Methods and apparatus for predicting antagonists, antagonists, media and equipment

Abstract

This disclosure relates to the field of artificial intelligence technology, and provides a method, apparatus, antagonist, medium, and device for predicting antagonists based on artificial intelligence. The method includes: obtaining simulated drug molecules from a generator network of a trained generative adversarial neural network, wherein the generative adversarial neural network is trained based on existing antagonists corresponding to drug-like molecules and target receptor molecules; determining a database of small molecules to be screened based on the simulated drug molecules; performing virtual screening on the small molecule database based on the target receptor molecules to screen out target drug small molecules; and obtaining the antagonist of the target receptor molecules by performing cell activity tests on the target drug small molecules. This technical solution can increase the diversity of the chemical spatial structure of the simulated drug molecules and reduce the cost and time of drug development.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more specifically, to a method and apparatus for predicting antagonists based on artificial intelligence, an antagonist, and a computer-readable storage medium and electronic device for implementing the above method. Background Technology

[0002] Antagonists are substances that bind to receptors but do not produce biological effects themselves, yet block the effects mediated by the agonists of the relevant receptors. For example, when glucagon acts as a receptor, glucagon antagonists can bind to it and reduce the cyclic adenosine monophosphate (cAMP) response activity of glucagon, thereby promoting its hypoglycemic activity.

[0003] Current methods for identifying antagonists involve obtaining small molecules from commercial or open-source small molecule databases, then screening these small molecules using molecular docking techniques to ultimately select those that meet specific criteria as target antagonists. However, the chemical spatial structures of compounds identified as target antagonists based on these techniques are limited, and antagonists identified in this way often have similar skeletal structures to known drug molecules.

[0004] It is evident that the diversity of antagonists identified based on relevant technologies needs to be improved.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure. Summary of the Invention

[0006] The purpose of this disclosure is to provide a method and apparatus for predicting antagonists based on artificial intelligence, antagonists, and computer-readable storage media and electronic devices for implementing the above method, thereby improving the diversity of antagonists to at least some extent.

[0007] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0008] According to one aspect of this disclosure, a method for predicting antagonists based on artificial intelligence is provided, comprising: obtaining a simulated drug molecule from a generator network of a trained generative adversarial neural network, wherein the generative adversarial neural network is trained based on existing antagonists corresponding to drug-like molecules and target receptor molecules; determining a small molecule database to be screened based on the simulated drug molecule; performing virtual screening on the small molecule database based on the target receptor molecule to screen out target drug small molecules from the small molecule database; and obtaining an antagonist of the target receptor molecule by performing a cell activity test on the target drug small molecule.

[0009] According to one aspect of this disclosure, an apparatus for predicting antagonists based on artificial intelligence is provided, comprising: a simulated drug molecule identification module, a small molecule database identification module, a virtual screening module, and an antagonist identification module.

[0010] The simulated drug molecule determination module is configured to: obtain simulated drug molecules based on the generator network of a trained generative adversarial neural network, wherein the generative adversarial neural network is trained based on existing antagonists corresponding to the drug-like molecules and the target receptor molecules; the small molecule database determination module is configured to: determine a small molecule database to be screened based on the simulated drug molecules; the virtual screening module is configured to: perform virtual screening on the small molecule database based on the target receptor molecules to screen out target drug small molecules from the small molecule database; and the antagonist determination module is configured to: obtain the antagonist of the target receptor molecules by performing cell activity tests on the target drug small molecules.

[0011] In some embodiments of this disclosure, based on the foregoing scheme, the above-mentioned device further includes: a training sample acquisition module, a molecular structure prediction module, a discrimination result determination module, and a parameter optimization module.

[0012] The training sample acquisition module is configured to: acquire drug-like molecules and existing antagonists corresponding to the target receptor molecules as training samples for the generative adversarial neural network; the molecular structure prediction module is configured to: input the drug-like molecules in the training samples into the generator network of the generative adversarial neural network, and perform molecular structure prediction on the drug-like molecules based on the generator network to obtain predicted drug molecules; the discrimination result determination module is configured to: input the predicted drug molecules and existing antagonists in the training samples into the discriminator network of the generative adversarial neural network; and the parameter optimization module is configured to: optimize the parameters of the generative adversarial neural network based on the output of the discriminator network.

[0013] In some embodiments of this disclosure, based on the foregoing scheme, the parameter optimization module is specifically configured as follows: the output of the discriminator network includes a first discrimination result regarding the predicted drug molecule and a second discrimination result regarding the existing antagonist, wherein the first discrimination result includes the probability that the predicted drug molecule output by the discriminator network is an antagonist of the target receptor molecule; the second discrimination result includes the probability that the existing antagonist output by the discriminator network is an antagonist of the target receptor molecule; and the parameters of the generative adversarial neural network are optimized based on the first discrimination result and the second discrimination result.

[0014] In some embodiments of this disclosure, based on the foregoing scheme, the parameter optimization module is specifically configured as follows: fixing the parameters of the generator and adjusting the parameters of the discriminator, so that the first discrimination result is: the probability that the predicted drug molecule is an antagonist of the target receptor molecule is not greater than 0.5, and the second discrimination result is: the probability that the existing antagonist is an antagonist of the target receptor molecule is greater than 0.5; fixing the parameters of the discriminator and adjusting the parameters of the generator, so that after inputting the predicted drug molecule output by the generator after parameter adjustment into the discriminator, the first discrimination result is: the probability that the predicted drug molecule is an antagonist of the target receptor molecule is greater than 0.5.

[0015] In some embodiments of this disclosure, based on the foregoing scheme, the number of times the data is downsampled in the generator network is the same as the number of times the data is upsampled. The downsampling process includes convolution and pooling, and the upsampling process includes deconvolution and feature supplementation.

[0016] In some embodiments of this disclosure, based on the foregoing scheme, the generator network uses tanh as the activation function.

[0017] In some embodiments of this disclosure, based on the foregoing scheme, the molecular structure prediction module is specifically configured to: perform graph encoding on the chemical formulas of the drug-like molecules in the training samples to obtain graph representations corresponding to the drug-like molecules, and input the graph representations corresponding to the drug-like molecules into the generator network; the discrimination result determination module is specifically configured to: perform graph encoding on the chemical formulas of the existing antagonists to obtain graph representations corresponding to the existing antagonists; and input the graph representations corresponding to the predicted drug molecules and the existing antagonists into the discriminator network.

[0018] In some embodiments of this disclosure, based on the foregoing scheme, the simulated drug molecule determination module is configured to: perform graph encoding on the chemical formula of the target drug molecule to obtain a graph representation corresponding to the target drug molecule; input the graph representation corresponding to the target drug molecule into a trained generator network, perform molecular structure prediction based on the generator network, and output a graph representation of the simulated drug molecule; and decode the graph representation of the simulated drug molecule to obtain a chemical formula of the simulated drug molecule that satisfies a preset format.

[0019] In some embodiments of this disclosure, based on the foregoing scheme, the aforementioned small molecule database determination module is specifically configured as follows:

[0020] The simulated drug molecules output by the generator network are transformed into three dimensions to obtain three-dimensional simulated drug molecules, thus obtaining the small molecule database mentioned above.

[0021] In some embodiments of this disclosure, based on the foregoing scheme, the small molecule database determination module further includes a decoding unit.

[0022] The decoding unit is configured to decode the encoded simulated drug molecules output by the generator network before the three-dimensional conversion unit performs three-dimensional conversion on the simulated drug molecules output by the generator network to obtain three-dimensional simulated drug molecules.

[0023] In some embodiments of this disclosure, based on the foregoing scheme, the above-mentioned apparatus further includes a preprocessing module.

[0024] The preprocessing module is configured to: before the virtual screening of the small molecule database based on the target receptor molecule, obtain the target receptor molecule from the existing database and perform one or more of the following treatments on the target receptor molecule: amino acid residue completion treatment, hydrogenation treatment, and charge treatment on the target receptor molecule.

[0025] In some embodiments of this disclosure, based on the foregoing scheme, the virtual screening is specifically configured as follows: using the target receptor molecule as the target site, and based on the shape, size, and electrostatic characteristics of the protein active pocket of the target site, an optimized search algorithm is used to calculate the affinity between the drug molecules in the small molecule database and the target site, so as to screen the target drug small molecule in the small molecule database as a candidate compound of the target receptor molecule.

[0026] In some embodiments of this disclosure, based on the foregoing scheme, the above-mentioned apparatus further includes: a format unification module.

[0027] The aforementioned format unification module is configured to unify the format of the target receptor molecules with the format of the drug molecules in the aforementioned small molecule database before the virtual screening process is performed on the aforementioned small molecule database based on the aforementioned target receptor molecules.

[0028] In some embodiments of this disclosure, based on the foregoing scheme, the antagonist determination module is specifically configured as follows: for each of the target drug small molecules, it is mixed with a binder and test cells containing a target receptor molecule, wherein the binder binds to the target receptor molecule to activate the target receptor molecule signaling pathway and releases the test target component, wherein the target drug small molecule plays an antagonistic role to inhibit the activity of the target receptor molecule; an antagonist drug of the target receptor molecule is selected as a positive control; the test target component corresponding to each of the target drug small molecules is obtained, and the half-inhibition concentration (WIC) value of each of the target drug small molecules is calculated based on the test target component; and the antagonist of the target receptor molecule is determined based on the WIC value.

[0029] In some embodiments of this disclosure, based on the foregoing scheme, the antagonist determination module is further configured to: determine the target drug small molecules with half-inhibition concentration values ​​less than a preset threshold as antagonists of the target receptor molecule; or, sort the target drug small molecules according to their half-inhibition concentration values, and determine the N target drug small molecules with smaller half-inhibition concentration values ​​as antagonists of the target receptor molecule, where N is a positive integer.

[0030] In some embodiments of this disclosure, based on the foregoing scheme, the binding agent is a glucagon peptide, the target component is the second messenger cyclic adenosine monophosphate (cAMP), and the antagonist determination module is specifically configured to: select an antagonist drug of hGCGR as a positive control; mix each target drug molecule, the glucagon peptide, and human embryonic kidney cells containing the target receptor molecule; the glucagon peptide binds to hGCGR to activate the hGCGR signaling pathway to release the second messenger cAMP, wherein the target drug molecule plays an antagonistic role in inhibiting the activity of hGCGR; detect the second messenger cAMP by homogeneous time-resolved fluorescence HTRF, and calculate the half-inhibition concentration (WIC) value of each target drug molecule based on the HTRF intensity value; and determine the antagonist of the target receptor molecule based on the WIC value.

[0031] According to one aspect of this disclosure, an antagonist is provided, comprising: the backbone molecule of the antagonist being: a simulated drug molecule generated based on a trained generative adversarial neural network, wherein the simulated drug molecule is determined after virtual screening based on a target receptor molecule.

[0032] According to one aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the artificial intelligence-based antagonist prediction method described in the first aspect above.

[0033] According to one aspect of this disclosure, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the artificial intelligence-based antagonist prediction method described in the first aspect above.

[0034] According to one aspect of this disclosure, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the artificial intelligence-based antagonist prediction method provided in the various embodiments described above.

[0035] As can be seen from the above technical solutions, the method, apparatus, computer-readable storage medium, and electronic device for predicting antagonists based on artificial intelligence in the exemplary embodiments of this disclosure have at least the following advantages and positive effects:

[0036] In some embodiments of this disclosure, a simulated drug molecule of a target receptor molecule (e.g., human glucagon) is obtained based on a deep learning model, and an antagonist for the target receptor molecule is further determined based on the model drug molecule. Since the deep learning model is trained on massive amounts of relevant sample data, it can modify the skeletal structure of the drug-like molecule, thereby increasing the diversity of the chemical spatial structure of the simulated drug molecule. Consequently, the skeletal structure of the antagonist determined based on the simulated drug molecule differs significantly from the skeletal structure of existing drugs, which is beneficial for improving drug diversity.

[0037] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0038] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0039] Figure 1 A schematic diagram of an exemplary system architecture to which the technical solutions of the embodiments of this disclosure can be applied is shown.

[0040] Figure 2 The diagram illustrates a flowchart of an artificial intelligence-based method for predicting antagonists in an exemplary embodiment of this disclosure.

[0041] Figure 3 A flowchart illustrating a method for predicting antagonists based on artificial intelligence in another exemplary embodiment of this disclosure is shown.

[0042] Figure 4This diagram illustrates the structure of a generator network for generating adversarial neural networks in an exemplary embodiment of this disclosure.

[0043] Figure 5 This diagram illustrates the structure of a discriminator network for a generative adversarial neural network in an exemplary embodiment of the present disclosure.

[0044] Figure 6 This diagram illustrates a flowchart of a method for training a generator network that generates an adversarial neural network in an exemplary embodiment of this disclosure.

[0045] Figure 7 This diagram illustrates a flowchart of a method for training a generator network that generates an adversarial neural network in another exemplary embodiment of this disclosure.

[0046] Figure 8 The diagram shows a flowchart of a virtual screening method in an exemplary embodiment of this disclosure.

[0047] Figure 9 This illustration shows a flowchart of a biopsy method for a target drug small molecule in an exemplary embodiment of this disclosure.

[0048] Figure 10 This diagram illustrates the structure of an apparatus for predicting antagonists based on artificial intelligence, according to an exemplary embodiment of this disclosure.

[0049] Figure 11 A schematic diagram of the structure of an electronic device in an exemplary embodiment of this disclosure is shown. Detailed Implementation

[0050] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be more thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art.

[0051] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0052] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0053] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0054] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0055] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0056] Computer vision (CV) is a science that studies how to enable machines to "see." More specifically, it refers to machine vision, which uses cameras and computers to replace human eyes in tasks such as target recognition, tracking, and measurement, and further performs image processing to create images more suitable for human observation or transmission to instruments. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems capable of extracting information from images or multidimensional data. Computer vision technologies typically include image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, 3D object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping (SLAM), and common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0057] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, transfer learning, inductive learning, and instructional learning.

[0058] The solutions provided in this disclosure relate to technologies such as computer vision and machine learning in artificial intelligence, and are specifically illustrated through the following embodiments:

[0059] This technical solution provides a method, apparatus, antagonist, medium, and device for predicting antagonists based on artificial intelligence. Firstly, Figure 1 A schematic diagram of an exemplary system architecture to which the technical solutions of the embodiments of this disclosure can be applied is shown.

[0060] like Figure 1 As shown, the system architecture 100 may include a terminal 110, a network 120, and a server 130. The terminal 110 and the server 130 are connected through the network 120.

[0061] Terminal 110 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Network 120 can be a communication medium of various connection types that can provide a communication link between terminal 110 and server 130, such as a wired communication link, a wireless communication link, or a fiber optic cable, etc., which are not limited herein. Server 130 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, and big data and artificial intelligence platforms.

[0062] In this technical solution, an antagonist is a compound that, upon binding to a receptor molecule, does not produce a biological effect itself but blocks the agonist-mediated action of that receptor molecule. A receptor molecule is a compound molecule that binds to an antagonist, thereby blocking agonist-mediated action. A drug-like molecule is a compound molecule that is similar to a known drug.

[0063] For example, a user can send the name or identifier of a target receptor molecule to the server from terminal 120. For instance, the target receptor molecule is human glucagon. Server 130 can provide training for a generative adversarial neural network model; specifically, server 130 acquires drug-like molecules and existing antagonists corresponding to human glucagon to train the generative adversarial neural network.

[0064] For example, the trained generative adversarial neural network model can also be stored on the server 130. The server 130 then obtains simulated drug molecules based on the generator network of the trained generative adversarial neural network. Since the aforementioned generative adversarial neural network is trained on a massive amount of relevant sample data, the deep learning model can modify the skeletal structure of the drug-like molecule, thereby increasing the diversity of the chemical spatial structure of the simulated drug molecules and overcoming the limitation of molecular chemical spatial structures in related technologies that rely solely on existing drug databases.

[0065] Furthermore, to enable small molecules to directly search for suitable conformations within the protein's active pocket, the following exemplary solution is provided. For example, server 130 determines a database of small molecules to be screened based on the simulated drug molecules described above. A virtual screening process is then performed on this small molecule database based on the target receptor molecule to select the target drug small molecule. This solves the problem that deep neural learning models struggle to grow 3D drug molecules within the target receptor's active pocket.

[0066] Ultimately, the antagonist of the target receptor molecule can be obtained by performing cell activity tests on the aforementioned target drug small molecules.

[0067] Furthermore, the artificial intelligence-based antagonist prediction method in this disclosure can also be applied to a terminal. This disclosure does not impose any specific limitations on this. This disclosure primarily illustrates the application of the above method to the server-side 130.

[0068] The following section introduces the artificial intelligence-based method for predicting antagonists provided in this technical solution. Among other things, Figure 2 This diagram illustrates a flowchart of a method for predicting antagonists based on artificial intelligence in an exemplary embodiment of this disclosure. (See reference...) Figure 2 The method for predicting antagonists based on artificial intelligence provided in this embodiment includes:

[0069] Step S210: A simulated drug molecule is obtained from the generator network of the trained generative adversarial neural network, wherein the generative adversarial neural network is trained based on the drug-like molecule and the existing antagonists corresponding to the target receptor molecule.

[0070] Step S220: Determine the database of small molecules to be screened based on the simulated drug molecules;

[0071] Step S230: Perform virtual screening on the small molecule database based on the target receptor molecule to screen out the target drug small molecule from the small molecule database;

[0072] Step S240: Obtain the antagonist of the target receptor molecule by performing a cell activity test on the target drug small molecule.

[0073] Figure 2 The AI-based antagonist prediction scheme provided in the illustrated embodiment, on the one hand, uses a deep learning model to obtain simulated drug molecules of the target receptor molecule, thereby increasing the diversity of the chemical spatial structure of the simulated drug molecules. Consequently, the antagonist scaffold structure determined based on these simulated drug molecules differs significantly from the scaffold structures of existing drugs, which is beneficial for improving drug diversity. On the other hand, a database of small molecules to be screened is determined based on the simulated drug molecules, and this database is virtually screened based on the target receptor molecule. This allows the target drug small molecule to directly search for suitable conformations in the protein activity pocket of the target receptor, thus helping to ensure the rationality of the designed antagonist.

[0074] In the following exemplary embodiments, the target receptor molecule is described using human glucagon as an example. Of course, the target receptor molecule can also be a specific type of receptor, which is not limited here.

[0075] For example, Figure 3 A flowchart illustrating a method for predicting antagonists based on artificial intelligence in another exemplary embodiment of this disclosure is shown. (See reference...) Figure 3 The method for predicting antagonists includes: a simulated drug molecule acquisition part 31, a virtual screening part 32 for simulated drug molecules, and a biopsy part 33 for the screened target drug molecules. The simulated drug molecule acquisition part 31 can be determined based on a trained Generative Adversarial Network (GAN). Specifically, graph encoding is performed based on the chemical formula of the drug-like molecule to obtain its corresponding graph representation. For example, atoms are represented as points on the graph, and covalent bonds are represented as lines connecting the points. Further, the graph representation corresponding to the drug-like molecule is input into the generator network of the trained GAN, and the graph representation corresponding to the simulated drug molecule can be obtained based on the output of the generator network.

[0076] In this exemplary embodiment, the network structure of the GAN used in this embodiment will be described first. Specifically, Figure 4 and Figure 5 The network structures of the generator network and the discriminator network of a GAN according to an embodiment of the present disclosure are shown respectively.

[0077] refer to Figure 4 The generator network provided in this embodiment may include convolutional layers, pooling layers, feature supplementation layers, deconvolutional layers, and feature normalization layers. In the neural network architecture, convolutional and pooling operations can be used to extract deep features from the aforementioned drug-like molecule graph representation, enabling the learning of logical topological relationships within the molecule graph through these deep features. For example, vertices in the molecule graph represent atoms, and the side lengths represent bond lengths. The model is trained to learn the variation patterns of vertices and side lengths, as well as the relationships between vertices and side lengths. However, compared to the input drug-like molecule graph representation, multiple convolutional and pooling operations reduce the feature image of the molecule graph, resulting in information loss. Therefore, in this embodiment, to reduce information loss, for each downsampling (e.g., ... Figure 4 In the middle, from 41 to 410), a corresponding upsampling is performed (i.e., feature supplementation layer, deconvolution layer, such as...). Figure 4 In this embodiment, the upsampling parameters and downsampling parameters in the generator network are equal, resulting in image downsampling during the upsampling stage and corresponding image upsampling during the downsampling stage. In other words, the generator network in this embodiment adopts a Unet network structure (e.g., ...). Figure 4 The dashed lines in the image form a "U" shape, which can reduce the loss of original information during network transmission. This improves the phenomenon that the image content output by the generator is inconsistent with the image content input to the generator due to the mismatch between the downsampling parameters and the upsampling process parameters, and ultimately helps to improve the prediction accuracy of simulated drug molecules.

[0078] In an exemplary embodiment, Figure 4 The generator network shown can use the tanh function as the activation function. Specifically, the tanh function can be used to map the output data upsampled by 420 to the range of -1 to 1.

[0079] In an exemplary embodiment, reference is made to Figure 5 The discriminator network of the GAN provided in this embodiment includes multiple convolutional layers (such as...). Figure 5 (501-510 in the example). It should be noted that Wasserstein GAN (WGAN) can also be used in this exemplary technical solution. In this case, the discriminator network does not use the sigmoid function as the activation function in downsampling 510, thus effectively avoiding the vanishing gradient loss of the generator network.

[0080] In an exemplary embodiment, Figure 4 and Figure 5The network structures of the generator and discriminator networks shown are for illustrative purposes only. In practice, the network structures can be adjusted according to actual needs, thus the network structure of the neural network used to generate simulated drug molecules in this disclosure is scalable.

[0081] Based on the network structure of the GAN described above Figure 6 The flowchart shown is a schematic diagram of a training method for a generative adversarial neural network in an exemplary embodiment of the present disclosure, including steps S610-S630.

[0082] In step S610, a drug-like molecule is obtained, and an existing antagonist corresponding to the target receptor molecule is obtained as a training sample for generating an adversarial neural network.

[0083] For example, in the case where the target receptor molecule is human glucagon (hGCGR), existing hGCGR antagonists are obtained, such as those that have already been reported. A large number of drug-like molecules are also obtained. For instance, one million drug-like molecules are obtained from a commercial or open-source small molecule database (e.g., ZINC). Further, the existing hGCGR antagonists and the aforementioned one million drug-like molecules are used as a training sample set.

[0084] Before inputting the training samples into the GAN, graph encoding is performed on each sample to obtain the corresponding graph representation. For example, to facilitate processing in the GAN, each drug-like molecule / existing antagonist in the training sample set is graph-encoded using an adjacency tensor (A) and an annotation matrix (X). For instance, a carbon atom of cyclopentane can be encoded as a vertex of a pentagon, with the edges of the pentagon corresponding to the covalent bonds in the cyclopentane molecule (such as...). Figure 3 (See section 31 for simulated drug molecule acquisition).

[0085] In step S620, the drug-like molecules in the training samples are input into the generator network in the generative adversarial neural network, and the molecular structure of the drug-like molecules is predicted based on the generator network to obtain the predicted drug molecules.

[0086] In an exemplary embodiment, Figure 7 A schematic diagram of the framework of a model training method according to an embodiment of the present disclosure is shown. (Reference) Figure 7 The generative adversarial neural network consists of a generator network G710 and a discriminator network D720.

[0087] The generator network G 710 is used to perform molecular structure prediction processing (e.g., convolution, deconvolution, and normalization) on drug-like molecule A (specifically, the graph representation corresponding to drug-like molecule A), and outputs the result as the predicted drug molecule A' (specifically, the graph representation corresponding to predicted drug molecule A'). The training objective of the generator network G 710 is to obtain a predicted drug molecule that the discriminator network D 720 classifies as "true" for drug-like molecule A. That is, the probability that the discriminator network D 720 predicts the predicted drug molecule A' as an antagonist of the target receptor molecule is greater than 0.5. In other words, the training objective of the generator network G 710 is to make the predicted drug molecule as close as possible to existing antagonists, achieving a near-perfect deception (the discriminator network D 720 classifies the predicted drug molecule A' as a "true" antagonist of the target receptor molecule).

[0088] Continue to refer to Figure 6 In step S630, the predicted drug molecule and the existing antagonists in the training samples are input into the discriminator network of the generative adversarial neural network, and the parameters of the generative adversarial neural network are optimized based on the output of the discriminator network.

[0089] In one exemplary embodiment, the predicted drug molecule A' (specifically, the graphical representation corresponding to the predicted drug molecule A') and the existing antagonist B (specifically, the graphical representation corresponding to the existing antagonist B) are input into the discriminator network D 720 to determine whether the predicted drug molecule A' is an antagonist of the target receptor molecule. Specifically, if the discriminator network D 720 determines that the predicted drug molecule A' is an antagonist of the target receptor molecule, it indicates that the predicted drug molecule A' generated by the current generator has high accuracy (i.e., it can achieve a convincing effect), and the optimization of the generator network parameters can be stopped. If the discriminator network D 720 determines that the predicted drug molecule A' is not an antagonist of the target receptor molecule, it indicates that the accuracy of the predicted drug molecule A' generated by the current generator needs to be further improved (i.e., it cannot currently achieve a convincing effect), and the optimization of the generator network parameters needs to continue.

[0090] In another exemplary embodiment, the output of the discriminator network includes a first discrimination result regarding the predicted drug molecule and a second discrimination result regarding the existing antagonist. The first discrimination result includes the probability, output by the discriminator network, that the predicted drug molecule A' is an antagonist of the target receptor molecule; the second discrimination result includes the probability, output by the discriminator network, that the existing antagonist B is an antagonist of the target receptor molecule.

[0091] Specifically, we will continue to refer to Figure 7The discriminator network D 720 receives the predicted drug molecule A' (specifically, the graph representation corresponding to the predicted drug molecule A') and the existing antagonist B (specifically, the graph representation corresponding to the existing antagonist B). The discriminator network D 720 predicts the probability that the predicted drug molecule A' is an antagonist of the target receptor molecule by comparing the predicted drug molecule A' and the existing antagonist B (i.e., the first discrimination result). If the probability value is greater than 0.5, the first discrimination result is: the predicted drug molecule A' is an antagonist of the target receptor molecule, that is, the discrimination result output by the discriminator network is: the predicted drug molecule A' is "true"; correspondingly, if the probability value is not greater than 0.5, the first discrimination result is: the predicted drug molecule A' is not an antagonist of the target receptor molecule, that is, the discrimination result output by the discriminator network is: the predicted drug molecule A' is "false". Simultaneously, the discriminator network D 720 also predicts the probability that the existing antagonist B is an antagonist of the aforementioned target receptor molecule (i.e., the second discrimination result mentioned above). If the probability value is greater than 0.5, it indicates that the second discrimination result is: the existing antagonist B is an antagonist of the aforementioned target receptor molecule, that is, the discrimination result output by the discriminator network is: the existing antagonist B is "true"; correspondingly, if the probability value is not greater than 0.5, it indicates that the second discrimination result is: the existing antagonist B is not an antagonist of the aforementioned target receptor molecule, that is, the discrimination result output by the discriminator network is: the existing antagonist B is "false".

[0092] The training objective of the discriminator network D720 is to: distinguish existing antagonists as "true" (the probability value of the above second discrimination result is greater than 0.5), and distinguish predicted drug molecules as "false" (the probability value of the above first discrimination result is not greater than 0.5).

[0093] In this embodiment, the parameters of the generative adversarial neural network are further optimized based on the first discrimination result and the second discrimination result.

[0094] In the exemplary embodiments, reference continues to be made to... Figure 7 The first and second discrimination results are determined based on the difference in logical topology between the predicted drug molecule A' and the existing antagonist B. The parameters of the generator network G 710 and the discriminator network D 720 are then optimized based on these results. Specifically:

[0095] In this embodiment, the following loss function can be defined:

[0096]

[0097] Where, x i z iThese are the data corresponding to the existing antagonist B and the data corresponding to the predicted drug molecule A', respectively. The value of i is a positive integer not greater than m, where m is the number of samples.

[0098] The optimization objective of the above loss function is:

[0099] min G max D Loss

[0100] In this embodiment, the parameters of the generator network G are first fixed, and the parameters of the discriminator network D are adjusted. The parameters of the discriminator D are optimized as follows:

[0101]

[0102] Where D(x) i ) represents the second discrimination result mentioned above (i.e., the probability that the existing antagonist B is an antagonist of the target receptor molecule), and the larger the value, the better; D(G(z) i )) represents the first discrimination result mentioned above (i.e., the probability that drug molecule A' is an antagonist of the target receptor molecule), and its value is better the smaller it is (that is, log(1-D(G(z))). i The bigger the better.

[0103] For example, through max D Loss causes the discriminator network D to output the following first discrimination result: the probability that drug molecule A' is an antagonist of the target receptor molecule is no greater than 0.5, and the second discrimination result: the probability that the existing antagonist B is an antagonist of the target receptor molecule is greater than 0.5.

[0104] Then, with the parameters of the discriminator D fixed, the parameters of the generator network G are adjusted. Since the generator network G does not involve the existing antagonist B (corresponding to data x),... i Therefore, the optimization of the generator network G parameters does not involve the processing of logD(x). i Specifically, the parameters of generator G are optimized as follows:

[0105]

[0106] Among them, D(G(z) i )) represents the first discrimination result mentioned above (i.e., the probability of predicting that drug molecule A' is an antagonist of the target receptor molecule), and its value is better the larger it is (that is, log(1-D(G(z))). i The smaller the better.

[0107] For example, by min GLoss inputs the predicted drug molecule A' output by the parameter-adjusted generator G into the discriminator D. The first discrimination result output by the discriminator D is: the probability that the above predicted drug molecule A' is an antagonist of the target receptor molecule is greater than 0.5.

[0108] Through continuous iterative calculations, the aforementioned loss value is ensured to meet the preset requirements. Furthermore, the trained GAN's generator network G (which can be denoted as a "molecular graph growth model") can grow simulated drug molecules similar to those in the training set.

[0109] In an exemplary embodiment, when using WGAN for the above model prediction, the EarthMover (EM) distance can be used as the loss function. Since the log value is no longer calculated in the loss function, the training can be more easily converged.

[0110] The specific implementation method for determining the simulated drug molecule based on the molecular graph growth model (i.e., the generator network of the trained GAN or WGAN) is as follows: obtain the chemical formula of the target drug molecule and encode it into a graph to obtain the corresponding graph representation; input the graph representation of the target drug molecule into the molecular graph growth model, predict the molecular structure based on the molecular graph growth model, and output the graph representation of the simulated drug molecule.

[0111] It should be noted that the molecular graph growth model above outputs a graph representation of the simulated drug molecule. In order to facilitate the virtual screening in the later stage, it is necessary to decode the simulated drug molecule to obtain a standard format, such as SMILES (Simplified Molecular Input Line Entry System, a linear symbol used to input and represent molecular reactions) or SDF (Structure Data File) format.

[0112] Therefore, the encoded simulated drug molecule output by the above molecular graph growth model is decoded to obtain the chemical formula of the simulated drug molecule that meets the preset format.

[0113] Continue to refer to Figure 3 After the simulated drug molecule acquisition section 31, the virtual screening section 32 for simulated drug molecules is entered. The following describes an example of virtual screening of simulated drug molecules.

[0114] For example, continue to refer to Figure 2 In step S220, a database of small molecules to be screened is determined based on the simulated drug molecules described above; and in step S230, the database of small molecules is subjected to virtual screening based on the target receptor molecules in order to screen out the target drug small molecules from the database of small molecules.

[0115] The aforementioned small molecule database is used for virtual screening. Specifically, 10 million simulated drug molecules are generated using the molecular graph growth model, and their standard format is obtained through decoding. Further, to facilitate the direct search for suitable conformations within the protein active pocket by the simulated drug molecules, the simulated drug molecule format needs to be batch-converted into three-dimensional simulated drug molecules (e.g., mol2 format). In an exemplary embodiment, two-dimensional simulated drug molecules are converted into three-dimensional molecules using force field transformation in chemical simulation software, resulting in the aforementioned small molecule database, which can be used as a drug database for virtual screening.

[0116] In an exemplary embodiment, for comparison, 100,000 non-simulated drug molecules can be randomly selected from an existing database (e.g., the ChemBridge database). Furthermore, these non-simulated drug molecules are incorporated into the aforementioned small molecule database to form a virtual screening drug database.

[0117] In an exemplary embodiment, given the principle of virtual screening—screening candidate compounds from a small molecule database based on the shape, size, electrostatic properties, and other characteristics of the protein active pocket of the target (target receptor molecule)—it is necessary to pre-treat the target receptor molecule before performing virtual screening on the aforementioned new molecule database to ensure its successful execution. This pre-treatment could include processes such as amino acid residue completion, hydrogenation, or charge addition.

[0118] For example, the human glucagon receptor molecule hGCGR (PDB number: 5EE7) is downloaded from an existing database (e.g., a PDB database) as the target receptor molecule. Further, the target receptor molecule hGCGR is pretreated using the following methods: The software completes the amino acid residues of the hGCGR protein molecule and performs hydrogenation and charge addition treatments on hGCGR. Furthermore, the pre-processed hGCGR protein molecule is saved in a three-dimensional format (e.g., mol2 format).

[0119] It is important to note that, in order to ensure the number of virtual screenings, the format of the target receptor molecules should be unified with the format of the drug molecules in the small molecule database, such as unifying them to the mol2 format.

[0120] In an exemplary embodiment, reference is made to Figure 8Based on molecular docking technology, a virtual screening process is performed on the aforementioned small molecule database and the aforementioned three-dimensional target receptor molecules. The molecular docking technology is based on the "lock and key" model principle, where the receptor protein 81 is equivalent to the "lock" and the small molecule 82 is equivalent to the "key." Virtual screening is achieved by checking whether the two can bind 83. Specifically, using the target receptor molecule hGCGR as the target site, an optimized search algorithm is used to calculate the affinity between drug molecules in the aforementioned small molecule database and the target site based on the shape, size, and electrostatic properties of the target's protein active pocket. This allows for the screening of target drug molecules from the small molecule database as the corresponding targets for the target receptor molecule hGCGR, i.e., candidate compounds.

[0121] Furthermore, based on whether the small molecules in the aforementioned small molecule database are potential target drugs, each small molecule is scored using a scoring function, and the small molecules are sorted according to their scores to obtain a number of small molecules with higher scores (e.g., 30, numbered C1-C30 from highest to lowest score) (denoted as "target drug small molecules").

[0122] Continue to refer to Figure 3 After the simulated drug molecule acquisition section 31 and the virtual screening section 32 for the simulated drug molecules, the next step is to perform biopsy on the screened target drug small molecules (as shown in Table 1). The following describes an example of performing biopsy on the screened target drug small molecules.

[0123] For example, continue to refer to Figure 2 In step S240, the antagonist of the target receptor molecule is obtained by performing a cell activity test on the target drug small molecule.

[0124] In an exemplary embodiment, for each target drug small molecule, it is mixed with a binding agent and test cells containing the target receptor molecule. The binding agent binds to the target receptor molecule to activate the target receptor molecule signaling pathway and releases the test target component. The target drug small molecule acts as an antagonist to inhibit the activity of the target receptor molecule. Simultaneously, an antagonist drug for the target receptor molecule is selected as a positive control. Further, the test target component corresponding to each target drug small molecule is obtained, and the half-maximum inhibitory concentration (WMC) value of each target drug small molecule is calculated based on the test target component. Finally, the antagonist of the target receptor molecule is determined in the cedar target drug small molecule based on the WMC value.

[0125] For example, in the case where the target receptor molecule is human glucagon, Figure 9 The following is a specific implementation method for testing the cell activity of the above-mentioned target drug small molecules:

[0126] In step S910, an antagonist drug for hGCGR is selected as a positive control. Each of the target drug small molecules, glucagon peptide fragments, and human embryonic kidney cells containing the target receptor molecule are mixed. The glucagon peptide fragments bind to hGCGR to activate the hGCGR signaling pathway and release the second messenger cyclic adenosine monophosphate (cAMP). The target drug small molecules play an antagonistic role to inhibit the activity of hGCGR.

[0127] In an exemplary embodiment, an hGCGR antagonist drug (e.g., L-168,049) was selected as a positive control. Given that human embryonic kidney cells (e.g., HEK293) have human glucagon receptors on their cell membranes, they can be used to test the drug activity of human glucagon antagonists. In this embodiment, the target drug molecule, a glucagon peptide, and HEK293 cells containing the target receptor molecule are mixed. The glucagon peptide binds to hGCGR to activate the hGCGR signaling pathway, thereby promoting the release of the second messenger cAMP. The target drug molecule antagonizes hGCGR activity.

[0128] In step S920, the second messenger cAMP is detected by homogeneous time-resolved fluorescence HTRF, and the half-inhibitory concentration (WIC) of each target drug molecule is calculated based on the intensity value of the HTRF.

[0129] In an exemplary embodiment, homogeneous time-resolved fluorescence (HTRF) technology was used to test the activity of each target drug molecule (as shown in Table 1, comprising 30 target drug molecules). Specifically, after the glucagon peptide activates the glucagon receptor, the intracellular second messenger cyclic adenosine monophosphate (cAMP) is released. HTRF can detect the second messenger cAMP, and the half-maximal inhibitory concentration (IC50) of the drug molecule can be calculated based on the fluorescence intensity.

[0130] Step S930: Determine the antagonist of the target receptor molecule in the target drug small molecule based on the half-inhibition concentration value.

[0131] In an exemplary embodiment, for the 30 target drug molecules shown in Table 1, 11 identical concentration gradients were set for each target drug molecule, and the activity of each target drug molecule was determined in triplicate using the HTRF second messenger cyclic adenosine monophosphate (cAMP) response assay. The test results are shown in Table 1.

[0132] Table 1

[0133]

[0134]

[0135] Table 2 shows the half-inhibitory concentrations (IC50) of the 30 target drug small molecules tested. For example, the compounds labeled C13 and C14 have relatively small IC50 values ​​of 446.5 nM and 422.5 nM, respectively, and can be considered as novel human glucagon antagonists.

[0136] This technical solution combines the advantages of deep learning and virtual screening. It uses a deep learning model to predict simulated drug molecules, increasing the diversity of their chemical spatial structures. Consequently, the antagonist skeleton structure determined based on these simulated drug molecules differs significantly from the skeleton structures of existing drugs, thus enhancing drug diversity. Simultaneously, a database of small molecules to be screened is established based on these simulated drug molecules, and this database is virtually screened using target receptor molecules. This allows the target drug small molecule to directly search for suitable conformations within the protein activity pocket of the target receptor, thereby ensuring the rationality of the designed antagonist. Therefore, this technical solution can reduce drug development costs and time while improving drug diversity.

[0137] This disclosure provides an antagonist whose backbone molecule is a simulated drug molecule generated based on a trained generative adversarial neural network, wherein the simulated drug molecule is determined after virtual screening based on a target receptor molecule. Specific embodiments corresponding to the simulated drug molecule generated based on the trained generative adversarial neural network have been described in the section on methods for predicting antagonists based on artificial intelligence, and will not be repeated here. Similarly, specific embodiments of virtual screening based on target receptor molecules have also been described in the section on methods for predicting antagonists based on artificial intelligence, and will not be repeated here either.

[0138] Those skilled in the art will understand that all or part of the steps of the above embodiments are implemented as a computer program executed by a processor (including a GPU / CPU). When the computer program is executed by the GPU / CPU, it performs the functions defined by the methods provided in this disclosure. The program can be stored in a computer-readable storage medium, such as a read-only memory, a magnetic disk, or an optical disk.

[0139] Furthermore, it should be noted that the above figures are merely illustrative representations of the processes included in the methods according to exemplary embodiments of this disclosure, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0140] The following is passed Figure 10The present disclosure provides an embodiment of an artificial intelligence-based device for predicting antagonists, which can be used to perform the training method of the target model described above.

[0141] Figure 10 A schematic diagram of the structure of an apparatus for predicting antagonists based on artificial intelligence in an exemplary embodiment of this disclosure is shown. Figure 10 As shown, the aforementioned artificial intelligence-based antagonist prediction device 1000 includes: a simulated drug molecule determination module 1001, a small molecule database determination module 1002, a virtual screening module 1003, and an antagonist determination module 1004.

[0142] The simulated drug molecule determination module 1001 is configured to: obtain simulated drug molecules based on the generator network of a trained generative adversarial neural network, wherein the generative adversarial neural network is trained based on existing antagonists corresponding to the drug-like molecules and the target receptor molecules; the small molecule database determination module 1002 is configured to: determine a small molecule database to be screened based on the simulated drug molecules; the virtual screening module 1003 is configured to: perform virtual screening on the small molecule database based on the target receptor molecules to screen out target drug small molecules from the small molecule database; and the antagonist determination module 1004 is configured to: obtain the antagonist of the target receptor molecules by performing cell activity tests on the target drug small molecules.

[0143] In some embodiments of this disclosure, based on the foregoing scheme, the above-mentioned artificial intelligence-based antagonist prediction device 1000 further includes: a training sample acquisition module, a molecular structure prediction module, a discrimination result determination module, and a parameter optimization module.

[0144] The training sample acquisition module is configured to: acquire drug-like molecules and existing antagonists corresponding to the target receptor molecules as training samples for the generative adversarial neural network; the molecular structure prediction module is configured to: input the drug-like molecules in the training samples into the generator network of the generative adversarial neural network, and perform molecular structure prediction on the drug-like molecules based on the generator network to obtain predicted drug molecules; the discrimination result determination module is configured to: input the predicted drug molecules and existing antagonists in the training samples into the discriminator network of the generative adversarial neural network; and the parameter optimization module is configured to: optimize the parameters of the generative adversarial neural network based on the output of the discriminator network.

[0145] In some embodiments of this disclosure, based on the foregoing scheme, the parameter optimization module is specifically configured as follows: the output of the discriminator network includes a first discrimination result regarding the predicted drug molecule and a second discrimination result regarding the existing antagonist, wherein the first discrimination result is the probability that the predicted drug molecule output by the discriminator network is an antagonist of the target receptor molecule; the second discrimination result includes the probability that the existing antagonist output by the discriminator network is an antagonist of the target receptor molecule; and the parameters of the generative adversarial neural network are optimized based on the first discrimination result and the second discrimination result.

[0146] In some embodiments of this disclosure, based on the foregoing scheme, the parameter optimization module is specifically configured as follows: fixing the parameters of the generator network and adjusting the parameters of the discriminator network, so that the first discrimination result is: the probability that the predicted drug molecule is an antagonist of the target receptor molecule is not greater than 0.5, and the second discrimination result is: the probability that the existing antagonist is an antagonist of the target receptor molecule is greater than 0.5; fixing the parameters of the discriminator network and adjusting the parameters of the generator network, so that after the predicted drug molecule output by the generator network with adjusted parameters is input into the discriminator network, the first discrimination result is: the probability that the predicted drug molecule is an antagonist of the target receptor molecule is greater than 0.5, and the second discrimination result is: the probability that the existing antagonist is an antagonist of the target receptor molecule is greater than 0.5.

[0147] In some embodiments of this disclosure, based on the foregoing scheme, the number of times the data is downsampled in the generator network is the same as the number of times the data is upsampled. The downsampling process includes convolution and pooling, and the upsampling process includes deconvolution and feature supplementation.

[0148] In some embodiments of this disclosure, based on the foregoing scheme, the generator network uses tanh as the activation function.

[0149] In some embodiments of this disclosure, based on the foregoing scheme, the molecular structure prediction module is specifically configured to: perform graph encoding on the chemical formulas of the drug-like molecules in the training samples to obtain graph representations corresponding to the drug-like molecules, and input the graph representations corresponding to the drug-like molecules into the generator network; the discrimination result determination module is specifically configured to: perform graph encoding on the chemical formulas of the existing antagonists to obtain graph representations corresponding to the existing antagonists; and input the graph representations corresponding to the predicted drug molecules and the existing antagonists into the discriminator network.

[0150] In some embodiments of this disclosure, based on the foregoing scheme, the simulated drug molecule determination module 1001 is configured to: perform graph encoding on the chemical formula of the target drug molecule to obtain a graph representation corresponding to the target drug molecule; input the graph representation corresponding to the target drug molecule into a trained generator network, perform molecular structure prediction based on the generator network, and output a graph representation of the simulated drug molecule; and decode the graph representation of the simulated drug molecule to obtain a chemical formula of the simulated drug molecule that satisfies a preset format.

[0151] In some embodiments of this disclosure, based on the foregoing scheme, the small molecule database determination module 1002 is specifically configured to: perform three-dimensional transformation on the simulated drug molecules output by the generator network to obtain three-dimensional simulated drug molecules, thereby obtaining the small molecule database.

[0152] In some embodiments of this disclosure, based on the foregoing scheme, the small molecule database determination module 1002 includes: a decoding unit.

[0153] The decoding unit is configured to decode the encoded simulated drug molecules output by the generator network before the three-dimensional conversion unit performs three-dimensional conversion on the simulated drug molecules output by the generator network to obtain three-dimensional simulated drug molecules.

[0154] In some embodiments of this disclosure, based on the foregoing scheme, the virtual screening module 1003 is specifically configured to: use the target receptor molecule as the target site, and calculate the affinity between the drug molecule in the small molecule database and the target site based on the shape, size, and electrostatic properties of the protein active pocket of the target site using an optimized search algorithm, so as to screen out the target drug small molecule as a candidate compound of the target receptor molecule in the small molecule database.

[0155] In some embodiments of this disclosure, based on the foregoing scheme, the above-mentioned artificial intelligence-based antagonist prediction device 1000 further includes a preprocessing module.

[0156] The preprocessing module is configured to: before the virtual screening module 1003 performs virtual screening on the small molecule database based on the target receptor molecule, obtain the target receptor molecule from the existing database and perform one or more of the following treatments on the target receptor molecule: amino acid residue completion treatment, hydrogenation treatment, and charge treatment.

[0157] In some embodiments of this disclosure, based on the foregoing scheme, the above-mentioned artificial intelligence-based antagonist prediction device 1000 further includes: a format unification module.

[0158] The aforementioned format unification module is configured to unify the format of the target receptor molecules with the format of the drug molecules in the small molecule database before the virtual screening module 1003 performs virtual screening on the small molecule database based on the target receptor molecules.

[0159] In some embodiments of this disclosure, based on the foregoing scheme, the antagonist determination module is specifically configured as follows: for each of the target drug small molecules, it is mixed with a binder and test cells containing a target receptor molecule, wherein the binder binds to the target receptor molecule to activate the target receptor molecule signaling pathway and releases the test target component, wherein the target drug small molecule plays an antagonistic role to inhibit the activity of the target receptor molecule; an antagonist drug of the target receptor molecule is selected as a positive control; the test target component corresponding to each of the target drug small molecules is obtained, and the half-inhibition concentration (WIC) value of each of the target drug small molecules is calculated based on the test target component; and the antagonist of the target receptor molecule is determined based on the WIC value.

[0160] In some embodiments of this disclosure, based on the foregoing scheme, the antagonist determination module 1004 is further configured to: determine the target drug small molecules with half-inhibition concentration values ​​less than a preset threshold as antagonists of the target receptor molecule; or, sort the target drug small molecules according to their half-inhibition concentration values, and determine the N target drug small molecules with smaller half-inhibition concentration values ​​as antagonists of the target receptor molecule, where N is a positive integer.

[0161] In some embodiments of this disclosure, based on the foregoing scheme, the binding agent is a glucagon peptide, the target component is the second messenger cyclic adenosine monophosphate (cAMP), and the antagonist determination module 1004 is specifically configured to: select an antagonist drug of hGCGR as a positive control; mix each target drug molecule, the glucagon peptide, and human embryonic kidney cells containing the target receptor molecule; the glucagon peptide binds to hGCGR to activate the hGCGR signaling pathway to release the second messenger cyclic adenosine monophosphate (cAMP), wherein the target drug molecule plays an antagonistic role to inhibit the activity of hGCGR; detect the second messenger cAMP by homogeneous time-resolved fluorescence HTRF, and calculate the half-inhibition concentration (WIC) value of each target drug molecule based on the HTRF intensity value; and determine the antagonist of the target receptor molecule based on the WIC value.

[0162] The specific details of each unit in the aforementioned AI-based antagonist prediction device have been described in detail in the AI-based antagonist prediction method, and therefore will not be repeated here.

[0163] Figure 11A schematic diagram of the structure of a computer system suitable for implementing the embodiments of the present disclosure is shown.

[0164] It should be noted that, Figure 11 The computer system 1100 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0165] like Figure 11 As shown, the computer system 1100 includes a processor 1101, which specifically includes a Graphics Processing Unit (GPU) and a Central Processing Unit (CPU). The processor 1101 can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 1102 or programs loaded from storage portion 1108 into Random Access Memory (RAM) 1103. The RAM 1103 also stores various programs and data required for system operation. The processor 1101, ROM 1102, and RAM 1103 are interconnected via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0166] The following components are connected to I / O interface 1105: an input section 1106 including a keyboard, mouse, etc.; an output section 1107 including a cathode ray tube (CRT), liquid crystal display (LCD), and speakers, etc.; a storage section 1108 including a hard disk, etc.; and a communication section 1109 including a network interface card such as a local area network (LAN) card and a modem, etc. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to I / O interface 1105 as needed. Removable media 1111, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 1110 as needed so that computer programs read from them can be installed into storage section 1108 as needed.

[0167] In particular, according to embodiments of this disclosure, the processes described below with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1109, and / or installed from removable medium 1111. When the computer program is executed by processor 1101, it performs various functions defined in the system of this application.

[0168] It should be noted that the computer-readable medium shown in the embodiments of this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0169] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0170] The units described in the embodiments of this disclosure can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the unit itself.

[0171] In another aspect, this application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the methods described in the above embodiments.

[0172] For example, the electronic device described above can achieve the following: Figure 2 As shown: Step S210, obtaining simulated drug molecules based on the generator network of the trained generative adversarial neural network, wherein the generative adversarial neural network is trained based on the drug-like molecules and existing antagonists corresponding to the target receptor molecules; Step S220, determining a small molecule database to be screened based on the simulated drug molecules; Step S230, performing virtual screening on the small molecule database based on the target receptor molecules to screen out target drug small molecules from the small molecule database; and Step S240, obtaining the antagonist of the target receptor molecules by performing cell activity tests on the target drug small molecules.

[0173] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0174] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of this disclosure.

[0175] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0176] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for predicting antagonists based on artificial intelligence, characterized in that, include: The simulated drug molecule is obtained from the generator network of the trained generative adversarial neural network, which is trained based on the existing antagonists corresponding to the drug-like molecule and the target receptor molecule. The simulated drug molecules output by the generator network are converted into three-dimensional simulated drug molecules to obtain a database of small molecules to be screened. The small molecule database is virtually screened based on the target receptor molecule in order to screen out the target drug small molecule from the small molecule database; The antagonist of the target receptor molecule is obtained by performing a cell activity test on the target drug small molecule; The generative adversarial neural network is trained in the following manner: Obtain drug-like molecules and existing antagonists corresponding to the target receptor molecules as training samples for generating adversarial neural networks; The drug-like molecules in the training samples are input into the generator network in the generative adversarial neural network, and the molecular structure of the drug-like molecules is predicted based on the generator network to obtain the predicted drug molecules. The predicted drug molecule and existing antagonists in the training samples are input into a discriminator network in a generative adversarial neural network to obtain the output of the discriminator network; wherein, the output of the discriminator network includes a first discrimination result regarding the predicted drug molecule and a second discrimination result regarding the existing antagonists; the first discrimination result includes the probability, output by the discriminator network, that the predicted drug molecule is an antagonist of the target receptor molecule, and the second discrimination result includes the probability, output by the discriminator network, that the existing antagonist is an antagonist of the target receptor molecule; The parameters of the generative adversarial neural network are optimized based on the first and second discrimination results to obtain the trained generative adversarial neural network.

2. The method according to claim 1, characterized in that, Optimizing the parameters of the generative adversarial neural network based on the first discrimination result and the second discrimination result includes: The parameters of the generator network are fixed, and the parameters of the discriminator network are adjusted so that the first discrimination result is: the probability that the predicted drug molecule is an antagonist of the target receptor molecule is not greater than 0.5, and the second discrimination result is: the probability that the existing antagonist is an antagonist of the target receptor molecule is greater than 0.

5. By fixing the parameters of the discriminator network and adjusting the parameters of the generator network, the first discrimination result is obtained by inputting the predicted drug molecule output by the generator network after parameter adjustment into the discriminator network: the probability that the predicted drug molecule is an antagonist of the target receptor molecule is greater than 0.

5.

3. The method according to claim 1, characterized in that, The number of times the data is downsampled in the generator network is the same as the number of times the data is upsampled. The downsampling process includes convolution and pooling, and the upsampling process includes deconvolution and feature supplementation.

4. The method according to claim 1, characterized in that, The generator network uses tanh as the activation function.

5. The method according to claim 1, characterized in that, The process of inputting drug-like molecules from the training samples into the generator network of the generative adversarial neural network includes: performing graph encoding on the chemical formulas of drug-like molecules from the training samples to obtain graph representations corresponding to the drug-like molecules, and inputting the graph representations corresponding to the drug-like molecules into the generator network; The process of inputting the predicted drug molecule and the existing antagonists in the training samples into the discriminator network of the generative adversarial neural network includes: graph encoding the chemical formula of the existing antagonists to obtain the graph representation corresponding to the existing antagonists; and inputting the graph representation corresponding to the predicted drug molecule and the graph representation corresponding to the existing antagonists into the discriminator network.

6. The method according to claim 1, characterized in that, Simulated drug molecules are obtained from the generator network of the trained generative adversarial neural network, including: The chemical formula of the target drug molecule is graphically encoded to obtain the graphical representation of the target drug molecule; The graph representation corresponding to the target drug molecule is input into the trained generator network, and molecular structure prediction is performed based on the generator network to output the graph representation of the simulated drug molecule. The graphical representation of the simulated drug molecule is decoded to obtain the chemical formula of the simulated drug molecule that meets a preset format.

7. The method according to claim 1, characterized in that, Virtual screening of the small molecule database based on the target receptor molecule includes: Using the target receptor molecule as the target site, and based on the shape, size, and electrostatic properties of the protein active pocket of the target site, an optimized search algorithm is used to calculate the affinity between the drug molecules in the small molecule database and the target site, so as to screen the target drug small molecule as a candidate compound of the target receptor molecule from the small molecule database.

8. The method according to claim 7, characterized in that, Before performing virtual screening of the small molecule database based on the target receptor molecule, the method further includes: For obtaining target receptor molecules from existing databases, the target receptor molecules are subjected to one or more of the following treatments: amino acid residue completion treatment, hydrogenation treatment, and charge treatment.

9. The method according to claim 7, characterized in that, Before performing virtual screening of the small molecule database based on the target receptor molecule, the method further includes: The format of the target receptor molecule is unified with the format of the drug molecule in the small molecule database.

10. The method according to any one of claims 1 to 9, characterized in that, Antagonists of the target receptor molecule are obtained by performing cellular activity assays on the target drug small molecule, including: For each of the target drug small molecules, it is mixed with a binding agent and a test cell containing a target receptor molecule, wherein the binding agent binds to the target receptor molecule to activate the target receptor molecule signaling pathway and releases the test target component, wherein the target drug small molecule has an antagonistic effect to inhibit the activity of the target receptor molecule; Antagonist drugs targeting the receptor molecule were selected as positive controls; Obtain the test target component corresponding to each target drug small molecule, and calculate the half-inhibitory concentration value of each target drug small molecule based on the test target component; The antagonist of the target receptor molecule is determined based on the half-inhibition concentration value.

11. The method according to claim 10, characterized in that, The antagonist of the target receptor molecule is determined based on the half-inhibitory concentration value, including: The target drug small molecules with half-inhibition concentration values ​​less than a preset threshold are identified as antagonists of the target receptor molecule; or, the target drug small molecules are sorted according to their half-inhibition concentration values, and the N target drug small molecules with smaller half-inhibition concentration values ​​are identified as antagonists of the target receptor molecule, where N is a positive integer.

12. The method according to claim 10, characterized in that, The target receptor molecule is human glucagon hGCGR, the binding agent is a glucagon peptide, and the test target component is the second messenger cyclic adenosine monophosphate (cAMP). The antagonist of the target receptor molecule is obtained by performing a cellular activity assay on the target drug molecule, including: An antagonist drug for hGCGR was selected as a positive control. Each of the target drug small molecules, glucagon peptide fragments, and human embryonic kidney cells containing the target receptor molecule were mixed. The glucagon peptide fragments bound hGCGR to activate the hGCGR signaling pathway to release the second messenger cAMP. The target drug small molecules antagonized hGCGR activity. The second messenger cAMP was detected by homogeneous time-resolved fluorescence HTRF, and the half-inhibitory concentration (WIC) of each target drug molecule was calculated based on the intensity of the HTRF. The antagonist of the target receptor molecule is determined based on the half-inhibition concentration value.

13. An antagonist, characterized in that, include: The backbone molecule of the antagonist is a simulated drug molecule generated based on a trained generative adversarial neural network. The simulated drug molecule is determined by virtually screening a small molecule database based on target receptor molecules. The small molecule database is obtained by converting the simulated drug molecules output by the generator network into three-dimensional simulated drug molecules. The trained generative adversarial neural network is obtained by training using the method described in any one of claims 1 to 12.

14. A device for predicting antagonists based on artificial intelligence, characterized in that, The device includes: The simulated drug molecule determination module is configured to obtain simulated drug molecules based on the generator network of a trained generative adversarial neural network, wherein the generative adversarial neural network is trained based on the drug-like molecule and the existing antagonists corresponding to the target receptor molecule. The small molecule database determination module is configured to convert the simulated drug molecules output by the generator network into three-dimensional simulated drug molecules to obtain a small molecule database to be screened. The virtual screening module is configured to perform virtual screening on the small molecule database based on the target receptor molecule, so as to screen out the target drug small molecule from the small molecule database; The antagonist determination module is configured to: obtain the antagonist of the target receptor molecule by performing a cell activity test on the target drug small molecule; The training sample acquisition module is configured to: acquire drug-like molecules and acquire existing antagonists corresponding to the target receptor molecules as training samples for generating adversarial neural networks; The molecular structure prediction module is configured to: input drug-like molecules from the training samples into the generator network in the generative adversarial neural network, and perform molecular structure prediction on the drug-like molecules based on the generator network to obtain predicted drug molecules; The discrimination result determination module is configured to: input the predicted drug molecule and the existing antagonists in the training samples into the discriminator network of the generative adversarial neural network, and obtain the output of the discriminator network; wherein, the output of the discriminator network includes a first discrimination result regarding the predicted drug molecule and a second discrimination result regarding the existing antagonists; the first discrimination result includes the probability output by the discriminator network that the predicted drug molecule is an antagonist of the target receptor molecule, and the second discrimination result includes the probability output by the discriminator network that the existing antagonist is an antagonist of the target receptor molecule; The parameter optimization module is configured to optimize the parameters of the generative adversarial neural network based on the first discrimination result and the second discrimination result, so as to obtain the trained generative adversarial neural network.

15. A computer-readable storage medium, characterized in that, It contains computer programs; When the computer program is executed by a processor, it implements the method for predicting antagonists based on artificial intelligence as described in any one of claims 1 to 12.

16. An electronic device, characterized in that, The electronic device includes: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the artificial intelligence-based method for predicting antagonists as described in any one of claims 1 to 12.

17. A computer program product, characterized in that, The computer program product includes a computer program stored in a computer-readable storage medium, wherein a processor of a computer device reads from the computer-readable storage medium and executes the computer program, causing the computer device to perform the method for predicting antagonists based on artificial intelligence as described in any one of claims 1 to 12.

Images