A method for predicting acute intraperitoneal toxicity based on maccs keys and machine learning

Abstract

This patent relates to an acute intraperitoneal toxicity prediction method based on MACCS bonds and a machine learning framework. This method collects and processes toxicity data of chemical compounds, calculates molecular descriptors using MACCS bonds, and constructs a gradient boosting decision tree model for toxicity prediction through feature selection and removal of highly correlated features. The model's hyperparameters are optimized using a grid search method, and the interpretability of the model is improved using SHAP value analysis. This achieves high-precision and highly interpretable toxicity prediction. The method achieves a coefficient of determination of 0.914, a mean squared error of 0.047, a root mean square error of 0.218, and a mean absolute error of 0.148 on the test set. This invention significantly improves the accuracy of toxicity prediction and has strong generalization ability, making it widely applicable to toxicity screening in new drug development, reducing reliance on traditional animal experiments, and lowering R&D costs and time.

Description

Technical Field

[0001] This invention belongs to the field of computer-aided molecular design and drug safety assessment, specifically involving a method for predicting acute intraperitoneal toxicity based on MACCS bonds and a machine learning framework, which is particularly suitable for toxicity prediction and molecular design in the drug development process. Background Technology

[0002] Acute intraperitoneal toxicity is a crucial indicator in drug safety assessment, particularly in new drug development, used to screen and evaluate the toxicity of chemical substances. Traditional toxicity testing methods typically rely on animal experiments, especially intraperitoneal injection studies in mice (LD50). While these experiments provide effective toxicity data, they suffer from drawbacks such as long processing times, high costs, and ethical concerns. Furthermore, animal test results are significantly influenced by species differences and cannot fully reflect human toxicity responses, thus limiting their predictive accuracy.

[0003] With the advancement of computer technology, machine learning methods have been widely applied in the field of toxicity prediction, especially in drug development, where they can significantly improve prediction efficiency and reduce experimental costs. Machine learning can extract the nonlinear relationship between molecular structure and toxicity from large amounts of historical toxicity data, thus providing a basis for predicting the toxicity of new chemical substances. Compared with traditional models based on physical and thermodynamic principles, machine learning methods have significant advantages in handling complex molecular structures and large-scale datasets.

[0004] Nevertheless, existing machine learning models still have certain limitations. For example, insufficient generalization ability and inappropriate feature selection can lead to unstable prediction results. Furthermore, many existing methods lack interpretability of their predictions, making it difficult for accurate models to provide valuable structural optimization suggestions for drug developers. Therefore, improving the interpretability and accuracy of machine learning models, especially in predicting acute intraperitoneal toxicity, has become a significant challenge in the field of toxicity assessment. Summary of the Invention

[0005] The purpose of this patent is to address the shortcomings of existing methods for predicting acute peritoneal toxicity in terms of accuracy, interpretability, and generalization ability. It provides an acute peritoneal toxicity prediction method based on the fusion of MACCS bonds and machine learning algorithms to accurately predict the toxicity of compounds, reduce the use of animal experiments, improve the efficiency of new drug development, and provide a scientific basis for toxicity assessment.

[0006] To achieve the above objectives, the technical solution of this patent is as follows: This patent provides a method for predicting acute peritoneal toxicity based on MACCS bonds and machine learning. The core of the method lies in constructing a machine learning framework based on a gradient boosting decision tree model, automatically optimizing hyperparameters through a grid search method, and enhancing the interpretability of the model using SHAP value analysis, ultimately achieving high-precision, high-generalization-ability, and interpretable prediction of acute peritoneal toxicity.

[0007] The gradient boosting decision tree model described in this patent is characterized by its ability to assess and predict acute intraperitoneal toxicity, and the steps are as follows: Step 1: Collect acute intraperitoneal toxicity (LD50) data of chemical compounds, convert them into standard units (mol / kg), and express them in log10 format; the database should contain 12,000-13,000 data points, and establish a SMILES string representation for each compound for subsequent molecular structure analysis.

[0008] Step 2: Perform feature processing on the original data constructed in Step 1: First, remove descriptors with missing values ​​or zero variance; then, use MACCS keys to calculate the initial molecular descriptors based on the SMILES representation of the molecules; finally, calculate the Pearson correlation coefficient between features, remove highly linearly correlated features with correlation coefficients greater than 0.94-0.96, and finally retain 100-130 effective features as model input.

[0009] Step 3: Construct a decision boosting tree model and use a grid search algorithm to automatically optimize the key hyperparameters of the model, including the number of trees, maximum depth, learning rate, etc.

[0010] Step 4: The learning rate of the decision boosting tree model described in Step 3 is 0.001-0.01; Step 5: The number of trees in the decision boosting tree model described in Step 3 is 80-100; Step 6: The maximum depth of the decision boosting tree model described in Step 3 is 10-20; Step 7: Split the ionic liquid data obtained in Step 1 into a training set and a test set with a ratio of 0.8:0.2. Step 8: The training and test sets from Step 7 are fed into the decision boosting tree model for learning. The trained model is then evaluated using the coefficient of determination R2, mean squared error MSE, mean absolute error MAE, and root mean square error RMSE as evaluation metrics. Step 9: Use the SHAP interpretability analysis method to interpret the decision boosting tree trained in Step 8 and quantify the contribution of each input feature to the prediction result.

[0011] Compared with the prior art, this patent has the following advantages: (1) This patent uses a gradient boosting decision tree model combined with MACCS bond features to accurately capture the relationship between molecular structure and toxicity, significantly improving prediction accuracy; (2) This patent uses the SHAP value analysis method to quantify the contribution of input features to the prediction results, improve the interpretability of the model, and provide guidance for drug structure optimization; (3) This patent optimizes hyperparameters through grid search, and the model can adapt to different chemical substances, avoid overfitting, and has strong generalization ability; (4) This patent can replace traditional animal experiments, reduce the experimental cycle and cost, and at the same time, through computer simulation and machine learning technology, it can screen toxic compounds in advance, reduce research and development time and cost, and meet the ethical requirements of modern drug research and development. Attached Figure Description

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

[0013] Figure 1 This is a diagram of the gradient boosting decision tree model structure of this patent; Detailed Implementation

[0014] It should be noted that if the embodiments of this patent involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0015] Furthermore, if the embodiments of this patent involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, if the word "and / or" appears throughout the text, it means including three parallel solutions; for example, "A and / or B" includes solution A, solution B, or a solution that simultaneously satisfies A and B. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of a person skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this patent.

[0016] refer to Figure 1 This patent embodiment provides a method for predicting acute peritoneal toxicity based on MACCS bonds and machine learning.

[0017] In one possible implementation, this patent collects acute intraperitoneal toxicity data for 12,480 chemical compounds. All data are converted to standard units (mol / kg) and represented in log10 format. Input features are 166-bit binary molecular descriptors calculated from the SMILES string using MACCS keys. Highly correlated features with a correlation coefficient greater than 0.95 are removed by calculating the Pearson correlation coefficient between features. Ultimately, 120 features are retained for subsequent modeling. During training, a grid search optimization algorithm is used to search for optimal hyperparameters, including 100 trees, a maximum depth of 15, and a learning rate of 0.01. Model parameters are optimized by minimizing the prediction residuals, ultimately obtaining a gradient boosting decision tree model with optimal generalization ability. The gradient boosting decision tree model structure used is as follows: Figure 1 As shown in the accompanying drawings, the technical solution of this patent has been described above with reference to the preferred embodiments. However, it will be readily understood by those skilled in the art that the scope of protection of this patent is obviously not limited to these specific embodiments. Without departing from the principles of this patent, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this patent.

Claims

1. A method for predicting acute intraperitoneal toxicity based on MACCS bonds and machine learning, characterized in that... The steps to implement this method include: Step 1: Collect acute intraperitoneal toxicity (LD50) data of chemical compounds, convert them to standard units (mol / kg), express them in log10 format, and create a SMILES string representation for each compound; Step 2: Remove features with missing values ​​or zero variance, calculate the initial descriptor of the molecule using MACCS bonds, and remove highly linearly correlated features based on the Pearson correlation coefficient between features; Step 3: Construct a decision boosting tree model and automatically optimize hyperparameters, including the number of trees, maximum depth, and learning rate, using a grid search algorithm; Step 4: Train and evaluate the decision boosting tree model using the training and test sets, and use R2, mean squared error (MSE), mean absolute error (MAE), and root mean square error (RMSE) as model performance metrics. Step 5: Use the SHAP value analysis method to quantify the contribution of each input feature to the prediction result.

2. The method for predicting acute peritoneal toxicity based on MACCS bonds and machine learning according to claim 1, characterized in that: The ionic liquid database includes a dataset containing more than 12,000-13,000 data points.

3. The method for predicting acute peritoneal toxicity based on MACCS bonds and machine learning according to claim 1, characterized in that: In step 2, after feature engineering, 100-130 effective features are ultimately retained as model input.

4. The method for predicting acute peritoneal toxicity based on MACCS bonds and machine learning according to claim 1, characterized in that: The optimization hyperparameters of the decision boosting tree model include a learning rate of 0.001-0.01, a number of trees of 80-100, and a maximum tree depth of 10-20.

5. The method for predicting acute peritoneal toxicity based on MACCS bonds and machine learning according to claim 1, characterized in that: The gradient boosting decision tree model employs hyperparameter grid search optimization to improve the model's generalization ability and avoid overfitting.

6. The method for predicting acute peritoneal toxicity based on MACCS bonds and machine learning according to claim 1, characterized in that: In step 5, the ratio of the training set to the test set is 0.8:0.2.

Images