Small molecule generation method, device, electronic device and storage medium

Abstract

The present disclosure provides a method, apparatus, electronic device and storage medium for generating small molecules, relating to the technical field of small molecule generation. The method includes: obtaining a small molecule safety knowledge graph, in which each node stores attribute data of an existing small molecule, and each first branch stores chemical relationship data between the existing small molecules corresponding to the nodes at both ends of the first branch; for each existing small molecule, extracting attribute data related to the toxicity type and function type of the existing small molecule from the corresponding node to construct an embedding vector; screening important nodes based on the embedding vector; and adjusting the generation preference of the small molecule generation model based on the important nodes so that it generates new small molecules that meet the set toxicity and function requirements. This method combines the small molecule safety knowledge graph with the knowledge preference optimization technology to ensure that the generated new molecules meet the pharmacodynamic standards and have high biological safety.

Description

Technical Field

[0001] The present disclosure relates to the technical field of small molecule generation, and in particular, to a small molecule generation method, device, electronic device, and storage medium. Background Art

[0002] In recent years, the design and optimization of small molecule drugs have become a research hotspot in the biomedical field, especially playing an important role in new drug development and personalized medicine. Traditional drug design methods rely on means such as empirical rules, chemical synthesis, and in vitro experiments, with high costs, low efficiency, and long R & D cycles.

[0003] In the design of small molecule drugs based on deep learning, the generation model can automatically identify and generate molecules with specific pharmacological activities and low toxicity, and is widely used in the design of small molecule drugs. For example, the solution disclosed in Chinese Patent CN116741309A converts small molecules into 3D molecular graphs, uses a diffusion model to generate new small molecules, and at the same time, uses a graph neural network to construct a discriminator to judge the generated small molecules to ensure that the generated molecules meet the structural requirements of real molecules. However, existing generation models, such as GANs, variational autoencoders (VAEs), etc., cannot guarantee the safety of generating new small molecules. Their inferences mainly focus on the rationality of molecular structures and pharmacological properties, but obviously insufficient consideration is given to the toxicity reactions and adverse biological activity evaluations that new small molecules may cause.

[0004] New small molecules may have unknown toxicity or produce adverse reactions with certain receptors in the human body, thus triggering health problems such as allergies and toxicity reactions. Therefore, how to real-time evaluate the potential safety during the generation of small molecules and ensure that the generated small molecules meet the expected biosafety standards has become a technical problem to be solved urgently. Especially when faced with complex small molecule structures and variable toxicity mechanisms, it is difficult to comprehensively evaluate the safety of generated small molecules. Summary of the Invention

[0005] The main objective of the present disclosure is to provide a small molecule generation method, device, electronic device, and storage medium to solve the technical problem that the new small molecules generated in related technologies do not have safety due to the failure to consider toxicity reactions and pharmacological functions simultaneously.

[0006] To achieve the above objective, the first aspect of the present disclosure provides a small molecule generation method, including:

[0007] Obtain a small molecule safety knowledge graph, wherein in the small molecule safety knowledge graph, attribute data of an existing small molecule is stored in each of a plurality of nodes, and chemical relationship data between the existing small molecules corresponding to the nodes at both ends of each of the plurality of first branches is stored in each of the first branches;

[0008] For each of the existing small molecules, extract the attribute data related to the toxicity type and function type of the existing small molecule from the corresponding node in the small molecule safety knowledge graph, and construct an embedding vector based on the extracted attribute data;

[0009] Based on the embedding vector, screen out important nodes of the small molecule safety knowledge graph; and,

[0010] Based on the important nodes, adjust the generation preference of a preset small molecule generation model so that the small molecule generation model generates new small molecules that meet the set toxicity requirements and set function requirements.

[0011] Further, the method further includes:

[0012] Predict the attribute data of the new small molecule based on the small molecule safety knowledge graph;

[0013] Based on the attribute data of the new small molecule obtained by the prediction, evaluate whether the new small molecule has potential toxicity characteristics and whether it meets the biosafety requirements;

[0014] If the new small molecule contains potential toxicity characteristics or does not meet the biosafety requirements, perform a rejection process on the new small molecule and generate the new small molecule again; and,

[0015] If the new small molecule does not contain potential toxicity characteristics and meets the biosafety requirements, output the new small molecule as the small molecule generation result.

[0016] Further, the attribute data is configured to include small molecule ontology terms, chemical structure data, toxicity type data, and action target data; and,

[0017] The chemical relationship data is configured to include first relationship data for characterizing whether there is a pharmacological effect between two adjacent existing small molecules, and second relationship data for characterizing whether one of two adjacent existing small molecules inhibits the other;

[0018] The existing small molecules are configured to include harmful small molecules and harmless small molecules.

[0019] Further, the obtaining of the small molecule safety knowledge graph includes:

[0020] Obtain multiple of the existing small molecules;

[0021] For each of the existing small molecules, respectively obtain the small molecule ontology term, toxicity type data, and action target data, construct nodes of the existing small molecules in the small molecule safety knowledge graph, and use the small molecule ontology term, the toxicity type data, and the action target data as child nodes in the nodes respectively;

[0022] Obtain the association edges between the small molecule ontology term and the toxicity type data and the action target data, and construct the second branches of the existing small molecules in the small molecule safety knowledge graph; and,

[0023] Obtain the chemical relationship data between the existing small molecules and other existing small molecules, and construct the first branches of the existing small molecules in the small molecule safety knowledge graph;

[0024] Generate the small molecule safety knowledge graph including all the nodes, the first branches, and the second branches.

[0025] Further, extracting the attribute data related to the toxicity type and function type of the existing small molecules from the corresponding nodes in the small molecule safety knowledge graph, and constructing an embedding vector based on the extracted attribute data, includes:

[0026] Embed the small molecule safety knowledge graph through a pre-selected knowledge graph embedding model to obtain an initial embedding vector for each of the existing small molecules, where the elements of the initial embedding vector include the attribute data related to the toxicity type and function type of the existing small molecules in the small molecule safety knowledge graph;

[0027] Obtain external knowledge related to the toxicity type and function type of the existing small molecules, and fuse the external knowledge with the initial embedding vector to obtain the finally constructed embedding vector.

[0028] Further, the knowledge graph embedding model is configured as a TransE model; and, screening out important nodes of the small molecule safety knowledge graph based on the embedding vector, includes:

[0029] Obtain the importance scores of each node in the small molecule safety knowledge graph through a pre-selected node importance scoring model, where the input of the node importance scoring model includes the embedding vector matching the node;

[0030] Screen out key nodes and nodes with importance scores exceeding a set threshold from all the nodes in the small molecule safety knowledge graph as the important nodes.

[0031] Further, the generation preference of a pre-set small molecule generation model is adjusted based on the important nodes, so that the small molecule generation model generates new small molecules that meet the set toxicity requirements and set function requirements, including:

[0032] Generating a pruned sub-knowledge graph based on the important nodes;

[0033] Obtaining the small molecule preference pairs in the sub-knowledge graph;

[0034] Adjusting the generation preference of the small molecule generation model according to the small molecule preference pairs, so that the new small molecules output by the small molecule generation model conform to the characteristics of harmless small molecules, wherein the small molecule generation model includes a pre-trained language model.

[0035] The second aspect of the present disclosure provides a small molecule generation device, including:

[0036] A graph acquisition unit configured to acquire a small molecule safety knowledge graph, wherein in the small molecule safety knowledge graph, attribute data of an existing small molecule is stored in each of a plurality of nodes, and chemical relationship data between the existing small molecules corresponding to the nodes at both ends of the first branch is stored in each of a plurality of first branches;

[0037] An embedding representation unit configured to, for each of the existing small molecules, extract the attribute data related to the toxicity type and function type of the existing small molecule from the corresponding node in the small molecule safety knowledge graph, and construct an embedding vector based on the extracted attribute data;

[0038] A node screening unit configured to screen out important nodes of the small molecule safety knowledge graph based on the embedding vector; and,

[0039] A small molecule generation unit configured to adjust the generation preference of a pre-set small molecule generation model based on the important nodes, so that the small molecule generation model generates new small molecules that meet the set toxicity requirements and set function requirements.

[0040] The third aspect of the present disclosure provides a computer-readable storage medium storing computer instructions for causing a computer to execute the small molecule generation method provided in any one of the solutions of the first aspect above.

[0041] A fourth aspect of the present disclosure provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to cause the at least one processor to execute the small molecule generation method provided by any one of the above first aspects.

[0042] In the small molecule generation method provided by the embodiments of the present disclosure, the knowledge graph is combined with reinforcement learning to provide a knowledge preference optimization enhanced controllable generation scheme based on the small molecule safety knowledge graph, effectively realizing the integration of safety evaluation and small molecule generation model. Combining a large-scale pre-trained model and a knowledge graph, through knowledge preference learning, it guides the safe and controllable generation of small molecules to ensure that the generated small molecules can avoid potential biosafety problems to the greatest extent while meeting the pharmacodynamic requirements. Especially when facing complex small molecule structures and variable toxicity mechanisms, traditional bioinformatics methods are difficult to comprehensively evaluate the safety of generated small molecules. The small molecule generation method provided by the embodiments of the present disclosure can generate new small molecules based on the small molecule safety knowledge graph, and can instantaneously evaluate and optimize the safety of small molecules during the generation process, ensuring the evaluation and control of potential toxicity risks during the generation process, thereby solving the technical problem that the new small molecules generated in the related art do not have safety due to the failure to consider both toxicity reactions and pharmacological functions at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the related art, the following will briefly introduce the drawings required to be used in the description of the specific embodiments or the related art. Obviously, the following drawings are only some embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 Schematic diagram of the steps of the small molecule generation method provided by the embodiments of the present disclosure;

[0045] Figure 2 Schematic diagram of the steps of obtaining the graph provided by the embodiments of the present disclosure;

[0046] Figure 3 Schematic diagram of the embedding representation step provided by the embodiments of the present disclosure;

[0047] Figure 4 Schematic diagram of the node screening step provided by the embodiments of the present disclosure;

[0048] Figure 5 Schematic diagram of the small molecule generation step provided by the embodiments of the present disclosure;

[0049] Figure 6 Block diagram of the small molecule generation device provided by the embodiment of the present disclosure;

[0050] Figure 7 Block diagram of the electronic device further provided by the embodiment of the present disclosure. Detailed implementation manners

[0051] In order to enable those skilled in the art to better understand the solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present disclosure.

[0052] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so as to describe the embodiments of the present disclosure here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0053] In the present disclosure, the orientation or positional relationship indicated by the terms "upper", "lower", "left", "right", "front", "rear", etc. is based on the orientation or positional relationship shown in the accompanying drawings. These terms are mainly used to better describe the present disclosure and its embodiments, and are not used to limit that the indicated devices, elements or components must have a specific orientation, or be constructed and operated in a specific orientation.

[0054] Moreover, in addition to being able to represent the orientation or positional relationship, some of the above terms may also be used to represent other meanings. For example, the term "upper" may also be used to represent a certain attachment relationship or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in the present disclosure can be understood according to specific circumstances.

[0055] In addition, the terms "arranged", "provided with", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, or there can be internal communication between two devices, components, or parts. For those of ordinary skill in the art, the specific meanings of the above terms in this disclosure can be understood according to specific circumstances.

[0056] The term "molecular safety knowledge graph" is a comprehensive knowledge system that integrates various knowledge in the field of molecular biology to form a systematic and queryable network. Such a graph not only helps to deeply understand safety issues at the molecular level but also provides strong support for fields such as drug development and disease diagnosis.

[0057] The term "toxicity type" mainly includes two categories: acute toxicity reaction and chronic toxicity reaction. In addition to these two main toxicity reactions, there are also some other types of toxicity reactions, such as thyrotoxicosis and syphilitic encephalitis.

[0058] The term "function type" is a classification based on pharmacological properties. For example, it can include agonists, antagonists, or inverse agonists acting as receptors.

[0059] The term "knowledge graph embedding method" is a technology that maps entities and relationships in a knowledge graph to a continuous vector space. Its main purpose is to address the challenges of knowledge graph representation, enabling entities and relationships in the knowledge graph to be efficiently operated on through vector representations, such as relation extraction.

[0060] The term "knowledge graph embedding model" is a model used in a knowledge graph embedding method, such as the transfer distance model and the TransE model. Among them, the main idea of the transfer distance model is to transform the problem of measuring the rationality of triples in the vectorized knowledge graph into the problem of measuring the distance between the head entity and the tail entity. The key point of this method is how to design a scoring function, which is often designed as a function that utilizes the relationship to transfer the head entity to the tail entity rationally. The TransE model is inspired by the translational invariance in word vectors and represents both entities and relationships as vectors. For specific details, see the following description.

[0061] The term "small molecule preference pair" refers to the tendency or selective behavior shown by certain small molecules under specific conditions. For example, in the field of immunology, a small molecule preference pair can be the higher activity or affinity shown by a certain enzyme or protein on a specific DNA sequence or structure. Small molecule preference pairs include harmful small molecules and harmless small molecules.

[0062] The term "Knowledge Preference Optimization Method", abbreviated as KPO, mainly involves how to optimize the objective function to make the small molecule generation model more in line with the preferences and needs of users.

[0063] The term "TransE model", the full English name of TransE is Translating Embedding, which is a widely used knowledge graph embedding method. Its main goal is to map entities and relationships in the knowledge graph into a low-dimensional continuous vector space. This method represents relationships through vector addition, aiming to solve the semantic calculation problem and data sparsity of the knowledge graph. The core idea is to regard the relationships in the knowledge base as a certain translation vector between entities. Specifically, it regards the relationship in each triple instance (head entity vector h , relationship vector r , tail entity vector t ) as vector addition. By continuously adjusting h , r and t , make h + r as close as possible to t , that is h + r = t . By embedding the entity vectors and relationship vectors of small molecules into the vector space for use in the small molecule generation model.

[0064] It should be noted that, without conflict, the embodiments in this disclosure and the features in the embodiments can be combined with each other. The following will detail this disclosure with reference to the drawings and in combination with the embodiments.

[0065] There is a technical problem in the related art that the generated new small molecules do not have safety due to the failure to consider both toxicity reactions and pharmacological functions at the same time.

[0066] To solve the above technical problem, the embodiments of this disclosure provide a small molecule generation method. Figure 1 It is a schematic diagram of the steps of the small molecule generation method provided by the embodiments of this disclosure. As Figure 1 shown, the small molecule generation method includes the following steps S11 to step S14.

[0067] Step S11: Obtain a small molecule safety knowledge graph. Among them, in the small molecule safety knowledge graph, each node among multiple nodes stores attribute data of an existing small molecule, and each first branch among multiple first branches stores chemical relationship data between the existing small molecules corresponding to the nodes at both ends of the first branch.

[0068] Step S12: For each existing small molecule, extract the attribute data related to the toxicity type and function type of the existing small molecule from the corresponding nodes in the small molecule safety knowledge graph, and construct an embedding vector based on the extracted attribute data.

[0069] Step S13: Based on the embedding vectors, screen out the important nodes in the small molecule safety knowledge graph.

[0070] Step S14: Based on the important nodes, adjust the generation preference of the pre-set small molecule generation model, so that the small molecule generation model generates new small molecules that meet the set toxicity requirements and set function requirements.

[0071] For step S11, the small molecule safety knowledge graph can be an existing graph obtained directly or constructed through sub-steps S111 to S115. The attribute data of the corresponding small molecule can be captured through the nodes in the small molecule safety knowledge graph, and the complex chemical relationship data between small molecules can be captured through the first branches in the small molecule safety knowledge graph.

[0072] For step S12, any knowledge graph embedding method in related technologies, such as the transfer distance model and the TransE model, can be used to extract the attribute data related to the toxicity type and function type of the existing small molecule, construct an embedding vector, and complete the embedding representation function of extracting the small molecule and its toxicity-related attributes from the small molecule safety knowledge graph.

[0073] For step S13, there are also many screening methods that can be used. In addition to the solutions described in sub-steps S131 to S132 of the subsequent steps, deep learning networks, support vector machines, etc. can also be used. The embodiments of the present disclosure do not limit this.

[0074] For step S14, adjusting the generation preference of the pre-set small molecule generation model based on the important nodes through the knowledge preference optimization method can be directly adjusting the parameters of the small molecule generation model based on the important nodes, or it can also be the solutions described in sub-steps S141 to S143 in the following text. The embodiments of the present disclosure do not limit this. The finally generated new small molecules meet the set toxicity requirements (also called safety drugs) and set function requirements (also called efficacy requirements).

[0075] It should be noted that in addition to the pre-trained language model, the small molecule generation model can also be GANs, variational autoencoders (VAEs), etc. The embodiments of the present disclosure do not limit this.

[0076] Some embodiments of the present disclosure also provide a small molecule generation device, an electronic device, and a storage medium corresponding to the above method.

[0077] The small molecule generation method provided by the embodiments of the present disclosure is applicable to any existing occasion of safe small molecule generation, as well as the occasion of knowledge graph application, or other required scenarios, and the embodiments of the present disclosure do not limit this. Compared with the prior art, applying the small molecule generation method provided by the embodiments of the present disclosure combines the knowledge graph with reinforcement learning, provides a knowledge preference optimization enhanced controllable generation scheme based on the small molecule safety knowledge graph, effectively realizes the integration of safety evaluation and the small molecule generation model, combines large-scale pre-trained models and knowledge graphs, and guides the safe and controllable generation of small molecules through knowledge preference learning to ensure that the generated small molecules can avoid potential biosafety problems to the greatest extent while meeting the pharmacodynamic requirements. Especially when facing complex small molecule structures and variable toxicity mechanisms, traditional bioinformatics methods are difficult to comprehensively evaluate the safety of generated small molecules. The small molecule generation method provided by the embodiments of the present disclosure can generate new small molecules based on the small molecule safety knowledge graph, can instantaneously evaluate and optimize the safety of small molecules during the generation process, ensures the evaluation and control of potential toxicity risks during the generation process, and thus solves the technical problem that the newly generated small molecules in the related art do not have safety due to the failure to consider both toxicity reactions and pharmacological functions at the same time.

[0078] Based on Figure 1 the method shown, as a preferred implementation manner, the method further includes steps S15 to S18 to complete the safety evaluation function of candidate small molecules.

[0079] Step S15: Predict the attribute data of the new small molecule based on the small molecule safety knowledge graph;

[0080] Step S16: Based on the attribute data of the new small molecule obtained by prediction, evaluate whether the new small molecule has potential toxicity characteristics and whether it meets the biosafety requirements;

[0081] Step S17: If the new small molecule contains potential toxicity characteristics or does not meet the biosafety requirements, perform a rejection process on the new small molecule and generate a new small molecule again; and,

[0082] Step S18: If the new small molecule does not contain potential toxicity characteristics and meets the biosafety requirements, output the new small molecule as the small molecule generation result.

[0083] Specifically, through step S15, the attribute data of the new small molecule is predicted by comparing and verifying with the existing data of the existing small molecules in the small molecule safety knowledge graph. Through step S16, it is checked whether the small molecule meets the toxicity safety standard and its possible biological reactions are evaluated. This module can perform accurate risk assessment by combining with a toxicity database or a bioactivity model. The generated small molecule is subjected to biological reaction and toxicity analysis to ensure its biological activity and biological safety, and to ensure that the generated new small molecule does not contain potential toxicity characteristics.

[0084] Based on Figure 1 the method shown, as another preferred embodiment, the above-mentioned attribute data is configured to include small molecule ontology terms, toxicity type data, action target data, and chemical structure data. The chemical relationship data is configured to include first relationship data for characterizing whether there is a pharmacological action between two adjacent existing small molecules, and second relationship data for characterizing whether one of two adjacent existing small molecules has an inhibitory effect on the other. The existing small molecules are configured to include harmful small molecules and harmless small molecules.

[0085] In some embodiments of the present disclosure, step S1 can be configured to construct a small molecule safety knowledge graph, which includes harmful small molecules, harmless small molecules and their related attributes, and captures the complex chemical relationships between small molecules.

[0086] Figure 2 It is a schematic diagram of the map acquisition step provided by the embodiments of the present disclosure. As Figure 2 shown, step S11 further includes the following sub-steps S111 to sub-step S115 to complete the function of constructing the small molecule safety knowledge graph.

[0087] Sub-step S111: Obtain a plurality of existing small molecules.

[0088] Sub-step S112: For each existing small molecule, obtain small molecule ontology terms, toxicity type data, and action target data respectively, construct a node of the existing small molecule in the small molecule safety knowledge graph, and use the small molecule ontology terms, toxicity type data, and action target data as sub-nodes in the node respectively.

[0089] Sub-step S113: Obtain the associated edges between the small molecule ontology terms, toxicity type data, and action target data, and construct the second branches of the existing small molecule in the small molecule safety knowledge graph.

[0090] Sub-step S114: Obtain the chemical relationship data between the current existing small molecule and other existing small molecules, and construct the first branches of the existing small molecule in the small molecule safety knowledge graph.

[0091] Sub-step S115: Generate a small molecule safety knowledge graph that includes all nodes, first branches, and second branches.

[0092] Specifically, in some embodiments of the present disclosure, sub-step S111 can collect small molecule data labeled as "toxicity" and "safety" (antigenicity) from public databases such as the ChEMBL database, PubChem database, and DrugBank database to form a set of harmful small molecules and a set of harmless small molecules; and collect and organize small molecules with toxicity or potentially harmful biological activity or antigenicity and their toxicity description information, and after deduplication and cleaning, construct a set of harmful small molecules. Sub-step S112 can utilize the small molecule ontology and its toxicity description information (small molecule annotation data), and extract small molecule ontology structure data and the association relationship between the small molecule and the toxicity ontology therefrom, construct sub-nodes such as small molecule ontology terms, toxicity type data, and action target data, and characterize the relationship between small molecules and their attributes by constructing triples or higher-order tuples. For example, <small molecule ontology term (h), relationship (r), toxicity type data or action target data (t)>, where the relationship can be "has pharmacological activity" or "has an inhibitory effect on a certain target". Sub-step S113 can construct an association edge between the small molecule ontology term and the toxicity type data and the action target data (pharmacological reaction) through relationship extraction. Sub-step S115 can form a knowledge graph that includes harmful small molecules and harmless small molecules, toxicity-related attributes, and the relationship between small molecules.

[0093] The small molecule safety knowledge graph constructed through sub-steps S111 to S115 can capture the chemical relationships between small molecules and their safety information, providing knowledge support for subsequent small molecule generation models.

[0094] Figure 3 Schematic diagram of the embedding representation step provided for the embodiments of the present disclosure. As Figure 3 shown, step S12 further includes the following sub-steps S121 to S122 to complete the function of extracting the embedding representation of the small molecule sequence.

[0095] Sub-step S121: Embed the small molecule safety knowledge graph through a pre-selected knowledge graph embedding model to obtain an initial embedding vector for each existing small molecule. Among them, the elements of the initial embedding vector include attribute data related to the toxicity type and function type of the existing small molecule in the small molecule safety knowledge graph.

[0096] Sub-step S122: Obtain external knowledge related to the toxicity type and function type (pharmacological characteristics) of the existing small molecule, and fuse the external knowledge with the initial embedding vector to obtain the finally constructed embedding vector.

[0097] Specifically, in some embodiments of the present disclosure, the knowledge graph embedding model in sub-step S121 is configured as a TransE model. By embedding the small molecule safety knowledge graph through the TransE model, the initial embedding vector of the small molecule can be obtained. . The TransE model embeds entities and relationships in the graph through the following formula:

[0098] (1)

[0099] In the formula, is the head entity vector in the small molecule safety knowledge graph (such as small molecule ontology terms), is the tail entity vector in the small molecule safety knowledge graph (such as attribute data related to the toxicity type of small molecules, attribute data related to the function type), is the relationship vector (embedding vector related to the toxicity type and function type). The goal is to minimize the distance between the head entity adjacent and the tail entity vectors through the addition operation of the relationship vector. The initial embedding vector includes the head entity vector , the tail entity vector , and the relationship vector . Among them, the attribute data related to the function type is also called pharmacological property data. And, preferably, an interval-based loss function is adopted to make the score of the correct triple lower than that of the wrong triple by a safety interval. When updating the model parameters, ensure that the correct embedding vectors are closer in the vector space.

[0100] By minimizing the distance between the entity and the relationship, an initial embedding vector with an effective representation is obtained to support the subsequent training of the small molecule generation model.

[0101] In some other embodiments of the present disclosure, sub-step S121 uses a language model to extract the initial embedding vector of the small molecule sequence, including: by masking the small molecule sequence, masking part of the chemical structure or function description to obtain a masked sequence; using a language model-based generation method (such as BERT) to encode the masked sequence to obtain the initial embedding vector of the small molecule.

[0102] In sub-step S122, based on the small molecule safety knowledge graph, external knowledge (such as the toxicity description and pharmacological function of small molecules) is transformed into low-dimensional embedding vectors through an encoder, and these low-dimensional embedding vectors and the initial embedding vector of the small molecule itself Perform weighted fusion to form a comprehensive and enhanced embedding vector that contains pharmacological and toxicity information. That is, the initial embedding vector based on the small molecule sequence and external knowledge are fused based on the small molecule safety knowledge graph, and knowledge embedding is performed to obtain an enhanced embedding vector. External knowledge is embedded into the final embedding vector based on the small molecule safety knowledge graph.

[0103] Figure 4 This is a schematic diagram of the node screening step provided by the embodiments of the present disclosure. As Figure 4 shown, step S13 further includes the following sub-steps S131 to sub-step S132 to complete the node screening function.

[0104] Sub-step S131: Obtain the importance score of each node in the small molecule safety knowledge graph through a pre-selected node importance scoring model. Among them, the input of the node importance scoring model includes the embedding vector matched with each node.

[0105] Sub-step S132: Screen out the key nodes and the nodes with importance scores exceeding the set threshold from all the nodes in the small molecule safety knowledge graph as important nodes.

[0106] Specifically, in some embodiments of the present disclosure, the input of the node importance scoring model further includes the first branch number and the second branch number of each node. Sub-step S131 calculates the importance score of each node based on the node importance scoring algorithm of the weighted index, considering the connectivity and functional relevance of the node. Nodes with importance scores exceeding the set threshold are screened out based on the connectivity and functional relevance of the node, and the key nodes can be set. Sub-step S132 screens out important nodes, retains key small molecule nodes and high-score toxicity-related nodes (nodes with the largest amount of information), and is used to generate a pruned sub-knowledge graph, which is used to generate small molecule preference pairs. The graph structure is simplified, and at the same time, the core safety information is retained.

[0107] Exemplarily, the importance score of each node is determined by its connectivity and functional relevance. Connectivity is used to measure the relationship strength between a node and other nodes, and functional relevance measures the importance of a node in small molecule functions and toxicity. The node importance score is defined as:

[0108] (2)

[0109] In the formula, and is a weighting factor, usually adjusted through experiments or cross - validation to balance the impacts of node connectivity and functional relevance on the final result.

[0110] Connectivity measures the connectivity of a node in the graph, that is, the number of edges between it and other nodes. In a knowledge graph, connectivity represents the strength of the relationship between a node and other nodes, and it can usually be calculated in the following way:

[0111] (3)

[0112] In the formula, is the degree of node , representing the number of branches (including the first branch and the second branch) directly connected to node .

[0113] In a small - molecule safety knowledge graph, nodes with higher connectivity usually mean that they have more relationships in the graph. Therefore, these nodes often contain more information and play a more important role in subsequent small - molecule generation and optimization.

[0114] Functional relevance measures the relevance of a node to the target function in the small - molecule generation task. This relevance is usually based on toxicity relevance and pharmacological relevance. Toxicity relevance is whether the node is closely related to harmful small molecules or toxic characteristics. Pharmacological relevance is whether the node is closely related to pharmacological properties such as pharmacological activity, targets, metabolic pathways, etc. Specifically, functional relevance can be quantitatively evaluated according to the relationships between small molecules and their characteristics such as targets, toxicity, and efficacy. For example, if node represents a small molecule and this small molecule has important pharmacological effects or specific toxic characteristics, then the functional relevance of this node is relatively high. Functional relevance can be calculated by the following formula:

[0115] (4)

[0116] In the formula, represents the relevance between node and a certain functional feature (such as toxicity, pharmacological effect, etc.), is the weight related to each functional feature, reflecting the importance of this function in the total relevance.

[0117] Figure 5 is a schematic diagram of the small - molecule generation steps provided by an embodiment of the present disclosure. As Figure 5As shown, step S14 further includes the following sub-steps S141 to S143 to complete the function of optimizing the generation model based on the knowledge preference optimization method.

[0118] Sub-step S141: Generate a pruned sub-knowledge graph based on important nodes.

[0119] Sub-step S142: Obtain the small molecule preference pairs in the sub-knowledge graph.

[0120] Sub-step S143: Adjust the generation preference of the small molecule generation model according to the small molecule preference pairs, so that the new small molecules output by the small molecule generation model conform to the characteristics of harmless small molecules. Among them, the small molecule generation model includes a pre-trained language model.

[0121] Specifically, in some embodiments of the present disclosure, sub-step S141 reduces the graph calculation complexity through graph node pruning, optimizes the small molecule safety knowledge graph, and only retains important nodes. In the small molecule safety knowledge graph, there are usually a large number of redundant and low-correlation nodes, and the calculation burden can be reduced through sub-step S141. Sub-step S142 constructs small molecule preference pairs based on the structural relationships and embedding vectors (latent embedding representations) in the small molecule safety knowledge graph, and generates small molecule preference pairs by identifying harmless small molecules with a relatively high correlation with harmful small molecules. Sub-step S143 adjusts the generation preference of the model according to the small molecule preference pairs (harmless small molecules and harmful small molecules) in the small molecule safety knowledge graph, so that the newly generated small molecules conform to the characteristics of harmless small molecules, avoiding the generation of new small molecules with toxic characteristics, and completing the optimization process of the small molecule generation model. The generation process is guided by designing knowledge preference pairs. The preference pairs include the pairing of harmless small molecules and harmful small molecules, which are used to train the small molecule generation model to ensure that the generated small molecules meet the safety requirements, and give priority to applying the characteristics of harmless small molecules to the generation process. The knowledge preference optimization method is used to fine-tune the generation model. This optimization method constructs preference pairs between harmless small molecules and harmful small molecules, and makes the newly generated small molecules by the small molecule generation model more inclined to the characteristics of harmless small molecules through optimizing the objective function, avoiding the generation of small molecules with toxicity or adverse reactions.

[0122] Sub-step S141 selects the most informative nodes according to the node importance scores , and the nodes with lower importance will be pruned, retaining the core nodes that are most critical for the small molecule generation task. The pruned sub-graph will only contain high-score nodes and key edges, thus reducing the calculation complexity.

[0123] Sub-step S143 uses the pruning-based sub-knowledge graph to train the small molecule generation model based on the learning algorithm of the graph model, ensuring that the small molecule generation model can focus on the most important information and improve the generation quality. Through the optimization objective function in the following formula, the Knowledge Preference Optimization (KPO) method is applied to make the small molecule generation model more inclined to generate harmless small molecules and avoid generating small molecules with toxic characteristics. The optimization objective of the small molecule generation model can be expressed by the following formula:

[0124] (5)

[0125] In the formula, is a harmless small molecule, is a harmful small molecule, represents the generation score of the small molecule in the parentheses, P is a set of small molecule preference pairs.

[0126] The design process of the optimization objective function is as follows: Based on the difference between the newly generated small molecule and the target harmless small molecule, a loss function is designed; the generated small molecule is combined with the safety evaluation result output by the safety evaluation model, and the reward function is used to guide the small molecule generation model to generate new small molecules that meet the safety standards. By guiding the generation model through the optimization objective function, it is ensured that the generated molecules are closer to the characteristics of harmless small molecules in terms of chemical structure.

[0127] The small molecule generation model is fine-tuned through small molecule preference pairs (harmless small molecules and harmful small molecules). During the generation process, the characteristics similar to harmless small molecules are preferentially selected to avoid generating molecules with potential toxicity. Through the optimization objective function, the small molecule generation model gradually learns the preferential generation method of harmless small molecule characteristics. Through the above knowledge preference optimization process, the generated small molecules will gradually meet the pharmacodynamic requirements and avoid generating molecules with toxicity or adverse reactions.

[0128] It should be noted that the construction of the small molecule generation model is omitted in the above steps, and the small molecule generation model can adopt existing pre-trained language models, etc. The above steps can be executed in a system such as a control chip, and although the logical order is shown, in some cases, the steps described above can be executed in an order different from here.

[0129] From the above description, it can be seen that the present disclosure achieves the following technical effects:

[0130] 1. This method combines a small molecule safety knowledge graph with knowledge preference optimization technology to ensure that the generated molecules not only meet the pharmacodynamic standards but also possess high biological safety during the small molecule generation process. By constructing a small molecule safety knowledge graph, key information such as the toxicity, pharmacological effects, and molecular targets of small molecules is effectively incorporated into the generation model, ensuring that the generated small molecules meet the high safety standards while satisfying the pharmacodynamic requirements. This method can effectively identify potential toxic small molecules and avoid generating compounds that do not meet the biological safety requirements.

[0131] 2. Use the knowledge preference optimization method to fine-tune the generation model, enabling the generated new small molecules to preferentially select the characteristics of harmless small molecules and avoid generating harmless small molecules. By constructing small molecule preference pairs and combining the knowledge preference optimization method, the generation model is guided to preferentially select the chemical characteristics of harmless small molecules during the generation process. This optimization process makes the generated molecules not only innovative but also able to avoid generating potentially toxic molecules, thus enhancing the safety and reliability of drug design.

[0132] 3. Generate new small molecules based on the optimized small molecule generation model and enhanced embedding vectors. The generated small molecules should not only meet the pharmacodynamic requirements but also comply with the biological safety standards.

[0133] 4. The small molecule generation model can be used to generate candidate new small molecules, and they can be further evaluated according to the safety requirements through steps S15 to S18. The evaluation methods include comparing the generated small molecules with the existing small molecules in the small molecule safety knowledge graph to detect whether the generated small molecules have potential toxic characteristics or adverse reactions.

[0134] 5. This method has strong flexibility and adaptability and can be applied to different types of small molecule generation tasks, including drug design, chemical screening, and other fields. Through this flexible generation mechanism, not only can new small molecules that meet the pharmacodynamics be generated, but their safety can also be ensured, making it suitable for multiple application scenarios such as biopharmaceuticals and environmental protection.

[0135] 6. This method is not limited to the traditional drug design field but can also be extended to the generation and optimization of other biological macromolecules, such as proteins, natural products, etc. In addition, with the continuous development of new small molecule databases and safety assessment technologies, it can easily adapt to the needs of these emerging fields and has broad application prospects.

[0136] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0137] An embodiment of the present disclosure also provides a small molecule generation device for implementing the above method embodiment. Figure 6 It is a block diagram of the small molecule generation device provided by the embodiment of the present disclosure. As Figure 6 shown, the small molecule generation device 20 includes a graph acquisition unit 21, an embedding representation unit 22, a node screening unit 23, and a small molecule generation unit 24.

[0138] The graph acquisition unit 21 is configured to acquire a small molecule safety knowledge graph. Among them, in the small molecule safety knowledge graph, attribute data of an existing small molecule is stored in each of the multiple nodes, and chemical relationship data between the existing small molecules corresponding to the nodes at both ends of the first branch is stored in each of the multiple first branches.

[0139] The embedding representation unit 22 is configured to, for each existing small molecule, extract attribute data related to the toxicity type and function type of the existing small molecule from the corresponding node in the small molecule safety knowledge graph, and construct an embedding vector based on the extracted attribute data.

[0140] The node screening unit 23 is configured to screen out important nodes based on the embedding vector.

[0141] The small molecule generation unit 24 is configured to adjust the generation preference of a preset small molecule generation model based on the important nodes, so that the small molecule generation model generates new small molecules that meet the set toxicity requirements and set function requirements.

[0142] In some embodiments of the present disclosure, the spectrum acquisition unit 21 can construct a small molecule safety knowledge graph, which includes chemical structure information of small molecules, harmful small molecule knowledge, harmless small molecule knowledge, and the association relationships between them. The embedding representation unit 22 can use the TransE model to convert the attribute data of small molecules into low-dimensional embedding vectors, embed the small molecule safety knowledge graph, generate initial embedding vectors of small molecules, minimize the distance between entities and relationships, thereby obtaining small molecule vectors with effective representations, and fuse external knowledge with the initial embedding vectors based on the small molecule safety knowledge graph to obtain enhanced embedding vectors. The node screening unit 23 is used to optimize the small molecule safety knowledge graph and reduce the computational burden. In the constructed small molecule safety knowledge graph, there are usually a large number of redundant and low-correlation nodes. The node screening unit 23 can screen out important nodes by calculating the importance scores of each node and based on the connectivity and functional relevance of the nodes. The small molecule generation unit 24 can fine-tune the small molecule generation model according to the preference pairs (harmless small molecules and harmful small molecules) in the important nodes, and optimize the learning process of the small molecule generation model. Through the knowledge preference optimization method, this module preferentially applies the characteristics of harmless small molecules to the generation process to ensure that the generated molecules are closer to the characteristics of harmless small molecules in terms of chemical structure. The small molecules generated by this module not only meet the pharmacodynamic requirements but also comply with the biosafety standards.

[0143] In some embodiments of the present disclosure, the small molecule generation device 20 includes a safety evaluation module. The safety evaluation module is configured to execute the solutions of steps S15 - S18. The evaluation method includes comparing the generated small molecules with the existing data in the small molecule safety knowledge graph to detect whether the generated small molecules have potential toxicity characteristics or adverse reactions. Specifically, it will check whether the small molecules meet the toxicity safety standards and evaluate the possible biological reactions they may cause. This module can perform accurate risk assessment by combining with a toxicity database or a bioactivity model.

[0144] The specific manners of the operations performed by each unit in the above device embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0145] Embodiments of the present disclosure also provide an electronic device, as Figure 7 shown, the electronic device 30 includes one or more processors 31 and a memory 32, Figure 7 Taking one processor 31 as an example.

[0146] The controller may further include: an input device 33 and an output device 34.

[0147] The processor 31, the memory 32, the input device 33, and the output device 34 may be connected through a bus or other means, Figure 5Take the bus connection as an example.

[0148] The processor 31 can be a central processing unit (CPU for short), and the processor 31 can also be other general-purpose processors, digital signal processors (DSP for short), application specific integrated circuits (ASIC for short), field-programmable gate arrays (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or a combination of the above types of chips. The general-purpose processor can be a microprocessor or any conventional processor.

[0149] As a non-transitory computer-readable storage medium, the memory 32 can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the control method in the embodiments of the present disclosure. The processor 31 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions, and modules stored in the memory 32, that is, implements the small molecule generation method in the above method embodiments.

[0150] The memory 32 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the processing device of the server operation, etc. In addition, the memory 32 can include high-speed random access memory, and can also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory 32 may optionally include a memory remotely set relative to the processor 31, and these remote memories can be connected to the network connection device through a network. Examples of the above networks include, but are not limited to, the Internet, enterprise intranets, local area networks, mobile communication networks, and combinations thereof.

[0151] The input device 33 can receive input digital or character information, and generate key signal inputs related to user settings and function controls of the processing device of the server. The output device 34 can include a display device such as a display screen.

[0152] One or more modules are stored in the memory 32, and when executed by one or more processors 31, they execute as Figure 1 shown in the method.

[0153] Those skilled in the art can understand that to implement all or part of the processes in the above method embodiments, it can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM for short), a random access memory (RAM for short), a flash memory (FM for short), a hard disk drive (HDD for short), or a solid-state drive (SSD for short), etc.; the storage medium can also include a combination of the above types of memories.

[0154] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for generating small molecules, characterized in that, Including: Obtain a small molecule safety knowledge graph. In the small molecule safety knowledge graph, attribute data of an existing small molecule is stored in each of multiple nodes, and chemical relationship data between the existing small molecules corresponding to the nodes located at both ends of each of multiple first branches is stored in each of the first branches; For each of the existing small molecules, extract the attribute data related to the toxicity type and function type of the existing small molecule from the corresponding node in the small molecule safety knowledge graph, and construct an embedding vector based on the extracted attribute data; Based on the embedding vector, screen important nodes of the small molecule safety knowledge graph, including: Obtain the importance score of each node in the small molecule safety knowledge graph through a pre-selected node importance scoring model. The input of the node importance scoring model includes the embedding vector matching the node, and the input of the node importance scoring model also includes the number of first branches and the number of second branches of each node; screen out key nodes and nodes with importance scores exceeding a set threshold from all the nodes in the small molecule safety knowledge graph as the important nodes; screen out nodes with importance scores exceeding the set threshold based on the connectivity and functional relevance of the nodes; by screening out important nodes, retain key small molecule nodes and high-score toxicity-related nodes for generating a pruned sub-knowledge graph, and this sub-knowledge graph is used to generate small molecule preference pairs; and, Based on the important nodes, adjust the generation preference of a pre-set small molecule generation model, so that the small molecule generation model generates new small molecules that meet the set toxicity requirements and set function requirements, including: Generate a pruned sub-knowledge graph based on the important nodes; obtain small molecule preference pairs in the sub-knowledge graph, construct small molecule preference pairs based on the structural relationship and embedding vector in the small molecule safety knowledge graph, and generate small molecule preference pairs by identifying harmless small molecules with a relatively high correlation with harmful small molecules; adjust the generation preference of the small molecule generation model according to the small molecule preference pairs, so that the new small molecules output by the small molecule generation model conform to the characteristics of harmless small molecules.

2. The small molecule generation method according to claim 1, wherein Also including: Predict the attribute data of the new small molecule based on the small molecule safety knowledge graph; Based on the attribute data of the new small molecule obtained by the prediction, evaluate whether the new small molecule has potential toxicity characteristics and whether it meets the biological safety requirements; If the new small molecule contains potential toxicity characteristics or does not meet the biological safety requirements, perform a deletion process on the new small molecule and generate the new small molecule again; and, If the new small molecule does not contain potential toxicity characteristics and meets the biological safety requirements, output the new small molecule as the small molecule generation result.

3. The small molecule generation method according to claim 1 or 2, characterized in that, The attribute data is configured to include small molecule ontology terms, chemical structure data, toxicity type data, and action target data; and, The chemical relationship data is configured to include first relationship data for characterizing whether there is a pharmacological effect between two adjacent existing small molecules, and second relationship data for characterizing whether one of two adjacent existing small molecules has an inhibitory effect on the other; and, The existing small molecules are configured to include harmful small molecules and harmless small molecules.

4. The small molecule generation method according to claim 1 or 2, characterized in that, Obtaining the small molecule safety knowledge graph includes: Obtaining a plurality of the existing small molecules; For each of the existing small molecules, respectively obtaining a small molecule ontology term, toxicity type data, and action target data, constructing a node of the existing small molecule in the small molecule safety knowledge graph, and respectively using the small molecule ontology term, the toxicity type data, and the action target data as sub-nodes in the node; Obtaining association edges between the small molecule ontology term and the toxicity type data and the action target data, and constructing second branches of the existing small molecule in the small molecule safety knowledge graph; and, Obtaining chemical relationship data between the existing small molecules and other existing small molecules, and constructing first branches of the existing small molecule in the small molecule safety knowledge graph; and, Generating the small molecule safety knowledge graph including all of the nodes, the first branches, and the second branches.

5. The small molecule generation method according to claim 1 or 2, characterized in that, Extracting the attribute data related to the toxicity type and function type of the existing small molecule from the corresponding node in the small molecule safety knowledge graph, and constructing an embedding vector based on the extracted attribute data includes: Embedding the small molecule safety knowledge graph through a pre-selected knowledge graph embedding model to obtain an initial embedding vector for each of the existing small molecules, where the elements of the initial embedding vector include the attribute data related to the toxicity type and function type of the existing small molecule in the small molecule safety knowledge graph; Obtaining external knowledge related to the toxicity type and function type of the existing small molecule, and fusing the external knowledge with the initial embedding vector to obtain the finally constructed embedding vector.

6. The small molecule generation method according to claim 5, wherein The knowledge graph embedding model is configured as a TransE model; and, screening out important nodes of the small molecule safety knowledge graph based on the embedding vector includes: Obtaining an importance score for each node in the small molecule safety knowledge graph through a pre-selected node importance scoring model, where the input of the node importance scoring model includes the embedding vector matching the node; Screening out key nodes and nodes with importance scores exceeding a set threshold from all the nodes in the small molecule safety knowledge graph as the important nodes.

7. The small molecule generation method according to claim 1 or 2, characterized in that, Adjusting the generation preference of a pre-set small molecule generation model based on the important nodes, so that the small molecule generation model generates new small molecules meeting set toxicity requirements and set function requirements includes: Generating a pruned sub-knowledge graph based on the important nodes; Obtaining small molecule preference pairs in the sub-knowledge graph; and, Adjust the generation preference of the small molecule generation model according to the small molecule preference, so that the new small molecules output by the small molecule generation model conform to the characteristics of harmless small molecules, wherein the small molecule generation model includes a pre-trained language model.

8. A small molecule generating device, characterized in that, Comprising: A map acquisition unit configured to acquire a small molecule safety knowledge map, wherein in the small molecule safety knowledge map, attribute data of an existing small molecule is stored in each of a plurality of nodes, and chemical relationship data between the existing small molecules corresponding to the nodes at both ends of the first branch is stored in each of a plurality of first branches; An embedding representation unit configured to, for each of the existing small molecules, extract the attribute data related to the toxicity type and function type of the existing small molecule from the corresponding node in the small molecule safety knowledge map, and construct an embedding vector based on the extracted attribute data; A node screening unit configured to screen important nodes of the small molecule safety knowledge map based on the embedding vector, including: Obtaining the importance score of each node in the small molecule safety knowledge map through a pre-selected node importance scoring model, wherein the input of the node importance scoring model includes the embedding vector matching the node, and the input of the node importance scoring model also includes the number of first branches and the number of second branches of each node; screening out key nodes and nodes with importance scores exceeding a set threshold from all the nodes in the small molecule safety knowledge map as the important nodes; screening out nodes with importance scores exceeding the set threshold based on the connectivity and functional relevance of the nodes; by screening out important nodes, retaining key small molecule nodes and high-score toxicity-related nodes for generating a pruned sub-knowledge map, and this sub-knowledge map is used to generate small molecule preference pairs; and, A small molecule generation unit configured to adjust the generation preference of a pre-set small molecule generation model based on the important nodes, so that the small molecule generation model generates new small molecules that meet the set toxicity requirements and set function requirements, including: Generating a pruned sub-knowledge map based on the important nodes; obtaining the small molecule preference pairs in the sub-knowledge map, constructing small molecule preference pairs based on the structural relationship and embedding vector in the small molecule safety knowledge map, and generating small molecule preference pairs by identifying harmless small molecules with high correlation with harmful small molecules; adjusting the generation preference of the small molecule generation model according to the small molecule preference pairs, so that the new small molecules output by the small molecule generation model conform to the characteristics of harmless small molecules.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to execute the small molecule generation method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the small molecule generation method according to any one of claims 1 to 7.

Images