A method, apparatus, medium, and product for reproductive toxicity probability prediction and risk assessment

Abstract

The application discloses a reproductive toxicity probability prediction and risk assessment method, device, medium and product, relates to the cross field of reproductive toxicology, computational toxicology and artificial intelligence, filters a plurality of reproductive toxicity characteristic parameters based on a reproductive toxicity outcome, a reproductive toxicity mechanism and cluster analysis, constructs a multi-dimensional reproductive toxicity characteristic vector through quantification and screening, and combines a to-be-tested exogenous compound to construct a standardized characteristic vector library, so that a reproductive toxicity probability prediction model is trained, reproductive toxicity probability prediction and risk assessment based on the multi-dimensional characteristic vector are realized, a reproductive toxicity probability of the to-be-tested compound is obtained based on the characteristic vector library, a judgment standard reproductive toxicity and an evaluation risk grade are obtained, and the health risk of an adverse reproductive outcome caused by the to-be-tested compound is evaluated, so that the safety evaluation demand is met.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of reproductive toxicology, computational toxicology and artificial intelligence, and in particular to a method, device, medium and product for predicting the probability and assessing the risk of reproductive toxicity. Background Technology

[0002] Currently, reproductive toxicity data for chemicals are mostly obtained through animal experiments. However, conducting animal experiments presents several significant challenges: 1. It requires substantial financial resources; 2. The experimental cycle is long; 3. It requires the sacrifice of a large number of experimental animals; 4. It is difficult to conduct replicative verification and mechanistic studies. This contradicts the currently strongly advocated "3R" principle: Reduction, Refinement, and Replacement. Therefore, from both a scientific and economic perspective, the calculation of toxicological substitution methods is of paramount importance for the hazard assessment and management of exogenous chemicals.

[0003] Traditional reproductive toxicity animal testing not only requires a huge number of animals but also involves lengthy testing cycles and high costs. The lengthy testing cycle directly results in the inability to promptly complete reproductive toxicity testing for the vast majority of newly added compounds, creating a significant data gap. In addition to the lengthy cycle, traditional animal testing also suffers from enormous consumption of human, material, and financial resources. The testing process requires strict control of the animal housing environment (temperature, humidity, light, etc., all with clearly defined standards), the use of professional caretakers and testing personnel, the investment of large amounts of reagents and equipment, and long-term monitoring and data recording, all while adhering to GLP standards, further increasing testing costs. This high-consumption, low-efficiency testing model cannot meet the rapidly growing demand of the global reproductive toxicology services market.

[0004] In summary, the global market demand for reproductive toxicity testing is huge and continues to rise. However, the inherent limitations of traditional animal testing (long cycle, low efficiency, and huge consumption) have resulted in a long-term lack of reproductive toxicology data for many newly added compounds, which cannot meet the needs of safety assessment and has become a key pain point restricting the development of the industry. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, medium, and product for predicting reproductive toxicity probability and assessing risk, in order to solve the problem that a large number of newly added compounds have long lacked reproductive toxicology data, which has failed to meet the needs of safety assessment.

[0006] To achieve the above objectives, this application provides the following solution.

[0007] Firstly, this application provides a method for predicting the probability and assessing the risk of reproductive toxicity, including: Multiple reproductive toxicity characteristic parameters were screened based on reproductive toxicity outcomes, reproductive toxicity mechanisms, and cluster analysis; The various reproductive toxicity characteristic parameters are quantified and screened to construct a multidimensional reproductive toxicity feature vector; A feature vector library was constructed based on the aforementioned multidimensional reproductive toxicity feature vector and the exogenous compounds to be tested; A reproductive toxicity probability prediction model was constructed based on the logistic classification function framework. The reproductive toxicity probability prediction model is trained based on the aforementioned feature vector library; The reproductive toxicity probability prediction model after training is used to obtain the probability of reproductive toxicity of the exogenous compound to be tested, and the reproductive toxicity is determined and the risk level is assessed according to the judgment criteria.

[0008] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the reproductive toxicity probability prediction and risk assessment method described in any one of the above.

[0009] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the reproductive toxicity probability prediction and risk assessment method described in any one of the above.

[0010] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the reproductive toxicity probability prediction and risk assessment method described in any one of the above descriptions.

[0011] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, device, medium, and product for predicting reproductive toxicity probability and assessing risk. Based on reproductive toxicity outcomes, mechanisms, and cluster analysis, various reproductive toxicity characteristic parameters are screened. A multidimensional reproductive toxicity feature vector is constructed through quantification and screening, effectively integrating the physicochemical properties, structural characteristics, toxicokinetics, and genotoxicity of exogenous compounds. Based on the multidimensional reproductive toxicity feature vector and the exogenous compound to be tested, a standardized feature vector library is constructed to train a reproductive toxicity probability prediction model, achieving reproductive toxicity probability prediction and risk assessment based on multidimensional feature vectors. The reproductive toxicity probability prediction model is trained based on the feature vector library to obtain the probability of reproductive toxicity of the compound to be tested. Reproductive toxicity is assessed according to judgment criteria, and the risk level is evaluated, along with the health risk of adverse reproductive outcomes. This application meets safety assessment requirements by introducing multiple reproductive toxicity characteristic parameters. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating a method for predicting and assessing the probability of reproductive toxicity in one embodiment of this application. Figure 2 The correlation analysis results between 15 characteristic parameters provided in one embodiment of this application and reproductive toxicity; Figure 3 This is a functional structure diagram of a probabilistic prediction model for the reproductive toxicity of exogenous compounds based on a supervised learning algorithm, provided in an embodiment of this application. Figure 4 This is a schematic diagram of the main interface of a probabilistic prediction model for the reproductive toxicity of exogenous compounds based on a supervised learning algorithm, provided in an embodiment of this application. Figure 5 The first and second training results of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in an embodiment of this application are shown in the figure. Figure 6 The training results of the third and fourth training iterations of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in an embodiment of this application are shown in the figure. Figure 7 A schematic diagram of the fifth training result and the model saving interface of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in an embodiment of this application; Figure 8 A schematic diagram of the parameter output interface of an exogenous compound reproductive toxicity probability prediction model based on a supervised learning algorithm provided in an embodiment of this application; Figure 9 ROC curves of reproductive toxicity probability and various characteristic parameters provided in an embodiment of this application; Figure 10 A schematic diagram of the probability prediction submenu of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in an embodiment of this application; Figure 11 A schematic diagram of the data input interface for a probabilistic prediction model of exogenous compound reproductive toxicity based on a supervised learning algorithm provided in an embodiment of this application; Figure 12 A schematic diagram of a single-item prediction output interface of an exogenous compound reproductive toxicity probability prediction model based on a supervised learning algorithm provided in an embodiment of this application; Figure 13 A schematic diagram of the batch prediction output interface of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in an embodiment of this application; Figure 14 The reproductive toxicity prediction result diagram of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in Embodiment 1 of this application; Figure 15 This is a diagram showing the reproductive toxicity prediction results of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in Embodiment 2 of this application. Figure 16 This is a diagram showing the reproductive toxicity prediction results of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in Embodiment 3 of this application; Figure 17 This is a schematic diagram of the parameter input interface for the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in Embodiment 4 of this application; Figure 18 This is a schematic diagram of the parameter output interface of the exogenous compound reproductive toxicity probability prediction model based on supervised learning algorithm provided in Embodiment 4 of this application; Figure 19 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] Chemical registration applications are required to provide data on multiple toxicological endpoints, with reproductive toxicity being one of the mandatory endpoints. Due to its long evaluation cycle and the large number of animals involved, it is one of the most resource-intensive and labor-intensive testing items among toxicological endpoints. In terms of the scale of newly added compounds, the Chemical Abstracts Service (CAS) REGISTRY database currently contains over 71 million unique chemical substances, with approximately 12,000 new additions daily, translating to about 4.38 million new compounds annually. This massive volume of new compounds has led to a geometric increase in the demand for reproductive toxicity testing, but the inefficiency of traditional animal testing has become a bottleneck in meeting this demand.

[0016] According to a 2017 report by the European Commission, 9.39 million animals are used annually in scientific research across the 28 EU member states, with approximately 100,000 specifically used for embryonic developmental toxicity studies. This represents only a portion of global animal testing usage, and with the continuous emergence of new compounds, the number of animals used is expected to increase further. my country's GB / T 21771-2025 standard, "Repeated Dose Toxicity Screening Tests for Chemicals Combined with Reproductive / Developmental Toxicity," stipulates that a single test requires at least dozens of rats, and a complete testing cycle typically takes 3-6 months or even longer. This lengthy testing period directly results in the inability to complete reproductive toxicity testing for the vast majority of newly developed compounds in a timely manner, creating a significant data gap.

[0017] Based on data from CAS REGISTRY regarding newly added compounds each year, over 90% of the more than 4 million new compounds added annually lack reproductive toxicity testing, resulting in a lack of crucial reproductive toxicology data. Besides the lengthy testing process, traditional animal testing also incurs enormous costs in terms of manpower, resources, and finances. This high-consumption, low-efficiency testing model cannot meet the rapidly growing demand in the global reproductive toxicology services market. Against this backdrop, developing and promoting a rapid, high-throughput, and easy-to-use method for reproductive toxicity assessment has become an urgent need for the industry.

[0018] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] This application discloses a method for predicting the probability and assessing the risk of reproductive toxicity. It utilizes characteristic data of known reproductive and non-reproductive toxic compounds collected from an open-source chemical database system to construct a feature vector with the electrophilic active group, hormone-active structure, heavy metal coordination structure, monocyclic / polycyclic aromatic hydrocarbon structure, lipophilicity, half-life, and mutagenicity of the exogenous compound to be tested as characteristic parameters. A supervised learning algorithm is employed to extract the patterns of reproductive toxicity occurrence from a large amount of compound data, determine whether a substance possesses reproductive toxicity potential, and explore the probabilistic mapping relationship between the feature vector and the binary classification outcome of reproductive toxicity. This establishes a probability prediction model for the reproductive toxicity of exogenous compounds, or identifies key molecular structures affecting reproductive toxicity through feature importance analysis, providing clues for toxicity mechanism research. By inputting the characteristic parameters of the exogenous compound to be tested, the model coefficients are calculated and optimized using maximum likelihood estimation, outputting the probability of reproductive toxicity of the exogenous compound and assessing the reproductive toxicity risk level of the exogenous compound to be tested. This achieves rapid, high-throughput prediction and risk assessment of the reproductive toxicity probability of exogenous chemicals, effectively determining the human reproductive toxicity of exogenous compounds and assessing the health risk of adverse reproductive outcomes.

[0020] By integrating reproductive toxicity-related data from open-source databases and combining multi-dimensional features such as molecular fingerprinting, physicochemical properties, genotoxicity, structural characteristics, and toxicokinetic characteristics, and quantifying parameters through feature engineering, a supervised learning algorithm is used to construct a predictive model for the reproductive toxicity of exogenous compounds. This model directly reflects the structure-toxicity relationship, indicating the health hazards of exogenous compounds to humans, and achieves high-throughput and rapid prediction of the reproductive toxicity of exogenous compounds. It overcomes the limitations of traditional reproductive toxicity assessment methods, which rely on animal experiments, are time-consuming, costly, have low throughput, and require advanced technology. This provides an efficient and reliable technical means for early screening, toxicological mechanism research, and safety evaluation of the reproductive toxicity of exogenous compounds. This method does not require the use of live animals, saving manpower, material resources, financial resources, and time. It is highly reproducible, easy to operate, and conforms to the 3R principle of compound toxicity assessment.

[0021] In one exemplary embodiment, such as Figure 1 As shown, a method for predicting the probability of reproductive toxicity and assessing risk is provided. This method is executed by computer equipment, specifically by a computer device such as a terminal or server alone, or by a terminal and server together. The method includes the following steps.

[0022] Step 1: Based on reproductive toxicity outcomes, reproductive toxicity mechanisms, and cluster analysis, various reproductive toxicity characteristic parameters were screened out.

[0023] Step 2: Quantify and screen the various reproductive toxicity characteristic parameters to construct a multidimensional reproductive toxicity feature vector.

[0024] Step 3: Construct a feature vector library based on the multidimensional reproductive toxicity feature vector and the exogenous compound to be tested.

[0025] Step 4: Construct a reproductive toxicity probability prediction model based on the logistic classification function framework.

[0026] Step 5: Train the reproductive toxicity probability prediction model based on the feature vector library.

[0027] Step 6: Based on the trained reproductive toxicity probability prediction model, obtain the probability of reproductive toxicity of the exogenous compound to be tested, and determine the reproductive toxicity and assess the risk level according to the judgment criteria.

[0028] In an exemplary embodiment, step 1 screens out a variety of reproductive toxicity characteristic parameters based on reproductive toxicity outcomes, reproductive toxicity mechanisms, and cluster analysis, specifically including the following steps.

[0029] Step 101: Collect reproductive toxicity characteristic parameters of exogenous compounds using academic literature databases and open-source database systems.

[0030] Reproductive toxicity characteristic parameters, as a set of risk factors highly correlated with the reproductive toxicity of exogenous compounds, are mainly obtained through two methods. First, academic literature databases (such as PubMed, Web of Science, CNKI, etc.) are used to retrieve the latest toxicological research reports, reviews, and epidemiological surveys, as well as domestic standards and papers on occupational health and chemical safety. This allows for the acquisition of characteristic parameters for exogenous compounds known to have reproductive toxicity, followed by further screening using comparative analysis with known non-reproductive toxic compounds. Second, open-source chemical databases (such as PubChem, ATSDR, ChEMBL, ToxCast, EPA CompTox, NITE, ECHA CHEM, and the State Administration of Work Safety's Catalogue of Hazardous Chemicals (2015 Edition)) are used to acquire characteristic parameters for exogenous compounds known to have reproductive toxicity. This allows for the comparison and analysis of structural characteristics, physicochemical properties, toxicity categories, mechanisms of action, and toxicokinetics of reproductive toxic compounds to screen for their reproductive toxicity characteristic parameters.

[0031] Step 102: Conduct a comprehensive analysis of the correlation, quantifiability, availability, interpretability, and stability of reproductive toxicity outcomes and mechanisms with the characteristic parameters of the exogenous compound, and determine the reproductive toxicity characteristic parameters in the literature database and the open-source database.

[0032] Step 103: Obtain the intersection of reproductive toxicity characteristic parameters from the literature database and the open-source database. Through cluster analysis, 15 reproductive toxicity characteristic parameters involving the structural characteristics, physicochemical properties, toxicokinetics, genotoxicity, and exposure characteristics of exogenous compounds are selected, as shown in Table 1.

[0033] Table 1. Summary of Reproductive Toxicity Characteristic Parameters of Exogenous Compounds

[0034] In an exemplary embodiment, step 2 quantifies and filters the 15 reproductive toxicity characteristic parameters to construct a multidimensional reproductive toxicity characteristic vector, which specifically includes the following steps.

[0035] Step 201: Establish the correspondence between reproductive toxicity classification variables and reproductive toxicity to characterize the reproductive toxicity risk intensity of each exogenous compound. Referring to GB30000.24—2013 "Classification and Labelling Specifications for Chemicals Part 24: Reproductive Toxicity", reproductive toxicity is divided into four categories: 1A, 1B, 2 and none. The corresponding values ​​of the reproductive toxicity classification variables are 3, 2, 1 and 0, respectively.

[0036] Based on the above process, each reproductive toxicity characteristic parameter is divided into four levels (3, 2, 1, and 0) according to its intensity, serving as a classification variable for reproductive toxicity. Open-source toxicology, chemistry, and pharmacology data resources are retrieved to obtain scoring criteria information for the reproductive toxicity characteristic parameters of exogenous compounds (e.g., the number of electrophilic active groups, whether it can mimic / antagonize sex hormones, etc.). The reproductive toxicity characteristic parameters of the exogenous compounds are scored according to Table 2. Based on the quantitative scoring table of exogenous compound reproductive toxicity characteristic parameters, 15 reproductive toxicity characteristic parameters of exogenous compounds are quantified, establishing an exogenous compound dataset containing 15 reproductive toxicity characteristic parameters and classification variables; the exogenous compound dataset includes both reproductive toxic compounds and non-reproductive toxic compounds.

[0037] Table 2. Quantitative Scoring Table for Reproductive Toxicity Characteristic Parameters of Exogenous Compounds

[0038] Step 202: Based on the exogenous compound dataset, seven reproductive toxicity characteristic parameters with a correlation higher than the set correlation threshold are screened out by correlation analysis, and a multidimensional reproductive toxicity feature vector is constructed. The multidimensional reproductive toxicity feature vector includes electrophilic active groups, hormone-active structures, heavy metal coordination structures, monocyclic / polycyclic aromatic hydrocarbon structures, lipophilicity, half-life, and germ cell mutagenicity.

[0039] Since the correlation between reproductive toxicity feature parameters and categorical variables directly affects the classification ability of the predictive model, a higher correlation leads to a higher accuracy in distinguishing between reproductively toxic and non-reproductive toxic compounds, and a higher Area Under Curve (AUC) value. Conversely, if the correlation of reproductive toxicity feature parameters is too low, the loss function may fail to converge, resulting in model construction failure. Therefore, correlation analysis is needed to identify reproductive toxicity feature parameters that have a high correlation with categorical variables.

[0040] In one embodiment, this application uses SPSS 22.0 software to perform correlation analysis on the exogenous compound dataset, and outputs the correlation analysis results between 15 reproductive toxicity characteristic parameters and reproductive toxicity, as shown below. Figure 2 As shown, the larger the correlation coefficient r, the higher the correlation. From Figure 2The results show that the reproductive toxicity parameters with strong correlations to the categorical variables are half-life (correlation coefficient r=0.96), cell mutagenicity (correlation coefficient r=0.95), electrophilic active group (correlation coefficient r=0.84), and heavy metal coordination structure (correlation coefficient r=0.81). The reproductive toxicity parameters with moderate correlations to the categorical variables are lipophilicity (correlation coefficient r=0.68), monocyclic / polycyclic aromatic hydrocarbon structure (correlation coefficient r=0.46), and hormone-active structure (correlation coefficient r=0.44). The correlations of the remaining reproductive toxicity parameters are weak or very weak.

[0041] Eight reproductive toxicity feature parameters with weak or very weak correlation were removed, and seven reproductive toxicity feature parameters with a correlation higher than the set correlation threshold were selected. Based on electrophilic active groups, hormone active structures, heavy metal coordination structures, monocyclic / polycyclic aromatic hydrocarbon structures, lipophilicity (logP), half-life, and cell mutagenicity, a multidimensional feature vector was constructed.

[0042] Among them, the electrophilic active group characteristics of the exogenous compound to be tested refer to the functional groups in the molecule that have electrophilic reactivity, mainly including epoxides, α,β-unsaturated carbonyl groups (enones / enal), aldehydes, haloalkyl groups, etc., which are obtained by searching relevant databases (such as: Mayr Reactivity Database, UCIrvine Functional Electrophilic Dataset, PUBCHEM).

[0043] The hormonal activity structural characteristics of the exogenous compound to be tested refer to the structural skeleton with estrogen, androgen, or thyroid hormone receptor binding activity. These mainly include phenolic hydroxyl groups, bisphenol A compounds, alkylphenols, polychlorinated biphenyls, phthalates, aniline / benzamides, 17α-ethynyl groups, diketone structures, polybrominated diphenyl ethers (PBDEs), azoles, phenidones, etc., which are obtained through structure-based QSAR model prediction or by searching relevant databases (such as EDKB, EASIS, PubMed, etc.).

[0044] The heavy metal coordination structure characteristics of the exogenous compound to be tested refer to heavy metal ions that can form stable coordination with glutathione, DNA, thiol groups, etc., including lead (Pb) which can form highly reproductively toxic coordination structures. 2+ ), cadmium (Cd 2+ ), mercury (Hg) 2+ ), methylmercury (MeHg) + Arsenic (As) ions can form highly reproductively toxic coordination structures. 3+ As 5+ Organic arsenic), chromium (Cr) 6+ Highly toxic; Cr 3+ Weak), Nickel (Ni) 2+ ), Tin (Sn)2+ Copper (Cu) can cause reproductive toxicity through "excessive intake". 2+ / Cu + These heavy metal coordination structures with reproductive toxicity potential are mainly obtained through coordination chemistry calculations or by searching relevant databases (such as Pubchem, CompTox, CTD, etc.).

[0045] The monocyclic / polycyclic aromatic hydrocarbon structural characteristics of the exogenous compound to be tested refer to the aromatic hydrocarbon structure containing one or two or more benzene rings in the molecule, including benzo[a]pyrene (BaP), 7,12-dimethylbenzo[a]anthracene (DMBA), 3-methylcholanthracene (3-MC), benzo[a]anthracene (BaA), Chrysene, naphthalene, benzene, styrene, etc., which are obtained through ring system analysis or by searching relevant databases (such as EPACompTox, CTD, PubChem, PubMed).

[0046] The lipophilicity of the exogenous compound to be tested refers to the partition coefficient (logP) of the compound in the n-octanol / water two-phase system. Lipid solubility (logP) is a key property that determines whether a compound can penetrate biological membranes, enter reproductive organs, and produce toxicity. It is obtained through experimental determination or by searching relevant databases (such as ChemSpider).

[0047] The half-life characteristic of an exogenous compound being tested refers to its elimination half-life (t1 / 2) in vivo. The longer the half-life, the longer it accumulates in the testes, ovaries, and placenta. t1 / 2 can be obtained through pharmacokinetic experiments, prediction using physiologically based pharmacokinetic (PBPK) models, or by searching relevant databases (such as PubChem and ChemSpider).

[0048] The mutagenicity of the exogenous compound to be tested refers to the genotoxicity of the exogenous compound to germ cells (sperm, spermatocytes, spermatogonia, oocytes, etc.) and somatic cells. It can be obtained through in vitro / in vivo genotoxicity experiments such as mouse sperm abnormality test, spermatogonial chromosome aberration test, spermatocyte chromosome aberration test, transgenic rodent gene mutation test, Ames test, etc., or by searching relevant databases (such as: ChEMBL, NITE).

[0049] In an exemplary embodiment, step 3, based on the multidimensional reproductive toxicity feature vector and the exogenous compound dataset, constructs a standardized feature vector library, specifically including: A two-dimensional feature matrix is ​​established based on each group of exogenous compounds to be tested and the corresponding reproductive toxicity feature vectors. The two-dimensional feature matrix is ​​then standardized by data cleaning, deduplication, whitespace removal, and logical checks to establish a standardized feature vector library, totaling 1190 exogenous compounds, including 202 reproductive toxicity positive compounds and 988 reproductive toxicity negative compounds.

[0050] In an exemplary embodiment, step 4, which constructs a reproductive toxicity probability prediction model based on the logistic classification function framework, specifically includes: Based on the reproductive toxicity characteristic parameters and reproductive toxicity label variables of the exogenous compound to be tested, a probabilistic mapping relationship between the feature vector and the binary classification outcome of reproductive toxicity is established using the logistic classification function.

[0051] A reproductive toxicity probability prediction model is constructed based on the probability mapping relationship. The weight coefficients of the reproductive toxicity characteristic parameters in the reproductive toxicity probability prediction model are calculated and optimized through maximum likelihood estimation. The probability of occurrence of reproductive toxicity of the test compound and the reproductive toxicity risk of the test compound are output.

[0052] The reproductive toxicity probability prediction model is as follows: .

[0053] in, This represents the probability that the exogenous compound being tested has reproductive toxicity. ,σ( ) is the Sigmoid activation function. For linear combination values, , For bias terms, Let be the weighting coefficient of the j-th reproductive toxicity characteristic parameter. Let n be the n-dimensional feature vector of the exogenous compound to be tested (including feature parameters such as electrophilic active groups, hormone-active structures, and heavy metal coordination structures), and let the weighting coefficients be... The absolute value of the value reflects the contribution of reproductive toxicity characteristic parameters to the prediction of reproductive toxicity. A positive value indicates increased toxicity. A negative value indicates inhibition of toxicity.

[0054] The weight coefficient vector is solved by maximizing the log-likelihood function using the maximum likelihood estimation method and the Newton-Raphson iterative algorithm. The preset standard is that the change in the weighting coefficient is less than 0.001.

[0055] The log-likelihood function is: .

[0056] Where N is the number of training samples. This represents the reproductive toxicity label for the i-th training sample (1 indicates positive / reproductive toxicity, 0 indicates negative / no reproductive toxicity). To predict the probability of reproductive toxicity in the model.

[0057] In another exemplary embodiment, the construction and application of a reproductive toxicity probability prediction model are implemented using Python 3.8 programming language, and the specific process is as follows.

[0058] 1. Using the feature vector library of exogenous compounds as the training set, data is cleaned through the data standardization processing module to generate standardized data. A supervised learning algorithm is used to fit the functional mapping relationship between the reproductive toxicity feature parameters (electrophilic active groups, heavy metal coordination structures, etc.) of exogenous compounds and the classification variables (presence or absence of reproductive toxicity) to establish a reproductive toxicity probability prediction model.

[0059] 2. The program is titled "Probability Prediction Model of Reproductive Toxicity of Exogenous Compounds Based on Supervised Learning Algorithm," or simply "Prediction Model." The Prediction Model consists of two parts: model training and model application. Its functional structure is as follows: Figure 3 As shown, the model training section includes modules for loading the training set, data standardization, supervised learning, parameter solving, and model saving, used for building and optimizing high-precision prediction models. The model application section includes modules for loading the trained model, inputting exogenous compound reproductive toxicity characteristic parameters, predicting reproductive toxicity probabilities (including single and batch predictions), and assessing reproductive toxicity. By inputting exogenous compound characteristic parameters, it outputs the reproductive toxicity information of exogenous compounds, enabling prediction and risk assessment of exogenous compound reproductive toxicity. Before running the prediction model program, the following dependencies need to be installed: scikit-learn, numpy, pandas, scipy, rdkit, mordred, and matplotlib.

[0060] 3. The main interface of the prediction model is shown below. Figure 4 Based on the input compound characteristic parameters, the system directly outputs the reproductive toxicity probability and risk level of the target compound. The main interface, from top to bottom, consists of function buttons, a program progress display area (showing the program execution progress and completion time), a prediction model parameter table output area (showing the prediction model parameter list), and a prediction model formula display area (showing the prediction model formula).

[0061] The specific interface functions are displayed as follows: The prediction model program's runtime log window displays the message "Training set loaded successfully, totaling ... records".

[0062] When the data standardization module processes the data, the log window displays the message "Data standardization process completed, no abnormalities". When the supervised learning module is started, the log window displays the message "Supervised learning process started...".

[0063] If the prediction model is trained successfully, the log window will display the message "Model training complete" and show the accuracy (Accuracy, Acc), sensitivity (Sensitivity, Se), and specificity (Specificity, Sp).

[0064] To display the prediction model and its parameters, you can start the parameter solver module. The prediction model parameters will then be displayed in the output area of ​​the prediction model parameter table, and the mathematical expression of the model will be displayed in the model formula display area.

[0065] 4. Start the prediction model, load the training set to train the model, and continuously iterate and optimize it through machine learning to ensure that the model's accuracy (Accuracy, Acc) is ≥95%, and its sensitivity (Sensitivity, Se) and specificity (Specificity, Sp) are both above 88%, thus completing the prediction model training.

[0066] If Acc < 90% and Se and Sp < 88%, failing to meet the predetermined standards, the program will activate an adaptive iterative training mechanism to increase the sample size. The sample size will be increased according to the original positive to negative sample ratio, and the model will continue to be trained through the training set loading module until Acc ≥ 95% and Se and Sp are both ≥ 88%, meeting the predetermined standards. The model training will then be completed, and the model will be saved.

[0067] 5. Training was completed through iterative optimization using supervised learning after multiple loadings of the exogenous compound feature vector library. The first training iteration loaded the exogenous compound feature vector library, containing feature parameter data for 240 compounds, including 40 positive reproductive toxicity compounds and 200 negative reproductive toxicity compounds. After data standardization and supervised learning, the model's Acc, Se, and Sp were 0.785, 0.744, and 0.795, respectively, all below the preset standard. Continuing to collect compound feature parameter data, the second training iteration increased the sample size of the compound feature vector library to 480 compounds, including 80 positive reproductive toxicity compounds and 400 negative reproductive toxicity compounds. After data standardization and supervised learning, the model's Acc, Se, and Sp were 0.821, 0.778, and 0.832, respectively, all below the preset standard. The output results of the first and second training iterations of the prediction model are shown below. Figure 5 As shown.

[0068] 6. Continue collecting feature parameter data for exogenous compounds. In the third training iteration, the sample size of the exogenous compound feature vector library was increased to 720 compounds, including 120 compounds with positive reproductive toxicity and 600 compounds with negative reproductive toxicity. After data standardization and supervised learning, the model's Acc, Se, and Sp were 0.861, 0.816, and 0.872, respectively, all below the preset standard. Continue collecting feature parameter data for exogenous compounds. In the fourth training iteration, the sample size of the compound feature vector library was increased to 960 compounds, including 160 compounds with positive reproductive toxicity and 800 compounds with negative reproductive toxicity. After data standardization and supervised learning, the model's Acc, Se, and Sp were 0.901, 0.856, and 0.912, respectively, still below the preset standard. The output results of the third and fourth training iterations are as follows: Figure 6 As shown.

[0069] 7. Continue collecting characteristic parameter data of exogenous compounds. In the fifth training iteration, the sample size of the compound feature vector library was increased to 1190 compounds, including 202 positive reproductive toxicity compounds and 988 negative reproductive toxicity compounds. After data standardization and supervised learning, the model's Acc, Se, and Sp values ​​were 0.956, 0.881, and 0.972, respectively, meeting the preset standards, indicating successful model construction. Click the "Save Model" button to save the model to the specified directory. The output interface of the model after iterative optimization and the output results of the fifth training iteration are shown below. Figure 7 As shown.

[0070] 8. The parameter solving module of the prediction model automatically calculates the model parameters, including the bias term β0, weight coefficients β1-β7, the model reproductive toxicity prediction probability P-value, and the odds ratio EXP(B) of the reproductive toxicity characteristic parameters.

[0071] Model parameter output interface as follows Figure 8 As shown, column B of the model parameter table output area displays the values ​​of β0, β1, β2, β3, β4, β5, β6, and β7 as -6.431070, 0.273242, 1.065004, 2.283460, 0.373676, 0.262986, 1.331301, and 2.242670, respectively, and the corresponding P-values ​​(see...). Figure 8The p-values ​​(p-values) were all less than 0.05, indicating that each reproductive toxicity characteristic parameter had a statistically significant effect on the reproductive toxicity of exogenous compounds. The Exp(B) column in the model parameter output area showed that the odds ratios for the four reproductive toxicity characteristic parameters—visible hormone active structure, heavy metal coordination structure, half-life, and cell mutagenicity—were 2.900850, 9.810570, 3.785967, and 9.418448, respectively, all greater than 2, suggesting that these reproductive toxicity characteristic parameters are high-risk factors for reproductive toxicity.

[0072] Based on the maximum likelihood estimation (criterion), the Newton-Raphson iterative algorithm was used to calculate the model parameters. The weight coefficients and constants of the independent variables (electrophilic activity, hormone activity, heavy metal coordination, monocyclic / polycyclic aromatic hydrocarbons, lipophilicity, half-life, and mutagenicity) during the iteration process and at the convergence state are shown in Table 3.

[0073] Table 3. Data on the model parameter iteration process

[0074] As shown in Table 4, the model gradually converged during the iteration process: it terminated after 8 iterations because the changes in all estimated parameter values ​​were less than 0.001, meeting the preset convergence criterion and indicating that the parameter estimation had reached stability. During the iteration process, the log-likelihood value continuously increased with the number of iterations, gradually optimizing from -283.011526 in the first iteration to -173.198670 after the sixth iteration, with no significant change in subsequent iterations, suggesting that the model's fitting effect continuously improved and eventually reached its optimum. The weight coefficients and constants of each independent variable gradually converged during the iterations. The first 5 iterations were a rapid adjustment phase, with large changes in parameter values, indicating that the model quickly fitted the data characteristics. After the sixth iteration, all coefficients and constants basically stabilized, and the changes in coefficients in the 7th and 8th iterations were all <0.001, verifying the convergence and stability of the parameter estimation, indicating that the final model parameters can be used for subsequent model prediction.

[0075] 9. For example Figure 8 As shown, after the model training is complete, the expression for the prediction model is output in the model formula display area below the prediction model parameter output interface as follows: logit(P|Y=1)=-6.431070+0.273242*X1+1.065004*X2+2.283460*X3+0.373676*X4+0.262986*X5+1.331301*X6+2.242670*X7.

[0076] Logarithmically converted to: .

[0077] 10. Performance evaluation of the predictive model: The Hosmer-Lemeshow test was used to evaluate the degree of agreement between the model's predicted values ​​and the actual values. ROC-AUC analysis was used to evaluate the model's predictive accuracy. These dual validation methods fully ensured the scientific rigor and practicality of the program. ROC curves were used to evaluate the model's ability to identify reproductive toxicants. The ROC curves for each reproductive toxicity characteristic parameter with respect to 1-specificity and sensitivity are shown below. Figure 9 As shown in (a)-(h) in the table. The results show that the predicted probability value P after the final iteration optimization of the Hosmer-Lemeshow test is 0.191, indicating that the model's predicted value is in good agreement with the actual value; the area under the ROC curve of the model is 0.976, which is close to 1 and significantly higher than the AUC of each individual reproductive toxicity characteristic parameter in Table 4, indicating that the model has a good ability to identify reproductive toxins.

[0078] Table 4. Probability of reproductive toxicity and AUC values ​​of various characteristic parameters

[0079] 11. This application uses 5-fold cross-validation. Samples are randomly selected from the feature vector library (containing 1190 samples) to establish 5 validation sets. Each validation set contains 240 samples to verify the prediction accuracy, sensitivity and specificity of the prediction model. The results are shown in Table 5 below.

[0080] Table 5. Results of Five-Fold Cross-Validation

[0081] The Acc, Se, and Sp values ​​of the training set for the aforementioned prediction model are 0.956, 0.881, and 0.972, respectively; the mean Acc, Se, and Sp values ​​of the five-fold cross-validation set in Table 6 reach 0.902, 0.873, and 0.908, respectively. The higher performance of the training set compared to the cross-validation set is a normal phenomenon in machine learning modeling, and the Acc, Se, and Sp values ​​of both the training set and the five-fold cross-validation set are at a relatively good level. Statistical tests on the model weight coefficients β show that all coefficients... P <0.05 indicates that the model has good statistical reliability and toxicological rationality.

[0082] 12. Input the multidimensional characteristic parameters of the compound to be tested into the above probability prediction model to obtain the probability P of reproductive toxicity, and determine the reproductive toxicity and risk level according to the following judgment criteria based on the classification threshold.

[0083] The judgment criteria are as follows: If P < 0.50, there is no reproductive toxicity, and the reproductive toxicity risk level is determined to be none or low.

[0084] If 0.50 ≤ P < 0.70, reproductive toxicity is present, and the reproductive toxicity risk level is determined to be medium.

[0085] If 0.70 ≤ P < 0.90, reproductive toxicity is present, and the reproductive toxicity risk level is determined to be high.

[0086] If P ≥ 0.90, reproductive toxicity is present, and the reproductive toxicity risk level is determined to be high.

[0087] 13. Model Application. The reproductive toxicity probability prediction model of this application supports feature vector input modes, either line-by-line or batch input. It can predict the reproductive toxicity probability of exogenous compounds individually or in batches, assess reproductive toxicity and risk levels, and achieve high-throughput prediction and assessment of reproductive toxicity. The probability prediction submenu of the reproductive toxicity probability prediction model is as follows: Figure 10 As shown.

[0088] The single-item prediction mode loads a pre-trained prediction model through the model loading module, then inputs each characteristic parameter of the exogenous compound one by one through the data input module. After saving the data, the single-item prediction submodule of probability prediction outputs the prediction results, realizing the prediction of the reproductive toxicity probability and risk level of the exogenous compound, and displaying the results. The data input interface of the reproductive toxicity probability prediction model is as follows: Figure 11 As shown, the single-item prediction output interface of the reproductive toxicity probability prediction model is as follows. Figure 12 As shown.

[0089] The batch prediction mode involves inputting the feature parameters of the compounds to be tested in batches into an Excel file, forming a feature vector dataset. After loading the dataset using the training set loading module and standardizing it using the data standardization module, the batch prediction submodule of the probability prediction module outputs the results in batches in tabular form. This enables batch prediction of the reproductive toxicity probability and risk level of the compounds, and displays the results. The batch prediction output interface of the reproductive toxicity probability prediction model is shown below. Figure 13 As shown.

[0090] Compared with the prior art, this application has the following beneficial effects.

[0091] 1) This application provides a reproductive toxicity probability prediction model and application method for exogenous compounds, as well as supporting programs for model training and application. It can accurately predict the reproductive toxicity probability and risk level of exogenous compounds and has been verified to have high accuracy, good sensitivity and specificity. It can be used to quickly predict whether chemicals have the risk of causing human reproductive toxicity, and to predict the reproductive toxicity of chemical raw materials or new chemical substances, indicating potential health risks. It is expected to solve the problems of long cycle, high cost and complicated operation of existing reproductive toxicity detection methods.

[0092] 2) Through the analysis and screening of a large number of reproductive toxicity compound characteristic parameters, seven reproductive toxicity characteristic parameters with high correlation to reproductive toxicity were extracted: electrophilic active group, hormone active structure, heavy metal coordination structure, polycyclic aromatic hydrocarbon structure, lipophilicity, half-life, and cell mutagenicity. Practical application has confirmed that the multidimensional feature vector constructed using this set of multidimensional characteristic parameters has good predictive ability for the reproductive toxicity of exogenous compounds.

[0093] 3) Through the construction and application of a probabilistic prediction model for the reproductive toxicity of exogenous compounds, a group of high-risk factors for reproductive toxicity were identified. These factors are hormone-active structure, heavy metal coordination structure, half-life, and cell mutagenicity. The odds ratios of these four characteristic parameters were 2.90, 9.81, 3.78, and 9.41, respectively, which have a high influence weight on the probability of reproductive toxicity and are considered high-risk factors for reproductive toxicity. This has significant scientific value for the toxicological safety evaluation and risk warning of compound reproductive toxicity.

[0094] 4) The reproductive toxicity probability prediction model for exogenous compounds clearly defines the quantitative relationship between each reproductive toxicity characteristic parameter and the reproductive toxicity probability. This is the core mathematical foundation for the model to achieve accurate prediction. It accurately correlates the quantitative scores of reproductive toxicity characteristic parameters with the reproductive toxicity probability, clarifies the influence weight and calculation rules of each parameter, and ensures the scientific nature and repeatability of the model prediction.

[0095] 5) Improve the efficiency of reproductive toxicity detection, shorten the risk assessment cycle, and achieve high-throughput detection. Traditional compound reproductive toxicity detection relies on animal and in vitro experiments, which are cumbersome and time-consuming. This predictive model can achieve high-throughput screening of massive amounts of compounds, quickly output reproductive toxicity prediction results, shorten the cycle of chemical reproductive toxicity detection and risk assessment, and can be applied to new drug development, environmental pollutant screening, and early risk identification of industrial chemicals, accelerating the process of reproductive toxicity evaluation.

[0096] 6) Develop a dedicated program for the model using Python to implement the two core functions of model training and model application, forming a standardized program execution framework.

[0097] 7) Through in-depth mechanism analysis and integration of reproductive toxicology, genetic toxicology, and toxicokinetics principles, and after repeated adjustments and verifications, a scientific and feasible quantitative scoring scheme for reproductive toxicity characteristic parameters is proposed, namely the compound reproductive toxicity characteristic parameter quantitative scoring table. This enables the scientific quantification of reproductive toxicity characteristic parameters and provides a reliable basis for constructing a probability prediction model for the reproductive toxicity of exogenous compounds.

[0098] 8) Reduce testing costs and barriers, saving social resources. Traditional toxicity testing requires specialized GLP laboratories, SPF-grade animal facilities, and sophisticated testing equipment, resulting in high hardware costs and limited operational conditions. This technology eliminates the need for dedicated experimental sites and specialized testing equipment, relying on existing publicly available open-source data to complete model training and toxicity prediction. This significantly reduces the economic cost and resource consumption of compound reproductive toxicity testing, effectively saving public resources.

[0099] 9) Mitigating ethical risks and aligning with the "3R" principles of green experimentation. Traditional in vivo animal experiments do not require rigorous ethical review, leading to certain ethical controversies. This technology uses computational models to replace some in vivo and in vitro animal experiments, eliminating the need for ethical review of experimental animals and mitigating ethical risks from the outset, thus aligning with the "3R" principles of animal experimentation: substitution, reduction, and optimization.

[0100] 10) Reduce reliance on animal testing for reproductive toxicity risk assessment. Reproductive toxicity risk assessments of various chemicals typically require extensive animal testing and consume a large number of laboratory animals. This application, based on in-depth analysis of existing data, utilizes open-source data for reproductive toxicity prediction, effectively reducing reliance on animal testing.

[0101] 11) Identifying warning structures affecting reproductive toxicity using supervised learning algorithms. This technology leverages supervised learning algorithms to deeply mine massive amounts of molecular structure and toxicity data of known compounds, constructing an efficient reproductive toxicity prediction model. This model can effectively capture the complex correlation between molecular structure and reproductive toxicity, accurately identify warning structures affecting the reproductive toxicity of compounds, and make toxicity prediction more scientific and accurate.

[0102] 12) Highly scalable and adaptable to various toxicity studies. The supervised learning framework used in this technology has good openness and scalability, and the model has strong compatibility. Further integration of multi-dimensional data such as transcriptomics, protein interactions, and ADME parameters can be implemented to build multi-task, multi-endpoint exogenous compound toxicity prediction models, further elucidating complex toxic mechanisms such as endocrine disruption and epigenetic effects, and continuously expanding the model's predictive scope and application capabilities.

[0103] 13) Enhance the foresight and scientific rigor of chemical safety assessments. This predictive model provides a scientific and novel reproductive toxicity screening and assessment tool for the registration and application of various chemical products, including pharmaceuticals, pesticides, food additives, and environmental chemicals, significantly enhancing the foresight and scientific rigor of chemical safety assessments.

[0104] Example 1: Prediction of the reproductive toxicity probability of nonylphenol.

[0105] 1) Based on Table 2, the reproductive toxicity characteristic parameters of exogenous compounds to nonylphenol were scored, and the scores of each reproductive toxicity characteristic parameter were substituted into the prediction model formula to calculate the reproductive toxicity probability of nonylphenol. The scoring results and weighting coefficients of the reproductive toxicity characteristic parameters of nonylphenol are shown in Table 6.

[0106] Table 6. Score results and weighting coefficients of reproductive toxicity characteristic parameters of nonylphenol.

[0107] Specifically, the contribution value of each feature parameter is obtained by multiplying the score value of each feature parameter in Table 6 by the corresponding weight coefficient. Then, the contribution values ​​of each feature parameter are summed and a bias term is added to obtain the linear combination value Z. The prediction result can be obtained by substituting the Z value into the prediction model formula.

[0108] = = =0.854064752≈85.41%.

[0109] 2) According to the aforementioned judgment method: the probability of nonylphenol's reproductive toxicity is 85.4%, indicating that it has reproductive toxicity and the risk level is relatively high.

[0110] 3) Alternatively, feature parameter scores can be input into the prediction model, and the single-item prediction submodule of the probability prediction module can be called to directly obtain the prediction result, such as... Figure 14 As shown.

[0111] 4) According to the "List of Hazardous Chemicals (2015 Edition)" issued by the Ministry of Emergency Management of the People's Republic of China, nonylphenol has strong reproductive toxicity and is classified as a presumed human reproductive poison. This is consistent with the prediction results of this model.

[0112] Example 2: Prediction of the reproductive toxicity probability of lead sulfate.

[0113] 1) Based on the characteristic parameters of the exogenous compound lead sulfate in Table 2, score them and substitute the score values ​​into the prediction model formula to calculate the probability of reproductive toxicity of lead sulfate. The score results and weighting coefficients of the characteristic parameters of lead sulfate are shown in Table 7.

[0114] Table 7. Scoring results and weighting coefficients of lead sulfate characteristic parameters

[0115] Specifically, the contribution value of each feature parameter is obtained by multiplying the score value of each feature parameter in Table 7 by the corresponding weight coefficient. Then, the contribution values ​​of each feature parameter are summed and a bias term is added to obtain the linear combination value Z. The prediction result can be obtained by substituting the Z value into the prediction model formula.

[0116] = = =0.988028858≈98.80%.

[0117] 2) According to the aforementioned judgment method: the probability of reproductive toxicity of lead sulfate is 98.80%, indicating that it has reproductive toxicity and the risk level is high.

[0118] 3) Alternatively, feature parameter scores can be input into the prediction model, and the single-item prediction submodule of the probability prediction module can be called to directly obtain the prediction result, such as... Figure 15 As shown.

[0119] 4) According to the "List of Hazardous Chemicals (2015 Edition)" issued by the Ministry of Emergency Management of the People's Republic of China, lead sulfate is a known human reproductive toxicant, belonging to the highest risk category for reproductive toxicity. This is consistent with the results of this prediction model.

[0120] Example 3: Prediction of the reproductive toxicity probability of sodium bicarbonate.

[0121] 1) Based on the characteristic parameters of the exogenous compound sodium bicarbonate in Table 2, score them and substitute the score values ​​into the prediction model formula to calculate the probability of reproductive toxicity of sodium bicarbonate. The score results and weighting coefficients of sodium bicarbonate characteristic parameters are shown in Table 8.

[0122] Table 8. Scoring results and weighting coefficients of characteristic parameters of sodium bicarbonate

[0123] Specifically, the contribution value of each feature parameter is obtained by multiplying the score value of each feature parameter in Table 8 by the corresponding weight coefficient. Then, the contribution values ​​of each feature parameter are summed and a bias term is added to obtain the linear combination value Z. The prediction result is obtained by substituting the Z value into the prediction model formula.

[0124] = = =0.001608136≈0.16%.

[0125] 2) According to the aforementioned judgment method: the probability of reproductive toxicity of sodium bicarbonate is 0.16%, there is no reproductive toxicity, and the risk level is none or extremely low.

[0126] 3) Alternatively, feature parameter scores can be input into the prediction model, and the single-item prediction submodule of the probability prediction module can be called to directly obtain the prediction result, such as... Figure 16 As shown.

[0127] 4) Based on GB30000.24-2013 "Classification and Labelling Specifications for Chemicals Part 24: Reproductive Toxicity" and the NITE database, the existing toxicological data are insufficient to classify sodium bicarbonate into any of the hazard categories "Reproductive Toxicity 1A, 1B, Category 2". Therefore, it is not determined to be a reproductive toxic hazard. This is consistent with the results of this predictive model.

[0128] Example 4: Batch prediction of compound reproductive toxicity.

[0129] 1) According to as follows Figure 17 The output interface for the reproductive toxicity probability prediction model parameters is formatted by entering the characteristic parameters of the exogenous compound into an EXCEL spreadsheet, including the compound number, compound name, electrophilic active group, hormone-active structure, heavy metal coordination structure, monocyclic / polycyclic aromatic hydrocarbon structure, lipophilicity, half-life, and germ cell mutagenicity.

[0130] 2) Load the EXCEL file using the training set loading module, perform data standardization using the data standardization module, load the model using the model loading module, and then batch calculate the results using the batch prediction submodule of the probability prediction module. The results will be displayed in a pop-up window, such as... Figure 18 As shown.

[0131] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 19 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database is used for reproductive toxicity probability prediction and risk assessment. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to perform reproductive toxicity probability prediction and risk assessment.

[0132] Those skilled in the art will understand that Figure 19The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0133] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0134] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0135] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0136] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0137] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0138] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0139] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0140] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the probability and assessing the risk of reproductive toxicity, characterized in that, include: Multiple reproductive toxicity characteristic parameters were screened based on reproductive toxicity outcomes, reproductive toxicity mechanisms, and cluster analysis; The various reproductive toxicity characteristic parameters are quantified and screened to construct a multidimensional reproductive toxicity feature vector; A feature vector library was constructed based on the aforementioned multidimensional reproductive toxicity feature vector and the exogenous compounds to be tested; A reproductive toxicity probability prediction model was constructed based on the logistic classification function framework. The reproductive toxicity probability prediction model is trained based on the aforementioned feature vector library; The reproductive toxicity probability prediction model after training is used to obtain the probability of reproductive toxicity of the exogenous compound to be tested, and the reproductive toxicity is determined and the risk level is assessed according to the judgment criteria.

2. The method for predicting and assessing reproductive toxicity probability according to claim 1, characterized in that, Based on reproductive toxicity outcomes, mechanisms of reproductive toxicity, and cluster analysis, various reproductive toxicity characteristic parameters were screened, including: Reproductive toxicity characteristic parameters of exogenous compounds were collected using academic literature databases and open-source database systems; A comprehensive analysis was conducted on the correlation, quantifiability, availability, interpretability, and stability of reproductive toxicity outcomes and mechanisms with the characteristic parameters of the exogenous compounds to determine the reproductive toxicity characteristic parameters in literature databases and open-source databases. The intersection of reproductive toxicity characteristic parameters from the literature database and the open-source database was obtained, and various reproductive toxicity characteristic parameters were screened out through cluster analysis.

3. The method for predicting and assessing the probability of reproductive toxicity according to claim 1, characterized in that, The various reproductive toxicity characteristic parameters are quantified and screened to construct a multidimensional reproductive toxicity feature vector, specifically including: Based on a quantitative scoring table of reproductive toxicity characteristic parameters of exogenous compounds, the various reproductive toxicity characteristic parameters are quantified to establish an exogenous compound dataset containing the quantitative relationships between various reproductive toxicity characteristic parameters and classification variables; the exogenous compound dataset includes reproductive toxic compounds and non-reproductive toxic compounds; Based on the aforementioned exogenous compound dataset, seven reproductive toxicity characteristic parameters with a correlation higher than a set correlation threshold were screened out through correlation analysis, and a multidimensional reproductive toxicity feature vector was constructed. The multidimensional reproductive toxicity feature vector includes electrophilic active groups, hormone-active structures, heavy metal coordination structures, monocyclic aromatic hydrocarbon structures, polycyclic aromatic hydrocarbon structures, lipophilicity, half-life, and germ cell mutagenicity.

4. The method for predicting and assessing reproductive toxicity probability according to claim 1, characterized in that, Based on the aforementioned multidimensional reproductive toxicity feature vectors and the exogenous compounds to be tested, a standardized feature vector library is constructed, specifically including: A two-dimensional feature matrix is ​​established based on each group of exogenous compounds to be tested and the reproductive toxicity feature vectors corresponding to the exogenous compounds to be tested; The two-dimensional feature matrix is ​​cleaned, deduplicated, despaced, and logically checked to establish a standardized feature vector library.

5. The method for predicting and assessing reproductive toxicity probability according to claim 1, characterized in that, A reproductive toxicity probability prediction model is constructed based on the logistic classification function framework, specifically including: Based on the reproductive toxicity characteristic parameters and reproductive toxicity label variables of the exogenous compound to be tested, a logistic classification function is used to establish a probabilistic mapping relationship between the feature vector and reproductive toxicity. A reproductive toxicity probability prediction model is constructed based on the probability mapping relationship. The weight coefficients of the reproductive toxicity characteristic parameters in the reproductive toxicity probability prediction model are calculated and optimized through maximum likelihood estimation. The probability of occurrence of reproductive toxicity of the test compound and the reproductive toxicity risk of the test compound are output.

6. The method for predicting and assessing reproductive toxicity probability according to claim 5, characterized in that, The reproductive toxicity probability prediction model is as follows: ； in, This represents the probability that the exogenous compound being tested has reproductive toxicity. ,σ( ) is the Sigmoid activation function. For linear combination values, , For bias terms, Let be the weighting coefficient of the j-th reproductive toxicity characteristic parameter. A positive value indicates increased toxicity. A negative value indicates inhibition of toxicity. Let be the score value corresponding to the j-th reproductive toxicity feature parameter, where j = 1, 2, ..., n, and n is the total number of feature parameters.

7. The method for predicting and assessing the probability of reproductive toxicity according to claim 6, characterized in that, The judgment criteria are as follows: If P < 0.50, it is determined to be non-reproductive toxicity, and the reproductive toxicity risk level is determined to be no risk or low risk. If 0.50 ≤ P < 0.70, it is determined to have reproductive toxicity, and the reproductive toxicity risk level is determined to be medium risk. If 0.70 ≤ P < 0.90, it is considered to have reproductive toxicity, and the reproductive toxicity risk level is determined to be a high risk level; If P ≥ 0.90, it is considered to have reproductive toxicity, and the reproductive toxicity risk level is determined to be high risk.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the reproductive toxicity probability prediction and risk assessment method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the reproductive toxicity probability prediction and risk assessment method as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the reproductive toxicity probability prediction and risk assessment method as described in any one of claims 1-7.

Images