Learning model, prediction method, and information processing device

Abstract

To provide an improved technique for predicting preservative properties of pharmaceuticals.SOLUTION: A learning model is provided, for causing a computer to function to output objective variables related to physical properties of formulations based on explanatory variables regarding the formulations, where the explanatory variables include feature quantities related to preservatives, preservative aids, and preservability reducing agents blended with the formulations, and the objective variables include preservability of the formulations.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a learning model, a prediction method, and an information processing device. [Background technology]

[0002] Manufacturers and distributors of cosmetics, quasi-drugs and pharmaceuticals (hereinafter also referred to as drugs) are required to prevent microbial contamination throughout their life cycle, from production and distribution, and from opening to final use. Since many drugs are produced in non-sterile environments and are often not packaged for single use, it is essential to prevent the proliferation of contaminating microorganisms during the product production stage and the use stage after opening. In order to control contaminating microorganisms, it is necessary to impart appropriate preservative properties to the formulation according to the product characteristics. The preservative properties of the formulation are mainly expressed by preservatives, but since preservatives also affect human skin, etc., the types and concentrations of preservatives that can be added to products are regulated in each country to avoid safety issues (e.g., Non-Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Non-Patent Document 1] H Araki, S Hamada, Y Toshima, Y Kawazoe, M Otani, R Suzuki, J Usukura, M Ookawa. Cosmetics&Toiletries, “An Offensive Defense: Minimizing Preservatives, Maximizing Protection and Resources” Allured Business Media, 2018 Summary of the Invention [Problem to be solved by the invention]

[0004] Formula developers and antiseptic and fungicide specialists must use a limited number of preservatives at the lowest possible concentration to design a formula that is both antiseptic and safe. To design an appropriate preservative formula, research is being conducted on the mechanism of action of preservatives. In product formulas, not only the action of the preservative on microorganisms, but also various other compounded ingredients affect the preservative and microorganisms. This makes the preservative design of formulas and prediction of preservative performance technically difficult. The preservative performance of formulas is generally evaluated by preservative effectiveness tests at the product development stage, but since the tests usually take a long time, about one month, failure of the tests may lead to delays in product development. Therefore, pharmaceutical manufacturers are developing rapid and alternative methods for preservative effectiveness tests, as well as methods for predicting preservative performance by calculation.

[0005] In view of the above circumstances, an object of the present disclosure is to improve the technology for predicting the preservative properties of drugs. [Means for solving the problem]

[0006] (1) A learning model according to an embodiment of the present disclosure is a learning model for causing a computer to function so as to output a response variable related to a physical property of a formulation based on an explanatory variable related to the formulation, The explanatory variables include feature quantities related to preservatives, preservative assistants, and preservative reducing agents incorporated in the formulation, The response variable includes the preservative properties of the formulation.

[0007] (2) A learning model according to an embodiment of the present disclosure is a learning model according to (1), The characteristic amount includes aqueous phase concentrations of a specific preservative and a specific preservative aid among the preservatives, preservative aids, and preservative-reducing agents blended in the formulation.

[0008] (3) A learning model according to an embodiment of the present disclosure is a learning model according to (1) or (2), The preservative includes at least one of methyl parahydroxybenzoate, propyl parahydroxybenzoate, butyl parahydroxybenzoate, 2-phenoxyethanol, 2-isopropyl-5-methylphenol, benzoic acid, salicylic acid, and sorbic acid.

[0009] (4) A learning model according to an embodiment of the present disclosure is a learning model according to any one of (1) to (3), The predetermined preservative includes at least one of methyl parahydroxybenzoate, propyl parahydroxybenzoate, butyl parahydroxybenzoate, 2-phenoxyethanol, and 2-isopropyl-5-methylphenol.

[0010] (5) A learning model according to an embodiment of the present disclosure is a learning model according to any one of (1) to (4), The preservative aid includes at least one of propanediol, 1,3-butylene glycol, dipropylene glycol, 1,2-pentanediol, ethanol, zinc oxide, a fragrance, and a chelating agent.

[0011] (6) A learning model according to an embodiment of the present disclosure is the learning model described in (2), The predetermined preservative aid includes at least one of propanediol, 1,3-butylene glycol, dipropylene glycol, 1,2-pentanediol, and ethanol.

[0012] (7) A learning model according to an embodiment of the present disclosure is a learning model according to any one of (1) to (6), The preservative-reducing agent includes at least one of a polyhydric alcohol other than a dihydric alcohol, a polysaccharide, and an extract.

[0013] (8) A learning model according to an embodiment of the present disclosure is a learning model according to any one of (1) to (7), The response variable is a binary value of pass or fail.

[0014] (9) A learning model according to an embodiment of the present disclosure is a learning model according to any one of (1) to (8), The response variable is a value determined based on the survival rate of Staphylococcus aureus.

[0015] (10) A prediction method according to an embodiment of the present disclosure is a prediction method executed by an information processing device, which outputs a response variable related to a physical property of a formulation based on explanatory variables related to the formulation, inputting explanatory variables related to the formulation into a learning model; Outputting a response variable related to a physical property of the formulation based on the learning model; Including, The explanatory variables include feature quantities related to preservatives, preservative assistants, and preservative reducing agents incorporated in the formulation, The response variable includes the preservative properties of the formulation.

[0016] (11) An information processing device according to an embodiment of the present disclosure includes a control unit, and is configured to output a response variable related to a physical property of a formulation based on an explanatory variable related to the formulation, The control unit is Inputting explanatory variables related to the formulation into a learning model; Outputting a response variable related to a physical property of the formulation based on the learning model; The explanatory variables include feature quantities related to preservatives, preservative assistants, and preservative reducing agents incorporated in the formulation, The response variable includes the preservative properties of the formulation.

[0017] (12) A learning model according to an embodiment of the present disclosure is a learning model for causing a computer to function so as to output a response variable related to a physical property of a formulation based on an explanatory variable related to the formulation, The explanatory variables include the aqueous phase concentration of a preservative or preservative aid incorporated in the formulation; The response variable includes the preservative properties of the formulation.

[0018] (13) A prediction method according to an embodiment of the present disclosure is a prediction method executed by an information processing device, which outputs a response variable related to a physical property of a formulation based on explanatory variables related to the formulation, inputting explanatory variables related to the formulation into a learning model; Outputting a response variable related to a physical property of the formulation based on the learning model; Including, The explanatory variables include the aqueous phase concentration of a preservative or preservative aid incorporated in the formulation; The response variable includes the preservative properties of the formulation.

[0019] (14) An information processing device according to an embodiment of the present disclosure includes a control unit, and is configured to output a response variable related to a physical property of a formulation based on an explanatory variable related to the formulation, The control unit is Inputting explanatory variables related to the formulation into a learning model; Outputting a response variable related to a physical property of the formulation based on the learning model; The explanatory variables include the aqueous phase concentration of a preservative or preservative aid incorporated in the formulation; The response variable includes the preservative properties of the formulation.

[0020] (15) A learning model according to an embodiment of the present disclosure is a learning model according to any one of (1) to (9), The learning model is trained using training data in which data imbalance is eliminated by adding artificially created failing data and / or passing data. Effect of the Invention

[0021] According to one embodiment of the present disclosure, the prediction technique for the preservative properties of drugs can be improved. [Brief description of the drawings]

[0022] [Figure 1] FIG. 1 is a block diagram showing a schematic configuration of an information processing device according to an embodiment of the present disclosure. [Diagram 2] FIG. 2 is a flow chart outlining a prediction method according to an embodiment of the present disclosure. [Diagram 3] FIG. 3 shows a confusion matrix generated from the prediction results according to the first embodiment. [Figure 4] FIG. 4 shows the accuracy verification result and the ROC curve of the prediction model according to the first embodiment. [Diagram 5] FIG. 5 is a diagram showing a comparison of prediction accuracy. [Figure 6] FIG. 6 is a diagram showing the distribution of pass and fail data. [Figure 7] FIG. 7 shows a confusion matrix generated from the prediction results according to the second embodiment. [Figure 8] FIG. 8 shows the accuracy verification result and the ROC curve of the prediction model according to the second embodiment. [Figure 9] FIG. 9 shows a confusion matrix generated from the prediction result according to the modified example of the second embodiment. [Figure 10] FIG. 10 shows the accuracy verification result and the ROC curve of the prediction model according to the modified example of the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] Hereinafter, a technique for predicting the preservative properties of a drug according to an embodiment of the present disclosure will be described with reference to the drawings. In this embodiment, a case where a drug to be predicted for preservative properties is a skin care product that does not need to be washed off is illustrated as an example, but the prediction target is not limited to this. Drugs to be predicted include any cosmetics, quasi-drugs, medicines, hygiene products, and daily miscellaneous goods.

[0024] In each drawing, the same or corresponding parts are denoted by the same reference numerals. In the description of this embodiment, the description of the same or corresponding parts will be omitted or simplified as appropriate.

[0025] First, an overview of this embodiment will be described. In the preservative prediction technology according to this embodiment, a learning model is used. The learning model causes a computer such as an information processing device to function so as to output a response variable related to the physical properties of a formulation based on explanatory variables related to the formulation. Here, the explanatory variables include feature quantities related to the preservatives, preservative assistants, and preservative reducing agents blended in the formulation. The response variables also include the preservative properties of the formulation.

[0026] In this way, in the drug preservative prediction technique according to the present embodiment, the drug preservative is predicted using a learning model including features related to the preservatives, preservative assistants, and preservative reducing agents blended in the drug formulation as explanatory variables. The drug preservative prediction technique is improved in that the learning model can predict the preservatives taking into account the preservatives, preservative assistants, and preservative reducing agents that have a large effect on the preservatives. Note that the preservatives are components used to suppress the growth of microorganisms and prevent the drug from spoiling, and are, for example, components listed in the list of ingredients restricted for blending (positive list). The preservative assistants are components that are not listed in the list of ingredients restricted for blending (positive list), but have a certain preservative effect and have the effect of reducing the amount of preservatives blended in the drug. The preservative reducing agents are components that may reduce the preservatives of the drug.

[0027] (Preservatives) The preservative includes at least one of methyl paraoxybenzoate, propyl paraoxybenzoate, butyl paraoxybenzoate, 2-phenoxyethanol, 2-isopropyl-5-methylphenol, benzoic acid, salicylic acid, and sorbic acid. Note that the preservative is not limited to these, and any other preservative may be included.

[0028] (Preservative aid) The preservative aid includes at least one of propanediol, 1,3-butylene glycol, dipropylene glycol, 1,2-pentanediol, ethanol, zinc oxide, fragrance, and a chelating agent. Note that the preservative aid is not limited to these, and may include any other preservative aid.

[0029] (Anti-corrosive agent) The above-mentioned preservative lowering agent includes at least one of polyhydric alcohols other than dihydric alcohols, polysaccharides, and extracts. Note that the preservative lowering agent is not limited to these, and may include any other preservative lowering agent.

[0030] (Features) The above characteristic amount includes the type and concentration in water of a specific preservative and a specific preservative aid among the preservatives, preservative aids, and preservative property lowering agents blended in the formulation (hereinafter also referred to as aqueous phase concentration). The aqueous phase concentration may be a theoretically calculated value or may be a value measured by an experiment. The characteristic amount also includes the type and concentration of a preservative and a preservative aid other than the specific preservative and the specific preservative aid. The specific preservative includes at least one of methyl paraoxybenzoate, propyl paraoxybenzoate, butyl paraoxybenzoate, 2-phenoxyethanol, and 2-isopropyl-5-methylphenol. The specific preservative includes at least one of propanediol, 1,3-butylene glycol, dipropylene glycol, 1,2-pentanediol, and ethanol. The characteristic amount also includes the type and concentration of a preservative property lowering agent. For preservatives or preservative aids whose blended concentration in a formulation and aqueous phase concentration are expected to deviate by a predetermined value or more based on the physicochemical properties of the preservatives or preservative aids, such as water solubility, hydrophilicity, hydrophobicity, chemical structure, etc., it is desirable to use the concentration of the preservative or preservative aid in water as a characteristic quantity.

[0031] (Objective variable) The above-mentioned objective variable may be a value indicating antiseptic property determined based on the survival rate of Staphylococcus aureus. The objective variable may be a binary value of pass or fail. Hereinafter, in this embodiment, the objective variable is a value indicating antiseptic property determined based on the survival rate of Staphylococcus aureus, and is a binary value of pass or fail. In other words, the learning model according to this embodiment is described as a binary classification model. Note that the objective variable is not limited to this. For example, the objective variable may be any index indicating two or more stages of antiseptic property. Specifically, for example, the objective variable may be an LRV value. The objective variable may be a value indicating antiseptic property determined based on the survival rate of bacteria other than Staphylococcus aureus.

[0032] (Configuration of information processing device) The information processing device that functions to output a prediction result by a learning model may be any device. For example, a personal computer, a server computer, a general-purpose electronic device, or a dedicated electronic device can be adopted as the information processing device. Each configuration of the information processing device will be described with reference to FIG. 1.

[0033] As shown in FIG. 1, the information processing device 10 includes a control unit 11, a storage unit 12, an input unit 13, and an output unit .

[0034] The control unit 11 includes at least one processor, at least one dedicated circuit, or a combination of these. The processor is a general-purpose processor such as a central processing unit (CPU) or a graphics processing unit (GPU), or a dedicated processor specialized for a specific process. The dedicated circuit is, for example, a field-programmable gate array (FPGA) or an application specific integrated circuit (ASIC). The control unit 11 executes processes related to the operation of the information processing device 10 while controlling each part of the information processing device 10.

[0035] The storage unit 12 includes at least one semiconductor memory, at least one magnetic memory, at least one optical memory, or a combination of at least two of these. The semiconductor memory is, for example, a random access memory (RAM) or a read only memory (ROM). The RAM is, for example, a static random access memory (SRAM) or a dynamic random access memory (DRAM). The ROM is, for example, an electrically erasable programmable read only memory (EEPROM). The storage unit 12 functions as, for example, a main storage device, an auxiliary storage device, or a cache memory. The storage unit 12 stores data used in the operation of the information processing device 10 and data obtained by the operation of the information processing device 10.

[0036] The input unit 13 includes at least one input interface. The input interface is, for example, a physical key, a capacitive key, a pointing device, or a touch screen integrally provided with a display. The input interface may also be, for example, a microphone that accepts voice input, or a camera that accepts gesture input. The input unit 13 accepts an operation to input data used for the operation of the information processing device 10. The input unit 13 may be connected to the information processing device 10 as an external input device instead of being provided in the information processing device 10. As a connection method, any method such as USB (Universal Serial Bus), HDMI (registered trademark) (High-Definition Multimedia Interface), or Bluetooth (registered trademark) can be used.

[0037] The output unit 14 includes at least one output interface. The output interface is, for example, a display that outputs information as a video. The display is, for example, a liquid crystal display (LCD) or an organic EL (electro luminescence) display. The output unit 14 displays and outputs data obtained by the operation of the information processing device 10. The output unit 14 may be connected to the information processing device 10 as an external output device instead of being provided in the information processing device 10. As a connection method, any method such as USB, HDMI (registered trademark), or Bluetooth (registered trademark) can be used.

[0038] The functions of the information processing device 10 are realized by executing a program according to this embodiment on a processor equivalent to the information processing device 10. That is, the functions of the information processing device 10 are realized by software. The program causes a computer to execute the operations of the information processing device 10, thereby causing the computer to function as the information processing device 10. That is, the computer functions as the information processing device 10 by executing the operations of the information processing device 10 in accordance with the program.

[0039] In this embodiment, the program can be recorded on a computer-readable recording medium. The computer-readable recording medium includes a non-transitory computer-readable medium, such as a magnetic recording device, an optical disc, a magneto-optical recording medium, or a semiconductor memory. The program is distributed, for example, by selling, transferring, or lending a portable recording medium such as a DVD (digital versatile disc) or a CD-ROM (compact disc read only memory) on which the program is recorded. The program may also be distributed by storing the program in the storage of an external server and transmitting the program from the external server to another computer. The program may also be provided as a program product.

[0040] A part or all of the functions of the information processing device 10 may be realized by a dedicated circuit corresponding to the control unit 11. In other words, a part or all of the functions of the information processing device 10 may be realized by hardware.

[0041] (Operation of information processing device) The operation of the information processing device 10 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a flowchart showing an example of a prediction method executed by the information processing device 10 according to this embodiment.

[0042] Step S100: The control unit 11 of the information processing device 10 inputs explanatory variables related to the formulation into the learning model. As described above, the explanatory variables include features related to the preservatives, preservative assistants, and preservative reducing agents blended in the formulation. The learning model may be a neural network model generated based on a multilayer perceptron consisting of an input layer, a hidden layer, and an output layer. The learning model may also be a linear model, a nonlinear model, a support vector machine, or the like. Alternatively, the learning model may be a model generated based on a machine learning algorithm such as a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), or other deep learning.

[0043] Step S200: The control unit 11 outputs the objective variables related to the physical properties of the formulation based on the learning model. As described above, the objective variables include the preservative properties of the formulation. Any method can be adopted as the mode of outputting the objective variables. For example, the control unit 11 may display and output the objective variables related to the physical properties of the formulation by the output unit 14.

[0044] In this way, in the drug preservative prediction technique according to the present embodiment, the drug preservative is predicted using a learning model including, as explanatory variables, feature quantities related to the preservatives, preservative aids, and preservative-reducing agents blended in the drug formulation. This learning model improves the drug preservative prediction technique in that it can predict preservatives while taking into account the preservatives, preservative aids, and preservative-reducing agents that have a large effect on preservatives.

[0045] Example 1 Hereinafter, an example of predicting the preservative properties of a drug using the prediction technique according to the above embodiment will be described. In this example, 39 types of leave-on skin care products with 249 formulations were used as training data and test data for training the learning model. The product categories (product classifications), formulations, formulation forms, and data points of the training data and test data used are shown in Tables 1 and 2, respectively.

[0046] [Table 1]

[0047] [Table 2]

[0048] In Tables 1 and 2, the formulations are either soluble, oil in water emulsion (O / W), water in oil emulsion (W / O), or water in oil in water emulsion (W / O / W). Of the 249 formulations, 237 formulations shown in Table 1 were used as training data. On the other hand, of the 249 formulations, 12 formulations shown in Table 2 were used as test data.

[0049] (Explanatory variables in the examples) Of the 217 ingredients in the formula, 53 ingredients that are likely to affect preservative properties were used as explanatory variables. Table 3 shows a list of the types of ingredients used as explanatory variables (preservatives, preservative assistants, preservative reducing agents, and purified water).

[0050] [Table 3]

[0051] In this embodiment, the blended components shown in Table 3 and their blended concentrations or aqueous phase concentrations were used as explanatory variables. Specifically, for blended components not marked with * in Table 3, blended concentrations were used as explanatory variables. On the other hand, for blended components marked with * in Table 3, aqueous phase concentrations were used as explanatory variables. In other words, blended components marked with * correspond to "prescribed preservatives" or "prescribed preservative aids" in this disclosure. Specifically, for methyl paraoxybenzoate, propyl paraoxybenzoate, butyl paraoxybenzoate (straight-chain parabens with 1, 3, and 4 carbon atoms), 2-phenoxyethanol, 2-isopropyl-5-methylphenol (IPMP), and preservative aids such as alkanediols and ethanol, aqueous phase concentrations (concentrations in water) rather than blended concentrations were used as explanatory variables.

[0052] The aqueous phase concentration of the preservative was theoretically calculated using the following procedure. Table 4 shows the distribution ratio of methylparaben between representative oils and water, that is, the distribution coefficient K of methylparaben, which is calculated by dividing the concentration of methylparaben in the oil component (oil phase) by the concentration of methylparaben in the water (aqueous phase). m Shows.

[0053] [Table 4]

[0054] The partition coefficient K of methylparaben mentioned above m Using this, the distribution coefficient K of preservative i for each oil phase component was calculated according to the following formula (1).

[0055]

number

[0056] Here, WS is the maximum solubility of methylparaben in water (25°C). WS i is the maximum solubility of preservative i in water (25°C). Table 5 shows the maximum solubility of each preservative.

[0057] [Table 5]

[0058] Since the distribution of the preservative into the aqueous phase is considered to depend on the solubility of the preservative in water, the distribution coefficients of various oils with respect to various preservatives were calculated based on the maximum solubility in water of methylparaben and various preservatives, and the proportional equation of the distribution coefficients of various oils with respect to methylparaben.

[0059] Next, the distribution ratio of the preservative i between the mixed oil of the formulation consisting of the oil components 1 to n and water (the distribution coefficient of the mixed oil) was calculated according to the following formula (2).

[0060]

number

[0061]

number

[0062] The aqueous phase concentration C of alkanediols in a formulation consisting of alkanediols 1 to n having straight-chain carbon numbers of 3 or 4 wk was calculated using the following formula (4).

[0063]

number

[0064] Concentration of 1,2-pentanediol in the aqueous phase C wlwas calculated using the following formula (5).

[0065]

number

[0066] Ethanol aqueous phase concentration C wm was calculated using the following formula (5).

[0067]

number

[0068] From the aqueous phase concentrations of alkanediols and ethanol, a value C is calculated that defines the aqueous phase concentrations of alkanediols and ethanol in the formulation according to the following formula (7): wn was calculated.

[0069]

number

[0070] For the other three preservatives, five preservative assistants, 34 preservative reducing agents, and purified water, the concentrations of the ingredients were used as explanatory variables.

[0071] As an example, Table 6 below shows the concentrations of preservative components in a certain O / W emulsion (oil phase: hydrocarbon 5%, ester oil 8%, water phase: purified water and others 85%).

[0072] [Table 6]

[0073] Since this formulation contains a high concentration of highly polar ester oil, it was presumed that the aqueous phase concentration of parabens, which are highly soluble in oil, is lower than the blended concentration, and conversely, 1,3-butylene glycol is higher than the blended concentration.

[0074] (Objective variables in the examples) In this example, the objective variable was the result of a preservative effectiveness test using S. aureus (pass or fail). The preservative effectiveness test was performed under test conditions conforming to ISO 11930. S. aureus NBRC13276 was used as the test bacteria, and the test bacteria solution was inoculated and mixed into the formulation. After storing for a certain period of time, the number of live bacteria in the formulation was measured, and the logarithmic reduction value (LRV) of the number of test bacteria was calculated. Preservatives that met the LRV of Criteria B of ISO 11930, i.e., an LRV of 3 or more on the 14th day of storage, were considered to be pass, and those that did not meet the LRV were considered to be fail. A list of objective variables related to the training data is shown in Table 7.

[0075] [Table 7]

[0076] (Predictive analysis and performance verification methods) Training data (Table 1) was provided to machine learning software Prediction One (registered trademark) (manufactured by SONY Network Communications), and a predictive analysis of the results of the preservative efficacy test using S. aureus, which is the objective variable, was performed from the explanatory variables, which were the types of preservatives, preservative auxiliary agents, and preservative reducing agents, and their aqueous phase concentrations or blended concentrations. As a learning model, a binary classification model was created that classifies the objective variable into one of two classes, pass or fail. In the case of a missing value of the explanatory variable, i.e., when the ingredient was not blended in the prescription, the aqueous phase concentration or blended concentration was set to 0. In addition, a 5-fold cross-validation was performed in which the data was divided into 5 parts, one of which was used as test data, and the remaining 4 were used as learning data to evaluate the model. The 5-fold cross-validation was used to verify the prediction accuracy of the model and to verify the ability to predict unknown data not given at the time of learning in addition to the data given at the time of learning, i.e., to verify the generalization performance. A confusion matrix was generated from the prediction results, and the true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN) were counted to calculate the accuracy rate (Accuracy, the proportion of correct predictions among all predictions), precision (the proportion of predicted failures that were actually failures), recall (the proportion of actual failures that were predicted to be failures), and F-measure (the harmonic mean of precision and recall) according to the following formula.

[0077]

number

[0078]

number

[0079]

number

[0080]

number

[0081] (Example Results) The confusion matrix generated from the prediction results according to the embodiment is shown in Fig. 3. The accuracy rate, precision rate, recall rate, and F-measure calculated from the TP, TN, FP, and FN shown in the confusion matrix in Fig. 3 and the above-mentioned formulas (8) to (11) are as follows: Accuracy rate: 0.962 ·Precision rate: 0.571 ·Recall rate: 0.400 F-value: 0.471 As shown in the above results, it can be seen that this embodiment has a sufficient accuracy rate, precision rate, recall rate, and F-measure. In particular, it can be seen that this embodiment has a learning model with a high accuracy rate.

[0082] Furthermore, Figure 4 shows the accuracy verification results of the prediction model and the ROC curve (Receiver Operating Characteristic curve). The ROC curve is a graph in which the true positive rate is plotted on the vertical axis and the false positive rate is plotted on the horizontal axis when the threshold is changed in the binary classification of failure (positive) and pass (negative). The AUC (Area Under the Curve) determined by the ROC curve was 0.696. Here, AUC is the area under the ROC curve created. The AUC value was also as high as that of the standard learning model.

[0083] Test data was input into the obtained learning model (binary classification model in this example), and the accuracy of the model was evaluated. Four prescriptions (#14, #27, #30, #38) with prescription frameworks similar to the prescriptions in the training data in terms of base type and concentration, and eight prescriptions (#7, #35, #36) with prescription frameworks completely different from the prescriptions in the training data in terms of base type and concentration (Table 2) were used as test data for evaluation. The results are shown in Table 8.

[0084] [Table 8]

[0085] As shown in Table 8, for all test data, the predicted values ​​of antiseptic activity against S. aureus matched the actual measured values, demonstrating high accuracy.

[0086] In this way, in the drug preservative prediction technique according to the present embodiment, the drug preservative is predicted using a learning model including, as explanatory variables, feature quantities related to the preservatives, preservative assistants, and preservative-reducing agents blended in the drug formulation. Therefore, the drug preservative prediction technique is improved in that the drug preservative can be predicted taking into account the preservatives, preservative assistants, and preservative-reducing agents that have a large effect on the drug preservative.

[0087] In particular, in machine learning, it is generally said that overlearning is likely to occur if the number of features is large relative to the number of data. If all of the formulation ingredients, which are prescription information, are used as features, the number will be 217, which is considered redundant relative to the number of data. In addition, ingredients with low formulation concentrations, except for preservatives and certain plant extracts, have little effect on preservative properties and are likely not suitable as features. Therefore, in this embodiment, the formulation ingredients were classified and selected according to their characteristics and used for learning. A comparison of prediction accuracy is shown with reference to FIG. 5. FIG. 5 shows the results of the incorrect answer rate when preservatives, preservative assistants, preservative-reducing components, and other components are used as features. The incorrect answer rate indicates the proportion of actual measurements that pass are predicted as fail, or actual measurements that fail are predicted as pass. As shown in FIG. 5, when preservatives, preservative assistants, and preservative-reducing components are trained as features, the incorrect answer rate of the model is lower than the others, so it can be said that using these as features improves prediction accuracy.

[0088] Although the present disclosure has been described based on the drawings and embodiments, it should be noted that those skilled in the art can easily make various modifications and corrections based on the present disclosure. Therefore, it should be noted that these modifications and corrections are included in the scope of the present disclosure. For example, the functions included in each means or step can be rearranged so as not to be logically inconsistent, and multiple means or steps can be combined or divided into one.

[0089] For example, in the present embodiment, the explanatory variables include feature quantities related to the preservative, preservative auxiliary, and preservative reducing agent blended in the formulation, but are not limited thereto. For example, the explanatory variables may include the aqueous phase concentration of the preservative component blended in the formulation. In other words, the learning model may be a learning model that includes the aqueous phase concentration of the preservative component blended in the formulation as an explanatory variable and includes the preservative property of the formulation as a target variable. Such preservative components may include at least one of the preservative and the preservative auxiliary.

[0090] Example 2 Hereinafter, Example 2 will be described in which the learning model is reconstructed after adding the rejected data to the training data and eliminating the imbalance of the data. That is, the learning model may be trained using training data in which the imbalance of the data is eliminated by adding at least one of the rejected data and the passed data that are artificially created. As described above, the accuracy rate, which is one of the prediction accuracy indicators of the learning model, showed a high value of 0.962, but the accuracy of other indicators is improved, with the precision rate of 0.571, the recall rate of 0.400, the F-score of 0.471, and the AUC of 0.696. This is thought to be due to the fact that the training data used imbalanced data, in which 227 prescriptions are passed data and 10 prescriptions are failed data, was used for training. That is, the learning model may be reconstructed after adjusting the ratio of passed data to failed data included in the training data (hereinafter also referred to as the data ratio).

[0091] The data ratio may be set appropriately. As shown in Table 1, there are 39 types of target prescriptions (hereinafter also referred to as frameworks) in the present disclosure. When there are multiple prescription frameworks, it is preferable to adjust the balance between pass data and fail data for each prescription framework. Therefore, in Example 2, the ratio of pass data and fail data included in the training data for each prescription framework was adjusted. Specifically, as shown in Table 9, 44 prescriptions in which the blending concentration or aqueous phase concentration of the preservatives and preservative aids in the data used for training was reduced to a certain level, and 2 prescriptions in which the concentration was increased to a certain level or higher were artificially created.

[0092] [Table 9] TIFF2025013261000022.tif32150

[0093] The recipe was created using the following procedure. It is empirically clear that the effectiveness of preservatives against S. aureus varies depending on the type. Therefore, when the effect of methyl parahydroxybenzoate was set to 1, the effect of propyl parahydroxybenzoate at the same concentration was set to 2.4, 2-phenoxyethanol to 0.24, and 2-isopropyl-5-methylphenol to 6, and the total concentration of the preservatives in the recipe was converted to the concentration of methyl parahydroxybenzoate according to the following formula (12).

[0094]

number

[0095] FIG. 6 shows the distribution of pass data and fail data, with the calculated concentration (converted value) of methyl paraoxybenzoate on the horizontal axis and the concentration of diol, an auxiliary preservative, on the vertical axis. With reference to this plot, the concentration of the preservative and / or auxiliary preservative in the pass or fail data was reduced or increased to artificially create fail data or pass data. It was confirmed that the artificially created fail or pass data was present near the area where the fail or pass data was plotted, respectively. Based on this confirmation, it was determined whether the artificially created data was fail data or pass data.

[0096] (Results of Example 2) Figure 7 shows the confusion matrix of the prediction results of the learning model reconstructed using a total of 283 prescriptions, including the 44 failed prescriptions and the 2 passed prescriptions. Note that the actual measured values ​​in Figure 7 include the results of the preservative efficacy test for the training data shown in Table 1 and the failed or passed data created artificially.

[0097] The accuracy rate, precision rate, recall rate, and F-measure calculated from the TP, TN, FP, and FN shown in the confusion matrix of FIG. 7 and the formulas (8) to (11) are as follows: Accuracy rate: 0.908 ·Precision rate: 0.712 ·Recall rate: 0.870 F-value: 0.783 As shown in the above results, it is understood that sufficient accuracy rate, precision rate, recall rate, and F-measure are obtained in Example 2. In particular, it is understood that in Example 2, the accuracy rate is equivalent to that of Example 1, and a learning model in which the precision rate, recall rate, and F-measure are all improved compared to Example 1 is obtained.

[0098] 8 shows the ROC curve of the prediction model according to Example 2. The AUC (Area Under the Curve) determined by the ROC curve in FIG.

[0099] Furthermore, Fig. 9 shows a confusion matrix of the prediction results of the learning model reconstructed using a total of 281 prescriptions, including only 44 rejected data, as a modified example of Example 2. Note that the actual measured values ​​in Fig. 9 include the results of the preservative efficacy test for the training data shown in Table 1 and the rejected data created artificially.

[0100] The accuracy rate, precision rate, recall rate, and F-measure calculated from TP, TN, FP, and FN shown in the confusion matrix of FIG. 9 and from equations (8) to (11) are as follows: Accuracy rate: 0.914 ·Precision rate: 0.773 ·Recall rate: 0.773 F-value: 0.773 As shown in the above results, even when reconstructing by adding only the failed data, sufficient accuracy rate, precision rate, recall rate, and F value are obtained. In particular, in this case, the accuracy rate is equivalent to that of Example 1, and it is found that a learning model with improved precision rate, recall rate, and F value compared to Example 1 is obtained.

[0101] Figure 10 shows the ROC curve of the learning model reconstructed using a total of 281 prescriptions, including the addition of only the 44 failed prescriptions. The AUC (Area Under the Curve) determined by the ROC curve in Figure 10 was 0.932, a significant improvement over Example 1. [Explanation of symbols]

[0102] 10. Information processing device 11 Control section 12 Storage section 13 Input section 14 Output section

Claims

1. A learning model for causing a computer to function so as to output a response variable related to a physical property of a formulation based on an explanatory variable related to the formulation, The explanatory variables include feature quantities related to preservatives, preservative assistants, and preservative reducing agents incorporated in the formulation, A learning model, wherein the objective variable includes the preservative properties of the formulation.

2. The learning model according to claim 1 , wherein the feature amount includes an aqueous phase concentration of a predetermined preservative and a predetermined preservative aid among preservatives, preservative aids, and preservative reducing agents contained in the formulation.

3. The learning model of claim 1 , wherein the preservative includes at least one of methyl paraoxybenzoate, propyl paraoxybenzoate, butyl paraoxybenzoate, 2-phenoxyethanol, 2-isopropyl-5-methylphenol, benzoic acid, salicylic acid, and sorbic acid.

4. The learning model according to claim 2 , wherein the predetermined preservative includes at least one of methyl paraoxybenzoate, propyl paraoxybenzoate, butyl paraoxybenzoate, 2-phenoxyethanol, and 2-isopropyl-5-methylphenol.

5. The learning model of claim 1 , wherein the preservative aid includes at least one of propanediol, 1,3-butylene glycol, dipropylene glycol, 1,2-pentanediol, ethanol, zinc oxide, a fragrance, and a chelating agent.

6. The learning model of claim 2 , wherein the predetermined preservative aid includes at least one of propanediol, 1,3-butylene glycol, dipropylene glycol, 1,2-pentanediol, and ethanol.

7. The learning model according to claim 1 , wherein the preservative-reducing agent includes at least one of a polyhydric alcohol other than a dihydric alcohol, a polysaccharide, and an extract.

8. The learning model according to claim 1 , wherein the objective variable is a binary value of pass or fail.

9. The learning model according to claim 1 , wherein the objective variable is a value determined based on a survival rate of Staphylococcus aureus.

10. A prediction method for outputting a response variable related to a physical property of a formulation based on explanatory variables related to the formulation, the method being executed by an information processing device, inputting explanatory variables related to the formulation into a learning model; Outputting a response variable related to a physical property of the formulation based on the learning model; Including, The explanatory variables include feature quantities related to preservatives, preservative assistants, and preservative reducing agents incorporated in the formulation, A prediction method, wherein the dependent variable includes the preservative property of the formulation.

11. An information processing device comprising a control unit, the information processing device outputting a response variable related to a physical property of a formulation based on an explanatory variable related to the formulation, The control unit is Inputting explanatory variables related to the formulation into a learning model; Outputting a response variable related to a physical property of the formulation based on the learning model; The explanatory variables include feature quantities related to preservatives, preservative assistants, and preservative reducing agents incorporated in the formulation, The information processing device, wherein the objective variable includes the preservative property of the formulation.

12. The learning model of claim 1 , wherein the learning model is trained using training data in which data imbalance is eliminated by adding at least one of artificially created failed data and passed data.

Images