A natural product molecule activity prediction method, device, equipment, medium and product based on molecular meta-structure

Abstract

The application discloses a natural product molecule activity prediction method and device based on molecular meta-structure, equipment, medium and product, and relates to the field of natural product molecule activity prediction. The natural product molecule is first split into fragments, and the structure of each fragment is taken as a meta-structure, and then the natural product molecule activity is predicted based on the meta-structure, so that the model can capture the key features of the natural product molecule more carefully through the local features and interactions reflected by the meta-structure, thereby improving the prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of predicting the molecular activity of natural products, and in particular to a method, apparatus, equipment, medium, and product for predicting the molecular activity of natural products based on molecular unit structure. Background Technology

[0002] In the field of predicting the activity of natural product molecules, traditional methods mainly rely on experimental testing and computational models based on the whole molecular structure. While experimental testing methods are accurate, they suffer from significant drawbacks such as high cost, long processing time, and high resource consumption, making it difficult to meet the needs of large-scale screening and rapid discovery of active ingredients. In recent years, with the rapid development of artificial intelligence and machine learning technologies, the use of deep learning models for molecular activity prediction has gradually become a research hotspot. These methods train models to learn the relationship between molecular structure and activity, thereby enabling the prediction of the activity of new molecules. However, most existing deep learning-based prediction methods still use the whole molecular structure as input, neglecting the local features and interactions of molecules, resulting in low prediction accuracy and an inability to reflect the true activity of natural product molecules. Summary of the Invention

[0003] The purpose of this application is to provide a method, apparatus, device, medium, and product for predicting the molecular activity of natural products based on molecular unit structure, so as to improve the accuracy of predicting the molecular activity of natural products.

[0004] To achieve the above objectives, this application provides the following solution.

[0005] In a first aspect, this application provides a method for predicting the molecular activity of natural products based on molecular unit structure, including:

[0006] The natural product molecule to be predicted is broken down into multiple fragments, and the structure of each fragment is taken as the metastructure.

[0007] Determine the SMILES string representation for each metastructure;

[0008] The SMILES string representation of each metastructure is input into the molecular property prediction model to obtain the activity prediction result of the natural product molecule to be predicted. The molecular property prediction model is obtained by training a large oracle model, which includes a sequence information representation module, a structural feature extraction module, and an MLP module. The sequence information representation module and the structural feature extraction module are both connected to the MLP module. The sequence information representation module is used to process the sequence information of each metastructure based on the SMILES string of each metastructure. The structural feature extraction module is used to process the structural features of each metastructure based on the SMILES string of each metastructure. The MLP module is used to connect the sequence information processing result and the structural feature processing result, and perform activity prediction to obtain the activity prediction result.

[0009] Secondly, this application provides a natural product molecular activity prediction device based on molecular unit structure. The device applies the aforementioned method for predicting the activity of natural product molecules based on molecular unit structure. The device includes:

[0010] The molecular decomposition module is used to decompose the natural product molecule to be predicted into multiple fragments, and to take the structure of each fragment as the metastructure.

[0011] The metastructure representation module is used to determine the SMILES string representation for each metastructure;

[0012] The activity prediction module is used to input the SMILES string representation of each metastructure into the molecular property prediction model to obtain the activity prediction result of the natural product molecule to be predicted. The molecular property prediction model is obtained by training a large oracle model, which includes a sequence information representation module, a structural feature extraction module, and an MLP module. The sequence information representation module and the structural feature extraction module are both connected to the MLP module. The sequence information representation module is used to process the sequence information of each metastructure based on the SMILES string of each metastructure. The structural feature extraction module is used to process the structural features of each metastructure based on the SMILES string of each metastructure. The MLP module is used to connect the sequence information processing result and the structural feature processing result, and perform activity prediction to obtain the activity prediction result.

[0013] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for predicting the molecular activity of natural products based on molecular unit structure.

[0014] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for predicting the activity of natural product molecules based on molecular unit structure.

[0015] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for predicting the activity of natural product molecules based on molecular unit structure.

[0016] According to the specific embodiments provided in this application, this application has the following technical effects.

[0017] This application provides a method, apparatus, device, medium, and product for predicting the activity of natural product molecules based on molecular metastructures. This application first breaks down the natural product molecule into fragments, and takes the structure of each fragment as a metastructure; then, it predicts the activity of the natural product molecule based on the metastructure. Through the local features and interactions reflected by the metastructure, the model can capture the key features of the natural product molecule in greater detail, thereby improving the accuracy of the prediction. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a method for predicting the molecular activity of natural products based on molecular unit structure, provided in one embodiment of this application.

[0020] Figure 2 This is a schematic diagram illustrating the principle of a method for predicting the molecular activity of natural products based on molecular unit structure, provided in an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of the architecture of a sequence information representation module provided in an embodiment of this application.

[0022] Figure 4 This is a schematic diagram of the architecture of a structural feature extraction module provided in an embodiment of this application.

[0023] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] As a component of a molecule, the molecular structure can reflect the local features and interactions of the molecule in greater detail, which is of great significance for improving the accuracy of predictions.

[0027] To address the numerous shortcomings of existing technologies in predicting the molecular activity of natural products, this application provides a method for predicting the molecular activity of natural products based on molecular unit structure. This method aims to improve the accuracy and efficiency of prediction, reduce R&D costs, and accelerate the discovery process of active ingredients, thus possessing significant research value and practical significance.

[0028] In one exemplary embodiment, such as Figure 1 As shown, a method for predicting the molecular activity of natural products based on molecular unit structure is provided, including the following steps 101 to 103.

[0029] Step 101: The natural product molecule to be predicted is broken down into multiple fragments, and the structure of each fragment is taken as the metastructure.

[0030] Step 102: Determine the SMILES string representation for each metastructure.

[0031] Step 103: Input the SMILES string representation of each metastructure into the molecular property prediction model to obtain the activity prediction result of the natural product molecule to be predicted; the molecular property prediction model is obtained by training a big oracle model, which includes a sequence information representation module, a structural feature extraction module, and an MLP module; the sequence information representation module and the structural feature extraction module are both connected to the MLP module. The sequence information representation module is used to process the sequence information of each metastructure based on the SMILES string of each metastructure, the structural feature extraction module is used to process the features of each metastructure based on the SMILES string of each metastructure, and the MLP module is used to connect the sequence information processing result and the structural feature processing result, and perform activity prediction to obtain the activity prediction result.

[0032] Implementing steps 101 to 103 above can improve the accuracy of predicting the molecular activity of natural products.

[0033] In another exemplary embodiment, step 101 specifically includes steps 201 and 202.

[0034] Step 201: By simulating the fragmentation process of the natural product to be predicted under mass spectrometry conditions, the chemical bond breaking combinations of the natural product to be predicted during the fragmentation process are identified.

[0035] Step 202: Based on the chemical bond breaking combination, the natural product molecule to be predicted is broken down into multiple fragments.

[0036] In another exemplary embodiment, step 103 above includes steps 301-305 as follows.

[0037] Step 301: Data collection and preprocessing.

[0038] This application requires the extensive collection of molecular structure data of natural products from publicly available databases and literature. This data should cover a wide range of natural products to ensure the generalization ability of subsequent models. The collected data needs to be preprocessed, including removing duplicates and cleaning noisy data, to ensure data quality.

[0039] The data in this application primarily comes from publicly available databases and literature, such as PubChem and ChEMBL. These databases provide a wealth of natural product molecular structure data, including molecular formulas, SMILES strings, InChI bonds, and other information. To ensure the diversity and representativeness of the data, this application collected over 400,000 natural product molecular structure data from multiple databases.

[0040] The BBBP (Blood-Brain Barrier Penetration), Tox21, ClinTox, BACE (Beta-secretase Cleavage Site Prediction), and SIDER (Side Effect Resource) datasets from the MoleculeNet database were used as high-throughput screening datasets. The BBBP task aims to predict whether small molecule drugs can cross the blood-brain barrier (BBB); Tox21 contains 12 independent target prediction subtasks to assess the potential impact of molecules on a range of toxic pathways, including nuclear receptor activity, stress response, and ion channel blockade; the ClinTox dataset contains two prediction tasks: predicting whether a molecule has clinical toxicity (whether an FDA-approved molecule has been withdrawn due to toxicity) and whether it has mutagenicity (Ames test); BACE1 is an enzyme associated with Alzheimer's disease, and the task aims to predict molecules with BACE1 inhibitory capabilities; the SIDER dataset contains 27 different side effect prediction subtasks, aiming to predict whether a given drug may cause specific side effects, which is crucial for assessing drug safety.

[0041] The collected data may contain duplicates, missing data, or inconsistent formats, thus requiring data cleaning. First, a hash algorithm is used to remove duplicate data items. Second, missing data is filled in using context information or interpolation. Finally, the data format is standardized, such as converting the SMILES string into a uniform representation for easier subsequent processing.

[0042] Step 302: Generation of molecular unit structures.

[0043] After data preprocessing, this application employs a specific algorithm to decompose natural product molecules into multiple molecular substructures. These molecular substructures should contain specific chemical information and connectivity relationships to reflect the local features and interactions of the molecules, enabling the model to capture key molecular features more precisely, thereby improving prediction accuracy.

[0044] This application employs a rule-based fragmentation algorithm. This algorithm simulates the fragmentation process of natural products under mass spectrometry conditions, identifies specific chemical bond breaking combinations, and then disassembles the molecule into multiple fragments based on these structural units. Each fragment retains some structural and chemical information of the original molecule; this structural and chemical information constitutes the metastructure.

[0045] The generated metastructures may contain redundant or irrelevant information, thus requiring screening and optimization. This application screens metastructures based on factors such as size, complexity, and relevance to molecular activity to ensure that each metastructure effectively reflects the local features and interactions of the molecule.

[0046] To input metastructures into the model, they need to be converted into an appropriate representation. In this embodiment, SMILES strings are used to represent metastructures. Each metastructure's SMILES string contains information such as its atomic connection order and chemical bond types.

[0047] Step 303: Construct a molecular property prediction model.

[0048] Based on the generated metastructure data, this application proposes a large oracle model architecture, including a sequence information representation module, a structural feature extraction module, and an MLP module. The sequence information representation module processes the sequence information of the molecular metastructure, while the structural feature extraction module focuses on processing the structural features of the molecular metastructure. By capturing the spatial relationships within and between metastructures, it further mines the structural information of the molecule. The two modules work together to achieve efficient encoding and decoding of the molecular metastructure.

[0049] The sequence information encoding module consists of 2-12 Transformer encoder units (i.e., the first Transformer encoder) and is responsible for processing the sequence information of molecular metastructures. It employs a self-attention mechanism to capture the sequential associations and interactions between metastructures. Specifically, this module converts the input SMILES string into an embedding vector sequence, which is then processed through multiple layers of self-attention layers and a feedforward neural network.

[0050] The first Transformer encoder focuses on processing the features of molecular metastructures. It takes the metastructure vectors obtained in the previous step as input and outputs a feature tensor of equal dimension. Then, global feature average pooling compresses the dimension of the number of metastructures to 1, and the final output is the feature representation of each metastructure.

[0051] The first Transformer encoder includes several transformer encoder blocks, which, combined with an atomic-level sequence segmenter, can encode any SMILES (Simplified molecular input line entry system) into a fixed-length feature vector.

[0052] like Figure 3 As shown, the sequence information representation module specifically includes:

[0053] 1. Tokenizer.

[0054] Input: The string SMILES.

[0055] Output: The tokenized SMILES sequence.

[0056] Calculation process: Each character or special symbol in the SMILES string is converted into a corresponding token (integer index) using a preset vocabulary.

[0057] 2. Embedding layer.

[0058] Input: A sequence of tokenized SMILES (length L, where L is the number of tokens in the sequence).

[0059] Output: A sequence of embedding vectors (each label corresponds to a d-dimensional embedding vector).

[0060] Calculation process: For each tag ti (i = 1, 2, ..., L) in the sequence, find its corresponding embedding vector through the embedding matrix E ∈ R | V | × d, where | V | is the size of the vocabulary and d is the dimension of the embedding vector.

[0061] The calculation formula is: ei = E[ti], where ei ∈ Rd is the embedding vector of the label ti.

[0062] 3. Transformer encoder layer.

[0063] Input: Embedded vector sequence.

[0064] Output: A sequence of encoded vectors processed by self-attention.

[0065] The calculation process is as follows:

[0066] Positional encoding: To incorporate positional information into the sequence, rotated positional embeddings are used. For each embedding vector ei, its positional encoding pi is calculated, and then the two are added together to obtain a vector xi = ei + pi containing positional information.

[0067] Self-attention mechanism: Calculate the query, key, and value matrix: Q = XWQ, K = XWK, V = XWV, where X = [x1, x2, ..., xL] is a stack of input vector sequences, WQ, WK, WV ∈ Rd×dk is a learnable parameter matrix, and dk is the dimension of the query, key, and value vectors.

[0068] Calculate attention score and output using a linear attention mechanism:

[0069]

[0070] Feedforward Neural Network: A two-layer feedforward neural network is applied to the output of the self-attention mechanism. The first layer is a linear transformation plus an activation function (such as ReLU), and the second layer is another linear transformation.

[0071] 4. Multiple first Transformer encoders stacked

[0072] Input: The output encoded vector sequence of the previous layer.

[0073] Output: A deeper sequence of encoded vectors.

[0074] Calculation process:

[0075] The computation process is the same as that of a single first Transformer encoder, but it is repeated multiple times (stacked layers). Each layer receives the output of the previous layer as input and produces a new output. By stacking multiple layers, the model is able to capture more complex features and dependencies.

[0076] The sequence information representation module described above uses the training method of Masked Language Model (MLM). During training, a portion of the labels in the input sequence are randomly masked. The model then calculates the loss (such as cross-entropy loss) for predicting these masked labels and updates the model parameters using the backpropagation algorithm.

[0077] like Figure 4 As shown, the second transformer encoder in the structural feature extraction module consists of several transformer encoder modules. It takes the complete metastructure encoding vectors of the molecule obtained in the previous step as input and outputs a specific task (such as activity prediction, toxicity prediction, etc.) to extract the hidden structure-function relationships in the metastructure feature vectors. Through processing by multiple encoding layers, it utilizes a multi-head self-attention mechanism to capture dependencies in the sequence, and further processes and transforms the feature representation through a feedforward neural network. The structure and working principle of its second transformer encoder are consistent with the first transformer encoder, and will not be described again here.

[0078] The MLP module uses a multivariate perceptron with a fully connected layer to connect the two feature representations. Then, it outputs the activity prediction results of the natural product molecule through a fully connected layer and a sigmoid activation function.

[0079] Step 304: Model training and validation.

[0080] After the model architecture was constructed, this embodiment trained the model using the collected molecular substructure data. During training, by adjusting model parameters and optimizing the algorithm, the model was able to accurately predict the active tags corresponding to the molecular substructures. To verify the model's predictive ability, this application also tested it on multiple independent datasets and compared it with existing methods. The test results show that the method proposed in this application performs excellently in terms of prediction accuracy and generalization ability.

[0081] 1. Dataset partitioning.

[0082] To train and validate the model, the sample dataset was divided into training, validation, and test sets in an 8:1:1 ratio. The training set was used to train the model parameters, the validation set was used to tune hyperparameters and prevent overfitting, and the test set was used to evaluate the final performance of the model.

[0083] 2. Loss function and optimization algorithm.

[0084] The loss function used in this embodiment is the cross-entropy loss function, which measures the difference between the model's predicted results and the true labels. The optimization algorithm employs the Adam optimizer, which combines the advantages of the momentum method and the RMSprop optimizer, exhibiting fast convergence speed and good robustness.

[0085] 3. Training process.

[0086] During training, mini-batch gradient descent was used to update model parameters. Each batch contained a certain number of molecular structure samples. An early stopping mechanism was implemented to prevent overfitting; training stopped when performance on the validation set no longer improved. Furthermore, a learning rate decay strategy was employed to gradually reduce the learning rate, thereby accelerating convergence and improving the model's generalization ability. Figure 2 As shown.

[0087] 4. Model validation.

[0088] To validate the model's performance, it was evaluated on a test set. The evaluation metric was the area under the receiver operating characteristic curve (AUROC). Furthermore, comparative experiments were conducted with other existing molecular activity prediction methods to demonstrate the superiority of the proposed method. The results are shown in Table 1.

[0089] Table 1. Performance comparison of the example model with other models (AUROC)

[0090] BBBP Tox21 ClinTox2 HIV BACE SIDER RF 71.4 76.9 71.3 78.1 86.70 68.4 SVM 72.9 81.8 66.9 79.2 86.20 68.2 MGCN 85.0 70.7 63.4 73.8 73.4 55.2 D-MPNN 71.2 68.9 90.5 75.0 85.3 63.2 DimeNet - 78.0 76.0 - - 61.6 Hu 70.8 78.7 78.9 80.2 85.9 65.2 N-gram 91.2 76.9 85.5 83 87.6 63.2 GraphMVP-C 72.4 74.4 77.5 77.0 81.2 63.9 GEM 72.4 78.1 90.1 80.6 85.6 67.2 ChemBERTa 64.3 - 90.6 62.2 - - MOLFormer 92.6 82.4 89.2 80.9 87.9 61.1 This embodiment 93.34 82.41 91.57 79.62 88.28 66.81

[0091] Step 305: Post-processing.

[0092] The trained model can be applied to predict the activity of natural product molecules. By inputting the molecular structure of a new natural product, the model can quickly output its activity prediction results. To further improve the practicality and reliability of the prediction results, this application also includes a post-processing step for the prediction results. For example, removing noisy data and screening for highly active molecules, to ensure the accuracy and application value of the prediction results, specifically including:

[0093] First, noisy data in the prediction results, namely those samples with extremely low or extremely high prediction probabilities, were removed.

[0094] Secondly, the prediction results are screened and sorted according to business needs in order to identify molecules with high activity.

[0095] Then, the selected molecules were further experimentally verified and analyzed.

[0096] Compared with existing technologies, the method for predicting the activity of natural product molecules based on molecular unit structure in this application has the following significant advantages:

[0097] Improving prediction accuracy: By introducing the concept of molecular metastructure and combining it with deep learning technology, this application is able to capture the local features and interactions of molecules in greater detail, thereby improving the accuracy of prediction.

[0098] Reduced R&D costs: Compared with traditional experimental testing methods, this application does not require a large number of experimental operations and material consumption, thereby significantly reducing R&D costs.

[0099] Improved prediction efficiency: By leveraging the powerful computing capabilities of deep learning models, this application can predict the activity of a large number of natural product molecules in a short time, significantly improving prediction efficiency.

[0100] Enhanced generalization ability: By constructing diverse molecular metastructure datasets and conducting thorough training and validation of the model, the method proposed in this application has strong generalization ability and can be applied to different types of natural product molecular activity prediction tasks.

[0101] Promoting New Drug Development: This application provides an efficient and accurate method for predicting molecular activity in fields such as new drug development, which helps to accelerate the process of new drug development.

[0102] In summary, the method for predicting the molecular activity of natural products based on molecular unit structure proposed in this application has significant technical advantages and broad application prospects, and is expected to bring about revolutionary changes in new drug development.

[0103] Based on the same inventive concept, this application also provides a molecular structure-based natural product molecular activity prediction device for implementing the above-mentioned method for predicting the activity of natural product molecules based on molecular structure. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the molecular structure-based natural product molecular activity prediction device provided below can be found in the limitations of the molecular structure-based natural product molecular activity prediction method described above, and will not be repeated here.

[0104] In one exemplary embodiment, a natural product molecular activity prediction device based on molecular unit structure is provided, comprising:

[0105] The molecular decomposition module is used to break down the natural product molecule to be predicted into multiple fragments, and to treat the structure of each fragment as the metastructure.

[0106] The metastructure representation module is used to determine the SMILES string representation of each metastructure.

[0107] The activity prediction module is used to input the SMILES string representation of each metastructure into the molecular property prediction model to obtain the activity prediction result of the natural product molecule to be predicted. The molecular property prediction model is obtained by training a large oracle model, which includes a sequence information representation module, a structural feature extraction module, and an MLP module. The sequence information representation module and the structural feature extraction module are both connected to the MLP module. The sequence information representation module is used to process the sequence information of each metastructure based on the SMILES string of each metastructure. The structural feature extraction module is used to process the features of each metastructure based on the SMILES string of each metastructure. The MLP module is used to connect the sequence information processing result and the structural feature processing result, and perform activity prediction to obtain the activity prediction result.

[0108] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 5As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting the molecular activity of natural products based on molecular unit structure.

[0109] Those skilled in the art will understand that Figure 5 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0110] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0111] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0112] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0113] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0114] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0115] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the molecular activity of natural products based on molecular unit structure, characterized in that, include: The natural product molecule to be predicted is broken down into multiple fragments, and the structure of each fragment is taken as the metastructure. Determine the SMILES string representation for each metastructure; The SMILES string representation of each metastructure is input into the molecular property prediction model to obtain the activity prediction result of the natural product molecule to be predicted. The molecular property prediction model is obtained by training a large oracle model, which includes a sequence information representation module, a structural feature extraction module, and an MLP module. The sequence information representation module and the structural feature extraction module are both connected to the MLP module. The sequence information representation module is used to process the sequence information of each metastructure based on the SMILES string of each metastructure. The structural feature extraction module is used to process the structural features of each fragment based on the SMILES string of each metastructure. The MLP module is used to connect the sequence information processing result and the structural feature processing result, and perform activity prediction to obtain the activity prediction result.

2. The method for predicting the molecular activity of natural products based on molecular unit structure according to claim 1, characterized in that, The natural product molecule to be predicted is broken down into multiple fragments, specifically including: By simulating the fragmentation process of the natural product to be predicted under mass spectrometry conditions, the chemical bond breaking and combination of the natural product to be predicted during the fragmentation process can be identified. Based on the aforementioned chemical bond breaking combinations, the natural product molecule to be predicted is broken down into multiple fragments.

3. The method for predicting the molecular activity of natural products based on molecular unit structure according to claim 1, characterized in that, The sequence information representation module includes: a plurality of first Transformer encoders connected in sequence; The first Transformer encoder includes multiple layers of self-attention layers and feedforward neural network layers.

4. The method for predicting the molecular activity of natural products based on molecular unit structure according to claim 1, characterized in that, The structural feature extraction module includes a global feature average pooling layer, an output layer, and multiple second Transformer encoders connected in sequence. The last of the multiple second Transformer encoders connected in sequence is connected to the global feature average pooling layer, which is in turn connected to the output layer.

5. The method for predicting the molecular activity of natural products based on molecular unit structure according to claim 1, characterized in that, The training process for the Big Prophecy model includes: Obtain structural and activity data of natural product molecules with known activity from the data source, and construct a dataset; A hash algorithm is used to remove duplicate data items from the dataset to obtain a deduplicated dataset; the data items include structural data and activity data of natural product molecules with known activity. Missing data in each data item of the deduplicated dataset is filled in using contextual information or interpolation to obtain the filled dataset; Each known-activity natural product molecule in the imputed dataset is split into multiple fragments, and the SMILES string representation of each fragment of each known-activity natural product molecule is used to form the sample input. The activity data of each known-activity natural product molecule is used as the sample label to construct the sample dataset. The sample dataset is divided into a training set, a test set, and a validation set; The big oracle model is trained, tested, and validated based on the training set, the test set, and the validation set to obtain a well-trained big oracle model, which serves as a molecular attribute prediction model.

6. The method for predicting the molecular activity of natural products based on molecular unit structure according to claim 5, characterized in that, During the training, testing, and validation of the Big Oracle model, the cross-entropy loss function is used to calculate the loss of the Big Oracle model obtained in each iteration of training. During the training, testing, and validation of the big oracle model, the parameters of the big oracle model are updated using mini-batch gradient descent. During the training, testing, and validation of the big oracle model, a learning rate decay strategy is adopted to gradually reduce the learning rate.

7. A device for predicting the molecular activity of natural products based on molecular unit structure, characterized in that, The molecular structure-based natural product molecular activity prediction device uses the molecular structure-based natural product molecular activity prediction method according to any one of claims 1-6, and the molecular structure-based natural product molecular activity prediction device comprises: The molecular decomposition module is used to decompose the natural product molecule to be predicted into multiple fragments, and to take the structure of each fragment as the metastructure. The metastructure representation module is used to determine the SMILES string representation for each metastructure; The activity prediction module is used to input the SMILES string representation of each metastructure into the molecular property prediction model to obtain the activity prediction result of the natural product molecule to be predicted. The molecular property prediction model is obtained by training a large oracle model, which includes a sequence information representation module, a structural feature extraction module, and an MLP module. The sequence information representation module and the structural feature extraction module are both connected to the MLP module. The sequence information representation module is used to process the sequence information of each metastructure based on the SMILES string of each metastructure. The structural feature extraction module is used to process the structural features of each metastructure based on the SMILES string of each metastructure. The MLP module is used to connect the sequence information processing result and the structural feature processing result, and perform activity prediction to obtain the activity prediction result.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method for predicting the molecular activity of natural products based on molecular unit structure as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for predicting the molecular activity of natural products based on molecular structure as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for predicting the molecular activity of natural products based on molecular structure as described in any one of claims 1-6.

Images