Derivative molecule of target molecule, generation method and system thereof and computer product

Abstract

The invention provides a derivative molecule of a target molecule, a generation method and system thereof and a computer product, and relates to the technical field of deep learning and molecule generation. The method comprises the following steps: S1, obtaining an expression normal form of a target molecule; s2, inputting the expression normal form of the target molecule into a trained molecule generation model to generate a derivative molecule of the target molecule; wherein the target molecule is selected from siderophores. Based on the method, a complex and high-quality siderophore molecular structure can be generated, the discovery efficiency of novel siderophore molecules is remarkably improved, the research period is shortened, the experiment cost is reduced, and an efficient and economical new technical approach is provided for design and development of bioactive molecules.

Description

Technical Field

[0001] The present application relates to the technical field of deep learning and molecule generation, and in particular, to derivative molecules of target molecules and methods, systems and computer products for generating the same. Background Art

[0002] Siderophores are a class of small organic compounds produced by microorganisms that have a high affinity for iron ions and can chelate iron ions from the environment and transport them into cells. As a special natural product, siderophores not only play a key role in the acquisition and transportation of iron in microorganisms, but also exhibit significant antibacterial activity, providing a new approach for the development of new antibiotics and drugs.

[0003] However, there are still few known types of siderophores, and their molecular structures are diverse and complex, which greatly limits the research on the functions and application potential of siderophores. Therefore, mining and exploring potential new siderophores is of great significance for expanding the application range of siderophores and developing new drugs. The discovery of potential siderophores mainly relies on biochemical experiments in the laboratory. This method is not only time-consuming and labor-intensive, but also costly and highly dependent on the expertise and intuition of experts. Since the chemical space is discrete and vast, it becomes impractical to comprehensively search for all possible molecular structures. In addition, there is currently no method for discovering siderophore molecules based on deep learning. Summary of the invention

[0004] In order to solve at least one of the existing problems, the present application proposes a method for producing iron carrier molecules, which is used to obtain more high-quality iron carrier molecules, increase the types of iron carriers, and reduce the cost of discovering new iron carrier molecules.

[0005] Specifically, this application provides the following technical solutions:

[0006] In the first aspect of the present application, the present application proposes a method for generating a derivative molecule of a target molecule, the method comprising: S1, obtaining an expression paradigm of the target molecule; S2, inputting the expression paradigm of the target molecule into a trained molecular generation model to generate a derivative molecule of the target molecule; wherein the target molecule is selected from an iron carrier.

[0007] Based on the above method, complex and high-quality iron carrier molecular structures can be generated, which significantly improves the efficiency of discovering new iron carrier molecules, shortens the research cycle, and reduces experimental costs, providing a new efficient and economical new technology approach for the design and development of bioactive molecules.

[0008] In the second aspect of the present application, the present application proposes a system for generating a derivative molecule of a target molecule. According to an embodiment of the present application, the system includes: an expression paradigm acquisition module for acquiring the expression paradigm of the target molecule; a derivative molecule generation module for inputting the expression paradigm of the target molecule into a trained molecule generation model to generate a derivative molecule of the target molecule; wherein the target molecule is selected from an iron carrier.

[0009] The aforementioned system can efficiently obtain the expression paradigm of the target molecule and generate its derivative molecules. Combined with the trained molecular generation model, the design of iron carrier molecules significantly improves the generation efficiency and molecular diversity, while ensuring the structural rationality and functional relevance of the generated molecules, providing an automated and intelligent solution for the development and application of iron carrier molecules.

[0010] In the third aspect of the present application, the present application provides an iron carrier derivative molecule. According to an embodiment of the present application, the iron carrier derivative molecule is produced by the method described in the first aspect.

[0011] The iron carrier derivative molecules generated based on the above method are diverse and novel, which expands the application range of iron carriers and provides more potential candidate molecules for antibiotic development, metal ion transport and drug delivery. Through customized design, these derivative molecules can meet specific functional requirements, such as enhanced iron ion chelation and targeting, and improved drug delivery efficiency.

[0012] In a fourth aspect of the present application, the present application proposes a computer program product. According to an embodiment of the present application, the computer program product comprises: computer instructions; when part or all of the computer instructions are run on a computer, the method for generating a derivative molecule of a target molecule as described in the first aspect is executed.

[0013] In a fifth aspect of the present application, the present application proposes a computing device. According to an embodiment of the present application, the computing device comprises: a processor and a memory; the memory is used to store a computer program; the processor is used to execute the computer program to implement the method for generating a derivative molecule of a target molecule as described in the first aspect.

[0014] In a sixth aspect of the present application, the present application proposes a computer-readable storage medium. According to an embodiment of the present application, the computer-readable storage medium stores computer instructions or programs, and when the computer instructions or programs are run on a computer, the method for generating a derivative molecule of a target molecule as described in the first aspect is executed.

[0015] The aforementioned computer program product, computing device, and computer-readable storage medium realize efficient automation and effectively improve the efficiency of high-quality molecule generation by automatically executing the method of generating a derivative molecule of a target molecule through computer instructions. Secondly, the characteristics of the instructions make the above method highly consistent and reliable in different application scenarios.

[0016] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0018] Figure 1 It is a schematic diagram of the process of generating a derivative molecule of a target molecule provided by some examples of the present application;

[0019] Figure 2 This is a schematic diagram of the Transformer-mLSTM model structure provided by some examples of this application;

[0020] Figure 3 It is a schematic diagram of the process of generating the derivative molecules of the siderophore molecules provided in some examples of the present application;

[0021] Figure 4 is a schematic diagram of the data preprocessing process provided by some examples of this application;

[0022] Figure 5 This is a schematic diagram of the Transformer-mLSTM model pre-training process provided by some examples of this application;

[0023] Figure 6 It is a schematic diagram of the fine-tuning process of the pre-trained model provided by some examples of this application;

[0024] Figure 7 is a schematic diagram of the process of generating derivative molecules provided by some examples of this application;

[0025] Figure 8 It is a schematic diagram of a derivative molecule generation system of a target molecule provided in some examples of the present application;

[0026] Fig. 9 It is a schematic diagram of an electronic device provided in some examples of this application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0028] In this article, unless otherwise specified, the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described here can be implemented in a sequence other than those illustrated or described here. In the present application, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices. In the description of the present application, unless otherwise specified, "multiple" refers to two or more than two.

[0029] In this article, unless otherwise specified, "derivative molecules" refer to a series of new molecules generated by using a specific algorithm or model based on the structural characteristics or expression paradigm of the target molecule. These molecules have a certain degree of similarity with the target molecule in terms of molecular skeleton, functional group or other chemical properties, but have new structural or functional properties and can be used to expand the application range of the target molecule or optimize its performance. For example, in this application, derivative molecules are potential new molecules generated on the basis of iron carrier molecules through molecular generation models, which are used to accelerate the discovery and design of new iron carriers.

[0030] The present application proposes a method for generating a derivative molecule of a target molecule, a system for generating a derivative molecule of a target molecule, a computer program product, an electronic device, and a computer-readable storage medium. The following describes each of them in detail:

[0031] Methods for generating derivative molecules of target molecules

[0032] In one aspect of the present application, the present application proposes a method for generating a derivative molecule of a target molecule, referring to Figure 1 , the method comprising:

[0033] S1, obtaining an expression pattern of a target molecule; the target molecule is selected from a siderophore;

[0034] S2, inputting the expression pattern of the target molecule into the trained molecular generation model to generate a derivative molecule of the target molecule.

[0035] In the task of molecular generation, commonly used models include recurrent neural networks (RNN), long short-term memory networks (LSTM), generative adversarial networks (GAN), variational autoencoders (VAE) and Transformer models, each with its own advantages but also facing different challenges. Among them, RNN processes molecular formulas through a cyclic structure and can capture the dependencies between the previous and next positions in the molecular sequence, but it is easy to have gradient vanishing or exploding problems on long sequences; LSTM improves RNN by introducing a gating mechanism and can better handle long-distance dependencies, but the computational complexity is high and the training time is long. GAN consists of a generator and a discriminator, and generates high-quality molecules through adversarial training, but the training process is unstable and prone to mode collapse. VAE maps input molecular data to the latent chemical space through an encoding-decoding framework, and decodes to generate new samples, thereby generating diverse and smooth molecules, but the generated molecular structures may lack diversity. Transformer processes sequence data in parallel through a self-attention mechanism, which can efficiently capture global dependencies. At the same time, it can learn rich features in different representation subspaces through a multi-head attention mechanism and position encoding, but it may encounter memory limitations when processing very long molecular formulas.

[0036] Due to the inadequacy of existing models in molecular generation and the high specificity and functionalization of the siderophore molecular structure, there is currently no method for generating derivative molecules of siderophore molecules based on deep learning in the art. To this end, the inventors of the present application proposed a molecular generation model, which obtains highly diverse and chemically rational derivative molecules of siderophore molecules by mixing different deep learning models and improving their internal layer structures.

[0037] In some examples of the present application, the aforementioned molecular generation model is obtained by training in the following manner:

[0038] I, preprocess multiple natural product data to obtain a standard expression paradigm suitable for model input;

[0039] In some examples of the present application, the aforementioned multiple natural product data may come from a public natural product database, such as the COCONUT natural product database.

[0040] In some examples of the present application, the aforementioned preprocessing includes: obtaining an expression paradigm of the target data, wherein the target data includes: natural product data; parsing the expression paradigm to determine an effective expression paradigm; standardizing and deduplicating the effective expression paradigm to determine a standard expression paradigm.

[0041] The aforementioned parsing process is used to check whether the obtained expression paradigm can be parsed into a valid molecular structure. In some feasible examples, verification can be performed by using the Chem.MolFromSmiles function in the chemical toolkit RDKit.

[0042] The aforementioned standardization and deduplication processing is used to ensure the consistency and standardization of all standard paradigms, including atomic numbers, bond types, stereochemical information, etc. Then the Morgan fingerprint of each molecule is calculated, and the duplicate molecular entries are identified and removed by comparing the similarity of the fingerprints.

[0043] In some examples of the present application, the aforementioned expression paradigm is selected from a SMILES (Simplified Molecular Linear Input Specification) string.

[0044] II. Pre-training the encoder-decoder model based on the standard expression paradigm obtained in step I, including: converting the standard expression paradigm into a vector representation based on the encoder embedding layer; encoding the vector representation using a multi-head attention layer, a normalization layer, a residual connection, and a feedforward neural network layer to obtain the potential characteristics of the molecular structure; inputting the vector after nonlinear transformation into the decoder to obtain a molecular coding output; using cross entropy loss to calculate the loss value between the molecular coding output and the standard expression paradigm; and using back propagation to update the model parameters to obtain a pre-trained model.

[0045] In some examples of the present application, the aforementioned encoder is selected from Transformer, and the decoder is selected from MatrixLSTM. By mixing Transformer and Matrix LSTM (abbreviated as Transformer-mLSTM), the model combines the efficient parallel processing capability of Transformer and the long sequence modeling capability of mLSTM, which is conducive to the derivation of high-quality iron carrier molecules.

[0046] In some examples of the present application, the above-mentioned pre-trained model can be further fine-tuned to more accurately generate derivative molecules that meet the characteristics of siderophore molecules, thereby improving the quality and applicability of the generated results. Fine-tuning can also accelerate the convergence of the model, improve training efficiency, and reduce computing costs.

[0047] Among them, the above-mentioned efficient fine-tuning methods include:

[0048] III, preprocessing a plurality of siderophore data to obtain a standard expression paradigm of siderophore molecules suitable for input of the pre-trained model;

[0049] In some examples of the present application, the aforementioned plurality of siderophore data are selected from the SIDERITE siderophore database.

[0050] In some examples of the present application, the aforementioned preprocessing method is the same as the natural product data, the difference being that the data used here is a siderophore database.

[0051] IV, training the pre-trained model using the siderophore molecule standard expression paradigm, and relearning the decoder parameter weights of the pre-trained model;

[0052] The cross entropy loss is used to calculate the loss value between the molecular encoding output of the pre-trained model and the standard expression paradigm of the siderophore molecule; and the model parameters are updated by back propagation to obtain a molecular generation model.

[0053] In some examples of the present application, in the S2 step, it includes: converting the initially labeled target molecule expression paradigm into a vector in the encoder; the decoder generates a probability distribution of the next character based on the converted vector; randomly sampling a character from the probability distribution, adding it as the next character to the generated sequence, and using it as the next input of the decoder for recursive generation.

[0054] In some examples of the present application, in the S2 step, it further includes: the recursively generated characters satisfy any of the following conditions as a standard expression paradigm of a derivative molecule of the target molecule: 1) generating an end mark; 2) reaching a predetermined length limit.

[0055] After generating the standard expression paradigm of the derivative molecule of the target molecule, further comprising generating a molecular structure of the standard expression paradigm of the derivative molecule of the target molecule. In some examples of the present application, the tools used to implement the standard expression paradigm to generate a molecular structure include but are not limited to RDKit, ChemDraw or Mol2Vec, etc.

[0056] In some examples of the present application, after obtaining the target molecular structure based on the above steps, it further includes evaluating the molecular properties of the molecular structure, wherein the evaluation is selected from at least one of chemical rationality, novelty, biological activity prediction or synthetic feasibility.

[0057] The aforementioned chemical rationality includes: molecular structure validity, consistency of bonds and atoms, stereochemical rationality, consistency of molecular weight and chemical formula, chemical reactivity, etc.

[0058] In some examples of the present application, if the previously generated molecular structure fails the molecular property evaluation, it is regenerated based on the target molecule.

[0059] Based on the above characteristics, this application takes one of the implementation methods as an example to describe in detail the generation of derivative molecules of siderophore molecules:

[0060] The model is selected from the encoder-decoder hybrid model, specifically Transformer-mLSTM. The molecules are represented as SMILES and input into the hybrid network model for learning. The model uses Transformer as an encoder to capture global dependencies with efficient parallel processing capabilities and self-attention mechanism; mLSTM is used as a decoder to enhance the modeling ability of long sequence data. First, the model is pre-trained on large-scale data from a molecular database to learn a wide range of chemical patterns and structural features; then, it is fine-tuned on a specific molecular database to adapt to the generation task of a specific molecule. By repeatedly sampling the decoder, potential molecular candidates learned by the model are obtained. These molecules are then analyzed and screened in terms of chemical rationality, novelty, biological activity prediction, and synthetic feasibility to ensure that the final molecular candidates are highly diverse and chemically rational.

[0061] Reference flow chart for the generation of derivative molecules of siderophore molecules Figure 3 , including data collection, data preprocessing, model design, model pre-training, model fine-tuning, and molecule generation stages.

[0062] i. Data preparation

[0063] The molecular data and related properties required for model training are collected from public databases and literature, including large-scale molecular data and data on specific molecules, which are used for the subsequent pre-training and fine-tuning stages of the model respectively.

[0064] ii. Data preprocessing

[0065] 1. Clean the data: Use RDKit to verify whether the SMILES string in the database can correctly represent a molecule. If the SMILES string cannot be parsed into a valid molecular structure, it will be removed from the dataset.

[0066] 2. Standardize SMILES strings: Standardize SMILES strings to ensure the consistency and standardization of all SMILES strings, including atomic numbers, bond types, stereochemical information, etc.

[0067] 3. Deduplication: Identify and remove duplicate molecular entries by calculating Morgan molecular fingerprints and comparing similarities to ensure that each molecule appears only once.

[0068] 4. Divide the data set: Divide the data set into training set, validation set and test set. The division can be based on actual needs, such as 80% of the data is used for training and 20% for validation.

[0069] The data preprocessing flow chart is as follows: Figure 4 shown.

[0070] iii. Model design

[0071] The model uses an encoder-decoder network architecture, combining the advantages of Transformer and mLSTM to achieve efficient parallel processing capabilities and powerful long sequence modeling capabilities. Specifically, Transformer is used as an encoder to capture the global dependencies of molecular structures; mLSTM is used as a decoder to generate new molecular sequences.

[0072] Specifically, the encoder consists of an embedding layer, a positional encoding, a multi-head attention layer, a feedforward neural network layer, a residual connection, and a layer normalization. To ensure that all SMILES input to the network have the same length, the data length needs to be aligned when inputting the embedding layer. For SMILES with a length less than the maximum length, a special padding character is added at the end <pad>For SMILES whose length exceeds the maximum length, it is truncated to the set maximum length. Each SMILES input to the embedding layer will also have <eos>Mark as the input sequence to clarify the end of the molecular sequence.

[0073] Furthermore, the embedding layer converts the processed SMILES into a vector representation of fixed dimension and adds position information to each embedded vector through position encoding to retain the position information of the elements in the sequence.

[0074] Furthermore, the embedding vector will learn the rich features of molecular structure from multiple different representation subspaces through the multi-head self-attention mechanism. The calculation formula of multi-head self-attention is:

[0075] MultiHead(Q,K,V)=Concat(head1,....head h )W O ;

[0076]

[0077]

[0078] Among them, Q, K, and V represent query, key, and value respectively;

[0079] is the parameter matrix of the i-th attention head, W O is the parameter matrix used to combine the outputs of all heads;

[0080] Furthermore, the vector output by the multi-head attention layer is passed through a normalization layer and a residual connection to improve the training stability of the model. The normalization formula is:

[0081]

[0082] Among them, μ and σ are the mean and standard deviation respectively, γ and β are learnable parameters, and ε is a small constant to prevent the denominator from being 0. The residual connection directly adds the input to the output, allowing the gradient to flow through the network more easily.

[0083] Further, each vector is further processed by the feedforward neural network layer. This layer consists of a GLU gated linear activation unit layer. Unlike the traditional feedforward neural network layer consisting of two layers of linear transformation and an activation function, GLU allows the model to learn how to dynamically adjust the flow of information based on the input data, thereby better capturing complex patterns. This helps improve the generalization ability and performance of the model.

[0084] The output of the feedforward neural network will pass through the normalization layer and residual connection to further improve the stability of the model, and finally output a fixed-dimensional vector representation that contains the important features of the input sequence.

[0085] The decoder mLSTM (Matrix LSTM) greatly enhances the model's memory and parallel processing capabilities by extending the vector operations in traditional LSTM to matrix operations. Each state of mLSTM is no longer a single vector, but a matrix, which allows it to capture more complex data relationships and patterns in a single time step. In addition, mLSTM introduces a covariance update mechanism for key-value pair storage, which significantly increases the model's storage capacity and improves the model's parallel processing capabilities.

[0086] The input of the decoder is also aligned in length, and then the embedding layer converts the processed SMILES into a vector representation of fixed dimension and incorporates the position information and context information from the encoder.

[0087] Furthermore, the vector will be input into the pre-LayerNorm residual structure embedded in the mLSTM unit. The model will project the vector twice, once for externalizing the output gate and once as the input of the mLSTM unit to capture more complex feature interactions and increase the transparency and adjustability of the model.

[0088] Furthermore, the vectors are learnably skip-connected through causal convolution and Swish activation functions to maintain the causal relationship of time series data and enhance the nonlinear expression ability of the model. At the same time, the information transmission path is optimized through learnable skip connections, ultimately achieving the purpose of improving model performance.

[0089] Furthermore, the model uses a block diagonal matrix to obtain the query, key, and value, where the value skips the causal convolution and Swish activation function to retain the original complete information.

[0090] Furthermore, the mLSTM unit receives the above input and selectively remembers or forgets information through its gating mechanism (forget gate, input gate, output gate) and multiplicative interaction terms, thereby better capturing long-term dependencies in the sequence.

[0091] Furthermore, the mLSTM unit output vector is normalized by GroupNorm, and the learnable jump input is added to the result.

[0092] Furthermore, the external output gate gates the above outputs to decide which information should be retained and passed through the network, and which information should be suppressed or ignored.

[0093] The last step is to down-project the results of the above processing, map the high-dimensional vector back to the original dimension, and obtain the probability distribution of the characters in the sequence through the softmax function. Figure 2 shown.

[0094] iv. Model pre-training

[0095] Inputting unlabeled data sets from large-scale databases into the network model helps the model capture common complex patterns and features in molecular structures, enhances its generalization ability, and enables it to better adapt to the generation task of target molecules. The pre-trained model architecture is based on the Transformer encoder and mLSTM decoder. Specifically, the input sequence is first converted to a vector representation through the embedding layer, and then the features are extracted through the Transformer encoder to obtain the output of the encoder. The decoder receives both the output of the encoder and the added <bos>The target sequence of characters is finally generated into a new molecular sequence. The cross entropy loss between the output and the true sequence is calculated, and the model parameters are updated through back propagation. The cross entropy loss function is used to measure the difference between the predicted SMILES string and the true SMILES string, and the formula is:

[0096]

[0097] Among them, y t is the tth character in the true sequence, is the probability distribution predicted by the model.

[0098] After each round of training, the model performance is evaluated on the validation set, the validation loss is recorded, and the optimal parameter model is finally saved. Figure 5 .

[0099] v. Model fine-tuning

[0100] The Transformer-mLSTM hybrid model that performed well in the pre-training stage was selected as the basic model. This model has been pre-trained on large-scale unlabeled data and has good molecular representation capabilities. Next, the weights of the pre-trained model are loaded as the initial weights, the original Transformer encoder parameters are kept unchanged, and its powerful self-attention mechanism continues to be used to capture long-distance dependencies in the molecular structure, and the decoder part is adjusted. The target molecule data (iron carrier database) is input into the network for training, and the parameter weights of the decoder are relearned to learn the specific patterns and features of the target molecule. The cross-entropy loss between the output and the true label is calculated, and the model parameters are updated through back propagation. The hyperparameters of each training, such as batch size, learning rate, optimizer parameters, etc., are manually adjusted to find the best hyperparameter combination that minimizes the cross-entropy loss function value. Multiple rounds of training are performed to retain the best parameter model. The fine-tuning flowchart is as follows Figure 6 shown.

[0101] vi. Molecular Generation

[0102] After pre-training and fine-tuning, the model has the ability to generate target molecules with specific biological activity. First, load the best model parameters after fine-tuning and determine the initial conditions for generation. <bos>Character marks the start of the input sequence. <bos>The character is converted into a vector representation through the embedding layer, and then the vector representation is passed as input to the mLSTM decoder. The output of the decoder generates the probability distribution of the next character through the Softmax layer. Then, according to the selected strategy (such as the Top k strategy), the next character is sampled from the probability distribution. The newly generated character is added to the current sequence and used as the next input of the decoder. Repeat the above steps until the end mark is generated. <eos>characters or the maximum length limit is reached. Molecular generation flowchart reference Figure 7 .

[0103] The above method combines the advantages of Transformer and mLSTM and can be used to generate complex, high-quality iron carrier molecular structures, greatly shortening the discovery cycle of new iron carriers and significantly reducing research costs.

[0104] Derivatization system for target molecules

[0105] On the other hand, the present application proposes a system for generating a derivative molecule of a target molecule, referring to Figure 8 The system includes: an expression pattern acquisition module 100 and a derivative molecule generation module 200.

[0106] 100 modules for obtaining an expression pattern of a target molecule selected from a siderophore.

[0107] Module 200 is used to input the expression pattern of the target molecule into the trained molecular generation model to generate a derivative molecule of the target molecule.

[0108] In some examples of the present application, the aforementioned modules are connected, and the aforementioned connection includes a physical connection or a network connection.

[0109] Those skilled in the art will appreciate that the features and advantages described above for the method for generating derivative molecules of target molecules are also applicable to the above-mentioned system and will not be described in detail here.

[0110] It should be understood that the system embodiment and the method embodiment may correspond to each other, and similar descriptions may refer to the method embodiment. To avoid repetition, they will not be described here. Specifically, Figure 8 The system shown can execute the above-mentioned embodiment of the method for generating a derivative molecule of the target molecule, and the operations and / or functions performed by each module in the system correspond to the method embodiment, which will not be described in detail here for the sake of brevity.

[0111] The system of the embodiment of the present application is described above from the perspective of the functional module in conjunction with the accompanying drawings. It should be understood that the functional module can be implemented in hardware form, can be implemented by instructions in software form, and can also be implemented by a combination of hardware and software modules. Specifically, the steps of the method embodiment in the embodiment of the present application can be completed by the hardware integrated logic circuit and / or software form instructions in the processor, and the steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or a combination of hardware and software modules in the decoding processor to perform. Optionally, the software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory, and completes the steps in the above method embodiment in conjunction with its hardware.

[0112] Siderophore-derived molecules

[0113] In yet another aspect, the present application provides an iron carrier derivative molecule produced by any of the aforementioned methods for producing a derivative molecule of a target molecule.

[0114] In some examples of the present application, the aforementioned siderophore derivative molecule has any structural formula as shown in Formula 1)-5);

[0115]

[0116]

[0117] The iron carrier molecules generated based on the derivative molecule generation method of the aforementioned target molecule can effectively enhance the iron binding ability and transport efficiency and have good antibacterial properties.

[0118] Computer program product, computing device, and computer readable storage medium

[0119] In another aspect, the present application provides a computer program product, a computing device and a computer-readable storage medium. Based on the computer program product, the computing device and the computer-readable storage medium, the method for generating a derivative molecule of the target molecule as described above is executed.

[0120] The so-called electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Computing devices may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are intended to be examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0121] like Fig. 9 As shown, the electronic device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 502 or a computer program loaded from a storage unit 508 to a RAM (Random Access Memory) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An I / O (Input / Output) interface 505 is also connected to the bus 504.

[0122] A number of components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0123] The computing unit 501 can be various general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, CPU (Central Processing Unit), GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as the derivative molecule generation method of the target molecule. For example, in some embodiments, the derivative molecule generation method of the target molecule can be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 500 via ROM502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by the computing unit 501, one or more steps of the method described above can be executed. Alternatively, in other embodiments, the computing unit 501 may be configured to execute the aforementioned method for generating a derivative molecule of the target molecule in any other appropriate manner (eg, by means of firmware).

[0124] In the present application, the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program may be printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or processing in other suitable ways as necessary, and then stored in a computer memory. The various computer-readable storage media described herein may represent one or more devices and / or other machine-readable storage media for storing information. The term "machine-readable storage medium" may include, but is not limited to, wireless channels and various other media capable of storing, containing, and / or carrying instructions and / or data.

[0125] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0126] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0127] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0128] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0129] Embodiments of the present invention will be described in more detail below, examples of which are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0130] Example 1: Construction and application of molecular generation model

[0131] 1. Construction of the siderophore molecular generation model

[0132] 1.1 Collect the SMILES strings and related properties of natural products and siderophore molecules required for model training from the COCONUT natural product database and the SIDERITE siderophore database, which are used in the subsequent pre-training and fine-tuning stages of the model. All collected SMILES strings are stored in .txt files for easy model reading.

[0133] 1.2 Use the Chem.MolFromSmiles function in the chemical toolkit RDKit to verify whether each SMILES string can correctly represent a molecule. If the SMILES string cannot be parsed into a valid molecular structure, it will be removed from the data set. Use the Chem.MolToSmiles function in the RDKit library to standardize the SMILES string to ensure the consistency and standardization of all SMILES strings, including atomic number, bond type, stereochemical information, etc. Then calculate the Morgan fingerprint of each molecule, and identify and remove duplicate molecular entries by comparing the similarity of the fingerprints. Finally, 400,000 molecules from the COCONUT natural product database and 600 molecules from the SIDERITE siderophore database were obtained. These molecules were divided into training sets and validation sets in a ratio of 8:2, which were used to train the model and evaluate the model training effect, respectively.

[0134] 1.3 Define the model architecture. The model adopts a hybrid structure based on Transformer encoder and mLSTM decoder. The Transformer encoder part uses a multi-head self-attention mechanism to learn features from multiple different representation subspaces and further process the features through a feedforward neural network layer. The mLSTM decoder part is responsible for generating new molecular sequences. The encoder receives the input sequence, where the input sequence is added at the end of the SMILES <eos>Mark to clarify the end of the molecular sequence. Then, the multi-head self-attention mechanism and feedforward neural network layer are used to extract features and output a fixed-dimensional vector representation. The decoder generates a new molecular sequence through mLSTM. The decoder not only receives the output of the encoder, but also receives the target sequence as input, where <bos>This ensures that the decoder uses the actual characters in the target sequence instead of the characters generated in the previous step when generating the next character at each step, so that it can better learn the correct sequence pattern during training, thereby improving the accuracy and diversity of generation.

[0135] 1.4 The model was pre-trained using data from COCONUT natural products to learn common patterns and features of natural products. During the pre-training process, the batch size was set to 64, the initial learning rate was 0.001, and the learning rate scheduler was used to dynamically adjust the learning rate. The number of training rounds was set to 40 rounds, and an early stopping strategy was adopted. Training was stopped when the performance on the validation set did not improve for 5 consecutive rounds. Each round of iteration traversed the training set. For each batch of data, the input sequence was first converted to a vector representation through the embedding layer, then the features were extracted through the Transformer encoder, and then a new molecular sequence was generated through the mLSTM decoder. The cross entropy loss between the output and the true sequence was calculated, and the model parameters were updated by backpropagation. After each round of training, the model performance was evaluated on the validation set and the validation loss was recorded. If the validation loss was lower than the historical best value, the current model parameters were saved as the best model. The saved best model parameters will be used for subsequent fine-tuning tasks.

[0136] 1.5 Load the pre-trained Transformer-mLSTM hybrid model. This model has been pre-trained on large-scale unlabeled data and has good molecular representation capabilities. During fine-tuning, only the parameters of the decoder are adjusted, while the parameters of the encoder remain frozen to retain the general feature representation learned by the pre-trained model. The model is fine-tuned using data from the SIDERITE siderophore database, with the batch size set to 32, the initial learning rate set to 0.0001, and the learning rate scheduler is used to dynamically adjust the learning rate. The number of training rounds is set to 20, and an early stopping strategy is adopted. Training is stopped when the performance on the validation set has not improved for 5 consecutive rounds. After each round of training, the model performance is evaluated on the validation set and the validation loss is recorded. If the validation loss is lower than the historical best value, the current model parameters are saved as the best model.

[0137] 1.6 will use a special start tag <bos>This token, which serves as the beginning of the input sequence, is converted into a vector representation through an embedding layer. The vector captures the position information of the token in the vocabulary and its semantic features. The converted vector is then fed into the mLSTM decoder. When the mLSTM decoder receives the input vector, it calculates the probability distribution of the next character based on the current hidden state and the input vector. This probability distribution reflects the likelihood of each possible character becoming the next character. Each time, the k characters with the highest probability are selected from the probability distribution as candidate characters, and then a character is randomly selected from them as the next character. The newly generated character is added to the current sequence and used as the next input to the decoder. This process is repeated until the end mark is encountered. <eos>Or until the preset maximum length limit is reached. The generated SMILES string is verified for chemical rationality by the chemical toolkit RDKit. For invalid molecular structures, regeneration is attempted.

[0138] 2. Formation of siderophore molecules

[0139] The natural siderophore molecular structure with the following molecular formula was selected and converted into a SMILES string format to suit the model input.

[0140]

[0141] Input the SMILES string into the siderophore molecule generation model trained and fine-tuned in step 1, output the derivative molecule SMILES string corresponding to the siderophore target molecule, and convert it into the structural formula shown in Formula 1)-5) (corresponding to Formula I)-Formula V)) through ChemDraw;

[0142]

[0143]

[0144] Example 2: Evaluation of Molecular Properties of Siderophore-Derived Molecules

[0145] The molecular properties of the siderophore-derived molecules obtained in Example 1 were evaluated and compared with the relevant properties of known siderophores. The specific steps are as follows:

[0146] 1. Evaluate the chemical plausibility of the derived molecules

[0147] Use the function Chem.rdmolops.SanitizeMol provided by RDKit to standardize and verify the molecule and evaluate whether the structure is reasonable. This step ensures that the number of connections for each atom matches its valence state and removes molecular structures with unreasonable bonds. Then use the Chem.GetSSSR function to obtain all minimum non-overlapping rings and check their sizes to ensure that the ring structure in the generated molecule is reasonable. The above two steps are integrated into a comprehensive evaluation function that accepts a SMILES string as input and returns a Boolean value indicating whether the molecule has passed all checks and removes derivative molecular structures that do not meet the set conditions.

[0148] 2. Evaluate the novelty of the derived molecules

[0149] The SMILES string was converted to a molecule object using the Chem.MolFromSmiles function in the RDKit library, and then converted back to the standardized SMILES format using the Chem.MolToSmiles function. Any redundant or irregular parts in the SMILES string were removed to ensure the consistency of the SMILES representations of all derivative molecules and known siderophore molecules. The same SMILES characters were removed to retain the derivative molecules with specific novel structures.

[0150] 3. Evaluate the similarity of the derivative molecules to known siderophore molecules

[0151] The Morgan fingerprint of the derivative molecule and the known siderophore molecule was calculated using the function AllChem.GetMorganFingerprintAsBitVect provided by RDKit. This is a variable-length binary vector based on the atomic environment that can capture the local structural features of the molecule. Then a similarity metric (Tanimoto coefficient) was used to quantify the similarity between the two molecular fingerprints. The Tanimoto coefficient measures the overlap of the two sets by calculating the ratio of the intersection to the union, thereby evaluating the molecular fingerprint similarity of the derivative siderophore molecule and the known siderophore molecule.

[0152] 4. Evaluation of the physicochemical properties of the derived molecules

[0153] The functions Chem.Descriptors and Chem.rdMolDescriptors provided by RDKit are used to obtain the relevant physicochemical properties of the derivative molecules, including molecular weight (Molecular Weight, MW), polar surface area (Polar Surface Area, PSA), distribution coefficient (LogP), number of hydrogen bond donors (NumHDonors, NHD) and number of hydrogen bond acceptors (NumHAcceptors, NHA), and number of rotatable bonds (NumRotatableBonds, NRB), so as to obtain the relevant physicochemical properties of the derivative molecules.

[0154] Taking the siderophore-derived molecules selected in Example 1 as an example, the corresponding evaluation data are shown as follows:

[0155] Similarity MW PSA LogP NHD NHA NRB Formula 1 0.867 671.749 255.090 -0.504 7 11 0 Formula 2 0.765 376.365 159.350 1.155 7 7 7 Formula 3 0.659 360.411 148.070 0.065 5 5 13 Formula 4 0.672 680.669 350.590 -5.971 12 14 23 Formula 5 0.757 220.228 84.910 0.411 2 4 2

[0156] Example 3: Testing of Iron Binding Capacity of Siderophore-Derived Molecules

[0157] The siderophore-derived molecule obtained in Example 1 was chemically synthesized and compared with the known siderophore molecules in terms of iron binding capacity.

[0158] Use an iron ion finger reagent (such as Chrome Azurol S, CAS) for colorimetric testing. The specific steps are as follows:

[0159] 1. Preparation of CAS reagents

[0160] Prepare Chrome Azurol S (CAS) solution: dissolve 0.06 g CAS in 50 mL deionized water, add 1 mL 10% HDTMA (hexadecyltrimethylammonium bromide) solution, and mix well.

[0161] Prepare iron (III) standard solution: prepare 200 μM FeCl solution and mix with 0.02 M HCl in a 1:1 ratio.

[0162] 2. Preparation of test solution

[0163] Test solutions of the derivatized molecules were prepared (at concentrations of 10 μM, 50 μM, 100 μM).

[0164] CAS solution, iron (III) solution and test solution were mixed in a 1:1:1 ratio and incubated at room temperature for 30 min.

[0165] 3. Data Collection

[0166] The absorbance of the mixed solution was measured at 630 nm using a UV-visible spectrophotometer.

[0167] The blank control group used only CAS and iron (III) solution without derivatized molecules.

[0168] 4. Data Analysis

[0169] Calculate the chelating capacity (Q):

[0170]

[0171] Among them, A0 is the absorbance of the blank control, A S is the absorbance of the sample.

[0172] 5. Results Analysis

[0173] The iron-binding capacity of the siderophore-derived molecule obtained in Example 1 is expected to be higher than that of the existing natural siderophore molecule.

[0174] Example 4: Antibacterial activity experiment of siderophore-derived molecules

[0175] The siderophore-derived molecules obtained in Example 1 were chemically synthesized. The antibacterial effect of the siderophore-derived molecules on Staphylococcus aureus was evaluated when iron resources were limited compared with natural siderophores. The specific experimental steps were as follows:

[0176] 1. Culture medium preparation

[0177] Prepare iron-restricted medium (M9 medium without adding iron source).

[0178] Different concentrations of the derivatized molecules were added (10 μM, 50 μM, 100 μM).

[0179] 2. Strain culture

[0180] Staphylococcus aureus was inoculated into M9 medium and cultured at 37°C overnight.

[0181] Dilute to OD600 = 0.1 (initial bacterial concentration is about 1×10 CFU / mL).

[0182] 3. Antibacterial test

[0183] The diluted bacterial solution was dispensed into a 96-well plate, 200 μL per well.

[0184] Different concentrations of the derivatized molecule solution were added.

[0185] A positive control group (no derivatized molecule added) and a negative control group (sterile culture medium) were set up.

[0186] 4. Determination of bacterial growth

[0187] Incubate at 37°C with shaking for 24 hours.

[0188] The absorbance at a wavelength of 600 nm (OD600) was measured using a microplate reader.

[0189] 5. Calculate the antibacterial rate

[0190]

[0191] Among them, OD A Table positive control; OD B Table sample.

[0192] 6. Results Analysis

[0193] The antibacterial ability of the siderophore-derived molecules obtained in Example 1 is expected to be higher than that of the existing natural siderophore molecules.

[0194] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the semantic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0195] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are illustrative and cannot be understood as limitations on the present application. Ordinary technicians in the field can change, modify, replace and modify the above embodiments within the scope of the present application without departing from the principles and purpose of the present application.< / eos> < / bos> < / bos> < / eos> < / eos> < / bos> < / bos> < / bos> < / eos> < / pad>

Claims

1. A method for generating a derivative molecule of a target molecule, characterized in that: include: S1, obtain the expression paradigm of the target molecule; S2, inputting the expression pattern of the target molecule into the trained molecular generation model to generate a derivative molecule of the target molecule; Wherein, the target molecule is selected from iron carriers.

2. The method according to claim 1, characterized in that: The molecular generation model is obtained by training in the following manner: I, preprocess multiple natural product data to obtain a standard expression paradigm suitable for model input; II. Pre-training the encoder-decoder model using the standard representation paradigm, including: Converting the standard expression paradigm into a vector representation based on the encoder embedding layer; Encoding the vector representation using a multi-head attention layer, a normalization layer, a residual connection, and a feed-forward neural network layer; Input the encoded vector into the decoder to obtain the output molecule of the model; The cross entropy loss is used to calculate the loss value between the output molecule and the standard expression paradigm; and back propagation is used to update the model parameters to obtain the pre-trained model.

3. The method according to claim 2, characterized in that The encoder is selected from Transformer, and the decoder is selected from Matrix LSTM.

4. The method according to claim 2 or 3, characterized in that: The training further comprises: III, preprocessing a plurality of siderophore data to obtain a standard expression paradigm of siderophore molecules suitable for input of the pre-trained model; IV, training the pre-trained model using the siderophore molecule standard expression paradigm, and relearning the decoder parameter weights of the pre-trained model; The cross entropy loss is used to calculate the loss value between the molecular encoding output of the pre-trained model and the standard expression paradigm of the siderophore molecule; and the model parameters are updated by back propagation to obtain a molecular generation model.

5. The method according to claim 4, characterized in that The pre-processing comprises: Acquiring an expression paradigm of target data, wherein the target data includes: natural product data and / or siderophore data; Analyzing the expression paradigm to determine a valid expression paradigm; Standardizing and removing duplicates from the effective expression paradigm to determine a standard expression paradigm; Preferably, the expression pattern is selected from a SMILES string.

6. The method according to claim 1, characterized in that The S2 step includes: Converting the initially labeled target molecule expression pattern into a vector in an encoder; The decoder generates the probability distribution of the next character based on the transformed vector; Randomly sample from the probability distribution, add the sampled character as the next character to the generated sequence, and use it as the next input of the decoder for recursive generation.

7. The method according to claim 6, characterized in that The recursively generated characters satisfy any of the following conditions and serve as the standard expression paradigm of the derivative molecule of the target molecule: 1) Generate an end tag; 2) The predetermined length limit is reached.

8. The method according to claim 7, characterized in that include: Molecular structure generation is performed on the standard expression paradigm of the derivative molecules of the target molecule.

9. The method according to claim 8, characterized in that include: Performing a molecular property evaluation on the molecular structure, wherein the evaluation is selected from at least one of chemical plausibility, novelty, physicochemical properties, or synthetic feasibility; Optionally, the method further comprises: regenerating an invalid molecular structure.

10. A system for generating a derivative molecule of a target molecule, characterized in that: include: An expression pattern acquisition module, used to obtain the expression pattern of the target molecule; A derivative molecule generation module, used for inputting the expression pattern of the target molecule into the trained molecule generation model to generate a derivative molecule of the target molecule; Wherein, the target molecule is selected from iron carriers.

11. A siderophore-derived molecule, characterized in that The target molecule derivative is produced by the method for producing a target molecule derivative according to any one of claims 1 to 9.

12. The siderophore-derived molecule according to claim 11, characterized in that Having any structural formula of formula 1)-5); 13. A computer program product, characterized in that include: Computer instructions; When part or all of the computer instructions are run on a computer, the method for generating a derivative molecule of a target molecule according to any one of claims 1 to 9 is executed.

14. A computing device, characterized in that include: Processor and memory; The memory is used to store computer programs; The processor is used to execute the computer program to implement the method for generating a derivative molecule of a target molecule according to any one of claims 1 to 9.

15. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions or programs, and when the computer instructions or programs are executed on a computer, the method for generating a derivative molecule of a target molecule according to any one of claims 1 to 9 is executed.

Images