A method for evaluating the eye irritation of cosmetic raw materials
By using a machine learning fusion model combined with an integrated learning algorithm and a similarity algorithm, the in vitro substitution test results and in vivo test results of cosmetic raw materials were fusion-evaluated, and the problem of low accuracy of prediction evaluation of eye irritation for cosmetic raw materials was solved, achieving a 100% prediction accuracy.
Patent Information
- Application Number
- CN202411872718.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-18
- Publication Date
- 2025-04-08
AI Technical Summary
The prior art cannot provide a method for high accuracy in the evaluation of eye irritation prediction of cosmetic raw materials.
A machine learning fusion model based on a combination of integrated learning algorithms and similarity algorithms is used to input the data set of in vitro substitution test results and in vivo test results of the obtained cosmetic raw materials into the fusion model, and verify and evaluate the predictive evaluation effect of the fusion model, so as to clarify whether cosmetic raw materials cause irritation to the eyes.
The high accuracy of cosmetic raw materials for eye irritation prediction evaluation is achieved, and the accuracy of irritation prediction evaluation of the subject is 100%, solving the problem of low evaluation accuracy in the prior art.
Smart Images
Figure CN119782948A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of cosmetic safety evaluation, and in particular to a method for evaluating the eye irritation of cosmetic raw materials. Background Art
[0002] During normal use, cosmetics may come into contact with the eyes, especially eye cosmetics such as eye shadow, mascara, eye cream, etc., as well as facial cosmetics that may splash into the eyes. If cosmetics are irritating to the eyes, they may cause discomfort symptoms such as eye pain, redness, swelling, and tearing, and may even damage the eyes in severe cases. The most important way to evaluate the irritation of cosmetics to the user's eyes is to study the irritation of cosmetic raw materials to the eyes. However, there is no in vitro alternative test scheme in the prior art that can accurately evaluate the irritation of cosmetic raw materials to the eyes. Summary of the invention
[0003] The purpose of the present application is to provide a method for evaluating the eye irritation of cosmetic raw materials, which solves the technical problem that the prior art cannot provide a method with a high accuracy rate for predicting and evaluating the eye irritation of cosmetic raw materials.
[0004] The technical solution of this application:
[0005] The present application provides a method for evaluating the eye irritation of cosmetic raw materials, which is executed by a processor and includes:
[0006] Obtain data on in vitro substitution test results and in vivo test results of target cosmetic raw materials to form a data set;
[0007] Execute the input fusion model on the dataset;
[0008] Determine whether there is the same result data based on the data set. If so, perform evaluation result output. If not, perform ensemble learning and similarity judgment.
[0009] Output the evaluation results.
[0010] In some embodiments, the in vitro alternative test items of the data set consist of six test results: fluorescein leakage test, neutral red uptake test, red blood cell hemolysis test, chicken embryo chorioallantoic membrane test, chicken embryo chorioallantoic membrane trypan blue staining test, and in vitro rabbit eye test, and the data set includes at least 186 test data of 84 cosmetic raw materials.
[0011] In some embodiments, the determining whether there is the same result data based on the data set, if yes, outputting the evaluation result, if no, performing ensemble learning and similarity determination, includes:
[0012] Determine whether the data set has the same result data and output the evaluation result;
[0013] Determine that there is no identical result data in the data set and perform ensemble learning;
[0014] After ensemble learning, determine whether evaluation can be performed. If not, output the result that cannot be evaluated. If evaluation is possible, perform similarity calculation.
[0015] Calculate the most similar data results in the data set, perform judgment on the results, and output the evaluation results when the similarity is greater than 50%. When the similarity is less than or equal to 50%, output the result that cannot be evaluated.
[0016] In some embodiments, the integrated learning includes at least a voting classifier and a machine learning classification algorithm, and the machine learning classification algorithm includes at least a support vector machine algorithm, a logistic regression algorithm, a decision tree algorithm, a random forest algorithm and a K nearest neighbor algorithm.
[0017] In some embodiments, the similarity calculated by the similarity is expressed as:
[0018] s / (s+d)×100%
[0019] The number of identical test results is denoted as s, and the number of different test results is denoted as d.
[0020] The technical solution of the present application has at least the following advantages and beneficial effects: the present application is based on a machine learning fusion model that combines an integrated learning algorithm and a similarity algorithm. The acquired data set is input into the fusion model, and the prediction and evaluation effect of the fusion model is verified and evaluated, thereby clarifying whether the cosmetic raw materials cause irritation to the eyes. The accuracy of the prediction and evaluation of the irritation of the test substance of the present application reaches 100%, which solves the technical problem that the prior art cannot provide a method with a high accuracy in predicting and evaluating the irritation of cosmetic raw materials to the eyes. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 A schematic diagram of the evaluation method provided for this application;
[0022] Figure 2 Schematic diagram of the structure of the fusion model in this embodiment;
[0023] Figure 3 Schematic diagram of a comprehensive testing strategy based on a hierarchical combination of alternative methods of a fusion model in an embodiment. DETAILED DESCRIPTION
[0024] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0025] Example
[0026] Please refer to Figure 1-Figure 3 This embodiment specifically uses a specific experimental example to evaluate the eye irritation of cosmetic raw materials, and illustrates a method for evaluating the eye irritation of cosmetic raw materials provided by this embodiment. The method is executed by a controller and includes:
[0027] Obtain data on in vitro substitution test results and in vivo test results of target cosmetic raw materials to form a data set;
[0028] Execute the input fusion model on the dataset;
[0029] Determine whether there is the same result data based on the data set. If so, perform evaluation result output. If not, perform ensemble learning and similarity judgment.
[0030] Output the evaluation results.
[0031] It is worth noting that the method for evaluating the eye irritation of cosmetic raw materials provided in this embodiment is based on a machine learning fusion model that combines an ensemble learning algorithm and a similarity algorithm. The acquired data set is input into the fusion model, and the predictive evaluation effect of the fusion model is verified and evaluated, thereby clarifying whether the cosmetic raw materials cause irritation to the eyes, thereby solving the technical problem that the prior art cannot provide a method with a high accuracy for predicting and evaluating the eye irritation of cosmetic raw materials.
[0032] Preferably, a quick-lookup table for predicting and evaluating in vitro test results based on a fusion model and a stratified combination evaluation strategy for in vitro alternative tests are provided, wherein a quick-lookup table for predicting and evaluating is configured in the fusion model to improve the convenience of the machine learning fusion model in practical applications, the quick-lookup table for predicting and evaluating is generated by computer simulation in combination with the fusion model, and the stratified combination evaluation strategy for in vitro alternative tests is obtained by analyzing and comparing the accuracy, negative coincidence rate and positive coincidence rate of the collected data sets.
[0033] Preferably, this embodiment uses Python language and scikit-learn library to model the fusion model, and uses the collected data set to train the fusion model.
[0034] Furthermore, the in vitro alternative test items of the data set consist of six test results: fluorescein leakage test, neutral red uptake test, red blood cell hemolysis test, chicken embryo chorioallantoic membrane test, chicken embryo chorioallantoic membrane trypan blue staining test, and in vitro rabbit eye test, and the data set includes at least 186 test data of 84 cosmetic raw materials.
[0035] In detail, 186 test data were obtained for the 84 cosmetic raw materials, including data for 30 cosmetic raw materials as follows:
[0036] As shown in Table 1:
[0037]
[0038]
[0039] Table 1
[0040] Based on the above, Table 1 shows some data on eye irritation of cosmetic raw materials. If there are multiple test results in the table, it means that there are multiple test combination results for the substance. For example, the in vivo test result of 1% cetrimonium bromide is moderate irritation, while there are two possible combinations of in vitro alternative tests, namely, the results of FLT, NRU, RBCH, HET-CAM, CAM-TBS, and IRET are (Grade II, Grade II, Grade II, Grade II, Grade II, Grade II) and (Grade II, Grade II, Grade II, Grade III, Grade III, Grade II) respectively. It should be noted that grades I, II, and III in the table represent non-irritation, moderate irritation, and severe irritation, respectively.
[0041] Furthermore, the method of determining whether there is the same result data based on the data set, and if so, outputting the evaluation result, and if not, performing ensemble learning and similarity judgment, includes:
[0042] Determine whether the data set has the same result data and output the evaluation result;
[0043] Determine that there is no identical result data in the data set and perform ensemble learning;
[0044] After ensemble learning, determine whether evaluation can be performed. If not, output the result that cannot be evaluated. If evaluation is possible, perform similarity calculation.
[0045] Calculate the most similar data results in the data set, perform judgment on the results, and output the evaluation results when the similarity is greater than 50%. When the similarity is less than or equal to 50%, output the result that cannot be evaluated.
[0046] It is worth noting that the fusion model is composed of two parts: the ensemble learning algorithm and the similarity algorithm. Its structure is as follows: Figure 2As shown. Based on the above fusion model, in the experimental examples provided in this embodiment, through the analysis of experimental data, it is found that the results of various in vitro eye irritation alternative tests cannot completely replace the in vivo test results. For example, the in vitro alternative test results of isopropyl alcohol and isobutyl alcohol FLT, NRU, RBCH, HET-CAM, CAM-TBS, IRET are all severe irritation, moderate irritation, non-irritation, severe irritation, severe irritation and severe irritation, but in fact, the in vivo test irritation of isopropyl alcohol is moderate irritation, while the in vivo test irritation of isobutyl alcohol is severe irritation. Therefore, the machine learning fusion model in this embodiment realizes the four classifications of cosmetic raw materials: non-irritation, moderate irritation, severe irritation and unpredictable evaluation, thereby improving the accuracy of predicting and evaluating eye irritation from the output design of the machine learning fusion model.
[0047] Furthermore, the evaluation indicators used in this embodiment for predicting and evaluating the eye irritation of cosmetic raw materials using the machine learning fusion model are:
[0048] (1) Accuracy: used to indicate the proportion of the fusion model that correctly predicts the test irritation in the test subject: Accuracy = (true negative + true positive) / (true negative + true positive + false negative + false positive) × 100%.
[0049] (2) Negative coincidence rate: used to indicate the proportion of correct predictions in the non-irritating results predicted and evaluated by the fusion model. Negative coincidence rate = true negative / (true negative + false negative) × 100%.
[0050] (3) Positive coincidence rate: used to indicate the proportion of correct predictions in the moderate or severe irritation results predicted and evaluated by the fusion model: Positive coincidence rate = true positive / (true positive + false positive) × 100%.
[0051] In some embodiments, the ensemble learning includes at least a voting classifier and a machine learning classification algorithm, and the machine learning classification algorithm includes at least a support vector machine algorithm, a logistic regression algorithm, a decision tree algorithm, a random forest algorithm, and a K nearest neighbor algorithm. Figure 2 As shown, in the fusion model, the integrated learning consists of a voting classifier and five machine learning classification algorithms, among which the support vector machine algorithm (SVM), logistic regression algorithm, decision tree algorithm, random forest algorithm and K nearest neighbor algorithm are classic classification model algorithms in machine learning.
[0052] In some embodiments, the similarity calculated by the similarity is expressed as:
[0053] s / (s+d)×100%
[0054] The number of identical test results is denoted as s, and the number of different test results is denoted as d.
[0055] In detail, in the fusion model, the similarity calculation method of the two test subjects is: in the aforementioned six test results of the two test subjects, the number of identical test results is recorded as s, and the number of different test results is recorded as d, then the similarity of the two test subjects is s / (s+d)×100%; for the design of the input and output data encoding of the fusion model, this embodiment uses 1, 2, and 3 to represent the non-irritation, moderate irritation, and severe irritation of the test results, and uses -1 to represent that there are no in vitro test results or the model cannot predict the evaluation.
[0056] Optionally, the development environment of the intrusion detection classification experiment in this embodiment is a notebook with an Intel Core i7-1255U CPU, 32GiB RAM, and an NVIDIA GeForce MX570 GPU, a Manjaro Linux 64-bit operating system, and a Sklearn machine learning library.
[0057] Furthermore, in order to verify the predictive evaluation effect of the machine learning fusion model, the collected data in this embodiment is divided into training data and test data at a ratio of 9:1, that is, the test data of 75 substances are used to establish the fusion model, and the test data of 9 substances with a total of 20 items are used to test and verify the model. Finally, the results of the predictive evaluation of the fusion model are compared with the Draizetest classification results as shown in Table 2 below.
[0058]
[0059] As can be seen from Table 2, in the external validation experiment of the fusion model, the accuracy of the fusion model in predicting the grading results was 100%, and the highest accuracy in predicting the presence or absence of eye irritation was 98.9%. Therefore, the predictive evaluation effect of the fusion model provided in this embodiment is significantly improved.
[0060] In some embodiments, the present embodiment can also generate a comprehensive test strategy based on the fusion model layered combination alternative method based on the fusion model, such as Figure 3 shown.
[0061] Based on the above strategy, the prediction and evaluation accuracy of in vitro alternative tests were analyzed using experimental data, and the results were as follows: In a single in vitro alternative test, the overall accuracy of different in vitro alternative methods for predicting and evaluating irritation was ranked from high to low as follows: IRET, NRU, FLT, CAM-TBS, RBCH, HET-CAM. This conclusion will provide a reference for the test sequence of hierarchical combination alternative methods. In detail, as shown in Table 3, in the prediction and evaluation results of a single in vitro alternative test, the initial alternative test for a test substance that is presumed to be non-eye irritant is the HET-CAM test; as shown in Table 4, in the prediction and evaluation results of a combined in vitro alternative test, the initial alternative test for a test substance that is presumed to be an eye irritant is a combined FLT and RBCH test; if the initial alternative test results cannot determine the eye irritation of the test substance, the remaining IRET, NRU, and CAM-TBS experiments are performed in sequence in combination with the machine learning fusion model of this study.
[0062]
[0063]
[0064] Table 3
[0065]
[0066]
[0067] Table 4
[0068] Furthermore, as shown in Table 5, as the number of combinations of in vitro alternative tests increases, the test results of the test substance that can be predicted and evaluated by the fusion model will continue to improve. Some data examples of the quick-check table for in vitro test evaluation based on the fusion model are shown in Table 6. Using the quick-check table for predictive evaluation, the quick check of test values facilitates the experimenter to quickly query the predicted irritation of the current in vitro test combination results. These are the test result values arranged in the order of FLT, NRU, HET-CAM, CAM-TBS, IRET, and RBCH in the in vitro alternative tests that have been conducted.
[0069] Test combination FLT NRU RBCH HET-CAM CAM-TBS IRET Proportion of evaluable results Three test combinations Test value - Test value Test value - - 0.37(10 / 27) Four test combinations Test value - Test value Test value - Test value 0.58(47 / 81) Five test combinations Test value Test value Test value Test value - Test value 0.73(178 / 243) Six test packages Test value Test value Test value Test value Test value Test value 0.89(647 / 729)
[0070] Table 5
[0071] Test value quick check Irritation Prediction Value FLT NRU RBCH HET-CAM CAM-TBS IRET Similar substances 112111 1 1 1 2 1 1 1 Macrogol 400, Glycerin 111112 1 1 1 1 1 1 2 Calcium thioglycolate, ammonium nitrate, Tween 20 322322 2 3 2 2 3 2 2 TritonX-100(5%), Triton 100(5%)
[0072] Table 6
[0073] Based on the above content, the final experimental results of this embodiment show that the machine learning fusion model proposed in this application performs well in predicting and evaluating the irritation of cosmetic raw materials based on in vitro surrogate test result data, and the accuracy rate of the irritation prediction and evaluation of the test substances in the verification experimental results reaches 100%.
[0074] So far, various embodiments of the present disclosure have been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed here.
[0075] Although some specific embodiments of the present disclosure have been described in detail by way of examples, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. It should be understood by those skilled in the art that the above embodiments may be modified or some technical features may be replaced by equivalents without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.
Claims
1. A method for evaluating the eye irritation of cosmetic raw materials, the method being executed by a processor, characterized in that: Obtain data on in vitro substitution test results and in vivo test results of target cosmetic raw materials to form a data set; Execute the input fusion model on the dataset; Determine whether there is the same result data based on the data set. If so, perform evaluation result output. If not, perform ensemble learning and similarity judgment. Output the evaluation results.
2. The evaluation method according to claim 1, characterized in that: The in vitro alternative test items of the data set consist of six test results: fluorescein leakage test, neutral red uptake test, red blood cell hemolysis test, chicken embryo chorioallantoic membrane test, chicken embryo chorioallantoic membrane trypan blue staining test, and in vitro rabbit eye test, and the data set includes at least 186 test data of 84 cosmetic raw materials.
3. The evaluation method according to claim 1, characterized in that: The method of judging whether there is the same result data based on the data set, and if so, outputting the evaluation result, and if not, performing integrated learning and similarity judgment, includes: Determine whether the data set has the same result data and output the evaluation result; Determine that there is no identical result data in the data set and perform ensemble learning; After ensemble learning, determine whether evaluation can be performed. If not, output the result that cannot be evaluated. If evaluation is possible, perform similarity calculation. Calculate the most similar data results in the data set, perform judgment on the results, and output the evaluation results when the similarity is greater than 50%. When the similarity is less than or equal to 50%, output the result that cannot be evaluated.
4. The evaluation method according to claim 3, characterized in that: The integrated learning includes at least a voting classifier and a machine learning classification algorithm, and the machine learning classification algorithm includes at least a support vector machine algorithm, a logistic regression algorithm, a decision tree algorithm, a random forest algorithm and a K nearest neighbor algorithm.
5. The evaluation method according to claim 3, characterized in that: The similarity calculated by the similarity is expressed as: s / (s+d)×100% The number of identical test results is denoted as s, and the number of different test results is denoted as d.