A method and system for intelligently predicting the molecular structure of microbial secondary metabolites
By using Pfam protein family annotation and the Transformer deep neural network model, the generalization ability and accuracy problems of the prediction of the molecular structure of microbial secondary metabolites in the prior art have been solved, realizing efficient and high-precision prediction of the molecular structure of microbial secondary metabolites, supporting the design of natural drugs and synthetic biology experiments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI INST OF ORGANIC CHEM CHINESE ACAD OF SCI
- Filing Date
- 2023-11-22
- Publication Date
- 2026-07-21
AI Technical Summary
Existing rule-based techniques for predicting the molecular structure of microbial secondary metabolites suffer from limited generalization ability and low accuracy in chemical structure, making it difficult to predict the molecular structure of microbial secondary metabolites efficiently and with high precision.
The HMMER tool with the Pfam protein family annotation database was used for protein family annotation. Combined with one-hot encoding and the Transformer deep neural network prediction model, the feature word sequence of biosynthetic gene clusters was converted into a binary vector, which was then input into a pre-trained neural network model to output the SMILES character sequence of the molecular structure of microbial secondary metabolites.
It achieves efficient and high-precision prediction of the molecular structure of unknown microbial secondary metabolites, with a wide coverage, breaking through the limitations of rule-based prediction tools, supporting synthetic biology experiments and natural drug design, and promoting the research and development of natural drugs.
Smart Images
Figure CN117558360B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method and system for intelligently predicting the molecular structure of microbial secondary metabolites, belonging to the field of information processing technology. Background Technology
[0002] Microbial secondary metabolites are small, complex chemical substances synthesized by microorganisms through secondary metabolism after they have grown to a certain stage, under specific conditions such as time and environment. Microbial secondary metabolites are an important source of natural drugs, such as antibiotics and statins, and are therefore a key research subject in the biomedical field.
[0003] Currently, the prediction of the molecular structure of microbial secondary metabolites is based on rule-based prediction techniques. The prediction process generally involves: ① extracting the core biosynthetic enzyme sequence and auxiliary biosynthetic enzyme sequence encoded by the biosynthetic gene cluster; ② using sequence similarity in a rule base to find the substrate structure corresponding to the core biosynthetic enzyme sequence; ③ using sequence similarity in the rule base to find the biosynthetic modification and trimming reaction template corresponding to the auxiliary biosynthetic enzyme sequence; ④ constructing a virtual biosynthetic reaction based on the substrate structure and the biosynthetic modification and trimming reaction template; ⑤ inferring the product structure of the substrate structure as a reactant based on the virtual biosynthetic reaction. The main drawbacks of this rule-based prediction technique are: 1) Due to limitations imposed by the rule base and reaction template, the types of biosynthetic enzymes that can be identified and the types of secondary metabolites that can be predicted are very limited, often only predicting the simplest secondary metabolite molecular structures; 2) Because the reaction sites on the substrate structure and the order and number of virtual biosynthetic reactions cannot be predicted, the accuracy of the predicted chemical structure of the secondary metabolites is also low.
[0004] In other words, existing rule-based prediction techniques suffer from limitations due to the constraints of rule bases and reaction templates, such as limited generalization ability and poor accuracy in predicting the chemical structure of secondary metabolites. There is an urgent need in this field to develop a prediction technique that can not only efficiently and accurately predict the molecular structure of microbial secondary metabolites, but also has a wide range of predictive functions. Summary of the Invention
[0005] In view of the above-mentioned problems and needs of existing technologies, the purpose of this invention is to provide a method and system for intelligently predicting the molecular structure of microbial secondary metabolites, so as to provide scientific reference for synthetic biology experiments and natural drug design, thereby accelerating and promoting the process of natural drug research and development.
[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:
[0007] A method for intelligently predicting the molecular structure of microbial secondary metabolites includes the following steps:
[0008] S1) Use the HMMER tool with the Pfam protein family annotation database to annotate the biosynthetic enzyme sequences encoded by the input biosynthetic gene clusters, and treat each protein family number obtained from the annotation as a word. Combine all the annotated protein family numbers in sequence to obtain the biosynthetic feature word sequence.
[0009] S2) Based on the pre-constructed biosynthetic gene cluster feature coding table, the protein family numbers in the biosynthetic feature word sequence are converted into numerical codes one by one, and the obtained numerical codes are combined in sequence to obtain the biosynthetic feature word coding sequence.
[0010] S3) Use the one-hot encoding method to convert the biosynthetic feature word codes one by one into binary vectors, and combine the resulting vectors in order to obtain the biosynthetic feature word vector sequence;
[0011] S4) Input the biosynthetic feature word vector sequence into the pre-trained Transformer deep neural network prediction model, and output the SMILES word encoding sequence of the molecular structure of microbial secondary metabolites through model prediction.
[0012] S5) Based on the pre-constructed SMILES coding table of microbial secondary metabolite molecular structures, the SMILES word coding sequence predicted by the model is converted into a SMILES character sequence, thereby obtaining the predicted microbial secondary metabolite molecular structure.
[0013] In a preferred embodiment, in step S1), the annotation threshold is set to e-value < 0.01.
[0014] One implementation scheme, the construction of the biosynthetic gene cluster feature coding table in step S2), includes the following steps:
[0015] S21) Download existing known biosynthetic gene cluster sequence files from the MIBiG (Minimum Information about a Biosynthetic Gene cluster) database;
[0016] S22) Extract the biosynthetic enzyme sequence data encoded by each biosynthetic gene cluster from the downloaded existing known biosynthetic gene cluster sequence file;
[0017] S23) Use the HMMER tool with the Pfam protein family annotation database to annotate all the biosynthetic enzyme sequences extracted in step S22);
[0018] S24) Remove duplicates from all protein family IDs obtained from step S23) and arrange them in order. The table consisting of the protein family IDs and their corresponding sequence numbers is the biosynthetic gene cluster feature coding table.
[0019] In one embodiment, the digital code in step S2) is the sequence number corresponding to the protein family number in the biosynthetic gene cluster feature coding table.
[0020] One implementation scheme, the training of the Transformer deep neural network prediction model described in step S4), includes the following steps:
[0021] S41) Randomly divide the dataset into a training dataset and a validation dataset in a ratio of 9:1;
[0022] S42) Input the training dataset and validation dataset into the Transformer deep neural network, with each batch containing 16 pairs of data.
[0023] S43) Using the AdamW optimizer and cross-entropy loss function, train and validate the Transformer deep neural network in batches. When the prediction performance of the neural network on the validation dataset does not improve for 10 consecutive batches, stop training and use the saved optimal neural network state parameters as the Transformer deep neural network prediction model.
[0024] A preferred approach is to have a learning rate of 0.0001.
[0025] In a preferred embodiment, the Transformer deep neural network has the following characteristics: 1020 input sequence features, 35 output sequence features, 512 hidden layers, 8 multi-head attention layers, 6 encoder layers, and 6 decoder layers.
[0026] One implementation scheme, the acquisition of the dataset in step S41), includes the following processing steps:
[0027] S411) Download data files from the MIBiG database, including biosynthetic gene cluster sequence files and biosynthetic gene cluster annotation files;
[0028] S412) Extract the biosynthetic enzyme sequence data encoded by each biosynthetic gene cluster from the biosynthetic gene cluster sequence file;
[0029] S413) SMILES sequence data of microbial secondary metabolite molecular structures were extracted from biosynthetic gene cluster annotation files;
[0030] S414) Based on the MIBiG database number, the biosynthetic enzyme sequence encoded by each biosynthetic gene cluster is paired with the SMILES sequence of the corresponding microbial secondary metabolite molecular structure. The combination of all data pairs is the dataset.
[0031] In one embodiment, the input data for training a Transformer deep neural network is a sequence of word vectors representing biosynthetic features, and the output target data for training a Transformer deep neural network is a sequence of word vectors representing the molecular structures of microbial secondary metabolites (SMILES).
[0032] A further implementation scheme, the acquisition of the SMILES word vector sequence of the molecular structure of the microbial secondary metabolites includes the following steps:
[0033] A) Based on the pre-constructed SMILES coding table of microbial secondary metabolite molecular structures, convert the SMILES sequences in the dataset into SMILES word coding sequences.
[0034] B) The SMILES word codes are converted into binary vectors one by one using the one-hot encoding method. The resulting vectors are then combined in sequence to obtain the SMILES word vector sequence of the molecular structure of the microbial secondary metabolites.
[0035] A further implementation scheme, the construction of the SMILES coding table for the molecular structures of microbial secondary metabolites, includes the following steps:
[0036] a) Download data files from the MIBiG database, including biosynthetic gene cluster annotation files;
[0037] b) Extract SMILES sequence data of microbial secondary metabolite molecular structures from biosynthetic gene cluster annotation files;
[0038] c) The SMILES sequence of the extracted microbial secondary metabolite molecular structure is segmented by character, then all characters appearing in the SMILES sequence are extracted, and the obtained SMILES characters are deduplicated and arranged in order.
[0039] d) The table consisting of SMILES characters and their corresponding sequence numbers is the SMILES encoding table of the molecular structure of microbial secondary metabolites.
[0040] In a further implementation scheme, the SMILES word encoding refers to the sequence number corresponding to each SMILES character in the SMILES encoding table.
[0041] A system for intelligently predicting the molecular structure of microbial secondary metabolites, comprising:
[0042] An annotation module is used to annotate the input biosynthetic gene cluster sequence as a biosynthetic feature word sequence;
[0043] An encoding module for converting biosynthetic feature word sequences into biosynthetic feature word encoded sequences;
[0044] An embedder for converting a sequence of biosynthetic feature word encodings into a sequence of biosynthetic feature word vectors;
[0045] A Transformer deep neural network prediction model was used to predict the SMILES word-encoded sequence of the output microbial secondary metabolite molecular structure based on the input biosynthetic feature word vector sequence; and
[0046] The decoding module is used to convert the SMILES word encoding sequence of the model prediction output into a SMILES character sequence to obtain the predicted molecular structure of microbial secondary metabolites.
[0047] In addition, the present invention also provides a storage medium storing one or more programs including execution instructions, which can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the above-described method for intelligently predicting the molecular structure of microbial secondary metabolites.
[0048] The present invention also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above for intelligently predicting the molecular structure of microbial secondary metabolites.
[0049] The present invention also provides a computer program product, the computer program product comprising a computer program stored on a storage medium, the computer program comprising program instructions, which, when executed by a computer, enable the computer to execute the above-described method for intelligently predicting the molecular structure of microbial secondary metabolites.
[0050] Compared with the prior art, the beneficial technical effects of the present invention are as follows:
[0051] This invention utilizes artificial intelligence technology to not only intelligently predict the molecular structure of unknown microbial secondary metabolites based on the characteristics of input biosynthetic gene clusters, but also eliminates the need for rule base construction and manual data collection. It has a wide coverage, overcomes the limitations of rule-based prediction tools, and significantly improves prediction efficiency and accuracy. It can provide scientific support for synthetic biology experiments and natural drug design, and has important application value for accelerating and promoting the research and development of natural drugs. Attached Figure Description
[0052] Figure 1 The diagram shown is a schematic diagram of a system for intelligently predicting the molecular structure of microbial secondary metabolites provided by the present invention. Detailed Implementation
[0053] The technical solution of the present invention will be further described in detail and completely below with reference to the embodiments and accompanying drawings.
[0054] Example
[0055] A method for intelligently predicting the molecular structure of microbial secondary metabolites includes the following steps:
[0056] S1) The HMMER tool with the Pfam protein family annotation database is used to annotate the biosynthetic enzyme sequences encoded by the input biosynthetic gene clusters (preferably, the annotation threshold is set to e-value < 0.01). Each protein family number obtained from the annotation is treated as a word, and all the annotated protein family numbers are combined sequentially to obtain a biosynthetic feature word sequence. The biosynthetic gene cluster refers to the combination of all genes responsible for the biosynthesis of a certain secondary metabolite in microorganisms, and these genes exist in clusters in the microbial genome. The biosynthetic enzyme refers to the enzyme encoded by the biosynthetic gene, which drives and catalyzes the biosynthetic reaction of secondary metabolites. The biosynthetic reaction refers to a series of biochemical reactions mediated by biosynthetic enzymes that occur in microbial cells, ultimately synthesizing secondary metabolites.
[0057] S2) Based on the pre-constructed biosynthetic gene cluster feature coding table, the protein family numbers in the biosynthetic feature word sequence are converted into numerical codes one by one, and the obtained numerical codes are combined in sequence to obtain the biosynthetic feature word coding sequence.
[0058] The construction of the biosynthetic gene cluster feature coding table includes the following steps:
[0059] S21) Download existing known biosynthetic gene cluster sequence files from the MIBiG (Minimum Information about a Biosynthetic Gene cluster) database;
[0060] S22) Extract the biosynthetic enzyme sequence data encoded by each biosynthetic gene cluster from the downloaded existing known biosynthetic gene cluster sequence file;
[0061] S23) Use HMMER (a commonly used sequence alignment tool based on the Hidden Markov Model (HMM) algorithm, which can be used to align different types of biological sequences, including protein sequences and nucleic acid sequences) with the Pfam protein family annotation database to annotate all biosynthetic enzyme sequences extracted in step S22).
[0062] S24) Remove duplicates from all protein family IDs obtained from the annotation in step S23) and arrange the protein family IDs in order. The table consisting of the protein family IDs and their corresponding sequence numbers is the biosynthetic gene cluster feature coding table.
[0063] The aforementioned numerical code refers to the sequence number of the protein family number in the biosynthetic gene cluster feature coding table;
[0064] S3) Use the one-hot encoding method to convert the biosynthetic feature word codes one by one into binary vectors, and combine the resulting vectors in order to obtain the biosynthetic feature word vector sequence;
[0065] S4) Input the biosynthetic feature word vector sequence into the pre-trained Transformer (a deep neural network architecture that uses a multi-head attention mechanism) deep neural network prediction model, and the model will predict and output the SMILES (Simplified molecular input line entry specification, a specification that explicitly describes the molecular structure using ASCII strings) word encoding sequence of the molecular structure of microbial secondary metabolites.
[0066] The training of the Transformer deep neural network prediction model includes the following steps:
[0067] S41) Randomly divide the dataset into a training dataset and a validation dataset in a ratio of 9:1;
[0068] S42) Input the training dataset and validation dataset into the Transformer deep neural network, with each batch of data consisting of 16 pairs; the Transformer deep neural network has: 1020 input sequence features, 35 output sequence features, 512 hidden layers, 8 multi-head attention layers, 6 encoder layers, and 6 decoder layers.
[0069] S43) Using the AdamW optimizer (learning rate set to 0.0001) and cross-entropy loss function, the Transformer deep neural network is trained and validated in batches. When the prediction performance of the neural network on the validation dataset does not improve for 10 consecutive batches, training is stopped, and the optimal neural network state parameters are saved as the Transformer deep neural network prediction model. The input data for training the Transformer deep neural network is the sequence of word vectors for biosynthetic features, and the output target data for training the Transformer deep neural network is the sequence of word vectors for SMILES, which represents the molecular structure of microbial secondary metabolites.
[0070] The acquisition of the SMILES word vector sequences of the molecular structures of microbial secondary metabolites includes the following steps:
[0071] A) Based on the pre-constructed SMILES coding table of microbial secondary metabolite molecular structures, convert the SMILES sequences in the dataset into SMILES word coding sequences.
[0072] B) Use the one-hot encoding method to convert the SMILES word codes one by one into binary vectors, and combine the resulting vectors in order to obtain the SMILES word vector sequence of the molecular structure of the microbial secondary metabolites.
[0073] S5) Based on the pre-constructed SMILES coding table of microbial secondary metabolite molecular structures, the SMILES word coding sequence predicted by the model is converted into a SMILES character sequence, thereby obtaining the predicted microbial secondary metabolite molecular structure.
[0074] The construction of the SMILES coding table for the molecular structures of microbial secondary metabolites includes the following steps:
[0075] a) Download data files from the MIBiG database, including biosynthetic gene cluster annotation files;
[0076] b) Extract SMILES sequence data of microbial secondary metabolite molecular structures from biosynthetic gene cluster annotation files;
[0077] c) The SMILES sequence of the extracted microbial secondary metabolite molecular structure is segmented by character, then all characters appearing in the SMILES sequence are extracted, and the obtained SMILES characters are deduplicated and arranged in order.
[0078] d) The table consisting of SMILES characters and their corresponding sequence numbers is the SMILES encoding table of the molecular structure of microbial secondary metabolites.
[0079] The SMILES word encoding mentioned above refers to the sequence number corresponding to each SMILES character in the SMILES encoding table.
[0080] One implementation scheme, the acquisition of the dataset in step S41), includes the following processing steps:
[0081] S411) Download data files from the MIBiG database, including biosynthetic gene cluster sequence files and biosynthetic gene cluster annotation files;
[0082] S412) Extract the biosynthetic enzyme sequence data encoded by each biosynthetic gene cluster from the biosynthetic gene cluster sequence file;
[0083] S413) SMILES sequence data of microbial secondary metabolite molecular structures were extracted from biosynthetic gene cluster annotation files;
[0084] S414) Based on the MIBiG database number, the biosynthetic enzyme sequence encoded by each biosynthetic gene cluster is paired with the SMILES sequence of the corresponding microbial secondary metabolite molecular structure. The combination of all data pairs is the dataset.
[0085] like Figure 1 As shown, the present invention provides a system for intelligently predicting the molecular structure of microbial secondary metabolites, comprising:
[0086] An annotation module is used to annotate the input biosynthetic gene cluster sequence as a biosynthetic feature word sequence;
[0087] An encoding module for converting biosynthetic feature word sequences into biosynthetic feature word encoded sequences;
[0088] An embedder for converting a sequence of biosynthetic feature word encodings into a sequence of biosynthetic feature word vectors;
[0089] A Transformer deep neural network prediction model was used to predict the SMILES word-encoded sequence of the output microbial secondary metabolite molecular structure based on the input biosynthetic feature word vector sequence; and
[0090] The decoding module is used to convert the SMILES word encoding sequence of the model prediction output into a SMILES character sequence to obtain the predicted molecular structure of microbial secondary metabolites.
[0091] The following section uses the prediction of the molecular structure of abscisic acid, a secondary metabolite of the microorganism Leptosphaeria maculans JN3, as an example to illustrate the implementation and realization process of the method described in this invention:
[0092] S1) Input the biosynthetic gene cluster data of abscisic acid, a secondary metabolite of the microorganism Leptosphaeria maculans JN3, into the annotation module. The annotation module uses the HMMER tool with the Pfam protein family annotation database to perform protein family annotation on the sequences of biosynthetic enzymes (cytochrome P450 enzyme 1, cytochrome P450 enzyme 2, beta-ketoacyl synthase, and cytochrome P450 enzyme 3) encoded by the input biosynthetic gene cluster (annotation threshold set to e-value < 0.01). Each protein family number obtained from the annotation is treated as a word, and all the annotated protein family numbers are combined sequentially to obtain the biosynthetic feature word sequence: ['PF00067.25', 'PF00067']. .25','PF00109.29','PF14765.9','PF08659.13','PF00698.24','PF02801.25','PF13602.9','PF00107.29','PF08242.15','PF16197. 8','PF13649.9','PF00106.28','PF13489.9','PF13847.9','PF08241.15','PF13561.9','PF00550.28','PF01209.21','PF00067.25'];
[0093] S2) Using the encoding module, the protein family numbers in the biosynthetic gene cluster feature encoding table are converted into numerical codes one by one. The numerical code is the sequence number corresponding to each protein family number in the biosynthetic gene cluster feature encoding table. The obtained numerical codes are combined sequentially to obtain the biosynthetic feature word encoding sequence: [2,11,11,19,888,605,118,345,821,17,575,905,828,16,793,848,574,814,95,185,11,3]. The first code "2" in this encoding sequence is the default starting code of the neural network model input sequence, and the last code "3" is the neural network... The default end encoding of the input sequence for the network model; additionally, it should be noted that the protein family numbers in the biosynthetic gene cluster feature encoding table are not duplicated because deduplication has been performed during construction; however, when annotating the biosynthetic enzyme sequence encoded by the input biosynthetic gene cluster during prediction, deduplication should not be performed, as the annotated biosynthetic feature word sequence may contain multiple identical protein family numbers. For example, in this embodiment, there are 3 protein family numbers 'PF00067.25'; and these identical protein family numbers may be grouped together or not. For example, in this embodiment, two of the protein family numbers 'PF00067.25' are grouped together, while the other is not grouped with the previous two.
[0094] S3) The embedder uses the one-hot encoding method to convert the above-obtained biosynthetic feature word codes into binary vectors one by one, and combines the resulting vectors in order to obtain the biosynthetic feature word vector sequence.
[0095] S4) Input the transformed biosynthetic feature word vector sequence into the Transformer deep neural network prediction model. The model can predict the output SMILES word encoding sequence of the molecular structure of microbial secondary metabolites: [2,24,13,11,24,12,11,24,12,24,11,24,22,24,24,11,24,12,22,24,24,11,29,12,22,29,12,11,29,12,24,11,22,29,12,24,11,24,12,24,13,3];
[0096] S5) The decoding module, based on the pre-constructed SMILES encoding table of microbial secondary metabolite molecular structures, can convert the SMILES word encoding sequence predicted by the model into a SMILES character sequence: C1(C)(C)C(C=CC(C)=CC(O)=O)(O)C(=O)C=C(C)C1, thereby obtaining the predicted molecular structure of the microbial secondary metabolite abscisic acid:
[0097] It has a completely consistent molecular structure with the real abscisic acid.
[0098] As can be seen from the above, this invention utilizes artificial intelligence technology and a large amount of existing known instance data in the MIBiG database. By pre-training the Transformer deep neural network prediction model and pre-constructing a biosynthetic gene cluster feature encoding table and a SMILES encoding table of microbial secondary metabolite molecular structures, it not only achieves intelligent prediction of the molecular structure of unknown microbial secondary metabolites based on the input biosynthetic gene cluster features, but also eliminates the need for rule base construction and manual data collection, and has a wide coverage. It overcomes the limitations of rule-based prediction tools and significantly improves prediction efficiency and accuracy. It can provide scientific support for synthetic biology experiments and natural drug design, and has important application value for accelerating and promoting the development of natural drugs.
[0099] It should be noted that the method described in this invention can be implemented by computer software programming and stored in a storage medium, and then executed by an electronic device integrating the storage medium.
[0100] The aforementioned program modules generally include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types.
[0101] In addition, the present invention also provides a storage medium storing one or more programs including execution instructions, which can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the above-described method for intelligently predicting the molecular structure of microbial secondary metabolites.
[0102] The present invention also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above for intelligently predicting the molecular structure of microbial secondary metabolites.
[0103] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the system for intelligently predicting the molecular structure of microbial secondary metabolites. Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the system for intelligently predicting the molecular structure of microbial secondary metabolites via a network. The network includes, but is not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0104] The electronic device can exist in various forms, including but not limited to:
[0105] (1) Mobile communication devices: These devices are characterized by having mobile communication functions and are primarily designed to provide voice and data communication. These terminals include: smartphones (such as iPhones), multimedia phones, feature phones, and low-end phones, etc.
[0106] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, have computing and processing functions, and generally also have mobile Internet access capabilities. These terminals include: PDA, MID and UMPC devices, such as iPad;
[0107] (3) Portable entertainment devices: These devices can display and play multimedia content. These devices include audio and video players (such as iPods), handheld game consoles, e-books, as well as smart toys and portable car navigation devices.
[0108] (4) Server: A device that provides computing services. The components of a server include a processor, hard disk, memory, system bus, etc. Servers are similar to general computer architectures, but because they need to provide highly reliable services, they have higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.
[0109] (5) Other electronic devices with data interaction functions.
[0110] This invention also provides a computer program product, comprising a computer program stored on a storage medium, the computer program including program instructions that, when executed by a computer, enable the computer to perform the method described above for intelligently predicting the molecular structure of microbial secondary metabolites. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or certain portions of the embodiments.
[0111] Finally, it should be pointed out that the above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. In the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other. Furthermore, for the foregoing method embodiments, for the sake of simplicity, they are all described as a combination of a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.
Claims
1. A method for intelligently predicting the molecular structure of microbial secondary metabolites, characterized in that, Includes the following steps: S1) Use the HMMER tool with the Pfam protein family annotation database to annotate the biosynthetic enzyme sequences encoded by the input biosynthetic gene clusters, and treat each protein family number obtained from the annotation as a word. Combine all the annotated protein family numbers in sequence to obtain a biosynthetic feature word sequence. S2) Based on the pre-constructed biosynthetic gene cluster feature coding table, convert the protein family numbers in the biosynthetic feature word sequence into numerical codes one by one, and combine the obtained numerical codes in order to obtain the biosynthetic feature word coding sequence. S3) Use one-hot encoding to convert the biosynthetic feature word codes one by one into binary vectors, and combine the resulting vectors in order to obtain the biosynthetic feature word vector sequence; S4) Input the biosynthetic feature word vector sequence into the pre-trained Transformer deep neural network prediction model, and the model predicts and outputs the SMILES word encoding sequence of the molecular structure of microbial secondary metabolites. S5) Based on the pre-constructed SMILES coding table of microbial secondary metabolite molecular structures, the SMILES word coding sequence predicted by the model is converted into a SMILES character sequence, thereby obtaining the predicted microbial secondary metabolite molecular structure.
2. The method according to claim 1, characterized in that, The construction of the biosynthetic gene cluster feature coding table described in step S2) includes the following steps: S21) Download existing known biosynthetic gene cluster sequence files from the MIBiG database; S22) Extract the biosynthetic enzyme sequence data encoded by each biosynthetic gene cluster from the downloaded existing known biosynthetic gene cluster sequence file; S23) Use the HMMER tool with the Pfam protein family annotation database to annotate all the biosynthetic enzyme sequences extracted in step S22). S24) Remove duplicates from all protein family IDs obtained from step S23) and arrange them in order. The table consisting of the protein family IDs and their corresponding sequence numbers is the biosynthetic gene cluster feature coding table.
3. The method according to claim 1, characterized in that, The training of the Transformer deep neural network prediction model described in step S4) includes the following steps: S41) Randomly divide the dataset into a training dataset and a validation dataset in a ratio of 9:1; S42) Input the training dataset and validation dataset into the Transformer deep neural network, with each batch containing 16 pairs of data. S43) Using the AdamW optimizer and cross-entropy loss function, train and validate the Transformer deep neural network in batches. When the prediction performance of the neural network on the validation dataset does not improve for 10 consecutive batches, stop training and use the saved optimal neural network state parameters as the Transformer deep neural network prediction model.
4. The method according to claim 3, characterized in that, The acquisition of the dataset mentioned in step S41) includes the following processing steps: S411) Download data files from the MIBiG database, including biosynthetic gene cluster sequence files and biosynthetic gene cluster annotation files; S412) Extract the biosynthetic enzyme sequence data encoded by each biosynthetic gene cluster from the biosynthetic gene cluster sequence file; S413) SMILES sequence data of microbial secondary metabolite molecular structures extracted from biosynthetic gene cluster annotation files; S414) Based on the MIBiG database number, the biosynthetic enzyme sequence encoded by each biosynthetic gene cluster is paired with the SMILES sequence of the corresponding microbial secondary metabolite molecular structure. The combination of all data pairs is the dataset.
5. The method according to claim 3, characterized in that: The input data for training the Transformer deep neural network is a sequence of word vectors representing biosynthetic features; the output target data for training the Transformer deep neural network is a sequence of word vectors representing the molecular structures of microbial secondary metabolites (SMILES).
6. The method according to claim 5, characterized in that, The acquisition of the SMILES word vector sequences of the molecular structures of microbial secondary metabolites includes the following steps: A) Based on the pre-constructed SMILES coding table of microbial secondary metabolite molecular structures, convert the SMILES sequences in the dataset into SMILES word coding sequences. B) Use the one-hot encoding method to convert the SMILES word codes one by one into binary vectors, and combine the resulting vectors in order to obtain the SMILES word vector sequence of the molecular structure of microbial secondary metabolites.
7. The method according to claim 1, characterized in that, The construction of the SMILES coding table for the molecular structures of microbial secondary metabolites includes the following steps: a) Download data files from the MIBiG database, including biosynthetic gene cluster annotation files; b) Extracting SMILES sequence data of microbial secondary metabolite molecular structures from biosynthetic gene cluster annotation files; c) The SMILES sequence of the extracted microbial secondary metabolite molecular structure is segmented by character, then all characters appearing in the SMILES sequence are extracted, and the obtained SMILES characters are deduplicated and arranged in order. d) The table consisting of SMILES characters and their corresponding sequence numbers is the SMILES encoding table of the molecular structure of microbial secondary metabolites.
8. A system for intelligently predicting the molecular structure of microbial secondary metabolites using the method described in any one of claims 1 to 7, characterized in that, The system includes: An annotation module is used to annotate the input biosynthetic gene cluster sequence as a biosynthetic feature word sequence; An encoding module for converting biosynthetic feature word sequences into biosynthetic feature word encoded sequences; An embedder for converting a sequence of biosynthetic feature word encodings into a sequence of biosynthetic feature word vectors; A Transformer deep neural network prediction model was used to predict the SMILES word-encoded sequence of the output microbial secondary metabolite molecular structure based on the input biosynthetic feature word vector sequence; and The decoding module is used to convert the SMILES word encoding sequence of the model prediction output into a SMILES character sequence to obtain the predicted molecular structure of microbial secondary metabolites.
9. A storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the program can implement the method described in any one of claims 1 to 7.
10. An electronic device comprising: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor; characterized in that: the at least one processor is capable of performing the method according to any one of claims 1 to 7.