A method and system for predicting the toxicity of ionic liquids to acetylcholinesterase

Abstract

The present invention relates to the field of biochemical material image recognition, and discloses a method for predicting the toxicity of ionic liquids to acetylcholinesterase, comprising the following steps: obtaining the basic information of ionic liquids; generating deep learning to-be-predicted images of the toxicity of ionic liquids to acetylcholinesterase; using a VGG19_BN convolutional neural network for deep learning training to obtain an end-to-end deep learning model of the known toxicity of ionic liquids to acetylcholinesterase; processing new ionic liquid samples and new acetylcholinesterase samples to obtain the spliced to-be-predicted images of the toxicity of unknown ionic liquids to acetylcholinesterase as a test set, and then inputting the to-be-predicted images of the test set into the trained VGG19_BN convolutional neural network for deep learning to output the predicted EC50 experimental value, thereby completing the toxicity prediction of ionic liquids to acetylcholinesterase.

Description

Technical Field

[0001] The present invention relates to the field of biochemical material image recognition, and particularly to a method and system for predicting the toxicity of ionic liquids to acetylcholinesterase. Background Art

[0002] Ionic liquids (ILs) are organic salts mainly composed of organic cations and organic / inorganic anions, and are called ionic liquids because they are liquid at room temperature or near room temperature. A large number of studies have shown that ILs can cause toxicity to various organisms, such as algae, bacteria, fish, plants, mammals, etc. Acetylcholinesterase is a key enzyme in biological nerve conduction, which can catalyze the decomposition of acetylcholine and other choline esters that act as neurotransmitters, thereby preventing the excitatory effect of neurotransmitters on the postsynaptic membrane and ensuring the normal transmission of nerve signals in organisms. Domestic and foreign studies have found that certain ILs can affect the activity of acetylcholinesterase (AChE). If AChE is inhibited, it will lead to biomedical problems such as agitation, miosis, and even severe neuromuscular diseases. Therefore, it is necessary to clarify the toxicity risk of different types of ILs before their large-scale application.

[0003] In the prior art, the inhibitory effect and mechanism of ionic liquids on acetylcholinesterase are obtained through biochemical experiments. However, traditional toxicity experiments are expensive, time-consuming and laborious, and highly dependent on the technical level of experimental personnel. On the other hand, traditional quantitative structure-activity relationship studies predict specific effects through machine learning methods. However, these machine learning methods only focus on the properties of small molecule compounds themselves, do not consider the overall interaction mode between ILs and biological systems (such as proteins), and the method of descriptor calculation is prone to the problem of unclear structural expression of molecules, making it difficult to be transplanted to the prediction of high molecular compounds and even more unable to be used to explain the effect of ionic liquids on acetylcholinesterase. For this reason, we propose a method and system for predicting the toxicity of ionic liquids to acetylcholinesterase. Summary of the Invention

[0004] (1) Technical Problems to be Solved

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method and system for predicting the toxicity of ionic liquids to acetylcholinesterase, and solves the above problems.

[0006] (2) Technical Solutions

[0007] To achieve the above object, the present invention provides the following technical solutions: A method for predicting the toxicity of ionic liquids to acetylcholinesterase, comprising the following steps:

[0008] The first step: Obtain the basic information of ionic liquids;

[0009] Step 2: Generate deep learning images to be predicted for the toxicity of ionic liquids to acetylcholinesterase;

[0010] Step 3: Use the VGG19_BN convolutional neural network for deep learning training to obtain an end-to-end deep learning model for the known toxicity of ionic liquids to acetylcholinesterase;

[0011] Step 4: Process new ionic liquid samples and new acetylcholinesterase samples to obtain the spliced images to be predicted for the toxicity of unknown ionic liquids to acetylcholinesterase as the test set. Then, input the images to be predicted in the test set into the trained VGG19_BN convolutional neural network for deep learning, and output the predicted EC50 experimental values, thereby completing the toxicity prediction of ionic liquids to acetylcholinesterase.

[0012] Preferably, the specific steps of the second step include the following:

[0013] S1: Standardize the SMILES of the collected ionic liquids to generate corresponding two-dimensional structure images;

[0014] S2: Obtain the corresponding cavity three-dimensional structure by combining acetylcholinesterase with ionic liquids, and slice the area with the most obvious cavity three-dimensional structure features to obtain two-dimensional color slice images of the cavity;

[0015] S3: Splice the two-dimensional structure color images of the cation and anion of the ionic liquid and the two-dimensional color slice images of the cavity to form training images.

[0016] Preferably, different atoms in the two-dimensional structure images are colored differently. Specifically, nitrogen elements are colored blue, oxygen elements are colored red, sulfur elements are colored yellow, carbon elements are colored black, fluorine elements are colored blue, phosphorus elements are colored orange, hydrogen elements are not colored and not displayed, and the chemical bonds between different atoms are divided into two and have the same color as the connected atoms.

[0017] Preferably, the specific steps of the third step include the following:

[0018] S1: For the data set composed of all training images and the corresponding EC50 experimental values of drug safety indicators, group the data set according to the ratio of 80% for the training set and 20% for the validation set. Use the images to be predicted as the input and the corresponding EC50 experimental values of drug safety indicators as the predicted output;

[0019] S2: Construct a VGG19_BN convolutional neural network;

[0020] S3: Use the training set for deep learning training on the VGG19_BN convolutional neural network, and use the consistent known EC50 experimental values to train the VGG19_BN convolutional neural network until the VGG19_BN convolutional neural network converges;

[0021] S4: Use the validation set on the VGG19_BN convolutional neural network trained to convergence in step S32, learn the training image features, and output the predicted EC50 experimental values for verification.

[0022] Preferably, the specific steps of S4 are as follows:

[0023] S1: Generate two-dimensional structural color images of the cations and anions of the new ionic liquid sample respectively;

[0024] S2: Combine the new acetylcholinesterase sample with the ionic liquid to obtain the corresponding three-dimensional cavity structure, and slice the most obvious part of the three-dimensional cavity structure features to obtain a two-dimensional color slice image of the cavity;

[0025] S3: Stitch the two-dimensional structural color images of the new ionic liquid cations and anions respectively and the two-dimensional color slice image of the cavity to form a training image;

[0026] S4: Then input the training image into the trained and verified VGG19_BN convolutional neural network for prediction, and output the predicted value of the toxicity of the corresponding new ionic liquid to the new acetylcholinesterase EC50 experiment.

[0027] Preferably, when the VGG19_BN convolutional neural network conducts toxicity prediction training, an external verification method is adopted. When the coefficient of determination R 2 > 74% and the root mean square error RMSE < 0.210, it indicates that the training of the VGG19_BN convolutional neural network is completed.

[0028] A toxicity prediction system for an ionic liquid to acetylcholinesterase, comprising a basic information module, an image generation module, and a convolutional neural network. The basic information module and the convolutional neural network are both connected to the image generation module;

[0029] The basic information module is used to obtain the ASCII string SMILES of the ionic liquid, obtain the known EC50 experimental values of the drug safety index of the ionic liquid to acetylcholinesterase, and obtain the information of the molecular structure of acetylcholinesterase;

[0030] The image generation module is used to, according to the information in the basic information module, perform atomic differential coloring on the ionic liquid sample of the toxicity of the unknown ionic liquid to acetylcholinesterase and perform three-dimensional cavity structure slicing processing on the acetylcholinesterase sample to obtain an image to be predicted;

[0031] The convolutional neural network is used to perform deep learning prediction on the toxicity of unknown ionic liquids to acetylcholinesterase based on the to-be-predicted images generated by the image generation module, and obtain the predicted values of the corresponding toxicity;

[0032] The convolutional neural network is the VGG19_BN convolutional neural network trained for predicting the toxicity of ionic liquids to acetylcholinesterase.

[0033] Preferably, the image generation module includes an ionic liquid image module for generating a two-dimensional structure from an ionic liquid sample, then differentiating and coloring the atoms of the two-dimensional structure to obtain the two-dimensional structure color images of the cations and anions in the ionic liquid respectively;

[0034] An image cavity image module for sequentially performing dehydration, deleting redundant ligands, and hydrogenating and modifying the MMFF94x force field on acetylcholinesterase, then converting the ionic liquid into a three-dimensional object, docking the ionic liquid with acetylcholinesterase, generating a cavity three-dimensional structure with a distance greater than the average distance from the center of the ionic liquid to the surrounding residues of acetylcholinesterase, and then performing slicing processing on the cavity three-dimensional structure to obtain the two-dimensional color slice images of the cavity;

[0035] An image splicing module for splicing the two-dimensional structure color images of the cations and anions in the ionic liquid and the two-dimensional color slice images of the cavity to form a to-be-predicted image of a 9-channel RGB tensor;

[0036] The ionic liquid image module and the cavity image module are respectively connected to the basic information module, the image splicing module is connected to the ionic liquid image module and the cavity image module, and the image splicing module is connected to the convolutional neural network.

[0037] (III) Beneficial effects

[0038] Compared with the prior art, the present invention provides a method and system for predicting the toxicity of ionic liquids to acetylcholinesterase, and has the following beneficial effects:

[0039] 1. The method for predicting the toxicity of ionic liquids to acetylcholinesterase uses image feature recognition of deep learning to achieve the prediction of the toxicity of ionic liquids to acetylcholinesterase. Compared with the existing method of obtaining through biochemical experiments, this embodiment avoids using expensive experimental instruments, greatly reduces the dependence on the technical level of testers, has higher automation level and prediction efficiency, and is more economical, efficient, and targeted. The structure, atomic information, the position information and spatial information of the binding to the receptor protein in acetylcholinesterase of the ionic liquid are included in the image, and the distinctiveness of the features is increased by recoloring, which is convenient for recognition by deep learning methods, avoids the bottleneck that traditional machine learning methods are difficult to apply to high-molecular compounds, clearly expresses the interaction between the ionic liquid and acetylcholinesterase, and opens up a new way to study the biological toxicity of ionic liquids.

[0040] 2. The system for predicting the toxicity of ionic liquids to acetylcholinesterase avoids using expensive experimental instruments, greatly reduces the dependence on the technical level of testers, and is more economical, efficient, and targeted. The structure, atomic information, the position information and spatial information of the binding to the receptor protein in acetylcholinesterase of the ionic liquid are included in the image, and the distinctiveness of the features is increased by recoloring, which is convenient for recognition by deep learning methods and avoids the bottleneck that traditional machine learning methods are difficult to apply to high-molecular compounds. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flowchart of a method for predicting the toxicity of an ionic liquid to acetylcholinesterase according to the present invention;

[0042] Figure 2 is Figure 1 a schematic structural diagram of the process experienced by the image stitching and input into a convolutional neural network for learning to obtain a prediction result;

[0043] Figure 3 is a structural block diagram of a system for predicting the toxicity of an ionic liquid to acetylcholinesterase according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] As Figure 1 and Figure 2 shown, a method for predicting the toxicity of an ionic liquid to acetylcholinesterase specifically includes the following steps:

[0046] S1. Obtain the basic information of ionic liquids; the specific process is as follows:

[0047] Collect the molecular structure ASCII strings SMILES of various types of ionic liquids (ILs) such as quaternary amines, piperidines, pyrrolidines, imidazoles, imidazoles, pyridines, morpholines, quaternary phosphines, and quinolines, the molecular structure information of acetylcholinesterase (AChE), and the experimental value data of the known drug safety index EC50 toxicity of these different types of ionic liquids when binding to acetylcholinesterase. Logarithmically process the experimental values, and then use the SMILES and the logarithmically processed experimental values as the basic information; in this embodiment, 95 ionic liquid samples with known basic information are preferably used. Each ionic liquid is collected by distinguishing organic cations and anions. To ensure reliability, all experimental values of the drug safety index EC50 toxicity are obtained by performing toxicity tests under the same conditions;

[0048] S2. Generate a deep learning image to be predicted for the toxicity of ionic liquids to acetylcholinesterase; the specific process is as follows:

[0049] S21. Standardize the SMILES of the ionic liquids collected in step S1 to generate corresponding two-dimensional structure images, and distinguish the coloring of different atoms in the two-dimensional structure images. Specifically, nitrogen elements are colored blue, oxygen elements are colored red, sulfur elements are colored yellow, carbon elements are colored black, fluorine elements are colored blue, phosphorus elements are colored orange, hydrogen elements are not colored and not displayed, and the chemical bonds between different atoms are divided into two and have the same color as the connected atoms. Convert the two-dimensional structure image into a color image in the RGB color gamut, and set the background to white; the two-dimensional structure color image contains the atomic information, molecular structure, and spatial position information of the ionic liquid.

[0050] In this embodiment, it is preferably to generate sdf files for the SMILES of the cations and anions of the ionic liquid respectively, then import them into the RDkit software to generate standardized SMILES, and then generate corresponding two-dimensional structure images in the RDkit software according to the standardized SMILES and then distinguish the coloring of atoms; in this embodiment, it is preferably to use image processing software to unify the position and direction of the two-dimensional structure in the image to generate a two-dimensional structure color image with a size of 224*224 and in the RGB color gamut.

[0051] S22. Obtain the corresponding cavity three-dimensional structure of acetylcholinesterase binding to the ionic liquid, and slice the most obvious part of the cavity three-dimensional structure characteristics to obtain a two-dimensional color slice image of the cavity;

[0052] In this embodiment, it is preferable to obtain the PDB file of acetylcholinesterase (4BDT) protease from the RSCB protein database, input the PDB file into the operating environment MOE of the molecular simulation software, and perform dehydration, deletion of redundant ligands, and hydrogenation to modify the MMFF94x force field on acetylcholinesterase in sequence. Then, convert the SMILES of the ionic liquid into a three-dimensional object, dock the ionic liquid with acetylcholinesterase using huprine W. ligand as the binding site (corresponding to the receptor protein of acetylcholinesterase), and save the pdb file of ILs-AChE as the output result of the molecular simulation software with the optimal pose; import the pdb file into the PyMOL software, generate a three-dimensional cavity structure with a size greater than the average distance from the center of the ionic liquid to the surrounding residues of acetylcholinesterase, set the cavity color to light blue and the transparency to 80%, and then select the position with the most obvious structural features in the three-dimensional cavity structure for slicing to obtain a two-dimensional color slice image of the cavity with a size of 224*224 and an RGB color gamut;

[0053] S23. Stitch the two-dimensional color images of the cation and anion of the ionic liquid and the two-dimensional color slice image of the cavity to form a training image;

[0054] In this embodiment, it is preferable to stitch the two-dimensional color image of the cation, the two-dimensional color image of the anion, and the two-dimensional color slice image of the cavity, these three images, into a training image of a 9-channel RGB tensor through the torch.Cat interface in the Pytorch machine learning library;

[0055] S3. Use the VGG19_BN convolutional neural network for deep learning training to obtain an end-to-end deep learning model for the toxicity of known ionic liquids to acetylcholinesterase; the specific process is as follows:

[0056] S31. For the dataset composed of all training images and the corresponding experimental values of the drug safety index EC50, group the dataset according to the ratio of 80% for the training set and 20% for the validation set, use the image to be predicted as the input, and the corresponding experimental value of the drug safety index EC50 as the predicted output;

[0057] S32. Construct a VGG19_BN convolutional neural network; in this example, preferably, the structure and connection relationship of the VGG19_BN convolutional neural network are specifically as follows: input layer - convolutional (9, 64) layer - BN layer - convolutional (64, 64) layer - BN layer - pooling layer - convolutional (64, 128) layer - BN layer - convolutional (128, 128) layer - BN layer - pooling layer - convolutional (128, 256) layer - BN layer - convolutional (256, 256) layer - BN layer - convolutional (256, 256) layer - BN layer - convolutional (256, 256) layer - BN layer - pooling layer - convolutional (256, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - pooling layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - pooling layer - fully connected layer - fully connected layer - fully connected layer - output layer, where the numbers in the parentheses are the specific setting methods and numbers of channels; the convolutional kernels in the sixteen convolutional layers are uniformly 3*3 and the strides are uniformly 1*1; the convolutional kernels of the pooling layer are uniformly 2*2 and the stride sizes are uniformly 2*2; to prevent overfitting, a Dropout term is added, with the Dropout rate set to 0.5, the learning rate Learning rate set to 0.01, the activation function to ReLU, the loss function to mean square root deviation MSE, the batch processing parameter batch size to 16, and the number of training epochs epochs to 1000;

[0058] S33. Use the training set for deep learning training on the VGG19_BN convolutional neural network, and use the consistent known EC50 experimental values to train the VGG19_BN convolutional neural network until the VGG19_BN convolutional neural network converges;

[0059] S34. Use the validation set on the VGG19_BN convolutional neural network trained to convergence in step S32 to learn the training image features and output the predicted EC50 experimental values for inspection; the validation set is inspected by an external validation method, and is scored by the coefficient of determination R 2 . Root mean square error RMSE;

[0060] In this example, preferably, when R 2 > 74% and RMSE < 0.210, the training of the VGG19_BN convolutional neural network is completed;

[0061] S4. Process the new ionic liquid sample and the new acetylcholinesterase sample to obtain the spliced image to be predicted of the toxicity of the unknown ionic liquid to acetylcholinesterase as the test set; then input the image to be predicted in the test set into the VGG19_BN convolutional neural network trained in step S3 for deep learning, and output the predicted EC50 experimental value, thereby completing the toxicity prediction of the ionic liquid to acetylcholinesterase. The specific process is as follows:

[0062] S41. Perform the same operations on the new ionic liquid sample as in step S21 to obtain the two-dimensional structural color images of the cation and anion respectively;

[0063] S42. Perform the same operations on the new acetylcholinesterase sample as in step S22 to obtain the two-dimensional color slice images of the three-dimensional cavity structure;

[0064] S43. Perform the same operations on the images obtained in steps S41 and S42 as in step S23 to obtain the image to be predicted, and then input the image to be predicted into the VGG19_BN convolutional neural network that has completed training and verification in step S33 for prediction, and output the predicted value of the toxicity of the corresponding new ionic liquid to the new acetylcholinesterase EC50 experiment.

[0065] As Figure 3 shown, a toxicity prediction system for an ionic liquid to acetylcholinesterase includes a basic information module, an image generation module, and a convolutional neural network connected in sequence.

[0066] The basic information module is connected to an external database, and is used to obtain the ASCII string SMILES of the ionic liquid, obtain the experimental value of the drug safety index EC50 of the known ionic liquid to acetylcholinesterase, and obtain the molecular structure information of acetylcholinesterase. When obtaining the ASCII string SMILES of the ionic liquid, the anion and cation are collected separately.

[0067] The image generation module includes an ionic liquid image module, a cavity image module, and an image splicing module. The ionic liquid image module and the cavity image module are respectively connected to the basic information module. The image splicing module is connected to the ionic liquid image module and the cavity image module, and the image splicing module is connected to the convolutional neural network. The image generation module is used to perform atomic differential coloring on the ionic liquid sample of the toxicity of the unknown ionic liquid to acetylcholinesterase and perform cavity three-dimensional structure slicing processing on the acetylcholinesterase sample according to the information in the basic information module to obtain the image to be predicted.

[0068] The ionic liquid image module is used to generate a two-dimensional structure based on the SMILES of an ionic liquid sample, and then differentially color the atoms of the two-dimensional structure to obtain the two-dimensional color images of the cations and anions in the ionic liquid respectively. The differential coloring of atoms specifically means that nitrogen atoms are colored blue, oxygen atoms are colored red, sulfur atoms are colored yellow, carbon atoms are colored black, fluorine atoms are colored blue, phosphorus atoms are colored orange, hydrogen atoms are not colored and not displayed, the chemical bonds between different atoms are split in half and have the same color as the connected atoms, in the RGB color gamut, the background is set to white, and the size is 224*224.

[0069] The cavity image module is used to perform dehydration, deletion of redundant ligands, and hydrogenation to modify the MMFF94x force field on acetylcholinesterase in sequence according to the molecular structure information of the acetylcholinesterase sample, and then convert the SMILES of the ionic liquid into a three-dimensional object, dock the ionic liquid with acetylcholinesterase using huprine W. ligand as the binding site to obtain the three-dimensional structure of the cavity formed by the ionic liquid binding to acetylcholinesterase, and then perform slicing processing on the three-dimensional cavity structure to obtain the two-dimensional color slice image of the cavity. When generating the cavity, with the ionic liquid as the center, generate a three-dimensional cavity structure larger than the average distance from the center of the ionic liquid to the surrounding residues of acetylcholinesterase, set the cavity color to light blue and the transparency to 80%; during slicing processing, select the position with the most obvious structural features in the three-dimensional cavity structure for slicing, and set it in the RGB color gamut, with a size of 224*224.

[0070] The image stitching module is used to stitch the two-dimensional color images of the cations and anions generated by the ionic liquid image module with the two-dimensional color slice image of the cavity generated by the cavity image module into a 9-channel RGB tensor of the image to be predicted.

[0071] The convolutional neural network is used to perform deep learning prediction on the toxicity of an unknown ionic liquid to acetylcholinesterase based on the image to be predicted generated by the image stitching module, and obtain the predicted value of the toxicity of the corresponding EC50 experiment. The VGG19_BN convolutional neural network is trained through the prediction of the toxicity of ionic liquids to acetylcholinesterase.

[0072] In this embodiment, the preferred convolutional neural network is specifically the VGG19_BN convolutional neural network. In this embodiment, the specific structure and connection relationship of the preferred VGG19_BN convolutional neural network are as follows: input layer - convolutional (9, 64) layer - BN layer - convolutional (64, 64) layer - BN layer - pooling layer - convolutional (64, 128) layer - BN layer - convolutional (128, 128) layer - BN layer - pooling layer - convolutional (128, 256) layer - BN layer - convolutional (256, 256) layer - BN layer - convolutional (256, 256) layer - BN layer - convolutional (256, 256) layer - BN layer - pooling layer - convolutional (256, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - pooling layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - convolutional (512, 512) layer - BN layer - pooling layer - fully connected layer - fully connected layer - fully connected layer - output layer, where the numbers in the parentheses represent the specific setting method and number of channels; the convolutional kernels in the sixteen convolutional layers are uniformly 3*3 and the stride is uniformly 1*1; the convolutional kernel of the pooling layer is uniformly 2*2 and the stride size is uniformly 2*2; to prevent overfitting, Dropout is added, with the Dropout rate set to 0.5, the learning rate Learning rate set to 0.01, the activation function ReLU, the loss function mean square root deviation MSE, the batch processing parameter batch size set to 16, and the number of training epochs epochs set to 1000; the output layer outputs the predicted value of the toxicity of the EC50 experiment.

[0073] When training the VGG19_BN convolutional neural network, first obtain the two-dimensional structure color images of the cation and anion respectively from the known ionic liquids, and obtain the two-dimensional color slice images of the cavity three-dimensional structure from the known ionic liquids combined with acetylcholinesterase; then perform image stitching to obtain the training images, and input the training images into the VGG19_BN convolutional neural network, and use the known ionic liquids to train the VGG19_BN convolutional neural network with the experimental values of the toxicity of the new acetylcholinesterase EC50 experiment.

[0074] The VGG19_BN convolutional neural network is pre-trained using an image training set containing information on the toxicity of ionic liquids to acetylcholinesterase and tested using a validation set. When the coefficient of determination R 2 > 74% and the root mean square error RMSE < 0.210, it is considered that the performance of the VGG19_BN convolutional neural network meets the requirements and the training is completed.

[0075] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the toxicity of ionic liquids to acetylcholinesterase, characterized in that, It includes the following steps: The first step: Obtain the basic information of the ionic liquid; The second step: Generate the deep learning to-be-predicted image of the toxicity of the ionic liquid to acetylcholinesterase; The third step: Use the VGG19_BN convolutional neural network for deep learning training to obtain an end-to-end deep learning model of the known toxicity of the ionic liquid to acetylcholinesterase; The fourth step: Process the new ionic liquid sample and the new acetylcholinesterase sample to obtain the spliced to-be-predicted image of the toxicity of the unknown ionic liquid to acetylcholinesterase as the test set, and then input the to-be-predicted image of the test set into the trained VGG19_BN convolutional neural network for deep learning to output the predicted EC50 experimental value, thereby completing the toxicity prediction of the ionic liquid to acetylcholinesterase; The specific steps of the second step include the following: S1: Standardize the SMILES of the collected ionic liquid to generate the corresponding two-dimensional structure image; S2: Combine acetylcholinesterase with the ionic liquid to obtain the corresponding cavity three-dimensional structure, and slice the most obvious part of the cavity three-dimensional structure features to obtain the two-dimensional color slice image of the cavity; S3: Splice the two-dimensional structure color images of the ionic liquid cation and anion respectively and the two-dimensional color slice image of the cavity to form the training image.

2. The method for predicting the toxicity of an ionic liquid to acetylcholinesterase according to claim 1, characterized in that: In the two-dimensional structure image, different atoms are colored differently. Specifically, nitrogen atoms are colored blue, oxygen atoms are colored red, sulfur atoms are colored yellow, carbon atoms are colored black, fluorine atoms are colored blue, phosphorus atoms are colored orange, hydrogen atoms are not colored and not displayed, and the chemical bonds between different atoms are split in half and have the same color as the connected atoms.

3. A method for predicting the toxicity of ionic liquids to acetylcholinesterase according to claim 1, characterized in that: The specific steps of the third step include the following: S1: A data set composed of all training images and the corresponding EC50 experimental values of drug safety indicators is grouped according to the ratio that the training set accounts for 80% and the validation set accounts for 20%. The to-be-predicted image is used as the input, and the corresponding EC50 experimental value of the drug safety indicator is used as the predicted output; S2: Construct a VGG19_BN convolutional neural network; S3: Use the training set for deep learning training on the VGG19_BN convolutional neural network, and use the consistent known EC50 experimental values to train the VGG19_BN convolutional neural network until the VGG19_BN convolutional neural network converges; S4: Use the validation set on the VGG19_BN convolutional neural network trained to convergence in step S3 to learn the training image features and output the predicted EC50 experimental value for inspection.

4. A method for predicting the toxicity of ionic liquids to acetylcholinesterase according to claim 1, characterized in that: The specific steps of the fourth step are as follows: S1: Generate the two-dimensional structure color images of the cation and anion of the new ionic liquid sample respectively; S2: Combine the new acetylcholinesterase sample with the ionic liquid to obtain the corresponding cavity three-dimensional structure, and slice the most obvious part of the cavity three-dimensional structure features to obtain the two-dimensional color slice image of the cavity; S3: Splice the two-dimensional structure color images of the new ionic liquid cation and anion respectively and the two-dimensional color slice image of the cavity to form the training image; S4: Then, input the training images into the VGG19_BN convolutional neural network after training and validation for prediction, and output the predicted values of the toxicity of the corresponding new ionic liquids to the new acetylcholinesterase EC50 experiment.

5. A method for predicting the toxicity of ionic liquids to acetylcholinesterase according to claim 1, characterized in that: When the VGG19_BN convolutional neural network is trained for toxicity prediction, an external test is adopted. When the coefficient of determination R 2 > 74% and the root mean square error RMSE < 0.210, it indicates that the training of the VGG19_BN convolutional neural network is completed.

6. An ionic liquid toxicity prediction system for acetylcholinesterase implementing the method according to claim 1, characterized in that, It includes a basic information module, an image generation module, and a convolutional neural network. Both the basic information module and the convolutional neural network are connected to the image generation module; The basic information module is used to obtain the ASCII string SMILES of the ionic liquid, obtain the experimental values of the known ionic liquids for the acetylcholinesterase drug safety index EC50, and obtain the information on the molecular structure of acetylcholinesterase; The image generation module is used to perform atomic differential coloring on the ionic liquid samples with unknown toxicity of ionic liquids to acetylcholinesterase and perform cavity three-dimensional structure slicing on the acetylcholinesterase samples to obtain the images to be predicted; The convolutional neural network is used to perform deep learning prediction on the toxicity of the unknown ionic liquids to acetylcholinesterase based on the images to be predicted generated by the image generation module, and obtain the predicted values of the corresponding toxicity; The convolutional neural network is the VGG19_BN convolutional neural network trained for predicting the toxicity of ionic liquids to acetylcholinesterase.

7. An ionic liquid toxicity prediction system for acetylcholinesterase according to claim 6, characterized in that: The image generation module includes an ionic liquid image module for generating a two-dimensional structure from the ionic liquid samples, then performing differential coloring on the atoms of the two-dimensional structure to obtain the two-dimensional structure color images of the cations and anions in the ionic liquid respectively; An image cavity image module for sequentially performing dehydration, deleting redundant ligands, and hydrogenation to modify the MMFF94x force field on acetylcholinesterase, then converting the ionic liquid into a three-dimensional object, docking the ionic liquid with acetylcholinesterase, generating a cavity three-dimensional structure with a distance greater than the average distance from the center of the ionic liquid to the surrounding residues of acetylcholinesterase, and then performing slicing processing on the cavity three-dimensional structure to obtain the two-dimensional color slices of the cavity; An image splicing module for splicing the two-dimensional structure color images of the cations and anions in the ionic liquid and the two-dimensional color slice images of the cavity to form the images to be predicted with a 9-channel RGB tensor; The ionic liquid image module and the cavity image module are respectively connected to the basic information module. The image splicing module is connected to the ionic liquid image module and the cavity image module, and the image splicing module is connected to the convolutional neural network.

Images