Compound organ toxicity prediction method and device and electronic equipment

Abstract

The invention provides a compound organ toxicity prediction method and device and electronic equipment. The method comprises the following steps: acquiring a to-be-predicted compound; extracting a first compound feature corresponding to the to-be-predicted compound; inputting the first compound feature into a pre-trained organ toxicity prediction model, and outputting an organ toxicity prediction result corresponding to the to-be-predicted compound; wherein the organ toxicity prediction model is obtained by training a target neural network through a compound organ toxicity training set; the target neural network is formed on the basis of a directed acyclic graph formed by the bad outcome event information corresponding to the toxicities of various specific organs. The method is a compound organ toxicity prediction mode based on a bad outcome pathway, and can effectively solve the technical problems of low modeling capability and poor reliability of a complex nonlinear relation in the prior art, thereby realizing more accurate and reliable compound organ toxicity prediction.

Description

Technical Field

[0001] The present application relates to the field of chemical technology, and in particular to a method, device and electronic equipment for predicting compound organ toxicity. Background Art

[0002] Compound safety evaluation is a very important part of the pharmaceutical and cosmetic industries. According to statistics, about 30% of new drug development failures are caused by compound safety issues. In order to improve the success rate of drug and cosmetic development, it is very important to conduct safety evaluation of compound organ toxicity in the early stages of research and development.

[0003] Traditional compound organ toxicity safety evaluation relies on compound organ toxicity prediction results, and the prediction process is mainly based on in vivo animal toxicology experiments and cell-based in vitro toxicology experiments. Animal-based in vivo toxicology experiments have the disadvantages of long cycles and high costs; cell-based in vitro toxicology experiments have the disadvantage of small evaluation scale. At present, with the accumulation of data, domestic and foreign research attempts to use machine learning algorithms to predict compound organ toxicity. The compound organ toxicity prediction based on machine learning relies on data modeling, which can quickly and large-scale compound toxicity assessment.

[0004] However, existing machine learning-based compound organ toxicity predictions still face the following problems: 1) Traditional machine learning algorithms have difficulty modeling complex nonlinear relationships; 2) Although deep learning algorithms can model complex nonlinear relationships, they are all black box models, making it difficult to understand the model decision-making process, and have low reliability in actual application scenarios. Summary of the invention

[0005] The purpose of the present application is to provide a compound organ toxicity prediction method, device and electronic device, which is a compound organ toxicity prediction method based on adverse outcome pathways, and can effectively solve the technical obstacles of low complex nonlinear relationship modeling capability and poor reliability in the prior art, thereby achieving more accurate and reliable compound organ toxicity prediction.

[0006] In a first aspect, the present application provides a method for predicting compound organ toxicity, the method comprising: obtaining a compound to be predicted; extracting a first compound feature corresponding to the compound to be predicted; inputting the first compound feature into a pre-trained organ toxicity prediction model, and outputting an organ toxicity prediction result corresponding to the compound to be predicted; wherein the organ toxicity prediction model is obtained by training a target neural network with a compound organ toxicity training set; and the target neural network is formed based on a directed acyclic graph consisting of adverse outcome event information corresponding to multiple specific organ toxicities.

[0007] Furthermore, the above-mentioned step of extracting the first compound feature corresponding to the compound to be predicted includes: performing the following steps based on the RDkit chemical informatics open source toolkit: converting the molecular linear input system SMILES of the compound to be predicted into a standardized SMILES; converting the standardized SMILES into a functional fingerprint FCFP with a radius of 6 and 2048 dimensions, and a 209-dimensional molecular descriptor; splicing the 2048-dimensional functional fingerprint FCFP and the 209-dimensional molecular descriptor to combine them into a 2257-dimensional first compound feature.

[0008] Furthermore, the training process of the above-mentioned organ toxicity prediction model is as follows: collect and screen adverse outcome event information, and form a directed acyclic graph of adverse outcome pathways for specific organ toxicity based on the adverse outcome event information; according to the hierarchical association of the directed acyclic graph of the adverse outcome pathways for specific organ toxicity, form a sparsely connected target neural network with knowledge embedded in the adverse outcome pathways; obtain a compound organ toxicity training set; the samples in the training set include: multiple compounds and their corresponding real toxicity labels; use the samples in the compound organ toxicity training set to train the target neural network, and adjust the network parameters through the loss value until convergence, so as to obtain an organ toxicity prediction model.

[0009] Furthermore, the above-mentioned step of collecting and screening adverse outcome event information, and forming a directed acyclic graph of an adverse outcome pathway for specific organ toxicity based on the adverse outcome event information, includes: screening multiple event information for specific organ toxicity from the harmful outcome pathway wiki; the multiple event information is information developed for mammals, has chemical stressors, and has a connection relationship; each event information includes: an event name, a biological ontology annotation corresponding to the event, and a harmful outcome pathway to which the event belongs; based on the association relationship between the multiple event information, an initial organ toxicity adverse outcome pathway directed graph is formed with the event name as a node; the initial organ toxicity adverse outcome pathway directed graph is processed according to the following rules to obtain a directed acyclic graph of the adverse outcome pathway for specific organ toxicity: the connection confidence is high or medium; the nodes with both out-degree and in-degree of 1 are completely deleted, and the upstream and downstream connections are maintained; if there is a loop, the relatively unimportant associations or redundant key events are deleted.

[0010] Furthermore, the target neural network includes: a molecular starting event layer, an intermediate key event layer, and an adverse outcome layer; the activation function of each layer is tanh, maintaining the activation value between -1 and 1, and each laser function is connected to a Batchnorm layer; for each sample i, the output of the intermediate key event t in the intermediate key event layer is expressed as: ; ; in, Represents the output result of the i-th sample at the intermediate key event t; Represents the processing function of the batch normalization layer; represents the compound representation after the input is processed by the linear layer; represents the network weight of the intermediate key event t; Represents the input features of the i-th sample at the intermediate key event t; represents the network bias; The output length of the output result is ;in, represents the degree of the intermediate key event t in the adverse outcome pathway network, and n represents the number of neurons representing the intermediate key event t.

[0011] Furthermore, the above-mentioned step of training the target neural network with samples in the compound organ toxicity training set and adjusting the network parameters by the loss value until convergence to obtain the organ toxicity prediction model includes: for each round of training set samples, extracting the second compound feature from the compound of the sample; inputting the second compound feature into the target neural network and outputting a predicted toxicity label; calculating the loss value according to the predicted toxicity label and the true toxicity label corresponding to the compound; updating the model parameters according to the loss value; when the model prediction accuracy on the validation set does not increase within a specified number of epochs, stopping the model training to obtain the organ toxicity prediction model; an epoch is a process of forward propagating and backpropagating the entire training set in the neural network once.

[0012] Furthermore, the step of calculating the loss value according to the predicted toxicity label and the actual toxicity label corresponding to the compound includes: calculating the loss value according to the following formula : , in, is the binary cross entropy function, represents the predicted toxicity label output by the intermediate key event t of the i-th sample, represents the true toxicity label of the i-th sample, and the parameter Set to 0.2, represents a linear layer, represents the predicted toxicity label output by the final node e of the i-th sample; N represents the total number of samples.

[0013] In a second aspect, the present application also provides a compound organ toxicity prediction device, which includes: a compound acquisition module, used to acquire the compound to be predicted; a feature extraction module, used to extract a first compound feature corresponding to the compound to be predicted; a model prediction module, used to input the first compound feature into a pre-trained organ toxicity prediction model, and output an organ toxicity prediction result corresponding to the compound to be predicted; wherein the organ toxicity prediction model is obtained by training a target neural network with a compound organ toxicity training set; the target neural network is formed based on a directed acyclic graph composed of adverse outcome event information corresponding to multiple specific organ toxicities.

[0014] In a third aspect, the present application further provides an electronic device, comprising a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method described in the first aspect above.

[0015] In a fourth aspect, the present application further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method described in the first aspect above.

[0016] In a compound organ toxicity prediction method, device and electronic device provided in the present application, the compound to be predicted is first obtained; the first compound feature corresponding to the compound to be predicted is extracted; then the first compound feature is input into a pre-trained organ toxicity prediction model, and the organ toxicity prediction result corresponding to the compound to be predicted is output; wherein the organ toxicity prediction model is obtained by training the target neural network with the compound organ toxicity training set; the target neural network is formed by a directed acyclic graph composed of adverse outcome event information corresponding to a plurality of specific organ toxicities. In this method, compound organ toxicity is predicted by using a directed acyclic graph network model composed of adverse outcome pathway event information, which can effectively solve the technical obstacles of low complex nonlinear relationship modeling capability and poor reliability in the prior art, thereby achieving more accurate and reliable compound organ toxicity prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A flow chart of a method for predicting organ toxicity of a compound provided in an embodiment of the present application; Figure 2 A flowchart of a model training process provided in an embodiment of the present application; Figure 3 A schematic diagram of a model network structure provided in an embodiment of the present application; Figure 4 A structural block diagram of a compound organ toxicity prediction device provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solution of the present application will be described clearly and completely below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0020] In response to the technical obstacles of low ability and poor reliability in modeling complex nonlinear relationships in the prior art, the embodiments of the present application provide a method, device and electronic device for predicting compound organ toxicity, which can achieve more accurate and reliable compound organ toxicity prediction by predicting compound organ toxicity based on a directed acyclic graph network model composed of adverse outcome pathway event information.

[0021] To facilitate understanding of this embodiment, a compound organ toxicity prediction method disclosed in the embodiment of the present application is first introduced in detail. Figure 1 A flow chart of a method for predicting organ toxicity of a compound provided in an embodiment of the present application, the method specifically comprises the following steps: Step S102, obtaining a compound to be predicted; Step S104, extracting the first compound feature corresponding to the compound to be predicted; In specific implementation, the following steps can be performed based on the RDkit cheminformatics open source toolkit: (1) Convert the molecular linear input system (SMILES) of the compound to be predicted into a standardized SMILES; SMILES stands for Simplified Molecular Input Line Entry System; (2) The standardized SMILES is converted into a functional class fingerprint FCFP with a radius of 6 and a dimension of 2048, as well as a molecular descriptor with a dimension of 209. The full name of FCFP is Functional-Class Fingerprint.

[0022] (3) The 2048-dimensional functional fingerprint FCFP and the 209-dimensional molecular descriptor are combined into a 2257-dimensional first compound feature.

[0023] The feature extraction process during model training is the same as described above.

[0024] Step S106, input the first compound feature into a pre-trained organ toxicity prediction model, and output the organ toxicity prediction result corresponding to the compound to be predicted; wherein the organ toxicity prediction model is obtained by training the target neural network with the compound organ toxicity training set; the target neural network is formed based on a directed acyclic graph composed of adverse outcome event information corresponding to a plurality of specific organ toxicities. The training process of the organ toxicity prediction model will be described in detail later.

[0025] For example, the compound to be predicted is the drug rifampicin, and the final prediction result is that it has liver toxicity. Organ toxicity may include liver toxicity, kidney toxicity, respiratory toxicity and cardiotoxicity.

[0026] The compound organ toxicity prediction method provided in the embodiment of the present application predicts compound organ toxicity by using a directed acyclic graph network model composed of adverse outcome pathway event information to achieve more accurate and reliable compound organ toxicity prediction.

[0027] The present application example also provides another compound organ toxicity prediction method, which is implemented on the basis of the above example; this example focuses on describing the model training process.

[0028] See also Figure 2 As shown in Figure 2, the training process of the above organ toxicity prediction model is as follows: Step S202, collecting and screening adverse outcome event information, and forming a directed acyclic graph of adverse outcome pathways for specific organ toxicity based on the adverse outcome event information; The specific implementation is achieved through several steps: (1) Screen multiple event information on specific organ toxicity from the adverse outcome pathway wiki; multiple event information is information developed for mammals, with chemical stressors and connections; each event information includes: event name, biological ontology annotation corresponding to the event, and the adverse outcome pathway to which the event belongs; In one embodiment, a list of specific organ toxicity adverse outcome pathways is first screened from AOP-wiki according to the following rules: A. Determine the corresponding organ toxicity based on the name and adverse outcome; B. Eliminate non-chemical stressors; C. Exclude development for non-mammals; D. Exclude those that have no connections.

[0029] An adverse outcome pathway AOP includes molecular initiating events, key events, and adverse outcomes; for example, AOP: 57; its name is AhR activation leading to liver fibrosis, the adverse outcome is liver fibrosis, the stressor is a compound that can cause receptor AhR activation, it was developed for mice, and it includes the relationship between multiple events, so AOP: 57 was retained.

[0030] Then, multiple event information for specific organ toxicity is obtained from the list of specific organ toxicity adverse outcome pathways. Each event information includes: event name, biological ontology annotation corresponding to the event (such as GO, cell type, etc.), and the adverse outcome pathway AOP to which the event belongs.

[0031] (2) Based on the correlation between multiple event information, a directed graph of the initial organ toxicity adverse outcome pathway is formed with the event name as the node; The association relationship is directional, and a directed graph can be formed based on the association relationship between events.

[0032] (3) The initial organ toxicity adverse outcome pathway directed graph is processed according to the following rules to obtain a directed acyclic graph of the adverse outcome pathway for specific organ toxicity: A. Connection confidence is high or medium; B. Nodes with both out-degree and in-degree of 1 are completely deleted, and the upstream and downstream connections are maintained; C. If a loop exists, delete relatively unimportant associations or redundant key events.

[0033] For example, among the multiple associations in AOP: 57, the association confidence between Event: 18 and Event: 451 is High, and Event: 18 and Event: 451 contain multiple directly connected upstream nodes and downstream nodes. Event: 18 and Event: 451 do not have loops with other events, so the two nodes Event: 18 and Event: 451 and the connection between the two nodes are retained.

[0034] Step S204, based on the hierarchical association of the directed acyclic graph of the adverse outcome pathway for specific organ toxicity, a sparsely connected target neural network in which the adverse outcome pathway knowledge is embedded is formed, such as Figure 3 As shown; The target neural network includes: a molecular starting event layer, an intermediate key event layer, and an adverse outcome layer; the activation function of each layer is tanh, maintaining the activation value between -1 and 1, and each laser function is connected to a Batchnorm layer; for each sample i, the output of the intermediate key event t in the intermediate key event layer is expressed as: ; ; in, Represents the output result of the i-th sample at the intermediate key event t; Represents the processing function of the batch normalization layer; represents the compound representation after the input is processed by the linear layer; represents the network weight of the intermediate key event t; Represents the input features of the i-th sample at the intermediate key event t; represents the network bias; The output length of the output result is ;in, represents the degree of the intermediate key event t in the adverse outcome pathway network, and n represents the number of neurons representing the intermediate key event t.

[0035] Step S206, obtaining a compound organ toxicity training set; the samples in the training set include: a plurality of compounds and their corresponding real toxicity labels; Step S208, using samples in the compound organ toxicity training set to train the target neural network, and adjusting the network parameters by the loss value until convergence, to obtain an organ toxicity prediction model.

[0036] Specifically, for each round of training set samples, the second compound feature is extracted from the compound of the sample; the second compound feature is input into the target neural network, and a predicted toxicity label is output; the loss value is calculated based on the predicted toxicity label and the true toxicity label corresponding to the compound; the model parameters are updated according to the loss value; when the model prediction accuracy on the validation set does not increase within a specified number of epochs, the model training is stopped to obtain an organ toxicity prediction model; an epoch is a process of forward propagating and backpropagating the entire training set in the neural network once.

[0037] Furthermore, the step of calculating the loss value according to the predicted toxicity label and the actual toxicity label corresponding to the compound includes: calculating the loss value according to the following formula : , in, is the binary cross entropy function, represents the predicted toxicity label output by the intermediate key event t of the i-th sample, represents the true toxicity label of the i-th sample, and the parameter Set to 0.2, represents a linear layer, represents the predicted toxicity label output by the final node e of the i-th sample; N represents the total number of samples.

[0038] According to the compound organ toxicity prediction method based on adverse outcome pathways provided in the embodiments of the present application, by collecting a list of specific organ toxicity adverse outcome pathways, a directed acyclic graph is formed based on the correlation between event information, the sparse connection of the neural network is guided, and the combination of neurons in the neural network represents a specific biological meaning, thereby making the internal structure of the deep learning model transparent, facilitating the subsequent tracing of the decision-making reasons of the model, and achieving more accurate and reliable compound organ toxicity prediction.

[0039] Based on the above method embodiment, the present application embodiment also provides a compound organ toxicity prediction device, see Figure 4 As shown, the device includes: a compound acquisition module 42, used to acquire the compound to be predicted; a feature extraction module 44, used to extract the first compound feature corresponding to the compound to be predicted; a model prediction module 46, used to input the first compound feature into a pre-trained organ toxicity prediction model, and output the organ toxicity prediction result corresponding to the compound to be predicted; wherein the organ toxicity prediction model is obtained by training the target neural network with the compound organ toxicity training set; the target neural network is formed based on a directed acyclic graph consisting of adverse outcome event information corresponding to multiple specific organ toxicities.

[0040] Furthermore, the feature extraction module 44 is used to perform the following steps based on the RDkit chemical informatics open source toolkit: converting the molecular linear input system SMILES of the compound to be predicted into a standardized SMILES; converting the standardized SMILES into a functional fingerprint FCFP with a radius of 6 and 2048 dimensions, and a 209-dimensional molecular descriptor; splicing the 2048-dimensional functional fingerprint FCFP and the 209-dimensional molecular descriptor to combine them into a 2257-dimensional first compound feature.

[0041] Furthermore, the above-mentioned device also includes: a model training module, which is used to execute the following training process of the organ toxicity prediction model: collecting and screening adverse outcome event information, forming a directed acyclic graph of adverse outcome pathways for specific organ toxicity based on the adverse outcome event information; forming a sparsely connected target neural network with knowledge embedded in the adverse outcome pathway according to the hierarchical association of the directed acyclic graph of the adverse outcome pathways for specific organ toxicity; obtaining a compound organ toxicity training set; the samples in the training set include: multiple compounds and their corresponding real toxicity labels; using the samples in the compound organ toxicity training set to train the target neural network, and adjusting the network parameters through the loss value until convergence, to obtain an organ toxicity prediction model.

[0042] Furthermore, the above-mentioned model training module is used to screen multiple event information for specific organ toxicity from the harmful outcome pathway wiki; the multiple event information is information developed for mammals, has chemical stressors, and has a connection relationship; each event information includes: event name, biological ontology annotation corresponding to the event, and the harmful outcome pathway to which the event belongs; according to the association relationship between the multiple event information, an initial organ toxicity adverse outcome pathway directed graph is formed with the event name as the node; the initial organ toxicity adverse outcome pathway directed graph is processed according to the following rules to obtain a directed acyclic graph of the adverse outcome pathway for specific organ toxicity: the connection confidence is high or medium; the nodes with both out-degree and in-degree of 1 are completely deleted, and the upstream and downstream connections are maintained; if there is a loop, the relatively unimportant associations or redundant key events are deleted.

[0043] Furthermore, the target neural network includes: a molecular starting event layer, an intermediate key event layer, and an adverse outcome layer; the activation function of each layer is tanh, maintaining the activation value between -1 and 1, and each laser function is connected to a Batchnorm layer; for each sample i, the output of the intermediate key event t in the intermediate key event layer is expressed as: ; ; in, Represents the output result of the i-th sample at the intermediate key event t; Represents the processing function of the batch normalization layer; represents the compound representation after the input is processed by the linear layer; represents the network weight of the intermediate key event t; Represents the input features of the i-th sample at the intermediate key event t; represents the network bias; The output length of the output result is ;in, represents the degree of the intermediate key event t in the adverse outcome pathway network, and n represents the number of neurons representing the intermediate key event t.

[0044] Furthermore, the above-mentioned model training module is used to extract the second compound feature from the compound of the sample for each round of training set samples; input the second compound feature into the target neural network and output a predicted toxicity label; calculate the loss value according to the predicted toxicity label and the actual toxicity label corresponding to the compound; update the model parameters according to the loss value; when the model prediction accuracy on the validation set does not increase within a specified number of epochs, stop model training and obtain an organ toxicity prediction model; epoch is the process of forward propagating and backpropagating the entire training set in the neural network once.

[0045] Furthermore, the above model training module is used to calculate the loss value according to the following formula: : , in, is the binary cross entropy function, represents the predicted toxicity label output by the intermediate key event t of the i-th sample, represents the true toxicity label of the i-th sample, and the parameter Set to 0.2, represents a linear layer, represents the predicted toxicity label output by the final node e of the i-th sample; N represents the total number of samples.

[0046] The device provided in the embodiment of the present application has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the embodiment of the device, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0047] The present application also provides an electronic device, such as Figure 5 As shown, it is a schematic diagram of the structure of the electronic device, wherein the electronic device includes a processor 51 and a memory 50, the memory 50 stores computer executable instructions that can be executed by the processor 51, and the processor 51 executes the computer executable instructions to implement the above method.

[0048] exist Figure 5 In the illustrated embodiment, the electronic device further includes a bus 52 and a communication interface 53 , wherein the processor 51 , the communication interface 53 and the memory 50 are connected via the bus 52 .

[0049] Among them, the memory 50 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 53 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 52 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 52 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0050] The processor 51 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 51. The above processor 51 can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor are combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor 51 reads the information in the memory and completes the steps of the method of the above embodiment in combination with its hardware.

[0051] An embodiment of the present application also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the above method. The specific implementation can be found in the aforementioned method embodiment, which will not be repeated here.

[0052] The computer program products of the methods, devices, and electronic devices provided in the embodiments of the present application include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be found in the method embodiments, which will not be repeated here.

[0053] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present application.

[0054] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0055] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0056] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for predicting organ toxicity of a compound, characterized in that: The method comprises: Obtaining compounds to be predicted; Extracting a first compound feature corresponding to the compound to be predicted; The first compound feature is input into a pre-trained organ toxicity prediction model, and the organ toxicity prediction result corresponding to the compound to be predicted is output; wherein the organ toxicity prediction model is obtained by training a target neural network with a compound organ toxicity training set; and the target neural network is formed by a directed acyclic graph consisting of adverse outcome event information corresponding to multiple specific organ toxicities.

2. The method according to claim 1, characterized in that The step of extracting the first compound feature corresponding to the compound to be predicted comprises: Based on the RDkit cheminformatics open source toolkit, perform the following steps: The molecular linear input system SMILES of the compound to be predicted is converted into a standardized SMILES; The standardized SMILES are converted into a functional fingerprint FCFP with a radius of 6 and a dimension of 2048, and a molecular descriptor with a dimension of 209; The 2048-dimensional functional fingerprint FCFP and the 209-dimensional molecular descriptor are concatenated to form a 2257-dimensional first compound feature.

3. The method according to claim 1, characterized in that The training process of the organ toxicity prediction model is as follows: Collecting and screening adverse outcome event information, and forming a directed acyclic graph of adverse outcome pathways for specific organ toxicity based on the adverse outcome event information; According to the hierarchical association of the directed acyclic graph of the adverse outcome pathways for specific organ toxicity, a sparsely connected target neural network is formed to embed the knowledge of the adverse outcome pathways; Obtaining a compound organ toxicity training set; the samples in the training set include: a plurality of compounds and their corresponding true toxicity labels; The target neural network is trained using samples in the compound organ toxicity training set, and the network parameters are adjusted by the loss value until convergence, thereby obtaining the organ toxicity prediction model.

4. The method according to claim 3, characterized in that The steps of collecting and screening adverse outcome event information, and forming a directed acyclic graph of adverse outcome pathways for specific organ toxicity based on the adverse outcome event information, include: Screening multiple event information on specific organ toxicity from the harmful outcome pathway wiki; the multiple event information is information developed for mammals, having chemical stressors and having connection relationships; each of the event information includes: an event name, a biological ontology annotation corresponding to the event, and a harmful outcome pathway to which the event belongs; According to the correlation between multiple event information, a directed graph of the initial organ toxicity adverse outcome pathway is formed with the event name as the node; The initial organ toxicity adverse outcome pathway directed graph is processed according to the following rules to obtain a directed acyclic graph of the adverse outcome pathway for specific organ toxicity: Connection confidence is high or medium; Nodes with both out-degree and in-degree of 1 are completely deleted, and the upstream and downstream connections are maintained; If a loop exists, delete relatively unimportant associations or redundant key events.

5. The method according to claim 3, characterized in that: The target neural network includes: a molecular starting event layer, an intermediate key event layer and an adverse outcome layer; the activation function of each layer is tanh, maintaining the activation value between -1 and 1, and each laser function is connected to a Batchnorm layer; for each sample i, the output of the intermediate key event t in the intermediate key event layer is expressed as: ； ； in, Represents the output result of the i-th sample at the intermediate key event t; Represents the processing function of the batch normalization layer; represents the compound representation after the input is processed by the linear layer; represents the network weight of the intermediate key event t; Represents the input features of the i-th sample at the intermediate key event t; represents the network bias; The output length of the output result is ;in, represents the degree of the intermediate key event t in the adverse outcome pathway network, and n represents the number of neurons representing the intermediate key event t.

6. The method according to claim 3, characterized in that The step of using the samples in the compound organ toxicity training set to train the target neural network and adjusting the network parameters by the loss value until convergence to obtain the organ toxicity prediction model comprises: For each round of training set samples, extracting a second compound feature from the compounds of the sample; inputting the second compound feature into the target neural network and outputting a predicted toxicity label; Calculating a loss value according to the predicted toxicity label and the true toxicity label corresponding to the compound; Update model parameters according to the loss value; When the model prediction accuracy on the validation set does not increase within a specified number of epochs, the model training is stopped to obtain the organ toxicity prediction model; the epoch is a process of forward propagating and backpropagating the entire training set once in the neural network.

7. The method according to claim 6, characterized in that The step of calculating the loss value according to the predicted toxicity label and the true toxicity label corresponding to the compound comprises: The loss value is calculated according to the following formula : ， in, is the binary cross entropy function, represents the predicted toxicity label output by the intermediate key event t of the i-th sample, represents the true toxicity label of the i-th sample, and the parameter Set to 0.2, represents a linear layer, represents the predicted toxicity label output by the final node e of the i-th sample; N represents the total number of samples.

8. A compound organ toxicity prediction device, characterized in that: The device comprises: A compound acquisition module, used to acquire compounds to be predicted; A feature extraction module, used to extract a first compound feature corresponding to the compound to be predicted; A model prediction module is used to input the first compound feature into a pre-trained organ toxicity prediction model, and output the organ toxicity prediction result corresponding to the compound to be predicted; wherein the organ toxicity prediction model is obtained by training the target neural network through the compound organ toxicity training set; the target neural network is formed based on a directed acyclic graph composed of adverse outcome event information corresponding to multiple specific organ toxicities.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images