WO2025084419A1 - Method for classifying skin type - Google Patents

Abstract

To provide a method for classifying the skin type of a subject who is a cosmetic user by using biological information from the skin.　Provided is a method for classifying the skin of a subject, the classification being into skin types generated from two clusters obtained by performing cluster analysis and grouping based on the expression state of genes contained in skin surface lipids of a plurality of healthy persons. The method includes a step for acquiring biological information from the skin of the subject and a step for determining from the biological information the skin type classification of the subject.

Description

How to classify skin types

The present invention relates to a method for classifying skin types using biometric information from the skin.

In recent years, a wide variety of cosmetics have been provided to suit the skin characteristics and skin concerns of the cosmetic user. For example, there are cosmetics for different skin types, such as for dry skin, combination skin, and oily skin, cosmetics to solve skin concerns such as blemishes, sagging skin, and wrinkles, as well as cosmetics that block ultraviolet rays, and a wide variety of other cosmetics.

Conventionally, in order to provide a cosmetic product suited to the individual skin characteristics of a user, counseling has been provided based on the results of a self-diagnosis questionnaire by the user and the measurement results of skin information such as texture, skin barrier function, moisture content, elasticity, and viscosity.
However, the user's self-diagnosis is a subjective judgment and does not accurately reflect the user's skin characteristics. Moreover, the objective information measured by the device is merely a temporary numerical representation of the measurable skin condition, which may change from day to day.

In recent years, technology has been developed to investigate the current and future physiological state of the human body by analyzing nucleic acids such as DNA and RNA in biological samples, and a method has been reported for evaluating human skin characteristics based on genetic factors using SNP analysis (Patent Document 1).

On the other hand, it has been discovered that RNA contained in skin surface lipids (SSL) can be used as a sample for biological analysis, and it has been reported that marker genes for the epidermis, sweat glands, hair follicles, and sebaceous glands can be detected from SSL (Patent Document 2).

[Patent Document 1] International Publication No. 2021/029332 [Patent Document 2] International Publication No. 2018/008319

The present invention relates to the following 1) and 2).
1) A method for classifying the skin of a subject, the classification being based on the expression state of genes contained in lipids on the skin surface of a plurality of healthy subjects and being generated from two clusters observed;
A method comprising: acquiring biometric information from the skin of a subject; and determining from the biometric information whether the subject is classified into one of the skin types.
2) A computing device for executing the method described in 1), the device having a means for inputting biometric information from the skin obtained from a subject, and executing a step of determining which of the skin types the subject is classified into from the biometric information in accordance with a program for executing a method for classifying the subject's skin.

Hierarchical clustering analysis using SSL-RNA. Hierarchical clustering analysis of keratinization and immune response using SSL-RNA.

Detailed Description of the Invention

The present invention relates to providing a method for classifying the skin type of a subject who is a cosmetic user, using biometric information from the skin.

The inventors collected SSL from several healthy female subjects and performed cluster analysis based on the expression state of the RNA contained in the SSL to classify it into two clusters, revealing that each cluster has its own characteristic gene expression. Furthermore, they found that it is possible to classify the skin type of subjects who are cosmetic users based on these.

The present invention allows for classification of the skin of cosmetic users, contributing to the selection of optimal cosmetics for each skin type or the provision of customized cosmetics.

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

In the present invention, "DNA" includes cDNA, genomic DNA, and synthetic DNA, and "RNA" includes total RNA, mRNA, rRNA, tRNA, non-coding RNA, and synthetic RNA.

In the present invention, the term "gene" refers to double-stranded DNA including human genomic DNA, as well as single-stranded DNA (positive strand) including cDNA, single-stranded DNA (complementary strand) having a sequence complementary to the positive strand, and fragments thereof, and refers to DNA in which some biological information is contained in the sequence information of the bases constituting the DNA.
Furthermore, the "gene" in question includes not only "genes" represented by a specific base sequence, but also nucleic acids encoding their homologues (i.e., homologs or orthologs), mutants such as gene polymorphisms, and derivatives.

The method for classifying skin of the present invention is a method for classifying a subject's skin into a skin type generated from two clusters recognized by a cluster analysis based on the expression states of genes contained in skin surface lipids of a plurality of healthy subjects,
The method includes a step of acquiring biometric information from the skin of a subject, and a step of determining, from the biometric information, to which of the skin types the subject is classified.

1. Generation of Skin Type In the method of classifying skin of the present invention, the expression states of genes contained in lipids on the skin surface of a plurality of healthy subjects are obtained in advance, and based on this, cluster analysis is performed to classify the skin into two clusters, and a skin type is generated based on the classified clusters.
Here, the subject "healthy person" is a person with good skin condition, preferably a person who has not been diagnosed with a skin disease by a doctor. In addition, although age and sex are not important, the expression state of genes contained in the lipids on the skin surface of multiple healthy people for each age, generation, and sex may be obtained according to the age and sex of the subject to be subjected to skin classification, and a skin type may be generated for each. For example, if the subject is a woman, the skin type is generated from the expression state of genes contained in the lipids on the skin surface of multiple healthy female people, if the subject is a man ... male people, if the subject is both male and female, the skin type is generated from the expression state of genes contained in the lipids on the skin surface of multiple healthy female people and healthy male people.
Here, "plurality" refers to an integer of 2 or more, preferably 10 or more, more preferably 100 or more, even more preferably 150 or more, and still more preferably 250 or more.

In the present invention, the biological sample used for gene expression analysis is skin surface lipids (SSL). "Skin surface lipids (SSL)" refers to the fat-soluble fraction present on the surface of the skin, and is also called sebum. In general, SSL mainly contains secretions from exocrine glands such as sebaceous glands in the skin, and exists on the skin surface in the form of a thin layer that covers the skin surface. SSL contains RNA expressed in skin cells (see Patent Document 2). In this specification, "skin" is a general term for the area including the epidermis, dermis, hair follicles, and tissues such as sweat glands, sebaceous glands, and other glands on the body surface, unless otherwise specified.

To collect SSL from the skin, any means used for recovering or removing SSL from the skin can be used. Preferably, an SSL absorbent material, an SSL adhesive material, or an instrument for scraping SSL from the skin, which will be described later, can be used. The SSL absorbent material or SSL adhesive material is not particularly limited as long as it is a material having affinity for SSL, and examples thereof include polypropylene, pulp, etc. More detailed examples of procedures for collecting SSL from the skin include a method of absorbing SSL into a sheet-like material such as oil blotting paper or oil blotting film, a method of adhering SSL to a glass plate, tape, etc., and a method of scraping SSL off and collecting it with a spatula, scraper, etc. In order to improve the adsorption of SSL, an SSL absorbent material that has been previously impregnated with a highly lipid-soluble solvent may be used. On the other hand, it is preferable that the SSL absorbent material contains a low content of highly water-soluble solvent or water, since the adsorption of SSL is inhibited if the SSL absorbent material contains a highly water-soluble solvent or water. The SSL absorbent material is preferably used in a dry state.
The site of skin from which SSL is collected is not particularly limited and may be any site of the body such as the head, face, neck, trunk, hands and feet, but is preferably from a site that secretes a lot of sebum, such as the skin of the face.

The collected RNA-containing SSL may be stored for a certain period of time. In order to minimize the degradation of the RNA contained therein, it is preferable to store the collected SSL under low-temperature conditions as soon as possible after collection. The temperature conditions for storing the RNA-containing SSL in the present invention may be 0°C or lower, and are preferably -20±20°C to -80±20°C, more preferably -20±10°C to -80±10°C, even more preferably -20±20°C to -40±20°C, even more preferably -20±10°C to -40±10°C, even more preferably -20±10°C, and even more preferably -20±5°C. The period for storing the RNA-containing SSL under the low-temperature conditions is not particularly limited, but is preferably 12 months or less, for example, 6 hours or more and 12 months or less, more preferably 6 months or less, for example, 1 day or more and 6 months or less, even more preferably 3 months or less, for example, 3 days or more and 3 months or less.

In the present invention, the expression state of a gene means an index showing the degree or tendency of gene expression, and the expression state of a gene contained in an SSL is obtained by measuring the expression level (expression amount) of RNA, specifically, by converting RNA into cDNA by reverse transcription, and then measuring the cDNA or its amplification product.
For extraction of RNA from SSL, a method that is usually used for extracting or purifying RNA from a biological sample, such as the phenol/chloroform method, the AGPC (acid guanidinium thiocyanate-phenol-chloroform extraction) method, or a method using columns such as TRIzol (registered trademark), RNeasy (registered trademark), or QIAzol (registered trademark), a method using special silica-coated magnetic particles, a method using Solid Phase Reversible Immobilization magnetic particles, or extraction using a commercially available RNA extraction reagent such as ISOGEN, can be used.

For the reverse transcription, a primer targeting a specific RNA to be analyzed may be used, but for more comprehensive nucleic acid storage and analysis, it is preferable to use a random primer. For the reverse transcription, a general reverse transcriptase or reverse transcription reagent kit can be used. Preferably, a highly accurate and efficient reverse transcriptase or reverse transcription reagent kit is used, and examples thereof include M-MLV Reverse Transcriptase and its variants, or commercially available reverse transcriptases or reverse transcription reagent kits, such as the PrimeScript (registered trademark) Reverse Transcriptase series (Takara Bio Inc.), the SuperScript (registered trademark) Reverse Transcriptase series (Thermo Scientific Inc.), and the like. SuperScript (registered trademark) III Reverse Transcriptase, SuperScript (registered trademark) VILO cDNA Synthesis kit (both manufactured by Thermo Scientific), etc. are preferably used.
The temperature of the extension reaction in the reverse transcription is preferably adjusted to 42° C.±1° C., more preferably 42° C.±0.5° C., and even more preferably 42° C.±0.25° C., while the reaction time is preferably adjusted to 60 minutes or more, more preferably 80 to 120 minutes.

The expression level of RNA can be measured, for example, by quantifying the cDNA or its amplified product by real-time PCR, multiplex PCR, microarray, sequencing, chromatography, etc. In one embodiment of the method of the present invention, a library containing DNA derived from expressed genes is prepared using multiplex PCR, the prepared library is sequenced by a next-generation sequencer, and calculation is performed based on the number of reads created (read count).

Multiplex PCR is a method for amplifying multiple gene regions simultaneously by using multiple primer pairs in a PCR reaction system. Multiplex PCR can be performed using a commercially available kit (e.g., Ion AmpliSeq Transcriptome Human Gene Expression Kit; Life Technologies Japan, Inc.) under conditions such as 99°C, 2 minutes → (99°C, 15 seconds → 62°C, 16 minutes) x 20 cycles → 4°C, hold.

The purification of the reaction products obtained by the PCR is preferably carried out by size separation of the reaction products. By size separation, the target PCR reaction products can be separated from primers and other impurities contained in the PCR reaction solution. Size separation of DNA can be carried out, for example, by using a size separation column, a size separation chip, magnetic beads that can be used for size separation, etc. A preferred example of magnetic beads that can be used for size separation is Solid Phase Reversible Immobilization (SPRI) magnetic beads such as Ampure XP.

For DNA sequencing, the purified PCR reaction products are prepared in an appropriate buffer solution, the PCR primer regions contained in the PCR-amplified DNA are cleaved, and an adapter sequence is further added to the amplified DNA. For example, the purified PCR reaction products are prepared in a buffer solution, the PCR primer sequences are removed from the amplified DNA and adapter ligation is performed, and the resulting reaction products are amplified as necessary to prepare a library for sequencing. These operations can be performed, for example, using the 5x VILO RT Reaction Mix included with the SuperScript (registered trademark) VILO cDNA Synthesis kit (Life Technologies Japan, Inc.) and the 5x Ion AmpliSeq HiFi Mix and Ion AmpliSeq Transcriptome Human Gene Expression Kit (Life Technologies Japan, Inc.) according to the protocols included with each kit.

Sequencing can be performed using a next-generation sequencer (e.g., Ion S5/XL system, Life Technologies Japan, Inc.), and the gene derived from each read sequence is determined by genetically mapping each read sequence obtained by sequencing to the hg19 AmpliSeq Transcriptome ERCC v1, which is the reference sequence of the human genome. RNA expression levels are calculated based on the number of reads (read count) generated by sequencing.

The read count value, which is expression level data obtained by sequencing, is appropriately corrected.
Here, examples of the corrected value for the read count value include an RPM value obtained by correcting the difference in the total number of reads between samples for the read count value, a value obtained by converting the RPM value to a logarithmic value with base 2 (Log ₂ RPM value), or a logarithmic value with base 2 with an integer 1 added (Log ₂ (RPM+1) value), or a count value corrected using DESeq2 (Love MI et al. Genome Biol. 2014) (Normalized count value), or a logarithmic value with base 2 with an integer 1 added (Log ₂ (count+1) value), with the Normalized count value corrected using DESeq2 being preferred.

Based on the obtained information on the expression state of genes contained in the lipids on the skin surface, cluster analysis is performed.The method of cluster analysis is not particularly limited as long as it can classify a plurality of objects into two or more clusters, and may be hierarchical clustering or non-hierarchical clustering.Cluster analysis can be performed by software capable of statistical analysis.
For example, as shown in the Examples described later, 281 SSL samples taken from the entire face of healthy female subjects aged 20 to 60 years were targeted, and based on the expression levels of RNA derived from the SSL (normalized count values), a correlation coefficient ρ was calculated by Spearman's rank correlation analysis between the samples, and a hierarchical clustering analysis was performed with the distance between the samples set to 1-ρ, resulting in classification into two clusters, Cluster 1 and Cluster 2, as shown in FIG. 1.

In the present invention, it is possible to generate skin types associated with each of the two clusters based on the gene expression information that characterizes the clusters.

When generating a skin type based on gene expression information, as shown in the Examples described below, when RNA is extracted in which the SSL-derived RNA expression ratio between the two clusters is 1.2 times or more and the corrected p-value (FDR) by a likelihood ratio test is less than 0.05, 614 genes whose expression is relatively higher in Cluster 1 compared to Cluster 2 and 855 genes whose expression is relatively higher in Cluster 2 compared to Cluster 1 are identified. It was confirmed by gene ontology (GO) enrichment analysis that the group of genes whose expression is relatively higher in Cluster 1 includes many genes related to epidermal keratinization, and the group of genes whose expression is relatively higher in Cluster 2 includes many genes related to immune response. That is, the keratinization-related genes with relatively high expression in Cluster 1 include genes annotated with GO terms such as skin development (GO: 0043588) and epidermis development (GO: 0008544), while the immune response-related genes with relatively high expression in Cluster 2 include genes annotated with GO terms such as immune system process (GO: 0002376) and response to cytokine (GO: 0034097). Note that GO terms (Gene Ontology) are common vocabulary for explaining the biological concept of a gene (biological process of a gene, cellular components and molecular functions), and GO terms are annotated (linked) to genes whose functions have been clarified. This allows each gene to be assigned a GO term that describes its function, and the GO term can be used to identify the gene to which the term is assigned.
As shown in the Examples described later, cluster analysis was performed based on information on the expression states of genes annotated with the keratinization-related GO terms skin development (GO:0043588) and epidermis development (GO:0008544), and genes annotated with the immune response-related GO terms immune system process (GO:0002376) and response to cytokine (GO:0034097). The clusters classified into two (Cluster 1' and Cluster 2', FIG. 2) were almost equivalent to Cluster 1 and Cluster 2 (Table 1). Therefore, it can be said that Cluster 1 is a group of high expression of keratinization-related genes, and Cluster 2 is a group of high expression of immune response-related genes. From these two clusters, two skin types are generated: a skin type with high expression of keratinization-related genes and a skin type with high expression of immune response-related genes, which are found in the skin of healthy women.

Furthermore, by analyzing various biological information from the skin of healthy individuals belonging to the Cluster 1 and Cluster 1' high keratinization-related gene expression group and the Cluster 2 and Cluster 2' high immune response-related gene expression group, it is possible to link the characteristic biological information of each group to the two skin types generated. As biological information from the skin, any information that can be obtained from the skin can be used, regardless of whether it is physiological information or physical information.

2. Classification of Skin Type of a Subject In the present invention, biological information is obtained from the skin of a subject, and it is determined from the biological information whether the subject is classified into one of the two skin types generated from the cluster analysis.
Here, the "subject" may be a cosmetic product user or a potential user considering using the cosmetic product, and there are no limitations on gender or age.
The term "biological information from the skin" is not particularly limited as long as it is information obtainable from the skin, regardless of whether it is physiological information or physical information. Specifically, examples include information on gene expression state in the SSL used in the above-mentioned cluster analysis, information on protein expression state and metabolites in the SSL, information on gene expression state, protein expression state and metabolites in biological samples that can be collected from the face (skin biopsy, stratum corneum, sweat, tears, saliva, etc.), information on gene expression state, protein expression state and metabolites of the flora of normal skin bacteria or bacteria-derived gene expression state, protein expression state and metabolites, and measurement of skin property values by instrumental measurement. Preferable instrumental measurements include transepidermal water loss, stratum corneum moisture content, sebum content, skin glycation degree, skin viscoelasticity, skin color, hemoglobin content, melanin content, skin blood flow, skin temperature, biophotons, etc.
When information on the gene expression state in SSL is used as biological information from the skin, collection of SSL and measurement of gene expression levels in SSL can be performed in the same manner as described above.

The skin type of the subject based on the measured gene expression state can be determined, for example, by 1) determining the skin type from the distance between the center of gravity of the two pre-generated clusters (Cluster 1 and Cluster 2, or Cluster 1' and Cluster 2') and the subject's SSL-derived RNA expression level, or by 2) determining the skin type based on a comparison with the expression level of a specific gene related to keratinization or immune response.
The method of determining the distance between the center of gravity of the two clusters and the expression level of SSL-derived RNA of the subject can be carried out by calculating the distance (similarity) between the two clusters and the expression level of SSL-derived RNA obtained from the subject.
For example, for two clusters (Cluster 1 and Cluster 2, or Cluster 1' and Cluster 2') that are populations, a center of gravity is calculated from a multidimensional vector with the SSL-derived RNA expression level as a variable, and the distance between the multidimensional vector based on the SSL-derived RNA expression level obtained from the subject and the center of gravity of the two clusters is calculated, and the cluster with a closer distance is determined as the type of the subject. As for the distance, it is also possible to calculate the Spearman's rank correlation coefficient ρ between the center of gravity and the subject's data as in the cluster analysis, and use 1-ρ. In addition, at this time, as the variable that determines the center of gravity or the RNA expression level obtained from the subject, it is also possible to narrow it down to genes that are relatively highly expressed in Cluster 1 compared to Cluster 2, or genes that are relatively highly expressed in Cluster 2 compared to Cluster 1, and genes that are annotated with GO terms related to keratinization or immune response.

In a method for determining skin type based on a comparison with the expression levels of specific genes related to keratinization or immune response, one or more gene species with no or less overlap in expression levels between two clusters (Cluster 1 and Cluster 2, or Cluster 1' and Cluster 2') that are the population are selected, a threshold for the expression level is set based on statistical values such as the average value and standard deviation, and the type can be determined from the SSL-derived RNA expression level of the relevant gene species obtained from the subject. Here, the specific gene related to keratinization or immune response is preferably selected from genes annotated with the keratinization-related or immune response-related GO terms, and from the viewpoint of the accuracy of distinguishing between the two clusters, it is more preferable to select from the 25 genes shown in Table 7 below.

In one embodiment, the expression level of one or more genes selected from the 25 genes is obtained from a subject, and the expression level is compared with a reference value to determine the skin type of the subject. Here, the reference value can be a threshold value in the Youden index of the ROC curve.

In another embodiment, the expression levels of one or more genes selected from specific genes related to keratinization or immune response are obtained from a subject, and the expression levels are substituted into a discriminant (discriminant model) for detecting skin type, thereby determining the skin type of the subject.
Specifically, a discriminant equation can be constructed by machine learning using an arbitrary group as a teacher sample, obtaining the expression level of a specific gene related to keratinization or immune response from each person in the teacher sample as an explanatory variable and using each person's skin type as a target variable, and then measuring the expression level of the specific gene in the subject and substituting the expression level into the discriminant equation to discriminate the skin type of the subject.
Here, the skin type of each person in the teacher sample can be determined by carrying out the above-mentioned cluster analysis based on information on the expression state of genes contained in the lipids on the skin surface of each person.

The specific genes related to keratinization or immune response used to construct the discriminant equation can be selected from genes annotated with the keratinization-related or immune response-related GO terms, but it is preferable to select one or more from the 25 genes shown in Table 7 below.

As an algorithm for constructing the discriminant, known algorithms such as algorithms used in machine learning can be used. Examples of machine learning algorithms include random forest, decision tree, gradient boosting, XGBossti (eXtreme Gradient Boosting), lightGBM, support vector machine with linear kernel (SVM linear), support vector machine with rbf kernel (SVM rbf), logistic regression, and regularized logistic regression. Examples of such a prediction model include a generalized linear model, a regularized linear discriminant analysis, a k-nearest neighbors method, a neural net, a multi-layer perceptron, a convolutional neural network, and a recurrent neural network. Verification data is input into the constructed prediction model to calculate a predicted value, and the model whose predicted value is most compatible with the actual measured value, for example, the model with the highest accuracy, can be selected as the optimal prediction model. In addition, the detection rate (Recall), accuracy (Precision), and their harmonic mean, the F-value, are calculated from the predicted and measured values, and the model with the largest F-value can be selected as the optimal prediction model.

When using information other than information on gene expression state in SSL as biometric information from the skin, it is preferable to use biometric information linked to the two clusters (Cluster 1 and Cluster 2, or Cluster 1' and Cluster 2') that have been generated. For the two clusters (Cluster 1 and Cluster 2, or Cluster 1' and Cluster 2') that are the population, the biometric information is acquired in advance, and a criterion for classifying into one of the two clusters, such as a threshold, is set, or a discriminant for classifying into one of the two clusters by machine learning is constructed. Then, based on the biometric information acquired from the subject, it can be determined which of the two clusters generated skin types the subject falls into.

The method of classifying the subject's skin of the present invention can be performed using a computing device (computer). Therefore, in another aspect, the present invention provides a computing device for executing the above-mentioned method, a program for causing the computing device to execute the above-mentioned method, and an information recording medium on which the program is recorded and which can be read by the computing device.

The computing device of the present invention has a means for inputting biometric information obtained from the skin of a subject, and includes a step of executing the above-mentioned method of determining which of the above-mentioned skin types the subject is classified into from the biometric information, in accordance with a program for executing the method of classifying the subject's skin of the present invention.

　Examples of information recording media readable by a computing device on which a program for executing the method of classifying a subject's skin of the present invention is recorded include magnetic disks, optical disks, magneto-optical disks, and flash memories. Note that in the present invention, "readable by a computing device" also includes cases where the program is distributed via a telecommunications line, etc.

In relation to the above-described embodiment, the present invention further discloses the following aspects.
<1> A method for classifying the skin of a subject, the classification being into skin types generated from two clusters obtained by performing cluster analysis based on the expression states of genes contained in lipids on the skin surface of a plurality of healthy subjects;
A method comprising: acquiring biometric information from the skin of a subject; and determining from the biometric information whether the subject is classified into one of the skin types.
<2> The method described in <1>, comprising a preparatory step of performing cluster analysis based on the expression state of genes contained in the skin surface lipids of multiple healthy subjects, dividing the subjects into two clusters, and generating skin types from each cluster.
<3> The method according to <1> or <2>, wherein the gene includes a gene annotated with at least one GO term selected from skin development (GO: 0043588), epidermis development (GO: 0008544), immune system process (GO: 0002376, and response to cytokine (GO: 0034097).
<4> The method according to any one of <1> to <3>, wherein the biological information from the skin is an expression state of genes contained in lipids on the skin surface of the subject.
<5> The method according to <4>, wherein the expression state of the gene is an expression state of a gene annotated with at least one GO term selected from skin development (GO: 0043588), epidermis development (GO: 0008544), immune system process (GO: 0002376, and response to cytokine (GO: 0034097).
<6> The method according to <4>, wherein the expression state of the gene is the expression state of at least one gene selected from 25 genes: ASPRV1, KRT17, KRT80, KRT79, DNASE1L2, SPRR1A, DSP, CDSN, CST6, LCP1, KRT6B, LCE1C, CALML5, CD83, JUP, TYROBP, SPINT1, CNFN, TMSB4X, NFKB1, B2M, MSN, CXCL16, CD58 and KRT72.
<7> The method according to any one of <1> to <6>, wherein the biological information from the skin of the subject is information selected from 1) information on the expression state of genes in the SSL used in the cluster analysis, the expression state of proteins in the SSL, or metabolites, 2) information on the expression state of genes, the expression state of proteins, or metabolites in a biological sample that can be taken from the face, 3) information on the flora of normal skin bacteria or the expression state of genes, the expression state of proteins, or metabolites derived from the bacteria, and 4) measurements of skin property values obtained by instrumental measurement.
<8> The method according to <7>, wherein the skin property values measured by an instrument are one or more selected from transepidermal water loss, stratum corneum moisture content, sebum content, skin glycation level, skin viscoelasticity, skin color, hemoglobin content, melanin content, skin blood flow, skin temperature, and biophotons.
<9> The method according to <7>, wherein the skin type of the subject based on the expression state of genes in SSL is determined by either 1) a method of determining from the distance between the center of gravity of two pre-generated clusters (Cluster 1 and Cluster 2, or Cluster 1' and Cluster 2') and the expression level of RNA derived from the subject's SSL, or 2) a method of determining based on a comparison with the expression level of a specific gene related to keratinization or immune response.
<10> The method according to <6>, further comprising obtaining an expression level of one or more genes selected from the 25 genes from a subject, and comparing the expression level with a reference value to determine the skin type of the subject.
<11> A calculation device for executing the method according to any one of <1> to <10>, the device having a means for inputting biometric information from the skin of a subject, and executing a step of determining which of the skin types the subject is classified into from the biometric information in accordance with a program for executing a method for classifying the skin of the subject.
<12> A program for causing a computer device to execute the method according to any one of <1> to <10>.
<13> An information recording medium on which the program according to <12> is recorded.

The present invention will be described in more detail below based on examples, but the present invention is not limited to these examples.
Example 1 Skin type classification analysis of SSL-RNA 1) SSL collection 281 sebum samples were collected from healthy female subjects aged 20 to 60 (24 samples in their 20s, 126 samples in their 30s, 29 samples in their 40s, and 102 samples over 50). Sebum was collected from the entire face of each subject using a sheet of oil blotting film (polypropylene, 5.0 cm x 8.0 cm, 3M).

2) RNA preparation and sequencing The oil blotting film from which sebum was collected in 1) above was cut to an appropriate size, and RNA was extracted using QIAzol Lysis Reagent (Qiagen) according to the attached protocol. Based on the extracted RNA, reverse transcription was performed at 42°C for 90 minutes using SuperScript VILO cDNA Synthesis kit (Life Technologies Japan, Inc.) to synthesize cDNA. The random primers included in the kit were used as primers for the reverse transcription reaction. A library containing DNA derived from the 20802 gene was prepared from the obtained cDNA by multiplex PCR. Multiplex PCR was performed using Ion AmpliSeqTranscriptome Human Gene Expression Kit (Life Technologies Japan, Inc.) under the conditions of [99°C, 2 minutes → (99°C, 15 seconds → 62°C, 16 minutes) × 20 cycles → 4°C, Hold]. The obtained PCR product was purified with Ampure XP (Beckman Coulter, Inc.), and then buffer reconstitution, digestion of primer sequences, adapter ligation and purification, and amplification were performed to prepare a library. The prepared library was loaded onto an Ion 540 Chip and sequenced using an Ion S5/XL system (Life Technologies Japan, Inc.). Each read sequence obtained by sequencing was genetically mapped to hg19 AmpliSeq Transcriptome ERCC v1, which is a reference sequence of the human genome, to determine the gene from which each read sequence originated.

3) Data Analysis The data of the expression amount (read count value) of the SSL-derived RNA of the subject obtained in 2) above was corrected using a method called DESeq2 (Love MI et al. Genome Biol. 2014). However, samples in which 4161 or more genes were not detected were excluded, and only 2605 types of genes for which expression amount data that was not a missing value was obtained in 90% or more of the sample subjects among the expression amount data of all samples after the exclusion were used in the following analysis. For the analysis, the count value corrected using the method called DESeq2 (Normalized count value) was used.
Based on the SSL-derived RNA expression level (normalized count value) obtained above, the correlation coefficient ρ was calculated by Spearman's rank correlation analysis between samples. At this time, the variance of the count value was corrected using a method called variance stabilizing transformation. Based on the correlation coefficient ρ, a hierarchical clustering analysis was performed in which the distance between samples was set to 1-ρ and the method of joining clusters was the Ward method. Based on the dendrogram obtained from the analysis, the samples were divided into two clusters (Cluster 1 and Cluster 2) (Figure 1).

By extracting RNA with an SSL-derived RNA expression ratio of 1.2 times or more between the two clusters and a corrected p-value (FDR) based on the likelihood ratio test of less than 0.05, 614 genes with relatively high expression in Cluster 1 and 855 genes with relatively high expression in Cluster 2 were identified.
A search for biological process (BP) was performed by gene ontology (GO) enrichment analysis using the public database PANTHER. As a result, 215 GO terms were obtained from the gene group with relatively high expression in Cluster 1, and 1287 GO terms were obtained from the gene group with relatively high expression in Cluster 2. Among these, there are terms related to keratinization and immune response, which are the main functions of the epidermal stratum corneum. In Cluster 1, the terms with the lowest FDR are "skin development (GO: 0043588, FDR: 5.23E-18)" and "epidermis development (GO: 0008544, FDR: 8.66E-17)". Three of the top five terms are related to keratinization. In Cluster 2, the terms with the lowest FDR are "immune system process (GO: 0002376, FDR: 1.84E-66)" and "response to cytokine (GO: 0034097, FDR: 7.22E-47)". " Four of the top five terms were related to immune response, with "

Example 2 Skin type classification analysis using SSL-RNA related to keratinization and immune response 1) Data used As in Example 1, based on the expression levels of SSL-derived RNA from the subjects (normalized count values), only 2605 types of genes for which non-missing expression level data was obtained in 90% or more of all samples were used in the following analysis.

2) Selection of keratinization/immune response-related genes From the 2,605 types of genes, 76 genes annotated with the GO term “skin development (GO:0043588)” and 83 genes annotated with “epidermis development (GO:0008544)” were extracted as genes related to keratinization, and 453 genes annotated with the GO term “immune system process (GO:0002376)” and 227 genes annotated with “response to cytokine (GO:0034097)” were extracted as genes related to immune response.

3) Data Analysis Based on the expression levels (normalized count values) of a total of 606 genes annotated with specific GO terms related to keratinization and immune response selected in 2) from the SSL-derived RNA, the correlation coefficient ρ was calculated by Spearman's rank correlation analysis between samples. At this time, the variance of the count values was corrected using a method called Valence Stabilizing Transformation. Based on the correlation coefficient ρ, a hierarchical clustering analysis was performed in which the distance between samples was set to 1-ρ and the method of joining clusters was the Ward method. Based on the dendrogram obtained by the analysis, the samples were divided into two clusters (Cluster 1' and Cluster 2') (Figure 2).

4) Comparison of cluster analysis results The two clusters observed in Example 1 were compared with the two clusters observed from the above data analysis. Of the 107 samples belonging to Cluster 1, 106 samples were included in Cluster 1', and of the 174 samples belonging to Cluster 2, 135 samples were included in Cluster 2', with a concordance rate of 85% or more (Table 1).

Example 3: Estimation of skin type using cluster centroid 1) Data used The expression data (read count value) of SSL-derived RNA obtained in Example 1 was converted to an RPM value corrected for the difference in the total number of reads between samples. However, only 2605 types of genes for which expression data that was not a missing value was obtained in 90% or more of all samples were used in the following analysis. In order to approximate the RPM value according to the negative binomial distribution to a normal distribution, the RPM value converted to the logarithmic value of base 2 (Log2RPM value) was used to calculate the centroid of each cluster.

2) Sample division The samples were divided into training samples (227 samples) and test samples (54 samples). The skin type of each sample was determined to be Cluster 1 or Cluster 2 as determined by the hierarchical clustering analysis in Example 1. The division was performed so that there was no bias in the proportion of Cluster 1 and Cluster 2 belonging to the training samples and the test samples, respectively (Table 2).

3) Calculation of the cluster centroid The expression amount data (Log2RPM value) (training data) of the gene to be analyzed of the SSL-derived RNA of the training sample was used as a variable, and the centroid (Centroid 1, Centroid 2) of each cluster (Cluster 1, Cluster 2) was calculated. In calculating the centroid, the expression amount data of each sample is expressed as a vector with the expression amount of each gene as a component, and the expression amount data is positioned as one data point in a 2605-dimensional space, which is the number of genes to be analyzed, and the average of the position vectors of the data points of all samples is used as the centroid. For example, if the data points belonging to Cluster 1 are x_1, x_2, ..., x_87, the centroid Centroid 1 of the cluster can be expressed as (1/87) * (x_1 + x_2 + ... + x_87). The centroid calculated in this way was used to determine the skin type of the test sample.

4) Skin type determination based on centroid The expression amount data (Log2RPM value) (test data) of the gene to be analyzed in the SSL-derived RNA of the test sample was used as a variable, and the Euclidean distance between Centroid 1 and Centroid 2 in the test sample was calculated. The distance between Centroid 1 and Centroid 2 of each test sample were compared, and the cluster that was the basis of the centroid with the closer distance was determined to be the skin type (Cluster 1 and Cluster 2) of each test sample.

5) Verification of Judgment Accuracy When the skin types of the test samples recognized in Example 1 were compared with the skin types judged based on the centroids of the training data, the accuracy rate (match rate) was 95% or higher (Table 3).

Example 4 Selection of Marker Genes for Determining Skin Type 1) Data Used and Sample Division Data on the expression level of SSL-derived RNA (read count value) obtained in Example 1 was used as data. As in Example 3, samples were divided into training samples (227 samples) and test samples (54 samples) (Table 2). The skin type of each sample was determined to be Cluster 1 or Cluster 2 as determined by the hierarchical clustering analysis in Example 1.

2) Analysis of training data The data on the expression levels (read count values) of the SSL-derived RNA of the training samples were corrected using DESeq2 to prepare training data. Only 2563 types of genes for which expression level data that was not a missing value was obtained in 90% or more of the sample subjects among the training data were used in the following analysis. For the analysis, count values corrected using a method called DESeq2 (normalized count values) were used.
By extracting RNA in which the SSL-derived RNA expression ratio of the training data was 1.2 times or more and the corrected p-value (FDR) by the likelihood ratio test was less than 0.05 between the two skin types of the training sample, 591 genes with relatively high expression in Cluster 1 and 823 genes with relatively high expression in Cluster 2 were identified.

3) Selection of differentially expressed keratinization/immune response-related genes Among the 591 genes that showed relatively high expression in Cluster 1 in the analysis of the training data in 2), 54 genes annotated with either the GO term "skin development (GO: 0043588)" or "epidermis development (GO: 0008544)" were extracted as genes related to keratinization. In addition, among the 823 genes that showed relatively high expression in Cluster 2, 313 genes annotated with either the GO term "immune system process (GO: 0002376)" or "response to cytokine (GO: 0034097)" were extracted as genes related to immune response.

4) Evaluation of the discrimination accuracy of individual genes in training samples For each of the 367 genes extracted in 3), an ROC curve was drawn using the expression data (Log2RPM value) as a variable, and genes more suitable for discriminating Cluster 1 or Cluster 2 of the training samples were extracted, and the discrimination accuracy was evaluated using the genes. When a sample is discriminated as being negative or positive using the magnitude of a certain continuous value as an index, the ROC curve plots the probability of a positive result in the positive group (sensitivity or true positive rate) on the vertical axis and the value obtained by subtracting the probability of a negative result in the negative group (specificity) from 1 (false positive rate) on the horizontal axis in the negative and positive groups. The larger the area under the curve (AUC) of the ROC curve, the higher the discrimination accuracy of the two groups is judged to be. In this case, the AUC of 367 genes was calculated and summary statistics were obtained (Table 4). Of the 367 genes, 93 genes with AUCs above the third quartile (0.808) were extracted.
Next, the gene expression level in the Youden index of the ROC curve for each of the 93 genes was set as a threshold value, and the skin type of the training sample was discriminated. The accuracy rate was calculated as an index of discrimination accuracy, and summary statistics were obtained (Table 5).

5) Evaluation of the discrimination accuracy of individual genes in test samples Among the 93 genes evaluated for discrimination accuracy in 4), 46 genes with better discrimination accuracy and a correct answer rate equal to or higher than the median value (0.795) in Table 5 were extracted. For the 46 genes, the threshold value set in the skin type discrimination of the training sample in 4) was used, and the threshold value was compared with the expression amount data (Log2RPM value) of the test sample to discriminate the skin type of the test sample. The correct answer rate was calculated as an index of discrimination accuracy, and summary statistics were obtained (Table 6). Among the 46 genes used for discrimination, 25 genes with better discrimination accuracy in the test sample and a correct answer rate equal to or higher than the median value (0.796) in Table 6 are optimal skin type determination marker genes that can discriminate Cluster 1 or Cluster 2 with higher accuracy.
Table 7 shows the gene names of the 25 genes, the AUC of the ROC curve, the threshold used to discriminate the skin type, the discrimination method, the accuracy rate in the training sample (training accuracy rate), and the accuracy rate in the test sample (test accuracy rate).

Example 5: Construction of a skin type prediction model using skin type determination marker genes 1) Data used and division of samples The corrected values (Log2RPM values) of the data of the expression amount of SSL-derived RNA used in Example 3 were used. The samples were divided into training samples (227 samples) and test samples (54 samples) in the same manner as in Example 3. The skin type of each sample was determined to be the skin type (Cluster 1 or Cluster 2) recognized by the hierarchical clustering analysis in Example 1.

2) Selection of Feature A few genes were randomly extracted from the 25 genes used as skin type determination markers in 5) of Example 4, and the extracted gene set was used as a feature.
Specifically, a feature amount obtained by extracting 2 genes from 25 genes was designated Feature 1, a feature amount obtained by extracting 3 genes from 25 genes was designated Feature 2, a feature amount obtained by extracting 4 genes from 25 genes was designated Feature 3, a feature amount obtained by extracting 5 genes from 25 genes was designated Feature 4, a feature amount obtained by extracting 10 genes from 25 genes was designated Feature 5, a feature amount obtained by extracting 15 genes from 25 genes was designated Feature 6, a feature amount obtained by extracting 20 genes from 25 genes was designated Feature 7, and a case in which all 25 genes were used as features was designated Feature 8. Extraction of features related to Features 1 to 8 was attempted 10 times for each Feature.
The genes extracted as feature quantities for each Feature are shown in Tables 8-1 to 8-7.

3) Model Construction A binary classification model was constructed using the expression amount data (Log2RPM value) of the feature (Feature) selected from the SSL-derived RNA of the training sample as the explanatory variable and the skin type (Cluster 1 or Cluster 2) as the objective variable. Logistic regression was used as the algorithm. Eight models were constructed using Features 1 to 8 as features. As mentioned above, Features 1 to 7 exist for the number of random extraction trials (10 times), so a total of 71 models were constructed.

4) Verification of the discrimination model The expression amount data (Log2RPM value) of the feature of the training sample was input to each of the 10 models (Models 1 to 10) constructed using Features 1 to 7 and the model constructed using Feature 8, and the skin type of the training sample was discriminated. The accuracy rate (match rate) of the skin type of the training sample discriminated from the model with respect to the skin type of the training sample recognized in Example 1 is shown as the training accuracy rate in Tables 8-1 to 8-7. In addition, the training accuracy rate in the case of the model constructed using Feature 8 (25 genes) was 1.00.
A high accuracy rate was achieved using either model, and it was confirmed that the discrimination model constructed in 3) can function as a discrimination model for skin types.

5) Identification of skin type of test sample The expression amount data (Log2RPM value) of the feature of the test sample was input to each of 10 models (Models 1 to 10) constructed using Features 1 to 7 and a model constructed using Feature 8, and the skin type of the test sample was determined. The accuracy rate (match rate) of the skin type of the test sample determined by the model with respect to the skin type of the test sample recognized in Example 1 is shown as the test accuracy rate in Tables 8-1 to 8-7. In addition, the test accuracy rate in the case of the model constructed using Feature 8 (25 genes) was 0.981.
As with the training samples, the accuracy rate was high for both models when the test samples were used, demonstrating that the 25 genes shown in Table 7 are useful as markers for discriminating skin types.

Claims (7)

A method for classifying the skin of a subject, the classification being into skin types generated from two clusters obtained by performing cluster analysis based on expression states of genes contained in lipids on the skin surface of a plurality of healthy subjects;
A method comprising: acquiring biometric information from the skin of a subject; and determining from the biometric information whether the subject is classified into one of the skin types. The method according to claim 1, which includes a preparatory step of dividing the skin into two clusters by cluster analysis based on the expression state of genes contained in lipids on the skin surface of a plurality of healthy subjects, and generating skin types from each cluster. The method according to claim 1 or 2, wherein the genes include genes annotated with at least one GO term selected from skin development (GO:0043588), epidermis development (GO:0008544), immune system process (GO:0002376, and response to cytokine (GO:0034097). The method according to any one of claims 1 to 3, wherein the biological information from the skin is the expression state of genes contained in lipids on the surface of the subject's skin. The method according to claim 4, wherein the expression state of the gene is an expression state of a gene annotated with at least one GO term selected from skin development (GO:0043588), epidermis development (GO:0008544), immune system process (GO:0002376, and response to cytokine (GO:0034097). The method according to claim 4, wherein the expression state of the gene is the expression state of at least one gene selected from the following 25 genes: ASPRV1, KRT17, KRT80, KRT79, DNASE1L2, SPRR1A, DSP, CDSN, CST6, LCP1, KRT6B, LCE1C, CALML5, CD83, JUP, TYROBP, SPINT1, CNFN, TMSB4X, NFKB1, B2M, MSN, CXCL16, CD58, and KRT72. A computing device for executing the method according to any one of claims 1 to 6, the device having a means for inputting biometric information obtained from the skin of a subject, and executing a step of determining which of the skin types the subject is classified into from the biometric information according to a program for executing a method for classifying the subject's skin.