd.). The stratum corneum moisture content was measured five times using a corneometer (Corneometer CM825; Courage+Khazaka electronic GmbH). TEWL was measured for 2 minutes using a Tewameter (Tewameter TM300; Courage+Khazaka electronic GmbH). Skin pH was measured once using a Skin-pH-meter (LAQUAact; HORIBA Ltd.). The skin color was measured three times using a Chromameter (Chromameter CR400 or Chromameter CM700d; KONICA MINOLTA JAPAN, INC.). CPT was measured once at 250 Hz and then 5 Hz using a Neurometer (Neurometer CPT/C; Neurotron, Inc.). The amount of sebum was measured once using a Sebumeter (Sebumeter SM815; Courage+Khazaka electronic GmbH). Five samples of the stratum corneum were collected from the same site using a tape.

4) Stinging Test Subjects were acclimatized in a constant temperature and humidity room (temperature 22±2° C., humidity 50±5%) after washing their faces for 15 minutes, and then subjected to the test according to the following procedure.
(i) 20 µL of the target substance solution was applied to the right cheek (or left cheek).
(ii) Immediate, 2.5 minutes and 5 minutes after application, subjects were asked to record the discomfort caused by the application according to the following criteria.
0 = No discomfort 0.5 = Slight discomfort, but not much 1.0 = Slight discomfort, but tolerable 1.5 = Slight to moderate discomfort 0 = Slight discomfort, but tolerable 2.5 = Slight to considerable discomfort 3.0 = Considerable discomfort, unbearable (iii) Moisten the application site with deionized water. It was gently wiped off with a damp cotton, and then wiped off with a dry cotton.
(iv) After waiting for 3 minutes, 20 μL of another target substance solution was applied to the cheek on the opposite side, and (ii) to (iii) were repeated.
By repeating (i)-(iv) above, up to 4 substances of interest were tested on the same day for one subject. Testing of the remaining target substances was performed the next day. Fourteen substances of interest were tested on each subject over four days. Table 1 shows the target substance solutions used in the test.

5) Calculation of the presence or absence of sensory irritation of the target substance 4) Based on the discomfort score of each test substance measured in (ii), the presence or absence of sensory irritation of each test substance on the skin of each subject was determined. . Specifically, if all three scores are 0 immediately after, 2.5 minutes and 5 minutes after application of one test substance obtained from one subject, the test substance is applied to the skin of the subject. When the score was 0.5 or more even once, it was regarded as having sensory irritation. A similar procedure was repeated to determine the presence or absence of organoleptic effects of each test substance on the skin of each subject.

2. Correlation between skin characteristics and skin substance sensitivity 1) Comparison between races When comparing skin characteristics and stinging test results between Japanese and German subjects, Japanese and German subjects Overall, they showed different properties in both skin properties and stinging test. Compared to Germans, the stratum corneum water content, skin pH, sebum content, and skin color b ^* values were significantly higher in Japanese, while TEWL, CPT, and skin color L ^* and a* values were significantly lower. rice field. In the stinging test, the Japanese are more sensitive to substances with higher organoleptic scores, while the Germans are more sensitive to substances with lower organoleptic scores. A trend was observed.

2) Hierarchical cluster analysis Hierarchical cluster analysis based on measurements of stratum corneum water content, TEWL, skin pH, CPT250, CPT5, sebum content, and skin color (L*, a* and b*) was conducted to determine the number of Japanese subjects and German subjects. Human subjects were each classified into five skin types (J-Type 1-5 and G-Type 1-5) (Table 2). As a result, two skin types (J and G-Type 1 and J and G-Type 2) common to Japanese and German were observed. Type 1 had atopic dry skin-like skin characteristics, namely, abnormal barrier function indicated by low stratum corneum moisture content and high TEWL, and dry skin indicated by low sebum content. , had a high skin pH, tended to be reddish and dull, and had a low threshold to electrical stimulation. Type 2 was apparently healthy skin based on skin color values, but was highly sensitive to electrical stimulation. Other skin types (J-Type 3-5, and G-Type 3-5) were unique to Japanese and Germans, respectively.
The results of the stinging test (target substances with high/low organoleptic score) were compared for each skin type. Japanese and German Type 1 (J and G-Type 1) and Type 2 (J and G-Type 2) tended to have high or low susceptibility to common substances, respectively. This suggests that individuals with similar skin characteristics have similar substance responsiveness, thus suggesting that skin characteristics are the main factors contributing to skin sensitivity to substances.

3) Analysis by skin temperature Fig. 2 shows the average skin temperature of the forehead, nose and cheeks of Japanese subjects for each skin type classified in 2). Type 1 had higher cheek temperature than other types. Type 2 tended to have low temperatures in the nose and cheeks, and low temperatures from the forehead to the tip of the chin. The skin temperature of Type3 and Type4 showed a contrasting trend, with Type3 having high skin temperature over the entire face, while Type4 had low skin temperature. Type5 had lower nasal temperatures than other sites.
As described above, the facial skin temperature distribution showed different characteristics for each of the skin types classified in 2), indicating that it can serve as an index of skin sensitivity to substances.

3. Construction of a model for analyzing the sensory irritation of substances to the skin by machine learning Skin characteristics of 200 subjects (120 Japanese, 80 Germans) a* and b*), CPT5 and CPT250, sebum content), and physicochemical properties of 14 target substances (boiling point, melting point, molecular weight, octanol-water partition coefficient, skin permeability coefficient, SP, HSP δd, HSP δp , HSP δh, ASP A, ASP B, ASP E, ASP L, ASP S, ASP V, IOB, and Chi Connectivity Indices (x0, x1, x2, xp3-xp10, xc3, xc4, xpc4, knotp), Kappa Shape 798 types of molecular descriptors such as Indices, Electrotopological State Indices, Topological descriptors) are used as explanatory variables, and the presence or absence of sensory irritation of the target substance (the value obtained in 1.5) above) is used as the objective variable. bottom. A discriminator was created that, when inputting the skin characteristics of a subject and the physicochemical characteristics of a substance, outputs whether or not the subject feels uncomfortable with the substance. Furthermore, a classifier was created in the same way by adding race (ethnicity) as a dummy variable to the explanatory variables. Table 3 shows the percentage of correct answers obtained by 10-fold cross-validation for the created discriminators. The random forest had the highest accuracy rate, and the naive Bayes classifier had the lowest. The presence or absence of dummy variables did not significantly affect the correct answer rate. This suggests that race is less important for skin sensitivity to substances, and that physicochemical properties of substances and skin characteristics are the main factors in skin sensitivity to substances.

Among the variables (skin properties and physicochemical properties) used in the random forest calculations, those with a large decrease in impurity were extracted as features of high importance. Fig. 3 shows the feature values and the amount of impurity reduction caused by them. Among the physico-chemical properties of substances are melting point, boiling point, skin permeability coefficient, molecular weight, octanol-water partition coefficient, solubility parameters (Hildebrand solubility parameter; SP, Hansen solubility parameter; HSP δh, δp and δd, and Abraham solvent Sum parameters; and skin L*, a* and b* values were important features contributing to the reduction of impurities.

Furthermore, by selecting about 20 of the feature quantities with high importance and using only the selected features by the random forest method, when the subject's skin characteristics and the physicochemical characteristics of the substance are entered, the substance is the relevant A discriminator was created that outputs whether or not the test subject's skin is uncomfortable. The performance of the created discriminator was evaluated by OOB (Out-Of-Bag). As a result of the evaluation, among the physicochemical properties of the quality, the melting point, molecular weight, octanol-water partition coefficient, and skin permeability coefficient, and among the skin properties, the stratum corneum moisture content, TEWL, and pH are particularly important features for identification. It turned out to be

Claims (7)

A method for analyzing the sensory irritation of a substance to the skin, comprising:
obtaining the physicochemical properties of the target substance;
Acquiring the skin characteristics of the skin, and analyzing the sensory irritation of the target substance to the skin by a machine learning model based on the physicochemical characteristics of the target substance and the skin characteristics of the skin;
including
the skin properties include stratum corneum water content, TEWL and pH of the skin;
the physico-chemical properties include melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient;
Method. The machine learning model is:
Skin characteristics of a subject for model building, including at least stratum corneum water content, TEWL and pH of the skin;
Physicochemical properties of the modeling test material, including at least melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient, wherein the modeling test material comprises at least pH modifiers, preservatives, polar oils and moisturizers. an agent; and
Measurement results of the sensory irritation of the model construction test substance on the skin of the model construction subject,
2. The method of claim 1, wherein the model is constructed by machine learning using A method for constructing a model for analyzing the sensory irritation of a substance to the skin, comprising the following:
Measurements of stratum corneum water content, TEWL and pH of skin for multiple subjects;
Melting points, molecular weights, octanol-water partition coefficients, and skin permeation coefficients for a plurality of substances; and results of measurements of the organoleptic properties of the substances on the skin of the plurality of subjects;
to create a classifier for identifying the presence or absence of sensory irritation to the skin of substances by machine learning with descriptors,
including,
Method. A program for calculating the sensory irritation of the target substance on the skin of the subject using a machine learning model from the data of the skin characteristics of the subject and the physicochemical properties of the target substance,
the skin properties include stratum corneum water content, TEWL and pH of the skin;
the physico-chemical properties include melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient;
program. The machine learning model is:
Skin characteristics of a subject for model building, including at least stratum corneum water content, TEWL and pH of the skin;
Physicochemical properties of the modeling test material, including at least melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient, wherein the modeling test material comprises at least pH modifiers, preservatives, polar oils and moisturizers. an agent; and
Measurement results of the sensory irritation of the model construction test substance on the skin of the model construction subject,
5. The program according to claim 4, which is a model constructed by machine learning using. A device for analyzing the sensory irritation of a target substance to the skin of a subject,
A calculation unit for calculating the sensory irritation of the target substance on the subject's skin using a machine learning model from data on the subject's skin characteristics and the physicochemical properties of the target substance, and
an output unit for outputting the result of the operation;
equipped with
the skin properties include stratum corneum water content, TEWL and pH of the skin;
the physico-chemical properties include melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient;
Device. The machine learning model is:
Skin characteristics of a subject for model building, including at least stratum corneum water content, TEWL and pH of the skin;
Physicochemical properties of the modeling test material, including at least melting point, molecular weight, octanol-water partition coefficient, and skin permeation coefficient, wherein the modeling test material comprises at least pH modifiers, preservatives, polar oils and moisturizers. an agent; and
Measurement results of the sensory irritation of the model construction test substance on the skin of the model construction subject,
7. The apparatus of claim 6, wherein the model is constructed by machine learning using

Images