Probiotic identification method and device, electronic equipment and storage medium

Abstract

The present disclosure provides a probiotic bacteria identification method and device, an electronic device and a storage medium. The probiotic bacteria identification method comprises: obtaining a to-be-tested sample extracted from a plant soil environment, and extracting a plurality of nucleotide sequences from the to-be-tested sample; calling a preset first neural network model to extract a representation feature of each nucleotide sequence; extracting a sequence feature corresponding to each nucleotide sequence, fusing the sequence feature and the corresponding representation feature of each nucleotide sequence to obtain a fusion feature of each nucleotide sequence; calling a pre-trained second neural network model to identify the fusion features corresponding to the plurality of nucleotide sequences to obtain a probiotic bacteria identification result of the to-be-tested sample. The embodiments of the present application can improve the accuracy of probiotic bacteria identification in a plant soil environment.

Description

Technical Field

[0001] This disclosure relates to the field of bioinformatics, specifically to a probiotic identification method and apparatus, electronic device and storage medium. Background Technology

[0002] Probiotics are a class of live microorganisms that colonize the host's internal environment and interact with the original microbial community. Probiotics can regulate the balance of the microecology and produce health benefits. For example, probiotics in the soil can promote plant growth, maintain plant health, and enhance plant resistance to diseases through various mechanisms. Therefore, probiotic identification is of great significance.

[0003] However, the methods for identifying probiotics in related technologies suffer from low accuracy. Summary of the Invention

[0004] The main objective of this application is to provide a probiotic identification method, device, electronic device, and storage medium, which aims to improve the accuracy of probiotic identification in plant soil environments.

[0005] To achieve the above objectives, a first aspect of this application proposes a probiotic identification method, the method comprising:

[0006] A test sample was obtained from the plant soil environment, and multiple nucleotide sequences were extracted from the test sample;

[0007] The characterization features of each nucleotide sequence are extracted by calling a preset first neural network model;

[0008] Extract the sequence features corresponding to each nucleotide sequence, and fuse the sequence features and corresponding characterization features of each nucleotide sequence to obtain the fusion feature of each nucleotide sequence;

[0009] The pre-trained second neural network model is invoked to identify the fusion features corresponding to multiple nucleotide sequences, thereby obtaining the probiotic identification result of the sample to be tested.

[0010] In some embodiments, the sequence features include continuous sequence features, and the extraction of sequence features corresponding to each nucleotide sequence includes:

[0011] For each nucleotide sequence, the nucleotide sequence is divided according to a preset continuous length to obtain multiple first sub-sequences corresponding to the nucleotide sequence;

[0012] Calculate the first repetition frequency of each of the first subsequences based on the number of repetitions of each of the first subsequences;

[0013] The continuous sequence features are constructed based on the first repetition frequency of multiple first subsequences.

[0014] In some embodiments, the sequence features further include non-continuous sequence features, and the extraction of sequence features corresponding to each nucleotide sequence includes:

[0015] For each nucleotide sequence, bases are extracted from the nucleotide sequence according to a preset interval length, and multiple second sub-sequences corresponding to the nucleotide sequence are obtained based on the extracted bases;

[0016] Calculate the second repetition frequency of each second subsequence based on the number of repetitions of each second subsequence;

[0017] The non-continuous sequence features are constructed based on the second repetition frequency of multiple second sub-sequences.

[0018] In some embodiments, fusing the sequence features and corresponding characterization features of each nucleotide sequence to obtain a fusion feature for each nucleotide sequence includes:

[0019] The sequence features and characterization features corresponding to each nucleotide sequence are converted into the same feature dimension;

[0020] For each nucleotide sequence, the weight of the sequence feature is determined based on the correlation between the sequence feature after feature dimension transformation and the characterization feature;

[0021] For each nucleotide sequence, the corresponding weight is weighted and calculated with the sequence feature to obtain the fusion feature.

[0022] In some embodiments, the second neural network model includes a first-level classifier and a second-level classifier;

[0023] The training process of the second neural network model includes:

[0024] Obtain microbial training samples; the microbial training samples include multiple pre-extracted sample nucleotide sequences, each sample nucleotide sequence corresponding to a sample fusion feature; the sample fusion feature is obtained by fusing pre-extracted sample characterization features and sample sequence features; the sample characterization features are obtained by calling the first neural network model to extract features from each sample nucleotide sequence;

[0025] The pre-trained first-level classifier is invoked to classify the sample fusion features of multiple sample nucleotide sequences as probiotics, thereby obtaining the first classification result for each sample nucleotide sequence, and the secondary characterization of multiple sample nucleotide sequences is determined based on the first classification result.

[0026] The second-level classifier is trained based on the second-level representation.

[0027] In some embodiments, the training process of the second-level classifier further includes:

[0028] For each of the microbial training samples, a portion of the sample nucleotide sequence is extracted from multiple sample nucleotide sequences corresponding to the microbial training sample, a first sample group is divided from the extracted portion of the sample nucleotide sequence, and the first sample group is divided into multiple sub-sample groups;

[0029] Obtain the first classification result of the nucleotide sequence of each sample in the sub-sample group output by the pre-trained first-level classifier;

[0030] The second-level classifier is trained based on the first classification result.

[0031] In some embodiments, the training process of the first-level classifier includes:

[0032] A second sample group is divided from the extracted portion of the sample nucleotide sequence, and the second sample group has a different sample nucleotide sequence from the first sample group;

[0033] The first-level classifier is trained based on the second sample group.

[0034] In some embodiments, obtaining microbial training samples includes:

[0035] Multiple target bacterial species were obtained from the plant soil environment as positive samples, and multiple non-target bacterial species and multiple harmful bacterial species were obtained from the plant soil environment as negative samples.

[0036] Multiple bacteria from non-plant soil environments whose similarity to the harmful bacteria species is within a preset threshold range are obtained as negative samples.

[0037] The positive samples and the negative samples are identified as the microbial training samples.

[0038] To achieve the above objectives, a second aspect of this application provides a probiotic identification device, the device comprising:

[0039] The sample extraction unit is used to acquire a test sample extracted from the plant soil environment and to extract multiple nucleotide sequences from the test sample.

[0040] The feature extraction unit is used to call a preset first neural network model to extract the characterization features of each nucleotide sequence;

[0041] The feature fusion unit is used to extract the sequence features corresponding to each nucleotide sequence, and fuse the sequence features and corresponding characterization features of each nucleotide sequence to obtain the fusion feature of each nucleotide sequence.

[0042] The identification unit is used to call a pre-trained second neural network model to identify the fusion features corresponding to multiple nucleotide sequences, and obtain the probiotic identification result of the sample to be tested.

[0043] To achieve the above objectives, a third aspect of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0044] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0045] To achieve the above objectives, a fifth aspect of this application provides a computer program product comprising a computer program that is read and executed by a processor of a computer device, causing the computer device to perform the method described in the first aspect.

[0046] The probiotic identification method, apparatus, electronic device, and storage medium proposed in this application extract the nucleotide sequence of the sample to be tested and fuse the characterization features and sequence features of the nucleotide sequence. A second neural network model is used to identify the fused features to obtain the probiotic identification result of the sample. Therefore, this application embodiment can realize probiotic identification of samples in the plant soil environment based on a model, thus enabling the application of this application embodiment to large-scale data. Furthermore, the second neural network model is based on fused features for probiotic identification, and the fused features can contain rich information in the sample to be tested, thereby improving the accuracy of probiotic identification of the sample. Attached Figure Description

[0047] The accompanying drawings are provided to further understand the technical solutions of this disclosure and constitute a part of the specification. They are used together with the embodiments of this disclosure to explain the technical solutions of this disclosure and do not constitute a limitation on the technical solutions of this disclosure.

[0048] Figure 1 This is a flowchart of the probiotic identification method provided in the embodiments of this application;

[0049] Figure 2 yes Figure 1 A flowchart of an embodiment of step S103;

[0050] Figure 3 yes Figure 1 Another embodiment of step S103 is shown in the flowchart.

[0051] Figure 4 yes Figure 1 Another embodiment of step S103 is shown in the flowchart.

[0052] Figure 5 This is a schematic diagram of the training process of the second neural network model provided in the embodiments of this application;

[0053] Figure 6 This is a flowchart of an embodiment of the training process of the second neural network model provided in this application;

[0054] Figure 7 This application provides a flowchart of an embodiment for constructing a dataset;

[0055] Figure 8 This is a flowchart of an embodiment of the first-level classifier training process provided in this application;

[0056] Figure 9 This is a flowchart of another embodiment of the training process of the second neural network model provided in this application;

[0057] Figure 10 This is a schematic diagram of the test results of the probiotic identification method provided in the embodiments of this application;

[0058] Figure 11 This is a schematic diagram of the probiotic identification device provided in the embodiments of this application;

[0059] Figure 12 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this disclosure.

[0061] Before providing a further detailed description of the embodiments of this application, the nouns and terms used in the embodiments of this application are explained, and the nouns and terms used in the embodiments of this application shall be interpreted as follows:

[0062] Open Reading Frame (ORF): An important term in molecular biology, referring to a continuous nucleotide sequence from a start codon (such as ATG) to a stop codon (such as TAA, TAG, or TGA).

[0063] Principal Component Analysis (PCA) is a linear dimensionality reduction technique that maps high-dimensional data to a low-dimensional space through orthogonal transformations while preserving the main features of the data. Kernel PCA is a nonlinear extension of PCA, representing kernel principal component analysis. Kernel PCA implicitly maps data to a high-dimensional space using kernel tricks, and then performs PCA in that high-dimensional space to capture nonlinear relationships.

[0064] min-max normalization is a data standardization method that linearly maps data to a specified range (usually [0,1] or [-1,1]) to eliminate dimensional differences.

[0065] Probiotics are a class of live microorganisms that colonize the host's internal environment and interact with the original microbial community. Probiotics can regulate the balance of the microecology and produce health benefits. For example, probiotics in soil can promote plant growth, maintain plant health, and enhance plant resistance to diseases through various mechanisms. Therefore, the identification of probiotics in plant soil is of great significance. Traditional probiotic identification methods typically compare the gene sequences of unlabeled samples with samples in a known database, labeling the unlabeled samples based on indicators such as sequence similarity, sequence coverage, and homology characteristics to identify probiotics in the unlabeled samples. However, this method relies on the reliability of the database, has a long comparison time, low processing efficiency, is not suitable for large-scale data, and cannot handle complex sample environments, limiting its application scope. In addition, related technologies also include computer-based probiotic identification methods, which identify probiotics based on trained models. While this method can handle large-scale data, it mainly focuses on environments such as animal intestines and food. Because the types and functions of microorganisms vary in different environments, this method may not guarantee identification performance in soil environments.

[0066] Based on this, embodiments of this application provide a probiotic identification method and apparatus, electronic device and storage medium, which can improve the accuracy of probiotic identification in plant soil environment.

[0067] The probiotic identification method provided in this application relates to the field of bioinformatics. This probiotic identification method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the probiotic identification method, but is not limited to the above forms.

[0068] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0069] The probiotic identification method provided in the embodiments of this application is described below.

[0070] Reference Figure 1 In some embodiments, the probiotic identification method provided in this application includes, but is not limited to, steps S101 to S105.

[0071] Step S101: Obtain the test sample extracted from the plant soil environment, and extract multiple nucleotide sequences from the test sample;

[0072] Step S102: Call the preset first neural network model to extract the characterization features of each nucleotide sequence;

[0073] Step S103: Extract the sequence features corresponding to each nucleotide sequence, and fuse the sequence features of each nucleotide sequence with the corresponding characterization features to obtain the fusion features of each nucleotide sequence;

[0074] Step S104: Call the pre-trained second neural network model to identify the fusion features corresponding to multiple nucleotide sequences and obtain the probiotic identification results of the sample to be tested.

[0075] Steps S101 to S104 of this embodiment involve extracting the nucleotide sequence of the sample to be tested and fusing the characterization features and sequence features of the nucleotide sequence. The fused features are then identified using a second neural network model to obtain the probiotic identification result of the sample. Therefore, this embodiment can achieve probiotic identification of samples in the plant soil environment based on a model, thus enabling its application to large-scale data. Furthermore, the second neural network model identifies probiotics based on fused features, which can contain rich information from the sample to be tested, thereby improving the accuracy of probiotic identification.

[0076] In step S101 of some embodiments, the sample to be tested may refer to a sample extracted from the plant soil environment, such as a bacterial strain extracted from the plant soil environment. Multiple nucleotide sequences (i.e., ORFs) are extracted from the sample to be tested based on the start codon and stop codon.

[0077] In step S102 of some embodiments, the first neural network model can refer to a large biological sequence model. Large biological sequence models provide a more advanced method to capture the complex patterns and semantic information of microbial genome sequences. Large biological sequence models can include Evo, HyenaDNA, DnaBert-2, etc., which learn from a large amount of biological sequence information through pre-training techniques, thereby capturing the global features of the sequence. Alternatively, the first neural network model can refer to the large language model Nucleotide Transformer. This large language model integrates information from 3202 human genomes and 850 different species genomes, providing pre-trained models with 50 million to 2.5 billion parameters, suitable for DNA characterization extraction from various species. Taking the large language model as an example, it can efficiently extract characterization features from each nucleotide sequence in the sample to be tested.

[0078] In step S103 of some embodiments, sequence features may refer to local patterns or statistical features used to describe each nucleotide sequence. For example, sequence features may include features based on k-mer (a continuous subsequence of length k extracted from a biological sequence or text, which can describe the local continuous pattern of the sequence) and features based on g-gap (a k-mer variant that allows g positional intervals in the sequence, which can describe the discontinuous functional association pattern of the sequence). For each nucleotide sequence, the sequence features and characterization features of the nucleotide sequence are fused to introduce knowledge into the characterization features, supplement more diverse information, and enhance interpretability.

[0079] Reference Figure 2 In some embodiments, the sequence features include continuous sequence features, and the step S103 of “calculating the sequence features corresponding to each nucleotide sequence” includes, but is not limited to, steps S201 to S203.

[0080] Step S201: For each nucleotide sequence, the nucleotide sequence is divided according to a preset continuous length to obtain multiple first sub-sequences corresponding to the nucleotide sequence;

[0081] Step S202: Calculate the first repetition frequency of each first subsequence based on the number of repetitions of each first subsequence;

[0082] Step S203: Construct continuous sequence features based on the first repetition frequencies of multiple first subsequences.

[0083] In step S201 of some embodiments, the preset continuous length can refer to k, and the first subsequence can be a k-mer. For example, assuming a nucleotide sequence is ATGCTA and k=2, the first subsequence includes AT, TG, GC, CT, and TA. It is understood that the specific value of k can be adaptively set according to actual needs, and this embodiment of the application does not specifically limit it.

[0084] In steps S202 to S203 of some embodiments, for each first subsequence, the first repeat frequency may refer to the normalized frequency, and the continuous sequence feature may refer to the frequency vector constructed based on the first repeat frequencies corresponding to multiple first subsequences. For example, for a certain nucleotide sequence, the genome of that nucleotide sequence may be defined as shown in Equation 1 below.

[0085] S = R1R2R3...R L (Equation 1)

[0086] Among them, R i(i = 1, 2, 3, ..., L) represents the i-th base in the nucleotide sequence S, and L represents the length of the corresponding nucleotide sequence. Thus, the continuous sequence characteristics can be represented as shown in Equation 2 below.

[0087] V k-mer =[f1 k-mer f2 k-mer ...f i k-mer ...f L k-mer (Equation 2)

[0088] Among them, f i k-mer This represents the first repetition frequency of the i-th first subsequence in the nucleotide sequence S. Specifically, the first repetition frequency can be calculated according to Equation 3 below.

[0089]

[0090] Where, n i This represents the frequency of the i-th first subsequence repeating in the nucleotide sequence S.

[0091] Reference Figure 3 In some embodiments, the sequence features also include non-continuous sequence features. Step S103, "calculating the sequence features corresponding to each nucleotide sequence", includes, but is not limited to, steps S301 to S303.

[0092] Step S301: For each nucleotide sequence, base extraction is performed on the nucleotide sequence according to the preset interval length, and multiple second sub-sequences corresponding to the nucleotide sequence are obtained based on the extracted bases;

[0093] Step S302: Calculate the second repetition frequency of each second subsequence based on the number of repetitions of each second subsequence;

[0094] Step S303: Construct non-continuous sequence features based on the second repetition frequencies of multiple second subsequences.

[0095] In step S301 of some embodiments, the preset interval length can refer to g, and the second subsequence can be a g-gap. For example, assuming a nucleotide sequence is ATGCTA and g = 2, the second subsequence includes AC, TT, and GA. It is understood that the specific value of g can be adaptively set according to actual needs, and this embodiment of the application does not specifically limit it.

[0096] In steps S302 to S303 of some embodiments, for each second subsequence, the second repeat frequency can refer to the normalized frequency, and the discontinuous sequence feature can refer to the frequency vector constructed based on the second repeat frequencies corresponding to multiple second subsequences. For example, for a certain nucleotide sequence, if the genomic definition of the nucleotide sequence is still as shown in Equation 1, then the discontinuous sequence feature can be expressed as shown in Equation 4 below.

[0097] V g-gap =[f1 g-gap f2 g-gap ...f j g-gap ...f L g-gap (Equation 4)

[0098] in, This represents the second repeat frequency of the j-th second subsequence in the nucleotide sequence S. Specifically, the second repeat frequency can be calculated according to Equation 5 below.

[0099]

[0100] Where, m j This represents the frequency of the j-th second subsequence repeating in the nucleotide sequence S.

[0101] Thus, for each nucleotide sequence in the sample to be tested, the sequence characteristic V of that nucleotide sequence is... s It can be represented as shown in Equation 6 below.

[0102]

[0103] Reference Figure 4 In some embodiments, step S103, "fusing the sequence features and corresponding characterization features of each nucleotide sequence to obtain the fusion feature of each nucleotide sequence", includes, but is not limited to, steps S401 to S403.

[0104] Step S401: Convert the sequence features and characterization features corresponding to each nucleotide sequence into the same feature dimension;

[0105] Step S402: For each nucleotide sequence, determine the weight of the sequence feature based on the correlation between the sequence features after feature dimension transformation and the characterization features;

[0106] Step S403: For each nucleotide sequence, the corresponding weight and sequence features are weighted and calculated to obtain the fusion feature.

[0107] In step S401 of some embodiments, in order to effectively fuse the sequence features and the representation features, it is necessary to preprocess the sequence features and the representation features, that is, to convert the sequence features and the representation features into the same feature dimension. For example, both the sequence features and the representation features can be converted into 100-dimensional vectors.

[0108] In steps S402 to S403 of some embodiments, various methods can be used to fuse the sequence features and representation features after feature dimension transformation. For example, the sequence features and representation features can be directly concatenated to obtain fused features. Alternatively, numerical calculations such as mean, median, maximum, and minimum values ​​can be performed on the sequence features and representation features to obtain fused features. Or, a model-based method can be used to fuse the sequence features and representation features after feature dimension transformation. This application does not specifically limit the fusion method.

[0109] Taking the cross-attention mechanism in the model approach as an example, as shown in Equation 7, cross-attention dynamically assigns a weight to each query by calculating the similarity between the query and the key.

[0110]

[0111] Where Q represents the Query matrix, K represents the Key matrix, V represents the Value matrix, and d k The dimension of the key vector is represented by Softmax(), which is the normalization exponential function.

[0112] In this embodiment, representational features are primary, while sequence features are secondary; that is, representational features are used as the "query," and sequence features are used as the "key" and "value." Thus, based on the cross-attention mechanism, the weights of the sequence features can be obtained, which reflect the degree of attention the representational features pay to the extracted sequence features. Combining this with Equation 7, a weighted sum of the sequence features based on the calculated weights can be obtained, resulting in a fused feature of the sequence features and representational features.

[0113] The benefits of steps S401 to S403 are that the cross-attention mechanism can dynamically adjust the weights of different feature sources according to the characteristics of the input data, thus flexibly adapting to different test samples. Furthermore, the weights can determine which sequence features contribute most to the model's decision, thereby enhancing the model's interpretability. In addition, the cross-attention mechanism can achieve information complementarity, enhancing the comprehensiveness of the representation.

[0114] In step S104 of some embodiments, the second neural network model can refer to a model capable of identifying whether a test sample is a probiotic based on fusion features. The fusion features corresponding to multiple nucleotide sequences (e.g., 100 nucleotide sequences) of the test sample are used as input data to the second neural network model. The second neural network model identifies the test sample based on these nucleotide sequences, obtaining a probiotic identification result indicating whether the test sample is a probiotic or not. In this way, probiotic identification of test samples in the plant soil environment can be achieved.

[0115] Understandably, while model-based methods exist for identifying probiotics in animal guts, the ecological complexity of plant and soil microorganisms, the interactions between microorganisms and plants, and the unique biological characteristics of the soil environment result in significant differences between data from plant and soil environments and data from animal gut environments. Therefore, even when using data from plant and soil environments to fine-tune models for animal gut environments, it is difficult to guarantee accuracy when identifying probiotics based on the fine-tuned model.

[0116] The training process of the second-generation neural network model is explained below.

[0117] Reference Figure 5 and Figure 6 In some embodiments, the second neural network model includes a first-level classifier and a second-level classifier, and the training process of the second neural network model includes, but is not limited to, steps S601 to S604.

[0118] Step S601: Obtain microbial training samples;

[0119] Step S602: Obtain microbial training samples; the microbial training samples include multiple pre-extracted sample nucleotide sequences, each sample nucleotide sequence corresponding to a sample fusion feature; the sample fusion feature is obtained by fusing pre-extracted sample characterization features and sample sequence features; the sample characterization features are obtained by calling the first neural network model to extract features from each sample nucleotide sequence;

[0120] Step S603: Call the pre-trained first-level classifier to classify the probiotics based on the sample fusion features of multiple sample nucleotide sequences, obtain the first classification result of each sample nucleotide sequence, and determine the secondary characterization of multiple sample nucleotide sequences based on the first classification result.

[0121] Step S604: Train the second-level classifier based on the second-level representation.

[0122] In step S601 of some embodiments, the microbial training sample may refer to the sample used to train the second neural network model.

[0123] Reference Figure 7 In some embodiments, step S601 includes, but is not limited to, steps S701 to S703.

[0124] Step S701: Obtain multiple target bacterial species as positive samples from the plant soil environment, and obtain multiple non-target bacterial species and multiple harmful bacterial species as negative samples from the plant soil environment.

[0125] Step S702: Obtain multiple bacteria from the non-plant soil environment whose similarity to harmful bacteria species is within a preset threshold range as negative samples.

[0126] Step S703: Multiple positive samples and multiple negative samples are identified as microbial training samples.

[0127] In step S701 of some embodiments, the target bacterial species can refer to bacterial species with probiotic functions, the non-target bacterial species can refer to bacterial species without probiotic functions, and the harmful bacterial species can refer to microorganisms that have a negative impact on the ecological environment. It is understood that probiotics and harmful bacterial species have a biologically opposing and dynamically balanced relationship. For example, 724 publicly released plant soil probiotic spores are obtained as positive samples from a public database (such as the National Center for Biotechnology Information, NCBI), and 384 non-spore-forming bacteria and 80 pathogenic bacteria from plant soil are obtained as negative samples from NCBI. It is understood that after obtaining the above positive and negative samples, these samples can be deduplicated to remove duplicate or redundant samples. Deduplication methods include sequence alignment deduplication, deduplication based on statistical characteristics (such as sequence length, GC content, coverage, etc.), and similarity threshold screening, etc., which are not specifically limited in this application example.

[0128] In step S702 of some embodiments, to increase the number of negative samples, data is collected from other environments (such as non-plant soil environments) as negative samples. Specifically, publicly available bacteria from other environments are collected from NCBI, and after deduplication, these bacteria are compared with 80 pathogens one by one using the Average Nucleotide Identity (ANI). Bacteria with a similarity within a preset threshold range (e.g., 95% to 99%) are used as negative samples. For example, based on the above deduplication and similarity comparison operations, 1468 bacteria can be obtained as new negative samples. It is understood that the specific value of the preset threshold range can be adaptively set according to actual needs, and this embodiment of the application does not specifically limit it.

[0129] In step S703 of some embodiments, multiple positive samples and multiple negative samples obtained based on the above embodiments can be used as microbial training samples. In addition, in order to avoid the influence of extreme data on the training of the second neural network model, the positive and negative samples in the training dataset can be filtered out, such as positive and negative samples with a total DNA sequence length of less than 1,000,000 bp or greater than 9,000,000 bp.

[0130] The advantage of steps S701 to S703 is that, since a large amount of data is collected from the non-plant soil environment, it would be time-consuming to determine whether each data point has probiotic function. In this embodiment, new negative samples are obtained from the non-plant soil based on pathogens, which essentially involves searching for data homologous to the pathogens in the non-plant soil environment. This not only improves the efficiency of obtaining new negative samples but also enhances their reliability.

[0131] In step S602 of some embodiments, because the sequences of some positive and negative samples are too long, directly inputting the positive and negative samples into the second neural network model would place high demands on the computing equipment and be time-consuming. Therefore, this embodiment uses the Prodigal tool (a software tool in bioinformatics specifically used to predict protein-coding genes in prokaryotic genomes; it efficiently and accurately predicts gene location, orientation, and function by identifying open reading frames and combining statistical models and dynamic programming algorithms) to extract the ORF of each positive and negative sample (i.e., the microbial training sample), obtaining the sample nucleotide sequence of each positive and negative sample. Furthermore, to reduce the impact of extreme data on the training of the second neural network model, the microbial training samples can be filtered based on the number of extracted sample nucleotide sequences, for example, filtering microbial training samples with fewer than 1000 or more than 8000 extracted sample nucleotide sequences. Moreover, for each remaining microbial training sample, the sample nucleotide sequences in these microbial training samples can also be filtered, such as filtering sample nucleotide sequences with a length less than 100 bp or greater than 2500 bp. Finally, multiple positive samples (e.g., 560 positive samples, totaling 2,294,287 nucleotide sequences) and multiple negative samples (e.g., 560 negative samples, totaling 2,669,941 nucleotide sequences) were randomly selected from the filtered data as microbial training samples.

[0132] For each microbial training sample, its nucleotide sequence is used as input data to the first neural network model to obtain the sample representation features of each nucleotide sequence. In this way, each microbial training sample can be converted into a numerical matrix. In this numerical matrix, the number of rows represents the number of nucleotide sequences contained in the corresponding microbial training sample, the number of columns represents the dimension of the representation features (e.g., 512 dimensions), and the numerical value corresponding to the matrix cell represents the value of the representation feature.

[0133] Furthermore, for each microbial training sample, the sample sequence features corresponding to each sample nucleotide in that microbial training sample can be calculated. The calculation method for the sample sequence features is described in steps S301 to S303 and S401 to S403 above, and will not be repeated here. For example, for each sample nucleotide sequence, the sample sequence features (e.g., a total of 132 dimensions) are obtained based on the extracted k-mer (including 1-mer, 2-mer, and 3-mer) features and g-gap (including 1-gap, 2-gap, and 3-gap) features.

[0134] Since the sample sequence features are extracted from the nucleotide sequences of each microbial training sample, for each sample nucleotide sequence, the sample representation features obtained based on the first neural network model can be mapped one-to-one with the sample sequence features, and these two features can be fused to obtain the sample fused features. Furthermore, before feature fusion, the sample representation features and sample sequence features can be preprocessed separately, such as performing feature dimension transformation. Specifically, for the sample representation features, a min-max normalization method can be used for global normalization, followed by Kernel PCA to reduce the sample representation features from 512 dimensions to 100 dimensions, and finally, a min-max normalization method can be used again for global normalization. For the sample sequence features, a min-max normalization method is first used for global normalization, followed by a feature selection method based on random forest ranking to select the top 100 features according to feature importance scores to form a new 100-dimensional vector, and finally, a min-max normalization method is used again for global normalization. In this way, the sample representation features and sample sequence features can maintain the same dimension and value range, allowing for the use of multiple fusion methods for feature fusion. The feature fusion method can be referred to the description of steps S402 to S403 above, and will not be repeated here.

[0135] In step S603 of some embodiments, the second neural network model can be a two-level classifier, that is, the second neural network model includes a first-level classifier (such as an Xtreme Gradient Boosting model) and a second-level classifier (such as a logistic regression model). The training objective of the first-level classifier is to enable it to identify whether any input nucleotide sequence (i.e., ORF) originates from probiotics, and the output data of the first-level classifier can be used as the input data of the second-level classifier. The training objective of the second-level classifier is to enable it to identify whether any sample is a probiotic.

[0136] Understandably, through the aforementioned feature fusion steps, each microbial training sample can be represented as a matrix composed of multimodal fusion representations. In the representation matrix, each row represents the fusion feature representation of a sample nucleotide sequence, each column represents the corresponding representation dimension, and each matrix cell represents the corresponding representation fusion value. For each representation matrix, based on whether the microbial training sample corresponding to the representation matrix is ​​a probiotic, a second probiotic label can be determined for each row of data in the representation matrix (i.e., the representation of each sample nucleotide sequence). The second probiotic label can indicate whether the sample nucleotide sequence corresponding to that row of data comes from probiotics. The first probiotic label can indicate whether the corresponding microbial training sample is a probiotic. Furthermore, in order to effectively train the second neural network model, the representation matrix corresponding to each microbial training sample can be compressed into vector form, and this vector form can be used as input data.

[0137] For each microbial training sample, the sample fusion features (i.e., multiple rows of data in the representation matrix) of multiple sample nucleotide sequences corresponding to that microbial training sample are input into a trained first-level classifier. The first-level classifier classifies these multiple sample fusion features as probiotics, that is, it identifies whether the sample nucleotide sequences corresponding to these multiple sample fusion features originate from probiotics. The first classification result output by the first-level classifier represents the score for each corresponding sample nucleotide sequence to originate from probiotics; the higher the score, the greater the probability that the corresponding sample nucleotide sequence originates from probiotics. Based on the scores corresponding to multiple sample nucleotide sequences, a score vector (i.e., a secondary representation) can be constructed.

[0138] In step S604 of some embodiments, for each microbial training sample, the secondary representation corresponding to the microbial training sample is input into the second-level classifier. The second-level classifier identifies probiotics in the sample based on this secondary representation, i.e., it identifies whether the corresponding microbial training sample comes from probiotics. The classification error of the second-level classifier can be determined based on the second classification result output by the second-level classifier (i.e., indicating whether the corresponding microbial training sample comes from probiotics) and the corresponding first probiotic label. The parameters of the second-level classifier can be adjusted based on this classification error, thereby achieving training of the second-level classifier.

[0139] The advantage of steps S601 to S605 is that, due to the long genome sequence of bacteria, it is impossible to input it into the model all at once, thus making it impossible to train the second neural network model to identify the entire bacterium. This embodiment designs the second neural network model as a two-level classifier, solving the problem of training and identification when the bacterial sample sequence is too long. The input data of the first-level classifier in this embodiment is the representation of the sample fusion features in the representation matrix, and the output data is the second-level representation. This is equivalent to converting the representation of the sample fusion features into a second-level representation, and representing the entire bacterial sample based on the second-level representation. During this training process, the first-level classifier can also identify whether a portion of the entire bacterial sample (i.e., the nucleotide sequence) originates from probiotics.

[0140] In some embodiments, the nucleotide sequences of the microbial training samples can be grouped, and the first-level classifier and the second-level classifier can be trained separately based on the grouping data. The training process of the second-level classifier will be described first below.

[0141] Reference Figure 8 In some embodiments, the training process of the second neural network model includes, but is not limited to, the following steps:

[0142] Step S801: For each microbial training sample, extract a portion of the sample nucleotide sequence from the multiple sample nucleotide sequences corresponding to the microbial training sample, divide the first sample group from the extracted portion of the sample nucleotide sequence, and divide the first sample group into multiple sub-sample groups.

[0143] Step S802: Obtain the first classification result of the nucleotide sequence of each sample in the sub-sample group output by the pre-trained first-level classifier;

[0144] Step S803: Train the second-level classifier based on the first classification result.

[0145] In step S801 of some embodiments, for each microbial training sample, a portion (e.g., a preset number) of sample nucleotide sequences are randomly selected from the microbial training sample, and this portion of the sample nucleotide sequences is divided into multiple groups. For example, for the characterization matrix corresponding to the microbial training sample, 1000 rows of data are randomly selected from the characterization matrix, and these 1000 rows of data are divided into two groups, such as a first sample group and a second sample group, each group may contain 500 rows of data. It is understood that the specific value of the preset number and the number of rows contained in each group can be adaptively set according to actual needs, and this embodiment of the application will not elaborate on this in detail.

[0146] Secondly, the first sample group can be further divided into multiple sub-sample groups. For example, the first sample group, which includes 500 rows of data, can be further divided into 5 sub-sample groups, each containing 100 rows of data.

[0147] It is understandable that the second sample group does not overlap with the first sample group. By dividing the second sample group and the first sample group, the first-level classifier and the second-level classifier are trained separately to ensure the training accuracy and improve the robustness of the model.

[0148] In step S802 of some embodiments, for each microbial training sample, each sub-sample group corresponding to the microbial training sample is input into a trained first-level classifier, which can output a corresponding first classification result. This first classification result can represent the score of each sample nucleotide sequence from probiotics among the multiple sample nucleotide sequences corresponding to each sub-sample group.

[0149] In step S803 of some embodiments, for each subsample group, a score vector (i.e., a secondary representation) can be constructed based on the first classification results corresponding to the nucleotide sequences of multiple samples in that subsample group. Specifically, since each subsample group includes 100 rows of data, and each row represents the representation of a sample nucleotide sequence, a 100-dimensional vector (i.e., a secondary representation) can be constructed based on the first classification results corresponding to the 100 rows of data. For each microbial training sample, the secondary representations corresponding to all subsample groups corresponding to that microbial training sample are input into a second-level classifier, which can output the corresponding second classification result. This second classification result can indicate whether the corresponding microbial training sample is a probiotic based on the secondary representation corresponding to each subsample group.

[0150] Based on the second classification result and the first probiotic label corresponding to each subsample group, the classification error of the second-level classifier for that subsample group can be determined. Based on the classification errors corresponding to all subsample groups, the parameters of the second-level classifier are adjusted, thereby realizing the group training of the second-level classifier.

[0151] The advantage of steps S801 to S803 is that by training the second-level classifier with grouped data, for any sample, regardless of whether the sample is complete, only 100 nucleotide sequences of the sample need to be extracted to identify whether the sample is a probiotic based on the first-level and second-level classifiers. Furthermore, different combinations of 100 nucleotide sequences from the same sample can be independently identified, and more reliable conclusions can be obtained based on multiple identification results.

[0152] It is understood that the aforementioned "100" nucleotide sequences are merely illustrative, and the number of nucleotide sequences can be adjusted according to actual needs. When the number of extracted nucleotide sequences is small, it's equivalent to identifying probiotics based on a limited number of nucleotide sequences, which can easily lead to unreliable identification results due to insufficient information. Conversely, when the number of extracted nucleotide sequences is large, if some samples have incomplete genome sequencing, insufficient nucleotide sequences may be extracted, resulting in unidentifiable samples. Therefore, the number of nucleotide sequences extracted is crucial.

[0153] In summary, in practical applications, only 100 ORFs of the sample to be tested can be extracted; that is, in step S101, only 100 nucleotide sequences can be extracted from the sample. Feature extraction is performed on these 100 nucleotide sequences based on the first neural network model to obtain the characterization features of each nucleotide sequence. The sequence features of each nucleotide sequence are calculated, and the sequence features and characterization features of each nucleotide sequence are fused, thus obtaining 100 fused features. The first-level classifier is called to obtain the secondary characterizations corresponding to these 100 fused features. The second-level classifier is then called to identify these secondary characterizations, thereby determining whether the corresponding sample to be tested is a probiotic.

[0154] The training process of the first-level classifier is explained below.

[0155] Reference Figure 9 In some embodiments, the training process of the first-level classifier includes, but is not limited to, steps S901 to S902.

[0156] Step S901: Divide the extracted partial sample nucleotide sequences into a second sample group. The second sample group has a different sample nucleotide sequence from the first sample group.

[0157] Step S902: Train the first-level classifier based on the second sample group.

[0158] In step S901 of some embodiments, based on the random selection and partitioning step of step S801, a second sample group (e.g., 500 rows) can be partitioned from the randomly selected sample nucleotide sequences (e.g., 1000 rows). Furthermore, the sample nucleotide sequences contained in the second sample group are different from those contained in the first sample group; that is, the second sample group and the first sample group do not overlap.

[0159] In step S902 of some embodiments, a first-level classifier is trained for each microbial training sample based on a second sample group of that microbial training sample. Specifically, the sample fusion features (i.e., row data) corresponding to the second sample group are used as input data for the first-level classifier to obtain a third classification result for the sample nucleotide sequence corresponding to each row data in the second sample group. This third classification result can represent the score of whether the corresponding sample nucleotide sequence comes from probiotics. The higher the score, the greater the probability that the corresponding sample nucleotide sequence comes from probiotics.

[0160] The second probiotic tag indicates whether the corresponding sample nucleotide sequence originates from probiotics. Based on the second probiotic tag and the third classification result, the classification error of the first-level classifier can be determined. The parameters of the first-level classifier are then adjusted based on this classification error, thereby achieving the training of the first-level classifier.

[0161] The probiotic identification method provided in this application enables probiotic identification of samples based on a model. By utilizing a cross-attention mechanism to fuse multiple representations (including representation features obtained based on a first neural network model and extracted sequence features), the model's ability to classify complex samples is increased, and the model's identification accuracy is improved. Furthermore, the model-based probiotic identification method enables the processing of large amounts of sequence data in a short time, thus improving data processing speed.

[0162] Understandably, depending on actual needs, in addition to fusing representational features and sequence features, features from other modalities (such as biological physicochemical features) can also be extracted and fused. Furthermore, besides using a two-dimensional classifier, a neural network model can be constructed as the classifier, and the representation matrix of each sample can be compressed into a representation vector based on max pooling or average pooling strategies as input data for the model training phase.

[0163] Reference Figure 10In one specific embodiment, this application was tested on publicly available data. The test data included 70 plant soil probiotic spore samples (totaling 292,282 ORFs) and 70 combinations of plant soil non-spore bacteria and pathogens samples (totaling 342,925 ORFs). Using Matthews Correlation Coefficient (MCC), F1 score, accuracy (ACC), recall (REC), and precision (PRE) as evaluation metrics, the following results were obtained: Figure 10 The evaluation results are shown.

[0164] Depend on Figure 10 It is evident that the model trained using multimodal fusion representations shows significant improvements across various evaluation metrics compared to the model trained solely using representations extracted from a large language model. This demonstrates that the method proposed in this application, which fuses sequence features with features extracted from a large language model (i.e., representation features), can effectively improve model performance. For the identification of probiotics in plant soil, a high accuracy of approximately 97% can be achieved, successfully identifying probiotics in plant soil based on the model. Furthermore, the maximum memory usage and runtime were recorded during testing when extracting representation features from 10 samples. With a batch size of 32, on a server with a 40GB A100 GPU, the maximum memory usage was 3227MB, and the runtime was 4 minutes and 39 seconds. This indicates that this application also possesses high data processing efficiency.

[0165] Reference Figure 11 This application also provides a probiotic identification device, which includes:

[0166] The sample extraction unit 1101 is used to acquire the test sample extracted from the plant soil environment and extract multiple nucleotide sequences from the test sample.

[0167] The feature extraction unit 1102 is used to call a preset first neural network model to extract the characterization features of each nucleotide sequence;

[0168] The feature fusion unit 1103 is used to extract the sequence features corresponding to each nucleotide sequence, fuse the sequence features of each nucleotide sequence with the corresponding characterization features, and obtain the fusion feature of each nucleotide sequence.

[0169] The recognition unit 1104 is used to call the pre-trained second neural network model to identify the fusion features corresponding to multiple nucleotide sequences and obtain the probiotic recognition result of the sample to be tested.

[0170] It is evident that the content of the above-described probiotic identification method embodiments is applicable to the embodiments of this probiotic identification device. The specific functions implemented by this probiotic identification device embodiment are the same as those of the above-described probiotic identification method embodiments, and the beneficial effects achieved are also the same as those achieved by the above-described probiotic identification method embodiments.

[0171] Reference Figure 12 , Figure 12 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0172] The processor 1201 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0173] The memory 1202 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1202 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1202 and is called and executed by the processor 1201 using the probiotic identification method of the embodiments of this application.

[0174] The input / output interface 1203 is used to implement information input and output;

[0175] The communication interface 1204 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0176] Bus 1205 transmits information between various components of the device (e.g., processor 1201, memory 1202, input / output interface 1203, and communication interface 1204);

[0177] The processor 1201, memory 1202, input / output interface 1203 and communication interface 1204 are connected to each other within the device via bus 1205.

[0178] This application also provides a computer program product, which includes a computer program. A processor of a computer device reads and executes the computer program, causing the computer device to perform the probiotic identification method described above.

[0179] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described probiotic identification method.

[0180] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0181] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the recurrence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0182] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0183] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0184] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0185] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0186] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0187] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0188] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0189] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0190] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0191] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for identifying probiotics, characterized in that, The method includes: A test sample was obtained from the plant soil environment, and multiple nucleotide sequences were extracted from the test sample; The characterization features of each nucleotide sequence are extracted by calling a preset first neural network model; Extract the sequence features corresponding to each nucleotide sequence, and fuse the sequence features and corresponding characterization features of each nucleotide sequence to obtain the fusion feature of each nucleotide sequence; The pre-trained second neural network model is invoked to identify the fusion features corresponding to multiple nucleotide sequences, thereby obtaining the probiotic identification result of the sample to be tested.

2. The method according to claim 1, characterized in that, The sequence features include continuous sequence features, and the extraction of sequence features corresponding to each nucleotide sequence includes: For each nucleotide sequence, the nucleotide sequence is divided according to a preset continuous length to obtain multiple first sub-sequences corresponding to the nucleotide sequence; Calculate the first repetition frequency of each of the first subsequences based on the number of repetitions of each of the first subsequences; The continuous sequence features are constructed based on the first repetition frequency of multiple first subsequences.

3. The method according to claim 2, characterized in that, The sequence features also include non-continuous sequence features, and the extraction of sequence features corresponding to each nucleotide sequence includes: For each nucleotide sequence, bases are extracted from the nucleotide sequence according to a preset interval length, and multiple second sub-sequences corresponding to the nucleotide sequence are obtained based on the extracted bases; Calculate the second repetition frequency of each second subsequence based on the number of repetitions of each second subsequence; The non-continuous sequence features are constructed based on the second repetition frequency of multiple second sub-sequences.

4. The method according to claim 1, characterized in that, The step of fusing the sequence features and corresponding characterization features of each nucleotide sequence to obtain the fusion feature of each nucleotide sequence includes: The sequence features and characterization features corresponding to each nucleotide sequence are converted into the same feature dimension; For each nucleotide sequence, the weight of the sequence feature is determined based on the correlation between the sequence feature after feature dimension transformation and the characterization feature; For each nucleotide sequence, the corresponding weight is weighted and calculated with the sequence feature to obtain the fusion feature.

5. The method according to claim 1, characterized in that, The second neural network model includes a first-level classifier and a second-level classifier; The training process of the second neural network model includes: A microbial training sample is obtained; the microbial training sample includes multiple pre-extracted sample nucleotide sequences, each sample nucleotide sequence corresponding to a sample fusion feature; the sample fusion feature is obtained by fusing pre-extracted sample characterization features and sample sequence features; the sample characterization feature is obtained by calling the first neural network model to extract features from each sample nucleotide sequence; The pre-trained first-level classifier is invoked to classify the sample fusion features of multiple sample nucleotide sequences as probiotics, thereby obtaining the first classification result for each sample nucleotide sequence, and the secondary characterization of multiple sample nucleotide sequences is determined based on the first classification result. The second-level classifier is trained based on the second-level representation.

6. The method according to claim 5, characterized in that, The training process for the second-level classifier also includes: For each of the microbial training samples, a portion of the sample nucleotide sequence is extracted from multiple sample nucleotide sequences corresponding to the microbial training sample, a first sample group is divided from the extracted portion of the sample nucleotide sequence, and the first sample group is divided into multiple sub-sample groups; Obtain the first classification result of the nucleotide sequence of each sample in the sub-sample group output by the pre-trained first-level classifier; The second-level classifier is trained based on the first classification result.

7. The method according to claim 6, characterized in that, The training process of the first-level classifier includes: A second sample group is divided from the extracted portion of the sample nucleotide sequence, the second sample group having a different sample nucleotide sequence from the first sample group; The first-level classifier is trained based on the second sample group.

8. The method according to claim 5, characterized in that, The acquisition of microbial training samples includes: Multiple target bacterial species were obtained from the plant soil environment as positive samples, and multiple non-target bacterial species and multiple harmful bacterial species were obtained from the plant soil environment as negative samples. Multiple bacteria from non-plant soil environments whose similarity to the harmful bacteria species is within a preset threshold range are obtained as negative samples. The positive samples and the negative samples are identified as the microbial training samples.

9. A probiotic identification device, characterized in that, The device includes: The sample extraction unit is used to acquire a test sample extracted from the plant soil environment and to extract multiple nucleotide sequences from the test sample. The feature extraction unit is used to call a preset first neural network model to extract the characterization features of each nucleotide sequence; The feature fusion unit is used to extract the sequence features corresponding to each nucleotide sequence, and fuse the sequence features and corresponding characterization features of each nucleotide sequence to obtain the fusion feature of each nucleotide sequence. The identification unit is used to call a pre-trained second neural network model to identify the fusion features corresponding to multiple nucleotide sequences, and obtain the probiotic identification result of the sample to be tested.

10. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the probiotic identification method according to any one of claims 1 to 8.

11. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the probiotic identification method according to any one of claims 1 to 8.

12. A computer program product comprising a computer program that is read and executed by a processor of a computer device, causing the computer device to perform the probiotic identification method according to any one of claims 1 to 8.

Images