Gene cluster identification method and device of bacterial genome, equipment and storage medium

Abstract

The invention discloses a gene cluster identification method and device of a bacterial genome, equipment and a storage medium. The method comprises the following steps: constructing a gene cluster recognition model by using a preprocessing unit, an ESM protein language unit, a cosine approximation calculation unit, a normalized pointwise mutual information unit, a term frequency-inverse document frequency unit, a graph convolution unit, a BERT unit and a softmax function unit; inputting a BGC sequence training sample of the bacterial genome into the gene cluster recognition model, training the gene cluster recognition model by using a cross entropy loss function, and performing iterative optimization on the training process by using an Adam optimizer; inputting the BGC sequence data set of the bacterial genome to be recognized into the target gene cluster recognition model for recognition; segments in the preliminary recognition result are corrected, and a final recognition result is obtained; by constructing an adjacent matrix and inputting the adjacent matrix into a graph convolution unit, global features contained in Pfam structural domain nodes in a BGC sequence are extracted, and the recognition and prediction performance of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bacterial genome gene cluster identification, and in particular to a bacterial genome gene cluster identification method, device, equipment and storage medium. Background Art

[0002] Antibiotics are an important drug in modern medicine. However, due to the overuse of antibiotics, microorganisms have developed resistance to antibiotics, which reduces or even inhibits the antibacterial effect of antibiotics and makes them lose their therapeutic effects. In order to cope with the increasing impact of antibiotic resistance, biologists and medical scientists around the world are actively developing new antibiotics. Among them, natural compounds with rich chemical structures and biological activities are a valuable resource. Some of them have antibacterial, anticancer or other immunomodulatory effects, and therefore are valued by drug developers. These natural compounds are encoded by biosynthesis gene clusters (BGCs), which are usually composed of a series of interrelated genes that encode a series of different enzymes for the co-synthesis of biologically active molecules. The BGCs of many natural microorganisms are the source of many drugs (mainly antibiotics), which are discovered by discovering their corresponding BGCs. Accurately locating BGCs in the genome and predicting their products and activities can be used for the discovery of new drugs. Current BGC prediction tools can generally be divided into two main methods: methods based on sequence similarity and machine learning methods that can learn more complex patterns.

[0003] Sequence similarity-based methods, such as antiSMASH and NP.searcher, only use existing databases and tools for comparison, which makes it difficult to capture the structural features in BGCs and have weak recognition capabilities for novel or dissimilar BGCs.

[0004] Existing machine learning-based methods, such as DeepBGC and BiGCARP, use deep learning models to learn the structural features contained in BGC and improve the ability to identify BGC. However, due to the small number of BGCs in the database, it is not possible to learn the features of BGC well, and existing machine learning methods only use local features in BGC sequences, but do not incorporate global features between BGC sequences. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a method, device, equipment and storage medium for identifying gene clusters in bacterial genomes to solve the problem that existing methods have weak ability to identify gene clusters in microorganisms.

[0006] In order to solve the above technical problems, the present application embodiment provides a method for identifying gene clusters in bacterial genomes, which adopts the following technical solutions, including:

[0007] Step 100, obtaining a preprocessing unit, an ESM protein language unit, a cosine approximation calculation unit, a normalized point-by-point mutual information unit, a term frequency-inverse document frequency unit, a graph convolution unit, and a BERT unit, and using the preprocessing unit, the ESM protein language unit, the cosine approximation calculation unit, the normalized point-by-point mutual information unit, the term frequency-inverse document frequency unit, the graph convolution unit, the BERT unit, and the softmax function unit to construct a gene cluster recognition model;

[0008] Step 200: input the BGC sequence training sample of the bacterial genome into the gene cluster recognition model, train the gene cluster recognition model using the cross entropy loss function, and iteratively optimize the training process using the Adam optimizer to obtain a target gene cluster recognition model;

[0009] Step 300: input the BGC sequence data set of the bacterial genome to be identified into the target gene cluster recognition model for identification to obtain a preliminary identification result;

[0010] Step 400: correct the fragments in the preliminary recognition result to obtain a final recognition result;

[0011] Among them, the cosine approximation calculation unit is used to calculate the similarity of the embedded vector of the Pfam domain node, the normalized point-by-point mutual information unit and the term frequency-inverse document frequency unit are used to construct the Pfam domain node adjacency matrix, and the graph convolution unit is used to extract the global features of the Pfam domain node.

[0012] Furthermore, the step 200 includes:

[0013] Step 210: using the preprocessing unit to convert the BGC sequence into a Pfam sequence;

[0014] Step 220, extracting the embedding vector of the Pfam sequence using the ESM protein language unit;

[0015] Step 230: Calculate the similarity of the embedding vectors according to the Pfam database and using the cosine approximation calculation unit.

[0016] Furthermore, the step 200 further includes:

[0017] Step 240: perform data enhancement on the BGC sequence according to the similarity to obtain an enhanced BGC sequence;

[0018] Step 250: construct a Pfam adjacency matrix according to the input BGC sequence and using the normalized point-by-point mutual information unit and the term frequency-inverse document frequency unit;

[0019] Step 260: Input the Pfam adjacency matrix into the graph convolution unit to extract Pfam global features.

[0020] Furthermore, the step 240 includes:

[0021] Step 241, using the Pfam database, obtaining all Pfam domains in the BGC sequence whose similarity is greater than a first specified threshold;

[0022] Step 242, obtaining the number of Pfam domains in the BGC, and screening out target BGC sequences greater than a second specified threshold;

[0023] Step 243, randomly replacing the Pfam domains in a specified number of BGCs in the target BGC sequence whose similarity is greater than the first specified threshold, to obtain the enhanced BGC sequence;

[0024] The enhanced BGC sequence includes a BGC sequence that has not replaced the Pfam domain and a BGC sequence that has replaced the Pfam domain.

[0025] Furthermore, the step 200 further includes:

[0026] Step 270: input the Pfam global feature into the BERT unit, and use the BERT unit to fuse the Pfam global feature with the local feature in the enhanced BGC sequence to obtain a first fused feature;

[0027] Step 280: Input the first fusion feature into the softmax function unit to obtain a preliminary recognition result.

[0028] Furthermore, the step 200 further includes:

[0029] Step 290: Calculate the function value of the cross entropy loss function using the preliminary recognition result and the true label;

[0030] Step 291: Update the gene cluster identification model parameters using the function value of the cross entropy loss function and the back propagation algorithm.

[0031] Furthermore, the step 400 includes:

[0032] Step 410, obtaining the predicted interval distance between any two consecutive segments constituting the Pfam domain in the BGC sequence;

[0033] Step 420: If the interval distance is less than a third specified threshold, merge the continuous segments into one segment;

[0034] Step 430: If the length of the continuous segment is less than a fourth specified threshold and there is no adjacent segment, the continuous segment is discarded.

[0035] In order to solve the above problems, a gene cluster identification device for a bacterial genome is also provided, which adopts the gene cluster identification method for a bacterial genome, comprising:

[0036] A construction module is used to obtain a preprocessing unit, an ESM protein language unit, a cosine approximation calculation unit, a normalized point-by-point mutual information unit, a term frequency-inverse document frequency unit, a graph convolution unit, and a BERT unit, and to construct a gene cluster recognition model using the preprocessing unit, the ESM protein language unit, the cosine approximation calculation unit, the normalized point-by-point mutual information unit, the term frequency-inverse document frequency unit, the graph convolution unit, the BERT unit, and the softmax function unit;

[0037] A training module, used to input the BGC sequence training sample of the bacterial genome into the gene cluster recognition model, train the gene cluster recognition model using a cross entropy loss function, and iteratively optimize the training process using an Adam optimizer to obtain a target gene cluster recognition model;

[0038] An identification module is used to input the BGC sequence data set of the bacterial genome to be identified into the target gene cluster identification model for identification to obtain a preliminary identification result;

[0039] The correction module is used to correct the fragments in the preliminary recognition result to obtain the final recognition result.

[0040] In order to solve the above problems, an embodiment of the present application further proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, it implements the steps of a method for identifying gene clusters of a bacterial genome.

[0041] In order to solve the above problems, an embodiment of the present application further proposes a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the steps of a method for identifying gene clusters of a bacterial genome are implemented.

[0042] Compared with the existing technology, the BGC sequence data is enhanced by using ESM protein language units and cosine similarity calculation units to calculate the Pfam domain proximity in BGC sequences, and randomly replacing BGC sequences with Pfam domain proximity greater than a specified threshold, so that the model can better learn the high-dimensional features contained in different types of BGCs. The adjacency matrix is ​​constructed by using normalized point-by-point mutual information units and term frequency-inverse document frequency units, and is input into the graph convolution unit to extract the global features contained in the Pfam domain nodes in the BGC sequence, thereby improving the recognition and prediction performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are 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.

[0044] Figure 1 It is a flowchart of an embodiment of a method for identifying gene clusters in a bacterial genome of the present application;

[0045] Figure 2 yes Figure 1 A flowchart of a specific implementation of S200;

[0046] Figure 3 yes Figure 2 A flowchart of a specific implementation method after S230;

[0047] Figure 4 yes Figure 3 A flowchart of a specific implementation of S240;

[0048] Figure 5 yes Figure 3 A flowchart of a specific implementation method after S260;

[0049] Figure 6 yes Figure 5 A flowchart of a specific implementation method after S280;

[0050] Figure 7 yes Figure 1 A flowchart of a specific implementation of S400;

[0051] Figure 8 It is a schematic diagram of a process of a method for identifying gene clusters in a bacterial genome of the present application;

[0052] Fig. 9This is a schematic diagram of the module structure of an embodiment of a bacterial genome gene cluster identification device of the present application;

[0053] Fig.10 It is a schematic diagram of the module structure of a computer device of the present application. DETAILED DESCRIPTION

[0054] The following will be combined with the accompanying drawings in the invention to clearly and completely describe the technical solutions in the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0055] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.

[0057] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0058] The purpose of the embodiments of the present application is to provide a method, device, equipment and storage medium for identifying gene clusters in bacterial genomes to solve the problem that existing methods have weak ability to identify gene clusters in microorganisms.

[0059] In order to solve the above technical problems, the present application embodiment provides a method for identifying gene clusters in bacterial genomes, which adopts the following technical solutions: Figure 1 , Figure 1 This is a flowchart of an embodiment of a method for identifying gene clusters in a bacterial genome of the present application; it includes:

[0060] S100, obtaining a preprocessing unit, an ESM protein language unit, a cosine approximation calculation unit, a normalized point-by-point mutual information unit, a term frequency-inverse document frequency unit, a graph convolution unit, and a BERT unit, and using the preprocessing unit, the ESM protein language unit, the cosine approximation calculation unit, the normalized point-by-point mutual information unit, the term frequency-inverse document frequency unit, the graph convolution unit, the BERT unit, and the softmax function unit to build a gene cluster recognition model.

[0061] In this embodiment, the cosine approximation calculation unit is used to calculate the similarity of the embedded vectors of the Pfam domain nodes, the normalized point-by-point mutual information unit and the term frequency-inverse document frequency unit are used to construct the Pfam domain node adjacency matrix, and the graph convolution unit is used to extract the global features of the Pfam domain nodes.

[0062] In this embodiment, by performing similarity calculation, performing data enhancement on the BGC sequence, constructing an adjacency matrix and using a graph convolution unit to extract global features, the recognition performance of the gene cluster recognition model is improved.

[0063] In this embodiment, the preprocessing unit is used to convert the original BGC sequence into a Pfam sequence, and the ESM protein language unit is used for the embedding vector of each Pfam domain node in the BGC sequence.

[0064] In this embodiment, the ESM protein language unit model is a deep learning model for processing protein tasks. It is based on the Transformer architecture, uses the Encoder Layer of stacked Transformers, and learns the dependencies between residues by randomly masking residues to predict the types of masked residues.

[0065] In this embodiment, the cosine similarity calculation unit is used to calculate the similarity between each Pfam domain node and the domain in the Pfam database, in preparation for enhancing the BGC data.

[0066] In this embodiment, the Normalized Pointwise Mutual Information (NPMI) unit model is used to determine the strength of association between Pfam domain nodes, and the term frequency-inverse document frequency unit model is used to calculate the frequency of occurrence of a certain Pfam domain node. In this embodiment, the adjacency matrix is ​​constructed using the Normalized Pointwise Mutual Information unit and the term frequency-inverse document frequency unit.

[0067] In this embodiment, the BERT unit is used to fuse the local features and global features of the Pfam domain, and the softmax function unit is used for classification and recognition.

[0068] S200, inputting the BGC sequence training sample of the bacterial genome into the gene cluster recognition model, training the gene cluster recognition model using the cross entropy loss function, and iteratively optimizing the training process using the Adam optimizer to obtain the target gene cluster recognition model.

[0069] In a preferred embodiment, Figure 2 , Figure 2 yes Figure 1 A flowchart of a specific implementation of S200; S200 includes: S210, using a preprocessing unit to convert a BGC sequence into a Pfam sequence; S220, using an ESM protein language unit to extract an embedding vector of the Pfam sequence; S230, calculating the similarity of the embedding vector based on the Pfam database and using a cosine approximation calculation unit.

[0070] In this embodiment, if Figure 8 , Figure 8 It is a flow chart of a method for identifying gene clusters in a bacterial genome of the present application; there are multiple BGCs in a BGC sequence, and each BGC is represented by multiple Pfam domain nodes, that is, a Pfam sequence.

[0071] In this embodiment, when the ESM protein language unit is used to extract the embedding vector of the Pfam domain node, the ESM protein language unit model first obtains the Pfam sequence, generates a corresponding embedding representation for each amino acid in the Pfam sequence, and then extracts the Pfam-related feature vector from the above embedding representation based on the definition of the Pfam family and the correspondence between the known protein sequence and the Pfam family.

[0072] In this embodiment, the cosine value of the Pfam node embedding vector in the BGC sequence and the related vector in the database is used to determine the similarity. A cosine value of 1 indicates that the two are exactly the same, and a cosine value of 0 indicates that the two vectors are orthogonal and have no similarity.

[0073] In a preferred embodiment, Figure 3 , Figure 3 yes Figure 2 A flowchart of a specific implementation method after S230; S200 also includes: S240, data enhancement of the BGC sequence according to the similarity to obtain an enhanced BGC sequence; S250, based on the input BGC sequence, and using the normalized point-by-point mutual information unit and the term frequency-inverse document frequency unit to construct a Pfam adjacency matrix; S260, inputting the Pfam adjacency matrix into the graph convolution unit to extract the Pfam global features.

[0074] In this embodiment, by performing similarity calculation, Pfam domains with high similarity in the BGC sequence are obtained, and then data enhancement is performed on the BGC sequence by replacement to alleviate the problem of insufficient sample quantity in the current BGC database.

[0075] In a preferred embodiment, Figure 4 , Figure 4 yes Figure 3 A flowchart of a specific implementation method of S240; S240 includes: S241, using the Pfam database to obtain all Pfam domains in the BGC sequence whose similarity is greater than a first specified threshold; S242, obtaining the number of Pfam domains in the BGC, and screening out the target BGC sequence whose similarity is greater than a second specified threshold; S243, randomly replacing the Pfam domains in a specified number of BGCs in the target BGC sequence whose similarity is greater than the first specified threshold, to obtain an enhanced BGC sequence; wherein the enhanced BGC sequence includes a BGC sequence in which the Pfam domain has not been replaced and a BGC sequence in which the Pfam domain has been replaced.

[0076] In this embodiment, the first specified threshold is preferably 95%, which is intended to screen out Pfam domains with a similarity greater than 95%. In this embodiment, the second specified threshold is preferably 10, which is intended to exclude BGCs with less than 10 Pfam domains. In BGC sequences with more than 10 Pfam domains, 5-10 BGCs are randomly selected for Pfam domain replacement to enhance the functional characteristics of BGC sequence data.

[0077] In this embodiment, when constructing the Pfam adjacency matrix, the TF-IDF value is first calculated, the term frequency (TF), for each Pfam domain in each protein sequence, the number of times the Pfam domain appears in the sequence is counted. For example, if Pfam domain A appears 3 times in sequence 1, and the total number of Pfam domain annotations in sequence 1 is 10, then the TF value of Pfam domain A in protein sequence 1 is 3 / 10=0.3. Inverse document frequency (IDF) value, in this embodiment, each Pfam sequence is regarded as a document. Calculate the IDF value of each Pfam domain. For example, there are a total of 100 Pfam sequences, and Pfam domain A appears in 20 of them, then IDF(A)=log(100 / 20)=log(5). The TF-IDF value is the TF value of the family in the sequence multiplied by its IDF value, and the normalized pointwise mutual information (NPMI) is calculated.

[0078] For any two Pfam domains A and B, calculate the NPMI value between them. Count the number of Pfam sequences that appear in both A and B (denoted as N(A,B)), the number of Pfam sequences that appear in A (denoted as N(A)), the number of Pfam sequences that appear in B (denoted as N(B)), and the total number of Pfam sequences (denoted as N). The formula for calculating the point-wise mutual information (PMI) is: PMI(A,B)=log(N(A,B)*N / (N(A)*N(B))). Then normalize the PMI to get the NPMI. The commonly used normalization formula is: NPMI(A,B)=PMI(A,B) / -log(N(A,B) / N).

[0079] Calculate the TF-IDF(A) value, TF-IDF(B) value, and NPMI(A,B) value, and multiply them together to obtain the elements of the adjacency matrix. Based on this principle, calculate the values ​​of all elements to obtain the adjacency matrix.

[0080] In a preferred embodiment, the graph convolution unit model adopts a three-layer graph convolutional neural network (GCN), which is a deep learning architecture for processing graph structured data. The core of the graph convolutional network is to update the feature representation of the node by aggregating the information of neighboring nodes. In each layer of graph convolution, the node calculates its own new features based on the features of its neighboring nodes and the weights of the edges connected to the neighboring nodes. The first layer: The input is the original node feature matrix and the Pfam node adjacency matrix of the graph. The second layer: The Pfam node features output by the first layer are used as input to further aggregate the information of more distant neighbors. The third layer: Based on the first two layers, the final information fusion and feature extraction are performed. The Pfam node features output by this layer can contain higher-level structural and semantic information of the graph, which is used for downstream Pfam node classification.

[0081] In a preferred embodiment, Figure 5 , Figure 5 yes Figure 3 A flowchart of a specific implementation method after S260 in the figure; S200 also includes: S270, inputting the Pfam global feature into the BERT unit, using the BERT unit to fuse the Pfam global feature with the local feature in the enhanced BGC sequence to obtain a first fused feature; S280, inputting the first fused feature into the softmax function unit to obtain a preliminary recognition result.

[0082] In this embodiment, if Figure 8, the BERT unit model obtains the global feature (Global Feature) from the graph convolution unit model (GCN), and fuses it with the local features of the BGC enhanced sequence to obtain the fused feature, which is output to the softmax function unit for classification and recognition.

[0083] In a preferred embodiment, Figure 6 , Figure 6 yes Figure 5 A flowchart of a specific implementation method after S280; S200 also includes: S290, using the preliminary recognition result and the true label to calculate the function value of the cross entropy loss function; S291, using the function value of the cross entropy loss function and the back propagation algorithm to update the gene cluster recognition model parameters.

[0084] S300, inputting the BGC sequence data set of the bacterial genome to be identified into the target gene cluster recognition model for identification, and obtaining a preliminary identification result.

[0085] In this embodiment, if Figure 8 , the prediction results are as follows:

[0086]

[0087] The product with the highest predicted probability is taken as the identified product, and then, on the next S axis, the fragment of the product is corrected to avoid missed reports and false positives.

[0088] In this embodiment, the above table is only an output example of the classification model and does not represent the actual prediction result.

[0089] S400: Correct the fragments in the preliminary recognition result to obtain a final recognition result.

[0090] In a preferred embodiment, if Figure 7 , Figure 7 yes Figure 1 A flowchart of a specific implementation method of S400; S400 includes: S410, obtaining the predicted interval distance between any two consecutive fragments constituting the Pfam domain in the BGC sequence; S420, if the interval distance is less than a third specified threshold, merging the consecutive fragments into one segment; S430, if the length of the consecutive fragment is less than a fourth specified threshold and there is no adjacent fragment, discarding the consecutive fragment.

[0091] In this embodiment, if Figure 8 The third threshold is preferably 10 units. When the distance between any two consecutive segments is less than 10 units, they are merged into one segment. The fourth threshold is preferably 5 units. When the length of a segment is less than 5 units, it is discarded.

[0092] By correcting the predicted Pfam domain fragments and filtering out missed and false positives, the recognition accuracy of the target gene cluster recognition model is further improved, and more accurate and reliable prediction results are obtained.

[0093] In one embodiment, if Figure 8 After obtaining the recognition result (determined to be a BGC sequence), the BGC sequence is then classified according to the category.

[0094] By using the ESM protein language unit and the cosine similarity calculation unit to calculate the Pfam domain proximity in the BGC sequence, and randomly replacing the BGC sequences with Pfam domain proximity greater than the specified threshold, the BGC sequence data is enhanced, so that the model can better learn the high-dimensional features contained in different types of BGCs. By using the normalized point-by-point mutual information unit and the term frequency-inverse document frequency unit to construct the adjacency matrix and input it into the graph convolution unit, the global features contained in the Pfam domain nodes in the BGC sequence are extracted, thereby improving the recognition and prediction performance of the model.

[0095] In order to solve the above problems, a bacterial genome gene cluster identification device is also provided, which adopts a bacterial genome gene cluster identification method, such as Fig. 9 , Fig. 9 This is a schematic diagram of the module structure of an embodiment of a bacterial genome gene cluster identification device 500 of the present application; it includes:

[0096] The construction module 501 is used to obtain a preprocessing unit, an ESM protein language unit, a cosine approximation calculation unit, a normalized point-by-point mutual information unit, a term frequency-inverse document frequency unit, a graph convolution unit and a BERT unit, and use the preprocessing unit, the ESM protein language unit, the cosine approximation calculation unit, the normalized point-by-point mutual information unit, the term frequency-inverse document frequency unit, the graph convolution unit, the BERT unit and the softmax function unit to construct a gene cluster recognition model; the training module 502 is used to input the BGC sequence training sample of the bacterial genome into the gene cluster recognition model, train the gene cluster recognition model using the cross entropy loss function, and iteratively optimize the training process using the Adam optimizer to obtain a target gene cluster recognition model; the recognition module 503 is used to input the BGC sequence data set of the bacterial genome to be recognized into the target gene cluster recognition model for recognition, and obtain a preliminary recognition result; the correction module 504 is used to correct the fragments in the preliminary recognition result to obtain a final recognition result.

[0097] In order to solve the above problems, an embodiment of the present application also proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, it implements the steps of a method for identifying gene clusters of bacterial genomes.

[0098] In order to solve the above problems, an embodiment of the present application further proposes a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the steps of a method for identifying gene clusters of a bacterial genome are implemented.

[0099] In order to solve the above problems, Fig.10 , Fig.10 A schematic diagram of a module structure of a computer device according to the present application. The present application also proposes a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of a method for identifying gene clusters of a bacterial genome are implemented.

[0100] The computer device may be a computer, server, workstation or other device, or a mobile phone, tablet, vehicle-mounted mobile terminal or other device with program execution capability. The internal structure diagram of the computer device may be as follows: Fig.10 As shown, Fig.10 It is a structural diagram of an embodiment of a computer device according to the present application. The computer device includes a processor, a memory and a communication module. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage medium stores an operating system, instructions or codes. The internal memory provides an environment for the operation of the operating system and instructions or codes in the non-volatile storage medium. When the instructions or codes are executed by the processor, the functions or steps of a method for fine-tuning the above-mentioned pre-trained model are implemented. The communication module of the computer device may include a network interface and / or a wireless communication module, and the computer device may communicate with other devices or service platforms through the communication module. In addition, the computer device may also include a display screen and an input device, etc.

[0101] The memory is used to store a computer program, the computer program includes program instructions, the processor is configured to call the program instructions, and the processor implements the following steps when executing instructions or codes: S100, obtaining a preprocessing unit, an ESM protein language unit, a cosine approximation calculation unit, a normalized point-by-point mutual information unit, a term frequency-inverse document frequency unit, a graph convolution unit and a BERT unit, and constructing a gene cluster recognition model using the preprocessing unit, the ESM protein language unit, the cosine approximation calculation unit, the normalized point-by-point mutual information unit, the term frequency-inverse document frequency unit, the graph convolution unit, the BERT unit and the softmax function unit; S200, inputting the BGC sequence training sample of the bacterial genome into the gene cluster recognition model, training the gene cluster recognition model using the cross entropy loss function, and iteratively optimizing the training process using the Adam optimizer to obtain a target gene cluster recognition model; S300, inputting the BGC sequence data set of the bacterial genome to be identified into the target gene cluster recognition model for identification, and obtaining a preliminary identification result; S400, correcting the fragments in the preliminary identification result to obtain a final identification result.

[0102] In order to solve the above problems, the embodiment of the present application further proposes a computer-readable storage medium, which adopts the following technical solution:

[0103] The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions. When the program instructions are executed by a processor, the processor executes a method for identifying gene clusters of a bacterial genome.

[0104] The present application also provides a computer-readable storage medium storing a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, Figures 1 to 7 The gene cluster identification method of a bacterial genome provided in each step can be specifically referred to the implementation methods provided in the above steps, which will not be repeated here.

[0105] The computer-readable storage medium may be a device for generating a video of a speaking person with a standardized lip shape provided by any of the aforementioned embodiments, or an internal storage unit of the terminal device, such as a hard disk or memory of a computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the computer device.

[0106] Furthermore, the computer-readable storage medium may also include both an internal storage unit of the computer device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0107] However, it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), digital signal processors (DSP), embedded devices, etc.

[0108] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0109] Compared with the existing technology, the BGC sequence data is enhanced by using ESM protein language units and cosine similarity calculation units to calculate the Pfam domain proximity in BGC sequences, and randomly replacing BGC sequences with Pfam domain proximity greater than a specified threshold, so that the model can better learn the high-dimensional features contained in different types of BGCs. The adjacency matrix is ​​constructed by using normalized point-by-point mutual information units and term frequency-inverse document frequency units, and is input into the graph convolution unit to extract the global features contained in the Pfam domain nodes in the BGC sequence, thereby improving the recognition and prediction performance of the model.

[0110] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.

[0111] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying gene clusters in a bacterial genome, comprising: Step 100, obtaining a preprocessing unit, an ESM protein language unit, a cosine approximation calculation unit, a normalized point-by-point mutual information unit, a term frequency-inverse document frequency unit, a graph convolution unit, and a BERT unit, and using the preprocessing unit, the ESM protein language unit, the cosine approximation calculation unit, the normalized point-by-point mutual information unit, the term frequency-inverse document frequency unit, the graph convolution unit, the BERT unit, and the softmax function unit to construct a gene cluster recognition model; Step 200: input the BGC sequence training sample of the bacterial genome into the gene cluster recognition model, train the gene cluster recognition model using the cross entropy loss function, and iteratively optimize the training process using the Adam optimizer to obtain a target gene cluster recognition model; Step 300: input the BGC sequence data set of the bacterial genome to be identified into the target gene cluster recognition model for identification to obtain a preliminary identification result; Step 400: correct the fragments in the preliminary recognition result to obtain a final recognition result; Among them, the cosine approximation calculation unit is used to calculate the similarity of the embedded vector of the Pfam domain node, the normalized point-by-point mutual information unit and the term frequency-inverse document frequency unit are used to construct the Pfam domain node adjacency matrix, and the graph convolution unit is used to extract the global features of the Pfam domain node.

2. The method for identifying gene clusters in bacterial genomes according to claim 1, characterized in that: The step 200 comprises: Step 210: using the preprocessing unit to convert the BGC sequence into a Pfam sequence; Step 220, extracting the embedding vector of the Pfam sequence using the ESM protein language unit; Step 230: Calculate the similarity of the embedding vectors according to the Pfam database and using the cosine approximation calculation unit.

3. The method for identifying gene clusters in bacterial genomes according to claim 2, characterized in that: The step 200 further includes: Step 240: perform data enhancement on the BGC sequence according to the similarity to obtain an enhanced BGC sequence; Step 250: construct a Pfam adjacency matrix according to the input BGC sequence and using the normalized point-by-point mutual information unit and the term frequency-inverse document frequency unit; Step 260: Input the Pfam adjacency matrix into the graph convolution unit to extract Pfam global features.

4. The method for identifying gene clusters in bacterial genomes according to claim 3, characterized in that: The step 240 includes: Step 241, using the Pfam database, obtaining all Pfam domains in the BGC sequence whose similarity is greater than a first specified threshold; Step 242, obtaining the number of Pfam domains in the BGC, and screening out target BGC sequences greater than a second specified threshold; Step 243, randomly replacing the Pfam domains in a specified number of BGCs in the target BGC sequence whose similarity is greater than the first specified threshold, to obtain the enhanced BGC sequence; The enhanced BGC sequence includes a BGC sequence that has not replaced the Pfam domain and a BGC sequence that has replaced the Pfam domain.

5. The method for identifying gene clusters in bacterial genomes according to claim 3, characterized in that: The step 200 further includes: Step 270: input the Pfam global feature into the BERT unit, and use the BERT unit to fuse the Pfam global feature with the local feature in the enhanced BGC sequence to obtain a first fused feature; Step 280: Input the first fusion feature into the softmax function unit to obtain a preliminary recognition result.

6. The method for identifying gene clusters in bacterial genomes according to claim 5, characterized in that: The step 200 further includes: Step 290: Calculate the function value of the cross entropy loss function using the preliminary recognition result and the true label; Step 291: Update the gene cluster identification model parameters using the function value of the cross entropy loss function and the back propagation algorithm.

7. The method for identifying gene clusters in bacterial genomes according to claim 1, characterized in that: The step 400 includes: Step 410, obtaining the predicted interval distance between any two consecutive segments constituting the Pfam domain in the BGC sequence; Step 420: If the interval distance is less than a third specified threshold, merge the continuous segments into one segment; Step 430: If the length of the continuous segment is less than a fourth specified threshold and there is no adjacent segment, the continuous segment is discarded.

8. A bacterial genome gene cluster identification device, using the bacterial genome gene cluster identification method according to any one of claims 1 to 7, characterized in that: include: A construction module is used to obtain a preprocessing unit, an ESM protein language unit, a cosine approximation calculation unit, a normalized point-by-point mutual information unit, a term frequency-inverse document frequency unit, a graph convolution unit, and a BERT unit, and to construct a gene cluster recognition model using the preprocessing unit, the ESM protein language unit, the cosine approximation calculation unit, the normalized point-by-point mutual information unit, the term frequency-inverse document frequency unit, the graph convolution unit, the BERT unit, and the softmax function unit; A training module, used to input the BGC sequence training sample of the bacterial genome into the gene cluster recognition model, train the gene cluster recognition model using a cross entropy loss function, and iteratively optimize the training process using an Adam optimizer to obtain a target gene cluster recognition model; An identification module is used to input the BGC sequence data set of the bacterial genome to be identified into the target gene cluster identification model for identification to obtain a preliminary identification result; The correction module is used to correct the fragments in the preliminary recognition result to obtain the final recognition result.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the steps of the method for identifying gene clusters of a bacterial genome as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the method for identifying gene clusters of a bacterial genome according to any one of claims 1 to 7 are implemented.

Images