Method and device for performing rat contamination experiment blood biochemical and blood routine index generative prediction on chemicals based on multi-modal fusion and GANs model

Abstract

The invention discloses a method for performing rat contamination experiment blood biochemical and blood routine index generative prediction on a chemical substance based on multi-modal fusion and a GANs model. The method comprises the following steps: S1, obtaining an SMILES formula of the chemical substance to be predicted; s2, acquiring data of three modes of the chemical substance, wherein the data comprises a molecular mechanism, biological activity and batch key molecular docking activity; s3, combining the contamination dose and contamination time of the three modal data, and inputting the combined data into a pre-trained alternative rat experiment prediction model; s4, an output result of the replacement rat experiment prediction model is generative prediction values of 25 indexes of blood biochemistry and blood routine under the set contamination dose and set contamination time of the data source chemicals. According to the method, the prediction precision and efficiency are greatly improved, the toxicity identification and management of novel chemicals can be effectively assisted, the time and energy of experimenters are effectively saved, animal resources are saved, and animal welfare is improved.

Description

Technical Field

[0001] The present invention relates to a method for generative prediction of blood biochemistry and blood routine indexes in rat poisoning experiments based on multimodal fusion and GANs models, and also relates to a corresponding device for generative prediction of blood biochemistry and blood routine indexes in rat poisoning experiments based on multimodal fusion and GANs models, belonging to the field of artificial intelligence-assisted prediction of chemical toxic effects, and belonging to multidisciplinary cross-research of computer science, toxicology, chemical informatics, and bioinformatics. Background Art

[0002] There is no doubt that synthetic chemicals have significantly improved food production and living standards, but the current situation is that the number of newly synthesized chemicals is much higher than the number of chemicals that have been included in the toxicological effect evaluation. According to the European Commission's study of public data, even if 2465 high-volume chemicals were examined more carefully, 21% of the compounds had no toxicological data and 65 compounds had insufficient data. At the same time, with the advancement of the ability to measure chemicals at low environmental concentrations, people's understanding of the range of effects of environmental pollutants on organisms has greatly increased, which means that the toxicity test research strategy should be adjusted to a certain extent. Chemical substances that may have adverse effects on human health and ecological safety without sufficient testing will inevitably enter the environment after production and use, and some of them behave like environmental organic pollutants.

[0003] The synthesis and diversification of man-made chemicals has increased dramatically over the past few decades, outpacing other well-known global challenges such as climate change. It is estimated that approximately 200-300 new chemicals are produced and marketed annually. Limited information based on laboratory toxicity data indicates that some emerging pollutants can cause harmful effects on exposed humans and mammals, such as cytotoxicity, developmental toxicity, hepatotoxicity, and endocrine disruption. Chemicals of particular concern include, but are not limited to, developmental neurotoxicants, endocrine disruptors, chemical herbicides, emerging pesticides, pharmaceutical wastes, and nanomaterials. These statistics emphasize the need for research to clarify the ecological impacts and health hazards of chemical pollution. Exposure to environmental organic pollutants has triggered significant ecological impacts and adverse health outcomes.

[0004] Based on artificial intelligence, the use of chemical databases, molecular descriptors, fingerprints and model algorithms are important factors in model development. With the development of information technology, multimodal methods have entered the field of scientific research. Multimodal deep learning has been proposed, and due to its high degree of nonlinearity, it has proven its advantages in representing multimodal data. If multimodal fusion provides a new breakthrough for computational toxicology prediction models, it can promote the development of intelligent systems in a more intelligent direction. Summary of the invention

[0005] The primary technical problem to be solved by the present invention is to provide a method for generative prediction of blood biochemistry and blood routine indicators in rat poisoning experiments based on multimodal fusion and GANs model.

[0006] Another technical problem to be solved by the present invention is to provide a device for generative prediction of blood biochemistry and blood routine indicators in rat poisoning experiments based on multimodal fusion and GANs model.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] According to a first aspect of an embodiment of the present invention, there is provided a generative prediction method for blood biochemistry and blood routine indexes in a rat poisoning experiment based on multimodal fusion and a GANs model, comprising the following steps:

[0009] Step S1, obtaining the SMILES formula of the chemical to be predicted;

[0010] Step S2, the device calls the internal file to calculate the data of three modes of the chemical, including molecular structure, biological activity and batch key molecule docking activity;

[0011] Step S3, combining the calculated three modal data with the poisoning dose and the poisoning time in a linear addition manner and inputting them into a pre-trained alternative rat experiment prediction model;

[0012] Step S4, the output result of the alternative rat experiment prediction model is the generative prediction value of 25 indicators of blood biochemistry and blood routine under the set poisoning dose and set poisoning time of the data source chemical;

[0013] Furthermore, the present invention can achieve detailed prediction of the developmental toxicity of chemicals through the following steps:

[0014] Step S5, further calling the self-written function to discretize the batch of key molecule docking data, and obtaining the gene corresponding to the variable with a value of 1;

[0015] Step S6: Screen out genes corresponding to proteins with higher docking scores for key molecule docking for annotation and functional analysis to achieve the purpose of predicting the affected pathways and output effect diagrams.

[0016] Preferably, the following processing is performed when calculating the required data:

[0017] Step S21, retrieve key genes related to human blood disease phenotypes through the HPO phenotype database, use Alpha Fold to predict a series of corresponding protein structure pdbqt files, remove ligands, add hydrogen, etc., and save them in the corresponding device;

[0018] Step S22, according to the SMILES formula of the chemical to be tested, using Python and other related libraries, calculate the 3D molecular structure of the chemical (chemical structure can be characterized by a set of numerical values, which are called molecular fingerprints or descriptors. They may characterize the properties of the molecule, such as log P, molecular weight, hydrogen bond donors, acceptors, rotatable bonds, etc., which can be linked to the experimental evidence of the molecule; for each level of molecular characterization, hundreds or thousands of structural features can be calculated. There are a variety of molecular descriptors and fingerprints that encode structural, topological, geometric, electrostatic, quantum chemical, thermodynamic, fragmentation features, etc.), and abstractly extract thousands of features, which is the first modal data;

[0019] Step S23, according to the SMILES formula of the chemical to be tested, using Python and other related libraries, calculate the biological activity data of the chemical (the biological activity data used in the present invention is calculated by the Chemical Checker tool. Chemical Checker extends the principle of small molecule similarity to all levels of biology. CC divides the data into five levels, from the chemical properties of the compound to the clinical results, and the complexity increases. The biological activity data is expressed in a universal vector format, and detailed information of the vectors of 25 CC spaces is obtained), and thousands of features are extracted, which is the second modal data;

[0020] Step S24: According to the SMILES formula of the chemical to be tested, Python and other related libraries are called to calculate the 3D molecular structure of the chemical, and the ligand removal and hydrogenation treatment are performed to generate a pbdqt file. The autodock vina package is called to perform batch molecular docking to generate a docking active sequence with a total of about 1k features, which is the third modal data;

[0021] Step S25, combining the three modes according to the requirements of the established algorithm model to merge the data of the poisoning dose and the poisoning time in a linear addition manner.

[0022] Preferably, when calculating the first modal data:

[0023] Step 221, the modal feature is saved as a 50×50 matrix, all values ​​are filled toward the center, and the edges are filled with random noise.

[0024] Preferably, when calculating the second modal data:

[0025] Step 231: The modal features are saved as a 25×128 matrix.

[0026] Preferably, when calculating the third modal data:

[0027] Step 241, the present invention pre-trains an additional LDA topic model based on text data mining of gene cards database for gene classification;

[0028] Step 242: According to the gene classification model of step 241, the data of the third mode is split into several sequences for storage.

[0029] Preferably, we process the calculated three modal data as follows:

[0030] The original feature data calculated in this study were processed with outliers, recoded, standardized, and resampled in turn.

[0031] Preferably, in S3, the three modal data are combined with the poisoning dose and the poisoning time, specifically including: the data is further standardized to between 0 and 1 to prevent errors caused by dimensional differences between different variables; the three modal data are trained by a separate network and further transmitted to the subsequent network; the data of the poisoning dose and the poisoning time are directly added to the subsequent network through full connection linearly.

[0032] Preferably, the alternative rat experiment prediction model is obtained through the following steps:

[0033] Step S31, obtain multiple groups of chemical SMILES numbers and 25 blood biochemistry and blood routine indicators, and calculate three modal data for each group of data according to the above method, and integrate these data into corresponding formats for modeling.

[0034] Step S32, obtain multiple groups of rat experimental data to train the pre-designed adversarial generative neural network (GANs) model structure, further perform parameter optimization, and obtain the optimal alternative rat experimental prediction model.

[0035] The optimal model for surrogate rat experiment prediction adopts a multimodal fusion structure and uses a Dropout layer for regularization. The model complexity is reduced by adjusting the number of nodes in the middle layer. The model uses an adversarial generative neural network, including a generator and a discriminator, which are trained together in an adversarial manner to achieve the effect of generating realistic data:

[0036] After the first modality data is passed in, the structure of the convolutional neural network (CNN) is used;

[0037] After the second modality data is passed in, the structure of the convolutional neural network (CNN) is used;

[0038] After the third modality data is passed in, the structure of the attention mechanism (Transformer) is used;

[0039] The above three modes are combined with the poisoning dose and the poisoning time and passed into the neural network. The model adopts an adversarial generative neural network, including a generator and a discriminator, which are trained together in an adversarial way to achieve the effect of generating realistic data.

[0040] Preferably, after making a prediction, the alternative rat experiment prediction model automatically calculates the distribution composition of each generated prediction indicator, taking into account laboratory errors, when the sum of the proportions of white blood cells (i.e., neutrophils, lymphocytes, monocytes, eosinophils, basophils) exceeds 105%, the data generated by the generator is identified as low-quality data.

[0041] Preferably, the alternative rat experiment prediction model has the function of subsequent processing and prediction of the affected pathways after making the prediction:

[0042] Step S51: After the indicator prediction is completed, the data of the third mode is automatically discretized according to a set threshold value, and the corresponding gene sequence is returned.

[0043] Step S61: Perform enrichment analysis based on the gene sequence returned by S51 to predict the pathways that may be affected.

[0044] According to a second aspect of an embodiment of the present invention, there is provided a device for generative prediction of blood biochemistry and blood routine indicators in a rat poisoning experiment based on multimodal fusion and a GANs model, comprising a processor and a memory, wherein the processor reads a computer program in the memory, and the result is displayed on a display, and is used to perform the following operations:

[0045] The data to be collected are obtained, and the data to be collected only include the SMILES formula, the poisoning dose, and the poisoning time of the chemical to be predicted.

[0046] Inputting the SMILES into the receiving interface of the whole device;

[0047] The output results of the alternative rat experiment prediction model are 25 blood biochemical and blood routine indicators, and the enrichment analysis results predict the affected pathways displayed on the display.

[0048] Beneficial effects of the present invention

[0049] The method and device for generative prediction of blood biochemistry and routine blood tests in rats exposed to chemicals based on multimodal fusion and GANs models provided by the present invention provide abundant relevant features for predicting blood biochemistry and routine blood tests in rats exposed to chemicals only through the SMILES formula of the chemical and the preset exposure dose and exposure time, thereby greatly improving the prediction accuracy and efficiency. On the one hand, it can effectively assist in the toxicity identification and management of new chemicals, and on the other hand, it can effectively save the time and energy of experimenters, and at the same time, it can also save animal resources and improve animal welfare. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic flow chart of a generative prediction method for blood biochemistry and blood routine indexes in a rat exposure experiment to chemicals based on multimodal fusion and GANs model provided in an embodiment of the present invention.

[0051] Figure 2 Technical route and data flow for in vitro biological activity calculation provided by embodiments of the present invention

[0052] Figure 3 Technical route and data flow for molecular descriptor calculation provided by embodiments of the present invention

[0053] Figure 4 Technical route and data flow for batch molecular docking calculation provided by embodiments of the present invention

[0054] Figure 5 Module design and block function distribution provided for the embodiment of the present invention

[0055] Figure 6 t-SNE dimension reduction clustering visualization diagram of built-in training data provided by the embodiment of the present invention

[0056] Figure 7 Training curve and loss function curve of the built-in model provided in the embodiment of the present invention

[0057] Figure 8 The results of the external validation experiment of the built-in model provided by the embodiment of the present invention

[0058] Fig. 9 Schematic diagram of the structure of the generative prediction device for blood biochemistry and blood routine indexes of rat poisoning experiments based on multimodal fusion and GANs model provided in the embodiment of the present invention DETAILED DESCRIPTION

[0059] The present invention will be further described below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto:

[0060] For emerging chemicals, in order to provide toxicity prediction values ​​for identification and management as early as possible, such as Figure 1As shown, the generative prediction method for blood biochemistry and blood routine indexes of rats exposed to chemicals based on multimodal fusion and GANs model provided in an embodiment of the present invention includes the following steps:

[0061] Step S1, obtaining the SMILES formula of the chemical to be predicted;

[0062] Step S2: The device calls the internal file to calculate the data of the three modes of the chemical, including molecular structure, biological activity and batch key molecular docking activity (such as Figure 5 - Module 2a);

[0063] When calculating the three modal data required by the model, after the chemical SMILES is imported, the following steps are included:

[0064] Step S21, retrieve key genes related to human blood disease phenotypes through the HPO phenotype database, use Alpha Fold to predict a series of corresponding protein structure pdbqt files, remove ligands, add hydrogen, etc., and save them in the corresponding device;

[0065] Step S22: According to the SMILES formula of the chemical to be tested, use Python and other related libraries to calculate the 3D molecular structure of the chemical, and abstractly extract thousands of features (preferably, refer to https: / / www.rdkit.org / docs / index.html for implementation). This is the first modal data (method as shown in Figure 2 shown);

[0066] Step 221, the modal feature is saved as a 50×50 matrix, all values ​​are filled toward the center, and the edges are filled with random noise.

[0067] Step S23: According to the SMILES formula of the chemical to be tested, use Python and other related libraries to calculate the biological activity data of the chemical and extract thousands of features, which is the second modal data (method as shown in Figure 3 As shown, the method and technology are derived from the literature Nat Biotechnol.2020;38(9):1087-1096);

[0068] Step 231: The modal features are saved as a 25×128 matrix.

[0069] Step 231, the present invention pre-trains an additional LDA topic model based on text data mining of gene cards database for gene classification;

[0070] Step S24: According to the SMILES formula of the chemical to be tested, call Python and other related libraries to calculate the 3D molecular structure of the chemical, and perform ligand removal and hydrogenation treatment to generate a pbdqt file. Call the autodock vina package to perform batch molecular docking and generate a docking active sequence with a total of about 1k features. This is the third modal data (method as shown in Figure 4 shown);

[0071] Step 241, the present invention pre-trains an LDA topic model based on text data mining of gene cards database for gene classification (wherein the LDA topic model works based on Bayesian statistical theory by constructing a three-layer structure (document-topic-vocabulary) generation model; its principle includes two parts: document generation process (first, a series of topics are randomly selected for each document; then, for each word in the document, a topic is randomly selected according to the topic distribution of the current position; finally, a word is randomly generated according to the vocabulary distribution under the selected topic) and model training (LDA model maximizes the probability of generating documents by adjusting the parameters of topic-vocabulary distribution and document-topic distribution. This is usually implemented using algorithms such as Gibbs Sampling or Variational Inference));

[0072] Step 242: According to the gene classification model of step 241, the data of the third mode is split into several sequences for storage.

[0073] Using the method of obtaining the three modal data of chemicals calculated in step S2, multiple groups of chemical SMILES numbers and 25 blood biochemistry and blood routine indicators are obtained, and each group of data is calculated according to the above method to obtain the three modal data. And these data are integrated into the corresponding format for modeling.

[0074] Step S3, combining the calculated three modal data with the poisoning dose and the poisoning time in a linear addition manner and inputting them into a pre-trained alternative rat experiment prediction model;

[0075] like Figure 5- As shown in module 2b, the pre-designed alternative rat experiment prediction model adopts a multimodal fusion method and uses a Dropout layer for regularization. The model complexity is reduced by adjusting the number of nodes in the middle layer. The last Dense layer uses a sigmoid activation function to output a probability for a binary classification problem, where: after the first modality data is passed in, the convolutional neural network (CNN) structure is used; after the second modality data is passed in, the convolutional neural network (CNN) structure is used; after the third modality data is passed in, the attention mechanism (Transformer) structure is used; the training data of the entire model is visualized after t-SNE dimension reduction as shown below: Figure 6 As shown, OPEN TG GATES database is the training set and DrugMatrixdatabase is the external validation set; Figure 7 The accuracy and loss function of the entire project during training are shown as they change with the training batches, revealing the good convergence of this study.

[0076] Step S4: Figure 5 - As shown in module 2c, the result of the alternative rat experiment prediction model is the generative prediction value of 25 blood biochemistry and blood routine indicators under the set toxic dose and set toxic time of the data source chemical; Figure 8 It shows the prediction consistency of the entire model results. The orange bar on the left is the consistency score between the training set and the validation set, and the blue bar on the right is the consistency score between the validation set and the predicted value, indicating that our prediction results are highly consistent with the actual values, and the model can give relatively accurate predictions.

[0077] Step S5, further calling the self-written function to discretize the batch of key molecule docking data, and obtaining the gene corresponding to the variable with a value of 1;

[0078] Step S6: Screen out genes corresponding to proteins with higher docking scores for key molecule docking for annotation and functional analysis to achieve the purpose of predicting the affected pathways and output effect diagrams.

[0079] The collected chemical SMILES formula obtained in step S1 is combined with the set exposure time and exposure dose and input into the receiving interface of the entire device, and the model will output the results of 25 blood biochemistry and blood routine indicators predicted by the generative formula. After making the prediction, the alternative rat experiment prediction model automatically calculates the distribution composition of each generated prediction indicator. Taking into account laboratory errors, when the sum of the proportions of white blood cells (i.e., neutrophils, lymphocytes, monocytes, eosinophils, and basophils) exceeds 105%, the data generated by the generator is identified as low-quality data.

[0080] In order to realize the method for generative prediction of blood biochemistry and blood routine indexes in rat poisoning experiments based on multimodal fusion and GANs model provided by the present invention, the present invention also provides a device for generative prediction of blood biochemistry and blood routine indexes in rat poisoning experiments based on multimodal fusion and GANs model. Fig. 9 As shown, the device for generative prediction of blood biochemistry and blood routine indicators in chemical poisoning experiments in rats based on multimodal fusion and GANs model includes a memory 31, a processor 32 and a display 33. The processor 32 reads the computer program in the memory 31 to enable the prediction device to execute the prediction model, and the results are displayed on the display 33.

[0081] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A method for generative prediction of blood biochemistry and blood routine indicators in rat poisoning experiments based on multimodal fusion and GANs model, characterized in that The steps include: Step S1, obtaining the SMILES formula of the chemical to be predicted; Step S2, obtaining data of three modes of the chemical, including molecular structure, biological activity and batch key molecule docking activity; Step S3, combining the three modal data with the poisoning dose and poisoning time, and inputting them into the pre-trained alternative rat experiment prediction model; Step S4, the output result of the alternative rat experiment prediction model is the generative prediction value of 25 indicators of blood biochemistry and blood routine under the set poisoning dose and set poisoning time of the data source chemical.

2. The method according to claim 1, characterized in that Obtaining data of the three modes of the test compound includes the following steps: Step S21, retrieve genes related to human blood disease phenotypes through the HPO phenotype database, use AlphaFold to predict a series of corresponding protein structure pdbqt files, remove ligands and hydrogenate them, and save them for subsequent calls; Step S22, according to the SMILES formula of the chemical to be tested, the 3D molecular structure of the chemical is calculated, and the features are abstractly extracted, which is the first modal data; Step S23, according to the SMILES formula of the chemical to be tested, the biological activity data of the chemical is calculated and the features are extracted, which is the second modal data; Step S24, according to the SMILES formula of the chemical to be tested, the 3D molecular structure of the chemical is calculated, and the ligand removal and hydrogenation treatment is performed to generate a pbdqt file; The pdbqt file of the protein macromolecule in step S21 is called together to perform batch molecular docking to generate a docking active sequence, which is the third modal data.

3. The method according to claim 2, characterized in that: When obtaining the first modal data: Step 221, the modal feature is saved as a 50×50 matrix, all values ​​are filled toward the center, and the edges are filled with random noise; When obtaining the second modal data: Step 231, the modal feature is saved as a 25×128 matrix; When obtaining the third modal data: Step 241, performing gene classification based on the LDA topic model mined from the gene cards database text data; the LDA topic model is obtained by the following steps: Ⅰ Randomly select a series of topics for each gene description document; Ⅱ For each word in the document, randomly select a topic based on the topic distribution of the current position; Ⅲ Randomly generate a word according to the word distribution under the selected topic; Ⅳ Use Gibbs sampling or variational inference algorithm to maximize the probability of generating documents by adjusting the parameters of topic-word distribution and document-topic distribution, thereby obtaining the LDA model; Step 242: According to the LDA topic model of step 241, the data of the third mode is split into several sequences for storage.

4. The method according to claim 2, characterized in that: In steps S22, S23 and S24, the original feature data is processed for outliers, re-encoded, standardized and re-sampled in turn.

5. The method according to claim 2, characterized in that In step S3, the three modal data are combined into the poisoning dose and the poisoning time, specifically including: the data is further standardized to between 0 and 1 to prevent errors caused by dimensional differences between different variables; the three modal data are trained by a separate network and further transmitted to the downstream network training layer; the data of the poisoning dose and the poisoning time are directly added to the subsequent network through full connection linearly.

6. The method according to claim 1, characterized in that: The alternative rat experiment prediction model is obtained through the following steps: Step S31, obtaining multiple groups of chemical SMILES formulas and 25 blood biochemistry and blood routine indicators, and each group of data obtains data of three modes according to step S2; Step S32, divide the multiple groups of data obtained in S31 into training sets and test sets, use the data of three modes as input and the corresponding 25 blood biochemistry and blood routine indicators as output to train the pre-designed adversarial generative neural network (GANs) model structure, further perform parameter optimization, and obtain the optimal alternative rat experiment prediction model.

7. The method according to claim 1 or 6, characterized in that: The pre-designed adversarial generative neural network (GANs) model structure of the alternative rat experiment prediction model adopts a multi-modal fusion structure and uses a Dropout layer for regularization. The model complexity is reduced by adjusting the number of nodes in the middle layer. The model adopts an adversarial generative neural network, including a generator and a discriminator, which are trained together in an adversarial manner to achieve the effect of generating realistic data: After the first modality data is passed in, the structure of the convolutional neural network CNN is used; After the second modality data is passed in, the structure of the convolutional neural network CNN is used; After the third modality data is passed in, the structure of the attention mechanism Transformer is used; The three modal data are combined with the poisoning dose and poisoning time and transmitted to the neural network.

8. The method according to claim 1 or 6, characterized in that: After making a prediction, the alternative rat experiment prediction model automatically calculates the distribution composition of each generated prediction indicator. When the sum of the white blood cell proportions exceeds 105%, the data generated by the generator is identified as low-quality data and discarded.

9. The method according to claim 1, characterized in that It also includes: Step S5, discretizing the key molecular docking data of the chemical batch to obtain the gene sequence corresponding to the variable with a value of 1; Step S6: According to the gene sequence returned by S5, GO or KEGG annotations are performed on the obtained related genes, and enrichment analysis is performed. The more genes enriched in a certain pathway, the more susceptible this pathway is, and the pathways that may be affected are subdivided and predicted.

10. A generative prediction device for blood biochemistry and blood routine indexes in rat poisoning experiments based on multimodal fusion and GANs model, characterized in that The device comprises a processor, a memory and a display, wherein the processor reads a computer program in the memory so that the prediction device executes the prediction model according to any one of claims 1 to 9, and the result is displayed on the display.

Images