A comprehensive identification method of grape downy mildew qh strain
By combining optical microscopy observation and image processing with molecular sequencing technology, morphological and molecular feature vectors are generated. A multimodal fusion model is used to identify the QH strain of *Botrytis cinerea*, which solves the problems of ambiguous identification results and poor reproducibility in existing technologies, and achieves efficient and reliable strain identification and source tracing analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- QINGHAI UNIVERSITY
- Filing Date
- 2026-03-31
- Publication Date
- 2026-06-26
AI Technical Summary
In the existing technology, the identification of Grape Downy mildew QH strains relies on optical microscopy observation, which has limited resolution and is highly subjective. The lack of effective integration of morphological data and molecular biological data leads to ambiguous identification results, poor reproducibility, and difficulty in standardization.
The morphological features were extracted by combining optical microscopy observation with image processing algorithms. Morphological parameters of hyphae, sporangia and conidiophores were extracted. The genomic DNA of *Peronospora spp.* was amplified by polymerase chain reaction, and the nucleotide sequence was obtained by sequencing. Molecular feature vectors were generated. A multimodal fusion model was used for comprehensive matching and judgment, and an identification report was generated for source tracing analysis.
This method enables standardized and digital identification of the QH strain of grape downy mildew, improving the objectivity and reliability of the identification results, solving the problems of ambiguous identification results and poor reproducibility in traditional methods, and providing scientific decision support for disease control.
Smart Images
Figure CN122278992A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of plant pathogen identification technology, and in particular to a comprehensive identification method for the QH strain of *Peronospora gracilis*. Background Technology
[0002] Grape downy mildew is a global grape disease caused by *Peronobacterium tumefaciens*, resulting in severe economic losses. This pathogen exhibits intraspecific genetic differentiation, with different strains (or physiological races) showing variations in pathogenicity, host specificity, and susceptibility to fungicides. The QH strain is a group of strains with unique biological characteristics recently discovered in greenhouse grape cultivation areas in Qinghai Province, my country.
[0003] Currently, traditional methods mainly rely on morphological observation, that is, manual identification of morphological features such as sporangia and sporangiophores using an optical microscope. However, ordinary optical microscopes have limited resolution and are difficult to clearly capture the fine structure of downy mildew. The observation results are often vague and easily affected by subjective judgment, especially for closely related species or intraspecific variations with similar morphology, making differentiation difficult. In addition, morphological data and molecular biological data are usually analyzed separately, lacking an effective fusion mechanism, which makes it impossible for multi-source information to corroborate each other. Identification conclusions often rely on the experience of the person identifying the funnel, resulting in strong subjectivity, poor reproducibility, and difficulty in standardization.
[0004] Therefore, a comprehensive identification method for the QH strain of *Peronospora spp.* is proposed to solve the above problems. Summary of the Invention
[0005] The main objective of this invention is to provide a comprehensive identification method for *Peronospora spp.* QH strains to address the problems mentioned in the background above.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a comprehensive identification method for *Peronomyces cerevisiae* strain QH, the method comprising the following steps:
[0007] S1. Obtain samples of grape downy mildew strains to be identified through field sampling and laboratory isolation techniques;
[0008] S2. Use an optical microscope to observe the microscopic morphology of the strain samples and extract the morphological characteristics of hyphae, sporangia and conidiophores.
[0009] S3. Based on morphological feature data, key morphological parameters are extracted through image processing algorithms to generate morphological feature vectors;
[0010] S4. Extract genomic DNA from the strain sample, amplify its housekeeping gene fragment by polymerase chain reaction, and obtain the nucleotide sequence by sequencing.
[0011] S5. Compare the sequencing sequence with the standard sequence of known QH strains, calculate the sequence similarity, and generate molecular feature vectors;
[0012] S6. Input the morphological feature vector and molecular feature vector into the multimodal fusion model, perform comprehensive matching and judgment, and output the identification result;
[0013] S7. Generate an identification report based on the identification results, and conduct source tracing analysis and risk warning in conjunction with strain information in the database.
[0014] Preferably, obtaining the sample of the grape downy mildew strain to be identified in step S1 includes the following steps:
[0015] S11. Collect grape leaves exhibiting typical downy mildew symptoms in greenhouse grape cultivation areas in Qinghai Province;
[0016] S12. After disinfecting the leaf surface, the pathogen was induced to grow under sterile conditions using the in vitro leaf culture method.
[0017] S13. A single strain was purified using the single sporangium isolation technique and then inoculated onto healthy grape leaves for propagation.
[0018] Preferably, the extraction of morphological features in S2 includes the following steps:
[0019] S21. After preparing the bacterial strain sample, observe it under an optical microscope;
[0020] S22. Collect high-resolution images of the morphology of hyphae emerging from stomata, the structure of conidiophores, and the attachment status of sporangia.
[0021] S23. Measure the size of the sporangia in the image and record their length, width and morphological characteristics.
[0022] Preferably, the generation of morphological feature vectors in step S3 includes the following steps:
[0023] S31. Extract the sporangium outline using an image segmentation algorithm, and calculate its aspect ratio, area, and perimeter.
[0024] S32. Encode the morphological variables into fixed-dimensional morphological feature vectors for subsequent fusion and determination.
[0025] Preferably, obtaining the nucleotide sequence in step S4 includes the following steps:
[0026] S41. Genomic DNA was extracted from the bacterial strain samples using the CTAB method, and its purity and concentration were detected by ultraviolet spectrophotometer.
[0027] S42. Design specific primers to amplify the endogenous transcriptional spacer region or β-tubulin gene fragment.
[0028] S43. The specificity of the amplified products was detected by agarose gel electrophoresis, and after recovery and purification, they were sent to the sequencing platform for Sanger sequencing to obtain the nucleotide sequence.
[0029] Preferably, generating the molecular feature vector in step S5 includes the following steps:
[0030] S51. Perform BLAST alignment of the nucleotide sequences obtained from sequencing with the standard sequences of QH strain, and calculate sequence consistency and coverage.
[0031] S52. Extract single nucleotide polymorphism sites and insertion / deletion markers from the sequence as molecular marker features;
[0032] S53. Combine sequence similarity scores with molecular marker features to generate molecular feature vectors.
[0033] Preferably, the generation of molecular marker features in S52 includes the following steps:
[0034] S521. Perform multiple sequence alignment between the nucleotide sequences obtained from sequencing and the standard sequences of QH strains, and use ClustalW or MAFFT algorithms to generate alignment results.
[0035] S522. Based on the comparison results, single nucleotide polymorphism sites and insertion / deletion markers were identified using mutation detection software, and sites that were conserved in QH strains and significantly different in closely related strains were screened out.
[0036] S523. Encode the selected polymorphic sites and insertion / deletion markers according to their positions on the genome to generate binary or numerical molecular marker feature vectors.
[0037] Preferably, the comprehensive matching and determination in S6 includes the following steps:
[0038] S61. Construct a multimodal fusion model and input the morphological feature vector and molecular feature vector into the support vector machine classifier;
[0039] S62. Train the model using QH strains and non-QH strains data from the training sample library and optimize the classification threshold;
[0040] S63. The matching probability between the output strain and the QH strain, and determine whether it is a QH strain based on the set threshold.
[0041] Preferably, optimizing the classification threshold in S62 includes the following steps:
[0042] S621. Collect at least 50 QH strain samples verified by both traditional morphology and molecular biology and 30 closely related non-QH strain samples, extract their morphological feature vectors and molecular feature vectors, and construct a training dataset.
[0043] S622. The support vector machine classifier is trained using cross-validation, and the kernel function parameters and penalty coefficient are optimized using a grid search algorithm.
[0044] S623. Calculate the classification accuracy, recall and F1 score at different thresholds on the validation set, and select the threshold that optimizes the overall performance as the final classification threshold.
[0045] Preferably, generating the identification report in step S7 includes the following steps:
[0046] S71. Generate an identification report based on the identification results, including morphological images, sequence alignment diagrams, matching probabilities, and judgment conclusions.
[0047] S72. Enter the identification results into the strain traceability database and associate them with geographical origin, collection time and host variety information;
[0048] S73. Based on the distribution of QH strains in the database, generate a regional risk warning map to assist in disease prevention and control decisions.
[0049] The present invention has the following beneficial effects:
[0050] 1. In this invention, based on the collection of leaves with typical downy mildew symptoms from greenhouse grape cultivation areas in Qinghai Province, a single strain was purified using in vitro leaf culture and single sporangium isolation technology. High-resolution images and precise size measurements were then taken using an optical microscope equipped with a microscopic imaging system to observe the morphology of hyphae emerging from stomata, the structure of conidiophores, and the attachment status of sporangia. Compared with existing technologies that rely solely on manual visual inspection and subjective experience for morphological description, this invention achieves standardized and digital acquisition of strain morphological characteristics. This solves the problems of blurred identification results and poor reproducibility caused by limited microscope resolution and subjective differences among observers in traditional morphological observation, laying a high-quality data foundation for subsequent feature quantification and improving the objectivity and reliability of the original data.
[0051] 2. In this invention, on the one hand, an image segmentation algorithm is used to extract the sporangium outline and calculate morphological parameters such as aspect ratio, area, and perimeter, encoding a fixed-dimensional morphological feature vector; on the other hand, genomic DNA is extracted using the CTAB method, and ITS or β-tubulin gene fragments are amplified using specific primers and then Sanger sequencing is performed. Single nucleotide polymorphism sites and insertion / deletion markers are further identified from the sequence as molecular marker features. Compared with existing technologies where morphological and molecular data are separated and analyzed independently, this invention extracts morphological quantitative features and molecular marker features in parallel and encodes them uniformly into a computable feature vector, achieving deep fusion of strain phenotypic and genotypic information. This solves the problem of multi-source information not being able to corroborate each other, providing complementary and dimensionally rich input data for subsequent multimodal fusion determination.
[0052] 3. In this invention, morphological feature vectors and molecular feature vectors are input into a multimodal fusion model constructed by a support vector machine for comprehensive matching and judgment. The classification threshold is optimized based on training samples of at least 50 QH strains and 30 closely related non-QH strains, and the final output is the matching probability and judgment conclusion. Compared with the existing technology that relies on the experience of the identifyr for comprehensive judgment, this invention realizes the automated fusion analysis of morphological and molecular features through a machine learning model, solving the problems of strong subjectivity and difficulty in standardization in traditional methods. At the same time, the identification results can be entered into the traceability database to associate geographical origin and host information, and generate regional risk warning maps. Compared with the existing technology where the identification results are only presented in the form of conclusions, this invention realizes a closed loop from single identification to identification, traceability and warning, providing scientific decision support for regional precision prevention and control of diseases. Attached Figure Description
[0053] Figure 1 This is a flowchart illustrating the comprehensive identification method for the QH strain of *Botrytis cinerea* according to the present invention. Detailed Implementation
[0054] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0055] Example 1, please refer to Figure 1 As shown: A comprehensive identification method for *Peronospora spp.* strain QH, the method includes the following steps:
[0056] Sample collection and microscopic observation:
[0057] S1. Obtain samples of grape downy mildew strains to be identified through field sampling and laboratory isolation techniques;
[0058] S2. Use an optical microscope to observe the microscopic morphology of the strain samples and extract the morphological characteristics of hyphae, sporangia and conidiophores.
[0059] Feature extraction and molecular sequencing:
[0060] S3. Based on morphological feature data, key morphological parameters are extracted through image processing algorithms to generate morphological feature vectors;
[0061] S4. Extract genomic DNA from the strain sample, amplify its housekeeping gene fragment by polymerase chain reaction, and obtain the nucleotide sequence by sequencing.
[0062] Molecular alignment and fusion determination:
[0063] S5. Compare the sequencing sequence with the standard sequence of known QH strains, calculate the sequence similarity, and generate molecular feature vectors;
[0064] S6. Input the morphological feature vector and molecular feature vector into the multimodal fusion model, perform comprehensive matching and judgment, and output the identification result;
[0065] S7. Generate an identification report based on the identification results, and conduct source tracing analysis and risk warning in conjunction with strain information in the database.
[0066] Obtaining samples of the *Peronobacterium* strain to be identified in S1 includes the following steps:
[0067] S11. Collect grape leaves exhibiting typical downy mildew symptoms from greenhouse grape cultivation areas in Qinghai Province, including the following steps:
[0068] Leaves with a distinct downy mildew layer, fresh and non-aging lesions were selected for collection. Each leaf was placed individually in a sterile sampling bag, and the collection time and location were marked. The leaves were then quickly transported back to the laboratory under refrigeration at 4°C. The number of leaves collected was determined based on the disease index, which was calculated using the following formula: To ensure that the collected samples cover leaves with different degrees of disease;
[0069] S12. After disinfecting the leaf surface, induce the growth of pathogens under aseptic conditions using an in vitro leaf culture method, including the following steps:
[0070] The collected leaves were surface disinfected in a clean bench. First, they were rinsed three times with sterile water to remove surface impurities. Then, they were soaked in 75% ethanol solution for 30 seconds and 2% sodium hypochlorite solution for 2 minutes for surface sterilization. Finally, they were rinsed five times with sterile water to remove disinfectant residue. After disinfection, the leaves were placed on sterile filter paper to absorb moisture. The growth of pathogens was induced by in vitro leaf culture method. The leaves were laid flat on 1% water agar medium with the back side facing up, with 3-5 leaves per plate. They were cultured in an incubator at 20℃ with 16 hours of light / 8 hours of darkness for 5-7 days. During this period, the growth of downy mildew on the back of the leaves was observed daily.
[0071] S13. A single strain was purified using single sporangial isolation technology and then inoculated onto healthy grape leaves for propagation, including the following steps:
[0072] After a fresh downy mildew layer forms on the underside of the leaf, a single sporangium is picked up under a stereomicroscope using a sterile inoculation needle. A single strain is then purified using a single sporangium isolation technique, where the single sporangium is transferred to sterile water to prepare a sporangium suspension. The formula for calculating the sporangium concentration is as follows: Where N is the number of sporangia in the counting chamber, V is the counting volume (mL), and the concentration is adjusted to approximately The suspension was added dropwise to the underside of healthy grape leaves, 5-10 drops per leaf, 10 μL per drop. The leaves were cultured under the same conditions for 7-10 days for propagation. Once fresh sporangia were produced on the leaves, the sporangia were collected as samples for purification of the strain.
[0073] The extraction of morphological features in S2 includes the following steps:
[0074] S21. After preparing the bacterial strain sample, observe it under an optical microscope, including the following steps:
[0075] Using a sterile inoculation needle, pick up a small amount of fresh sporangia from the back of the cultured leaf and place them in a sterile water droplet on a glass slide. Gently cover with a coverslip to make a temporary slide. Place the slide on the stage of an optical microscope with a microscopic imaging system. First, use a 10x objective to find the field of view, and then switch to a 40x objective for detailed observation. Adjust the focus and aperture to make the image clear and ensure that the fine structures of hyphae, sporangia, and sporangiophores can be distinguished.
[0076] S22. Collect high-resolution images of the mycelial morphology emerging from stomata, the structure of conidiophores, and the attachment status of sporangia, including the following steps:
[0077] Under a microscope, locate typical fields of view where hyphae emerge from leaf stomata. Adjust the fine focus knob to make both the hyphae and the stomatal edges clear simultaneously, and acquire images using a microscopic imaging system. Locate fields of view where the branching structure of conidiophores is clear, focusing on acquiring images of the angle between the main stem and branches of the conidiophores and the attachment pattern of the terminal sporangia. Locate fields of view where sporangia are naturally attached, acquiring images of the attachment morphology of sporangia at the ends of the conidiophores. Acquire at least 10 different fields of view for each type of image, save them in JPEG or TIFF format, and set the image resolution to at least 1920×1080 pixels.
[0078] S23. Measure the size of the sporangia in the image and record their length, width, and morphological characteristics, including the following steps:
[0079] Open the acquired sporangium image using ImageJ image analysis software. First, calibrate the scale using the micrometer image on the stage, setting the conversion relationship between pixels and actual length. Select the sporangium outline in the software and measure its major axis length and minor axis width. Repeat the measurement three times for each sporangium and take the average value. Measure at least 30 sporangia. The formula for calculating the average sporangium length is: ,in The total number of sporangia measured. The length of the i-th sporangium is measured, and the average width is calculated using the following formula: , The width of the i-th sporangium is measured, and the aspect ratio is calculated simultaneously. The morphological characteristics of sporangia, such as shape (elliptical, oval, or nearly spherical), color (pale yellow or colorless), and surface smoothness, were recorded.
[0080] Example 2, Feature Extraction and Molecular Sequencing: Generating morphological feature vectors in S3 includes the following steps:
[0081] S31. Extract the sporangium outline using an image segmentation algorithm, and calculate its aspect ratio, area, and perimeter, including the following steps:
[0082] The collected sporangium images were imported into the MATLAB Image Processing Toolbox. First, the RGB images were converted to grayscale. Median filtering was used to remove image noise, with the filter window size set to 3×3 pixels. Then, the Otsu adaptive thresholding algorithm was used to binarize the image, separating the sporangium region from the background and extracting the binarized contours of the sporangium. Each sporangium contour was marked, and its area (number of pixels) and perimeter (number of boundary pixels) were calculated. The major and minor axis lengths were calculated using the minimum bounding rectangle method. The aspect ratio of the sporangium was calculated using the following formula: ,in Let be the length of the major axis of the smallest bounding rectangle of the sporangium outline. The area is calculated using the formula: (The formula is missing from the provided text.) ,in The height of the image (number of rows). The width of the image (number of columns). The perimeter is calculated using the pixel value of the sporangium region in the binary image (1 represents a sporangium, 0 represents the background). ,in The boundary point coordinate sequence;
[0083] S32. Encode the morphological variables into fixed-dimensional morphological feature vectors for subsequent fusion and determination, including the following steps:
[0084] The morphological parameters extracted from each sporangium were summarized and statistically analyzed. The mean, standard deviation, and coefficient of variation of all measured sporangia (no less than 30) were calculated, and the mean aspect ratio was selected. Average area (pixels), area standard deviation , average perimeter (pixels) and aspect ratio variation coefficient These five key morphological feature variables are combined in a fixed order to construct a five-dimensional morphological feature vector. To eliminate differences in the dimensions of different feature variables, the feature vectors are standardized using Z-score. The standardization formula is as follows: ,in These are the original eigenvalues. and These are the mean and standard deviation of the corresponding features in the training sample library, respectively. The standardized feature vectors are then saved as input data for the subsequent multimodal fusion model.
[0085] Obtaining the nucleotide sequence in S4 involves the following steps:
[0086] S41. Extract genomic DNA from the bacterial strain sample using the CTAB method, and determine its purity and concentration using a UV spectrophotometer, including the following steps:
[0087] Take about 50 mg of fresh sporangia from S3 culture, place them in a mortar, add liquid nitrogen and grind them quickly into a fine powder. Transfer the powder to a 2 mL centrifuge tube, add 800 μL of CTAB extraction buffer preheated to 65℃ (containing 2% CTAB, 100 mM Tris-HCl, 20 mM EDTA, and 1.4 M NaCl), shake well to mix, and incubate at 65℃ for 60 minutes, inverting and mixing every 10 minutes. Add an equal volume of chloroform-isoamyl alcohol (24:1) mixture, gently invert and mix for 10 minutes, centrifuge at 12000 rpm for 10 minutes, and transfer the supernatant to a new centrifuge tube. Add 0.6 volumes of pre-cooled isopropanol, gently mix, and incubate at -20℃ for 30 minutes to precipitate DNA. Centrifuge at 12000 rpm for 10 minutes and discard the supernatant. Wash the precipitate twice with 75% ethanol, dry at room temperature, and dissolve in 50 μL of TE buffer. Take 2 μL of DNA solution and measure the absorbance at 260 nm and 280 nm wavelengths using a UV spectrophotometer. The formula for calculating DNA purity is: An R value between 1.8 and 2.0 indicates that the purity meets the requirements; the formula for calculating DNA concentration is... (Unit: ng / μL), record the concentration value for later use;
[0088] S42. Design specific primers to amplify the endogenous transcriptional spacer region or β-tubulin gene fragment, including the following steps:
[0089] Reference sequences of the ITS region and β-tubulin gene of *Persona chinensis* were downloaded from the NCBI database. Specific primers were designed using Primer Premier 5 software. The primer sequences for the ITS region are as follows: and The expected amplified fragment length is approximately 750 bp; The gene primer sequence is and The expected amplified fragment length is approximately 500 bp; the PCR reaction system is 25 μL, containing 12.5 μL of 2×Taq PCR MasterMix, 1 μL each of forward and reverse primers (10 μM), 2 μL of DNA template (approximately 50 ng), and ddH2O to bring the total to 25 μL; the PCR amplification program is as follows: 94℃ pre-denaturation for 5 minutes, 94℃ denaturation for 30 seconds, 55℃ annealing for 30 seconds, 72℃ extension for 1 minute, for a total of 35 cycles; final extension at 72℃ for 10 minutes;
[0090] S43. The specificity of the amplified products was detected by agarose gel electrophoresis, and after recovery and purification, they were sent to a sequencing platform for Sanger sequencing to obtain the nucleotide sequence, including the following steps:
[0091] Mix 5 μL of PCR amplification product with 1 μL of 6× loading buffer and spot it onto the wells of a 1.5% agarose gel (containing 0.5 μg / mL ethidium bromide). Electrophoresis is performed at a constant voltage of 120V for 30 minutes in 1×TAE buffer. After electrophoresis, observe the gel under a gel imaging system. The appearance of a single, clear target band indicates amplification specificity. Spot the remaining 20 μL of PCR product onto a preparative agarose gel. After electrophoresis, quickly cut the target band under UV light and purify it using an agarose gel extraction kit. Follow the kit instructions for specific steps. Finally, elute the DNA with 30 μL of elution buffer. Perform bidirectional Sanger sequencing on the purified PCR product using an ABI 3730xl sequencer. The sequencing primers are the same as the amplification primers. After sequencing, use SeqMan software to splice and proofread the bidirectional sequencing results, remove low-quality sequences at both ends, and obtain the final nucleotide sequence, which is then saved in FASTA format.
[0092] Example 3, Molecular Alignment and Fusion Determination: The generation of molecular feature vectors in S5 includes the following steps:
[0093] S51. Perform BLAST alignment of the nucleotide sequences obtained from sequencing with the standard sequences of QH strain, and calculate sequence identity and coverage, including the following steps:
[0094] Using the obtained FASTA format nucleotide sequence as the query sequence, and downloading the QH strain standard sequence (such as ITS or β-tubulin sequence) from a local database or NCBI as the target sequence, perform alignment using the NCBI BLAST+ tool (blastn program), setting the parameters as: expected value The character length is 28, the matching reward is 2, the mismatch penalty is -3, and other parameters are default. After the alignment is completed, sequence consistency and coverage are extracted from the output. The sequence consistency calculation formula is as follows: Where m is the number of matching bases in the alignment region, and L is the total length of the alignment region; the coverage calculation formula is: Where Q is the full length of the query sequence, and the values of I and C are recorded as two preliminary molecular features;
[0095] S52. Extracting single nucleotide polymorphism sites and insertion / deletion markers from the sequence as molecular marker features includes the following steps:
[0096] S521. Perform multiple sequence alignment between the nucleotide sequences obtained from sequencing and the standard sequences of the QH strain, and generate alignment results using ClustalW or MAFFT algorithms, including the following steps:
[0097] Collect the nucleotide sequences of the target strain, standard sequences of at least 5 known QH strains (obtained from previous validation samples), and sequences of at least 5 closely related non-QH strains (such as other physiological races or closely related species of *Botrytis cinerea*). Save all sequences in FASTA format in the same file. Launch ClustalW software, import the sequence file, and set the alignment parameters as follows: nick opening penalty 15, nick extension penalty 6.66, DNA conversion weight 0.5. If using MAFFT software, use the automatic selection strategy (-auto parameter), and leave other parameters as default. Run the multiple sequence alignment program to generate the alignment result file (ALN or FASTA format). The reliability of the alignment results is evaluated by the alignment quality score, calculated using the following formula: ,in To compare the total length, The average match score for each alignment column (match = 1, mismatch = 0, gap = 0), a Q value greater than 0.8 indicates good alignment quality;
[0098] S522. Based on the alignment results, single nucleotide polymorphism sites and insertion / deletion markers are identified using variant detection software, and sites that are conserved in QH strains and significantly different in closely related strains are screened, including the following steps:
[0099] Import the generated alignment result file into MEGA or DNAsp software. In MEGA, select the Find Conserved Sites and Find Variable Sites functions, set the minimum allele frequency threshold to 0.05, and identify all variant sites. Export the variant site list, which includes the position of each variant site on the reference sequence, base type, base composition of each sample, etc., and then perform site screening: First, define the conservation condition, that is, the site is identical in bases in at least 80% of QH strain samples, and denote the conserved base as . Secondly, define the condition for significant difference, that is, in non-QH strain samples, the bases and For samples with different proportions ≥70%, the list of variant sites is filtered using Excel or a custom Python script. Sites that meet both conditions are selected as specific molecular markers, and the number of selected marker sites is denoted as N.
[0100] S523. Encode the selected polymorphic sites and insertion / deletion markers according to their positions on the genome to generate binary or numerical molecular marker feature vectors, including the following steps:
[0101] The selected N marker sites are sorted according to their coordinates on the standard sequence (from smallest to largest). For each site, a binary encoding rule is used: if the base of the test sequence at that site is a conserved base of the QH strain... If the sequence matches, it is encoded as 1; otherwise, it is encoded as 0. For insertion / deletion markers, if the InDel exists in the sequence under test, it is encoded as 1; otherwise, it is encoded as 0. This generates an N-dimensional binary feature vector. ,in To enhance the discriminative power of the feature vectors, they can be further converted into numerical feature vectors. The conversion formula is as follows: ,in The weighting coefficient for this site can be calculated based on its conservation score or information entropy value in QH strains. The weighting formula is as follows: , The frequency of the j-th base at the i-th site is used to save the generated molecular marker feature vector, which is then concatenated with the obtained sequence consistency score and coverage to form a complete molecular feature vector.
[0102] S53. Combine sequence similarity scores with molecular marker features to generate molecular feature vectors, including the following steps:
[0103] The obtained consistency score I and coverage C are used as two continuous numerical features, along with the generated N-dimensional binary feature vector. The features are concatenated to form a molecular feature vector with dimension N+2. To eliminate the influence of dimensions, I and C are normalized using a min-max method. The normalization formula is as follows: ,in and To find the minimum and maximum values of the corresponding features in the training sample library, the final normalized molecular feature vector is saved as the input data for the subsequent multimodal fusion model.
[0104] The comprehensive matching and judgment process in S6 includes the following steps:
[0105] S61. Construct a multimodal fusion model by inputting morphological feature vectors and molecular feature vectors into a support vector machine classifier, including the following steps:
[0106] morphological feature vectors (Five-dimensional) and molecular feature vectors (N+2 dimensions) are concatenated end to end to form a fused feature vector. Its total dimension is Support Vector Machine (SVM) is chosen as the classifier, and the kernel function is Radial Basis Function (RBF), whose expression is: ,in This is the fused feature vector of the i-th and j-th samples. Given kernel parameters, the decision function of SVM is: ,in For symbolic functions, The number of support vectors, Here, b is a Lagrange multiplier, and b is a bias term. To define the true label for the i-th training sample, the SVC class from the open-source machine learning library scikit-learn is used during model construction. (Probability output enabled), other parameters are currently at their default values;
[0107] S62. Train the model using QH strains and non-QH strains data from the training sample library, and optimize the classification threshold, including the following steps:
[0108] S621. Collect at least 50 QH strain samples verified by both traditional morphology and molecular biology, and 30 closely related non-QH strain samples. Extract their morphological feature vectors and molecular feature vectors to construct a training dataset, including the following steps:
[0109] At least 50 QH strains, confirmed by both traditional morphological identification (sporangium morphology, size, etc.) and molecular biological identification (ITS / β-tubulin sequence alignment), and at least 30 closely related non-QH strains (including other physiological races or closely related species of the genus *Peronospora*) confirmed by the same method were selected from the laboratory strain library. Morphological feature vectors were extracted from each sample. and molecular eigenvectors The two are concatenated into a fused feature vector. And record the corresponding tags. (QH strains are labeled as 1, and non-QH strains are labeled as 0); combine the fused feature vectors and labels of all samples to form the original dataset, with the total number of samples denoted as M (M≥80). Then, divide the dataset according to... The proportions are randomly divided into training and validation sets, with the number of samples in the training set being... Number of validation set samples Stratified sampling is used during the partitioning process to ensure that the proportions of the two classes of samples in the training and validation sets are consistent with those in the original dataset.
[0110] S622. The support vector machine classifier is trained using cross-validation, and the kernel function parameters and penalty coefficients are optimized using a grid search algorithm, including the following steps:
[0111] Parameter optimization was performed using the GridSearchCV tool from the scikit-learn library. An SVM classifier was defined, with the radial basis function (RBF) chosen as the kernel function. The hyperparameter grid to be optimized was set as follows: candidate values for the penalty coefficient C were [0.1, 1, 10, 100], and candidate values for the kernel parameter γ were [0.001, 0.01, 0.1, 1]. A five-fold cross-validation strategy was adopted, randomly dividing the training set into 5 equal parts. Four parts were used for training and one part for validation in each iteration, repeated 5 times. The average validation accuracy was calculated. The cross-validation accuracy was calculated using the following formula: ,in These represent the number of true positive, true negative, false positive, and false negative samples, respectively, for all... and A grid search is performed using a combination of parameters to select the combination that maximizes the average cross-validation accuracy. ,in The optimized penalty coefficient is... To determine the optimal RBF kernel parameters, if multiple combinations have the same accuracy, the one with the smaller C is selected to avoid overfitting. After determining the optimal parameters, the model is retrained using all training set data to obtain the final SVM classifier.
[0112] S623. Calculate the classification accuracy, recall, and F1 score at different thresholds on the validation set, and select the threshold that optimizes the overall performance as the final classification threshold, including the following steps:
[0113] The fused feature vectors of the validation set samples are input into the trained SVM model, and the probability output of each sample belonging to the QH strain is obtained by Platt scaling. The threshold t is set from 0.1 to 0.9, with a step size of 0.05, resulting in 17 candidate thresholds. For each threshold t, the following steps are performed: Samples that meet the criteria are classified as QH strains (positive class); otherwise, they are classified as non-QH strains (negative class). Calculate the classification performance index at this threshold: accuracy. Recall rate accuracy F1 score Plot the threshold-F1 curve and select the threshold that maximizes the F1 score. As the final classification threshold, if the maximum F1 score corresponds to multiple thresholds, the threshold that yields the highest accuracy is selected; if the accuracy is still the same, the larger threshold is selected to improve the confidence of the judgment. The optimized threshold is then used. Save for the final determination of the test strain;
[0114] S63. The matching probability between the output strain and the QH strain, and the determination of whether it is a QH strain based on a set threshold, including the following steps:
[0115] For each bacterial strain sample to be tested, morphological and molecular feature vectors are extracted and concatenated into a fused feature vector, which is then input into a trained SVM model. The model outputs the probability that the sample belongs to the QH strain. , compare p with the optimized threshold Comparison: If If so, the strain is identified as a QH strain; if If the result is not QH strain, it is determined to be a non-QH strain, and the matching probability p is output as a confidence index of the identification result for subsequent report generation and decision-making reference.
[0116] Generating an authentication report in S7 involves the following steps:
[0117] S71. Generate an identification report based on the identification results, including morphological images, sequence alignment diagrams, matching probabilities, and judgment conclusions.
[0118] S72. Enter the identification results into the strain traceability database and associate them with geographical origin, collection time and host variety information;
[0119] S73. Based on the distribution of QH strains in the database, generate a regional risk warning map to assist in disease prevention and control decisions.
[0120] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.
Claims
1. A comprehensive identification method of QH strain of Uncinula necator, characterized in that, The method includes the following steps: S1. Obtain samples of grape downy mildew strains to be identified through field sampling and laboratory isolation techniques; S2. Use an optical microscope to observe the microscopic morphology of the strain samples and extract the morphological characteristics of hyphae, sporangia and conidiophores. S3. Based on morphological feature data, key morphological parameters are extracted through image processing algorithms to generate morphological feature vectors; S4. Extract genomic DNA from the strain sample, amplify its housekeeping gene fragment by polymerase chain reaction, and obtain the nucleotide sequence by sequencing. S5. Compare the sequencing sequence with the standard sequence of known QH strains, calculate the sequence similarity, and generate molecular feature vectors; S6. Input the morphological feature vector and molecular feature vector into the multimodal fusion model, perform comprehensive matching and judgment, and output the identification result; S7. Generate an identification report based on the identification results, and conduct source tracing analysis and risk warning in conjunction with strain information in the database.
2. The method of claim 1, wherein, The steps for obtaining the Grape Downy mildew strain sample to be identified in S1 include: S11. Collect grape leaves exhibiting typical downy mildew symptoms in greenhouse grape cultivation areas in Qinghai Province; S12. After disinfecting the leaf surface, the pathogen was induced to grow under sterile conditions using the in vitro leaf culture method. S13. A single strain was purified using the single sporangium isolation technique and then inoculated onto healthy grape leaves for propagation.
3. The method of claim 1, wherein, The extraction of morphological features in S2 includes the following steps: S21. After preparing the bacterial strain sample, observe it under an optical microscope; S22. Collect high-resolution images of the morphology of hyphae emerging from stomata, the structure of conidiophores, and the attachment status of sporangia. S23. Measure the size of the sporangia in the image and record their length, width and morphological characteristics.
4. The method of claim 1, wherein, The generation of morphological feature vectors in S3 includes the following steps: S31. Extract the sporangium outline using an image segmentation algorithm, and calculate its aspect ratio, area, and perimeter. S32. Encode the morphological variables into fixed-dimensional morphological feature vectors for subsequent fusion and determination.
5. The method according to claim 1, characterized in that, Obtaining the nucleotide sequence in step S4 includes the following steps: S41. Genomic DNA was extracted from the bacterial strain samples using the CTAB method, and its purity and concentration were detected by ultraviolet spectrophotometer. S42. Design specific primers to amplify the endogenous transcriptional spacer region or β-tubulin gene fragment. S43. The specificity of the amplified products was detected by agarose gel electrophoresis, and after recovery and purification, they were sent to the sequencing platform for Sanger sequencing to obtain the nucleotide sequence.
6. The method according to claim 1, characterized in that, The process of generating molecular feature vectors in S5 includes the following steps: S51. Perform BLAST alignment of the nucleotide sequences obtained from sequencing with the standard sequences of QH strain, and calculate sequence consistency and coverage. S52. Extract single nucleotide polymorphism sites and insertion / deletion markers from the sequence as molecular marker features; S53. Combine sequence similarity scores with molecular marker features to generate molecular feature vectors.
7. The method according to claim 6, characterized in that, The generation of molecular marker features in S52 includes the following steps: S521. Perform multiple sequence alignment between the nucleotide sequences obtained from sequencing and the standard sequences of QH strains, and use ClustalW or MAFFT algorithms to generate alignment results. S522. Based on the comparison results, single nucleotide polymorphism sites and insertion / deletion markers were identified using mutation detection software, and sites that were conserved in QH strains and significantly different in closely related strains were screened out. S523. Encode the selected polymorphic sites and insertion / deletion markers according to their positions on the genome to generate binary or numerical molecular marker feature vectors.
8. The method according to claim 1, characterized in that, The comprehensive matching and determination process in S6 includes the following steps: S61. Construct a multimodal fusion model and input the morphological feature vector and molecular feature vector into the support vector machine classifier; S62. Train the model using QH strains and non-QH strains data from the training sample library and optimize the classification threshold; S63. The matching probability between the output strain and the QH strain, and determine whether it is a QH strain based on the set threshold.
9. The method according to claim 8, characterized in that, The optimization of the classification threshold in S62 includes the following steps: S621. Collect at least 50 QH strain samples verified by both traditional morphology and molecular biology and 30 closely related non-QH strain samples, extract their morphological feature vectors and molecular feature vectors, and construct a training dataset. S622. The support vector machine classifier is trained using cross-validation, and the kernel function parameters and penalty coefficient are optimized using a grid search algorithm. S623. Calculate the classification accuracy, recall and F1 score at different thresholds on the validation set, and select the threshold that optimizes the overall performance as the final classification threshold.
10. The method according to claim 1, characterized in that, The process of generating the identification report in S7 includes the following steps: S71. Generate an identification report based on the identification results, including morphological images, sequence alignment diagrams, matching probabilities, and judgment conclusions. S72. Enter the identification results into the strain traceability database and associate them with geographical origin, collection time and host variety information; S73. Based on the distribution of QH strains in the database, generate a regional risk warning map to assist in disease prevention and control decisions.