Molecular generation method based on SMILES character string
Through the autoregressive generation method based on SMILES strings, the shortcomings of the existing technology in molecular generation and physical and chemical properties optimization are solved, efficient and accurate molecular generation is achieved, and the recovery rate and user requirements are improved.
Patent Information
- Application Number
- CN202510400113.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-01
- Publication Date
- 2025-06-24
AI Technical Summary
The existing molecular generation technology has shortcomings in the optimization of physical and chemical properties when generating compounds with practical application value, and the model based on SMILES strings performs weakly in the recovery rate indicator.
A method of autoregressive generation based on SMILES strings is proposed, supporting unconditional generation and conditional generation. By introducing a conditional generation strategy and attention mechanism, the accuracy and diversity of the generated results are improved, and fragment connection, skeleton replacement and optimization are performed when generating molecular structures.
It significantly improves the efficiency and diversity of molecular generation, reduces the consumption of computing resources, improves the accuracy and recovery rate of the generated results, and ensures that the generated molecular structure is more in line with user needs.
Smart Images
Figure CN120199366A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical fields of drug design and chemoinformatics, and particularly relates to a method for generating molecules based on SMILES strings. Background Art
[0002] The technology of molecule generation is a key link in promoting new drug research and development. Traditional methods mainly rely on rule-based systems or template-based search methods. These methods generate new molecular structures through pre-defined rule libraries or template libraries, using exhaustive or heuristic search. In recent years, with the development of deep learning technology, artificial intelligence-based molecule generation methods have gradually emerged. For example, the ResGen model proposed by the team of Professor Tingjun Hou at Zhejiang University is a 3D molecule generation model based on protein pockets, which adopts a parallel multi-scale modeling strategy, can capture the complex interactions between protein targets and ligands, and significantly improves the computational efficiency. In addition, important progress has also been made in the method of generating molecules based on SMILES strings. For example, the FragGPT model, by introducing the FU-SMILES representation method, gets rid of the traditional left-to-right sequential representation, effectively reduces the error accumulation problem in atom-by-atom generation, and significantly improves the efficiency and diversity of molecule construction.
[0003] However, the existing technologies still face many challenges in the process of molecule generation. Traditional rule- and template-based methods rely on manually formulated rule libraries, which are difficult to cover all possible molecular structures, and the generation process is complex, computationally intensive, and inefficient. Even in deep learning models such as ResGen, although breakthroughs have been made in the generation of 3D molecules in protein pockets, there are still deficiencies in generating compounds with practical application value, especially in the optimization of physicochemical properties. In addition, although the FragGPT model based on SMILES strings performs well in terms of the diversity and efficiency of molecule generation, it performs weakly in terms of the recovery rate index, indicating that it still has limitations in generating molecules with specific properties. Summary of the Invention
[0004] To solve the above technical problems, the present invention proposes a method for generating molecules based on SMILES strings, which realizes functions such as de novo design of molecules, fragment connection, and backbone replacement through autoregressive generation technology, and supports multiple generation strategies, including unconditional generation and conditional generation. This method not only improves the efficiency and diversity of molecule generation, but also significantly reduces the consumption of computing resources, bringing significant technological progress and economic benefits to the field of drug design.
[0005] The present invention introduces a conditional generation strategy into the autoregressive generation model, which can introduce conditional information such as molecular properties and target activities in the encoder module and generate SMILES strings that meet the conditions in the decoder module. In addition, the present invention also adds an optimization process for the molecular structure, including energy minimization and stereochemical configuration optimization, to ensure that the generated molecular structure is not only chemically reasonable but also closer to the actual application requirements in terms of physicochemical properties.
[0006] Through the attention mechanism and conditional generation strategy of the autoregressive generation model, the present invention significantly improves the accuracy and diversity of the generation results. During the generation process, the model can generate a molecular structure that matches the conditions according to the input specific template or fragment, thus significantly improving the recovery rate index. In addition, the present invention further enhances the specific attribute matching ability of the generated molecules through the fragment connection and skeleton replacement functions, ensuring that the generated molecular structure better meets the user's needs.
[0007] The present invention proposes a method for generating molecules based on SMILES strings, including:
[0008] Obtaining user input data;
[0009] Processing the user input data using an autoregressive generation model to generate a new SMILES string; wherein, the autoregressive generation model supports two strategies: unconditional generation and conditional generation;
[0010] Performing structural verification on the new SMILES string;
[0011] Converting the verified SMILES string into a molecular structure, and performing fragment connection and skeleton replacement during the generation of the molecular structure;
[0012] Outputting the generated molecular structure.
[0013] Preferably, the user input data includes initialization parameters, multiple SMILES string fragments, and the SMILES string of the template molecule.
[0014] Preferably, the autoregressive generation model is constructed using deep learning technology and includes:
[0015] An encoder module for encoding the input SMILES string into a high-dimensional feature vector;
[0016] A decoder module for generating a new SMILES string according to the encoded high-dimensional feature vector.
[0017] Preferably, the autoregressive generation model supports a conditional generation strategy, including:
[0018] Introduce conditional information into the encoder module, where the conditional information includes molecular properties and target activities;
[0019] Generate a SMILES string that meets the conditions in the decoder module according to the conditional information.
[0020] Preferably, when generating the SMILES string, the autoregressive generation model adopts an attention mechanism.
[0021] Preferably, the process of fragment connection and backbone replacement when generating the molecular structure includes:
[0022] Generate a molecular structure containing fragments according to the information of multiple SMILES string fragments input;
[0023] Replace the backbone part in the molecular structure according to the SMILES string backbone information of the input template molecule to generate a new molecular structure.
[0024] Preferably, it further includes the optimization process of the molecular structure:
[0025] Perform energy minimization on the generated molecular structure;
[0026] Perform stereochemical configuration optimization on the generated molecular structure to ensure that the generated molecular structure is reasonable in space.
[0027] In a second aspect, the present invention also provides a method for generating molecules based on SMILES strings applied to drug design, including:
[0028] Generate a molecular structure with potential activity according to the biological target information of the target disease;
[0029] Perform virtual screening and experimental verification on the generated molecular structure to screen out candidate drug molecules with high activity and low toxicity.
[0030] In a third aspect, the present invention also discloses a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the method described in the first aspect.
[0031] In a fourth aspect, the present invention also discloses a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the method described in the first aspect.
[0032] Compared with the prior art, the present invention has the following advantages and technical effects:
[0033] The present invention proposes a method for generating molecules based on SMILES strings. First, obtain user input data; secondly, use an autoregressive generation model to process the user input data to generate new SMILES strings; wherein, the autoregressive generation model supports two strategies: unconditional generation and conditional generation; then, perform structural verification on the new SMILES strings; further, convert the verified SMILES strings into molecular structures, and perform fragment connection and backbone replacement when generating the molecular structures; finally, output the generated molecular structures.
[0034] Through the autoregressive generation technology of SMILES strings, the present invention realizes functions such as de novo design of molecules, fragment connection, and backbone replacement, supports multiple generation strategies, including unconditional generation and conditional generation, and can efficiently and accurately generate diverse molecular structures according to the input templates or fragments, providing strong technical support for drug design and having broad application prospects and practical value. Brief Description of the Drawings
[0035] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:
[0036] Figure 1 It is the flowchart of the de novo design of molecules in the embodiments of the present invention;
[0037] Figure 2 It is the flowchart of fragment connection in the embodiments of the present invention;
[0038] Figure 3 It is the flowchart of backbone replacement in the embodiments of the present invention. Detailed Embodiments
[0039] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.
[0040] It should be noted that the steps shown in the flowchart of the 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.
[0041] Embodiment 1
[0042] This embodiment provides a method for generating molecules based on SMILES strings, including:
[0043] S1. Input user input data;
[0044] The user input data includes initialization parameters, multiple SMILES string fragments, and the SMILES string of the template molecule.
[0045] S2. Process the input SMILES strings using an autoregressive generation model to generate new SMILES strings; among them, the autoregressive generation model supports two strategies: unconditional generation and conditional generation.
[0046] Furthermore, data preprocessing includes:
[0047] SMILES standardization: Standardize the input SMILES strings to ensure they conform to chemical rules.
[0048] Data augmentation: Enhance the diversity of training data by randomly inserting, deleting, or replacing characters.
[0049] Furthermore, the autoregressive generation model is constructed using deep learning techniques, including:
[0050] An encoder module for encoding the input SMILES strings into high-dimensional feature vectors.
[0051] A decoder module for generating new SMILES strings based on the encoded high-dimensional feature vectors.
[0052] The autoregressive generation model uses a large amount of known SMILES string data during training to optimize the model parameters by minimizing the difference between the generated SMILES strings and the true SMILES strings.
[0053] Furthermore, the autoregressive generation model supports the conditional generation strategy, including:
[0054] Introduce conditional information into the encoder module, where the conditional information includes molecular properties and target activities.
[0055] Generate conditional SMILES strings in the decoder module according to the conditional information.
[0056] Furthermore, when generating SMILES strings, the autoregressive generation model uses an attention mechanism (Attention Mechanism) to improve the accuracy and diversity of the generation results.
[0057] S3. Perform structural verification on the new SMILES strings.
[0058] Verify and optimize the generated SMILES strings to ensure that the generated molecular structures conform to chemical rules and design requirements.
[0059] S4. Convert the verified SMILES string into a molecular structure;
[0060] The process of fragment connection and backbone replacement includes:
[0061] Generate a molecular structure containing fragments based on the information of multiple SMILES string fragments input;
[0062] Replace the backbone part in the molecular structure according to the SMILES string backbone information of the input template molecule to generate a new molecular structure.
[0063] S5. Output the generated molecular structure.
[0064] Furthermore, it also includes the optimization process of the molecular structure: perform energy minimization on the generated molecular structure; optimize the stereoconfiguration of the generated molecular structure to ensure that the generated molecular structure is reasonable in space.
[0065] When generating the molecular structure in this embodiment, it can automatically predict the molecular properties, specifically including: predicting the physical and chemical properties of the generated molecular structure, such as solubility, LogP value, etc.; predicting the biological activity of the generated molecular structure, such as the binding ability with the target protein.
[0066] This embodiment realizes functions such as de novo design, fragment connection, and backbone replacement of molecules through the autoregressive generation technology of SMILES strings, supports multiple generation strategies, including unconditional generation and conditional generation, and can generate diverse molecular structures according to the input templates or fragments.
[0067] First, the de novo design of molecules, such as Figure 1 shown, includes the following steps:
[0068] 1. Input initialization:
[0069] The user inputs initialization parameters through an input device (such as a keyboard, touch screen, etc.), including the selection of the generation strategy (unconditional generation or conditional generation), generation targets (such as molecular weight, specific functional groups, etc.).
[0070] The initialization parameters are transmitted to the data processing unit through the interface module.
[0071] 2. SMILES string generation:
[0072] After receiving the initialization parameters, the data processing unit starts the autoregressive generation model 103.
[0073] The autoregressive generation model generates an initial SMILES string based on the initialization parameters. This model uses deep learning techniques, such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), or Transformer architectures, to model the SMILES string.
[0074] The autoregressive generation model supports two strategies: unconditional generation and conditional generation, and can generate diverse molecular structures based on the input templates or fragments.
[0075] Unconditional generation: Without providing any prior information, the model can randomly generate diverse molecular structures;
[0076] Conditional generation: Based on the input specific templates or fragments, the model generates molecular structures that match the input conditions.
[0077] 3. Molecular structure verification:
[0078] The generated SMILES string is structurally verified through a verification module to ensure that the generated molecular structure conforms to chemical rules.
[0079] The verification module includes a chemical rule library and a structure parsing algorithm to parse and verify the SMILES string.
[0080] 4. Output of the molecular structure:
[0081] The verified SMILES string is converted into a molecular structure through an output module and displayed graphically on the user interface.
[0082] The user can further edit or optimize the generated molecular structure as needed.
[0083] Figure 1 Among them, the interface module is used to receive the initialization parameters input by the user.
[0084] The data processing unit is responsible for processing and transmitting the initialization parameters.
[0085] The autoregressive generation model generates a SMILES string based on the initialization parameters.
[0086] The verification module performs structural verification on the generated SMILES string.
[0087] The output module converts the verified SMILES string into a molecular structure.
[0088] The user interface displays the generated molecular structure.
[0089] Second, fragment connection, as Figure 2 shown, includes the following steps:
[0090] 1. Input fragment:
[0091] The user inputs two or more SMILES string fragments through an input device, and these fragments can be known drug molecule fragments or user-defined fragments.
[0092] The input fragments are transmitted to the data processing unit through the interface module.
[0093] 2. Fragment connection strategy selection:
[0094] The user selects a fragment connection strategy, such as direct connection, connection through bridging atoms, etc.
[0095] The selected strategy is transmitted to the data processing unit through the interface module.
[0096] 3. SMILES string generation:
[0097] After receiving the fragments and the connection strategy, the data processing unit starts an autoregressive generation model.
[0098] The autoregressive generation model generates a concatenated SMILES string based on the input fragments and the connection strategy.
[0099] 4. Molecular structure verification:
[0100] The generated SMILES string is structurally verified through a verification module to ensure that the concatenated molecular structure conforms to chemical rules.
[0101] 5. Output molecular structure:
[0102] The verified SMILES string is converted into a molecular structure through an output module and displayed graphically on the user interface.
[0103] Figure 2 Among them, the interface module is used to receive the fragments and connection strategies input by the user.
[0104] The data processing unit is responsible for processing and transmitting the input fragments and connection strategies.
[0105] The autoregressive generation model generates a SMILES string based on the input fragments and connection strategies.
[0106] The verification module performs structural verification on the generated SMILES string.
[0107] The output module converts the verified SMILES string into a molecular structure.
[0108] The user interface displays the generated molecular structure.
[0109] III. Skeleton replacement, such asFigure 3 As shown, it includes the following steps:
[0110] 1. Input the template molecule:
[0111] The user inputs the SMILES string of a template molecule through an input device.
[0112] The input template molecule is transmitted to the data processing unit through the interface module.
[0113] 2. Select the replacement skeleton:
[0114] The user selects the type of skeleton to be replaced, such as an aromatic ring, an aliphatic ring, etc.
[0115] The selected skeleton type is transmitted to the data processing unit through the interface module.
[0116] 3. Generate the SMILES string:
[0117] After receiving the template molecule and the replacement skeleton type, the data processing unit starts the autoregressive generation model.
[0118] The autoregressive generation model generates the replaced SMILES string based on the template molecule and the replacement skeleton type.
[0119] 4. Verify the molecular structure:
[0120] The generated SMILES string is structurally verified through the verification module to ensure that the replaced molecular structure conforms to chemical rules.
[0121] 5. Output the molecular structure:
[0122] The verified SMILES string is converted into a molecular structure through the output module and displayed graphically on the user interface.
[0123] Figure 3 Among them, the interface module is used to receive the template molecule and the replacement skeleton type input by the user.
[0124] The data processing unit is responsible for processing and transmitting the input template molecule and the replacement skeleton type.
[0125] The autoregressive generation model generates the SMILES string based on the input template molecule and the replacement skeleton type.
[0126] The verification module performs structural verification on the generated SMILES string.
[0127] The output module converts the verified SMILES string into a molecular structure.
[0128] The user interface displays the generated molecular structure.
[0129] Through the above detailed description, those skilled in the art of this technology can clearly understand and implement the molecular generation method based on SMILES strings, and apply it to drug design to achieve functions such as de novo design of molecules, fragment connection, and backbone replacement. This method not only improves the efficiency and accuracy of molecular design, but also provides new tools and ideas for drug research and development.
[0130] Example Two
[0131] This example also provides a molecular generation method based on SMILES strings for drug design, including:
[0132] Generating a molecular structure with potential activity according to the biological target information of the target disease;
[0133] Performing virtual screening and experimental verification on the generated molecular structure to screen out candidate drug molecules with high activity and low toxicity.
[0134] This example has all the advantages of the molecular generation method based on SMILES strings provided in Example One.
[0135] Example Three
[0136] This example also discloses a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in Example One are implemented.
[0137] Example Four
[0138] This example also discloses a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method described in Example One are implemented.
[0139] The above is only the preferred specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in this application should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.
Claims
1. A molecule generation method based on SMILES string, characterized in that: The following steps are involved: Get user input data; Using an autoregressive generation model to process the user input data to generate a new SMILES string; wherein the autoregressive generation model supports two strategies: unconditional generation and conditional generation; Perform structural validation on new SMILES strings; Convert the verified SMILES string into a molecular structure, and perform fragment connection and backbone replacement when generating the molecular structure; Output the generated molecular structure.
2. The method according to claim 1, characterized in that: The user input data includes initialization parameters, multiple SMILES string fragments, and a SMILES string of a template molecule.
3. The method according to claim 1, characterized in that The autoregressive generative model is constructed using deep learning technology, including: The encoder module is used to encode the input SMILES string into a high-dimensional feature vector; The decoder module is used to generate a new SMILES string based on the encoded high-dimensional feature vector.
4. The method according to claim 3, characterized in that The autoregressive generative model supports conditional generative strategies, including: introducing condition information into the encoder module, the condition information comprising molecular properties and target activity; According to the condition information, a SMILES character string that meets the condition is generated in the decoder module.
5. The method according to claim 1, characterized in that The autoregressive generation model adopts an attention mechanism when generating SMILES strings.
6. The method according to claim 1, characterized in that The process of fragment connection and backbone replacement when generating a molecular structure includes: Generate a molecular structure containing fragments according to the input multiple SMILES string fragment information; According to the SMILES string skeleton information of the input template molecule, the skeleton part in the molecular structure is replaced to generate a new molecular structure.
7. The method according to claim 1, characterized in that It also includes the optimization process of molecular structure: Perform energy minimization on the generated molecular structure; The generated molecular structure is optimized in stereo configuration to ensure that the generated molecular structure is reasonable in space.
8. A SMILES string-based molecular generation method for drug design, characterized in that: include: Generate molecular structures with potential activity based on biological target information of target diseases; The generated molecular structures are virtually screened and experimentally verified to screen out candidate drug molecules with high activity and low toxicity.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.