Method for screening drug resistance variation characteristics of bacteria based on whole genome level and application

Abstract

The invention relates to the technical field of bioinformatics, and particularly discloses a method for screening drug resistance variation characteristics of bacteria based on a whole genome level and application. According to the method, a machine learning prediction model is established by using a method for associating variation detection and drug resistance characteristics in a whole genome range, the correlation between variation of genes in a coding region and a drug resistance phenotype is considered, and the correlation between variation of a non-coding region and a drug resistance phenotype is also considered, so that the method is more comprehensive; meanwhile, during screening of variation sites, a threshold value is set, the biological significance of the sites is considered and compared, and the result is more reliable and accurate.

Description

Technical Field

[0001] The present application relates to the field of bioinformatics technology, and specifically to a method and application for screening bacterial drug resistance variation characteristics based on the whole genome level. Technical Background

[0002] The problem of bacterial resistance poses a huge threat to global public health. The mortality rate of hospital infections caused by drug-resistant pathogens is much higher than that of general bacterial infections, and it also leads to a significant increase in medical costs. Timely acquisition of pathogen types and drug sensitivity information is of great significance for clinical treatment.

[0003] With technological advancements and the development and improvement of various bacterial databases in recent years, NGS technology has enabled the rapid identification of infectious pathogens. However, the systematic assessment of bacterial resistance characteristics requires further optimization. Due to the complexity of pathogen resistance mechanisms, in addition to the generally accepted existence of a specific resistance gene that produces a corresponding resistance phenotype, it also includes mutations in genes encoding resistance enzymes, SNPs in resistance-determining regions, and insertions and deletions. These mutations occur not only in the coding regions of key genes, but also in non-coding regions. Existing molecular methods often rely on published resistance signature genes or existing resistance gene databases and often overlook variations in non-coding regions upstream and downstream of key genes, many of which are associated with resistance.

[0004] Therefore, by scanning the mutations across the entire genome, screening out important features associated with antibiotic resistance, building a comprehensive and accurate database of mutation features, and establishing a resistance prediction model using machine learning methods, the timeliness and accuracy of pathogen drug sensitivity information can be greatly improved.

[0005] Based on this, this application proposes a method and application for screening bacterial resistance variation characteristics based on the whole genome level. Summary of the Invention

[0006] To solve the above technical problems, the present application is directed to a certain bacteria-antibiotic combination. Based on the cultured and isolated strains and their drug sensitivity test result data collected from public databases (such as NCBI NDARO or BC-BRC libraries) or local hospitals, a sensitive strain is selected as a reference strain. The genome assembly sequence of each strain is directly compared with the reference genome, and the variation information of all bacterial genomes relative to the reference sequence is detected and counted. At the same time, based on the gene prediction information of the reference genome, the detected variant sites are annotated (including the positioning of CDS coding regions and non-coding regions, whether the SNP on the coding region is a synonymous or non-synonymous variation, and whether it causes the destruction of the starter or premature termination of the sequence, whether the InDel on the coding region causes the coding frame to shift, etc.). Then, the drug sensitivity result information of each strain is compared, and the consistency rate of each variant point with drug resistance is statistically calculated (ie, PPV = the number of strains with the variation detected and the phenotype being drug-resistant / the total number of strains with the variation detected), and the variant site annotation information is combined to select the variant site with a higher PPV (default>=0.8) and biological significance as the preliminary candidate drug resistance feature. Next, machine learning methods were used to correlate candidate variant resistance signatures with resistance phenotype data, identifying key features significantly associated with resistance phenotypes. Finally, all features significantly associated with resistance were summarized to construct a pathogen susceptibility prediction model, and the model's effectiveness was evaluated.

[0007] Therefore, the present application first provides a method for screening bacterial resistance variation characteristics based on the whole genome level, comprising the following steps:

[0008] 1) Acquisition of bacterial genome and drug sensitivity data;

[0009] 2) Quality control of bacterial genome and drug sensitivity data;

[0010] 3) Compare and obtain whole genome variation information;

[0011] 4) Association analysis between variant information and drug sensitivity phenotype;

[0012] 5) Variant feature annotation and grouping;

[0013] 6) LASSO regression model second round screening;

[0014] 7) Drug resistance annotation of important variant sites.

[0015] In some embodiments, the step 1) specifically includes: collecting genome sequences and phenotypic data of strains with drug sensitivity test results from the NCBIDARD and PATRIC databases;

[0016] Preferably, during the collection process, the number of strains is greater than 100, the ratio of resistant strains to sensitive strains is balanced, the number of strains collected continuously from the same region, the same literature source, or the same research team does not exceed 50%, and monoclonal strains are avoided;

[0017] In some embodiments, step 2) specifically includes: evaluating the genome size based on sequence alignment of the 16s gene and / or housekeeping gene, and removing strains that do not have the 16s gene and / or housekeeping gene aligned;

[0018] Preferably, the quality control further comprises: removing strains with abnormal genome size; calculating the average nucleotide identity (ANI) of the bacterial genome, and retaining the genomes of strains with an ANI threshold greater than 95%;

[0019] More preferably, the quality control further comprises: for strains whose drug sensitivity data only contain drug sensitivity test values but no drug sensitivity conclusions, determining drug sensitivity conclusions based on the drug sensitivity values according to the CISA drug sensitivity determination criteria, and filtering out strains without drug sensitivity data;

[0020] In some embodiments, step 3) specifically includes: selecting a reference genome of a target pathogen target antibiotic-sensitive strain from the NBCI genome database as a reference sequence; aligning all downloaded genomes with the reference sequence to obtain all variation information on the entire genome, including InDel and SNP information in CDS and non-CDS regions; filtering the alignment results and retaining the best alignment;

[0021] In some embodiments, step 4) specifically includes: performing an association analysis on all variant information contained in the pathogen genome corresponding to a certain antibiotic and the bacterial drug sensitivity phenotype, and calculating the positive consistency rate (PPV) of each variant feature relative to the drug resistance and drug sensitivity phenotype of the antibiotic;

[0022] Preferably, the PPV=the number of strains in which the current gene is detected and is drug-resistant / the number of all strains in which the current gene is detected;

[0023] In some embodiments, step 5) specifically includes: locating and grouping variant sites based on CDS regions and non-CDS regions on the whole genome obtained by gene prediction: CDS regions and non-CDS regions; dividing the CDS regions into synonymous mutation subgroups and non-synonymous mutation subgroups; setting a threshold to screen variant sites with a PPV>0.9 and a detection frequency>3 in drug-resistant bacteria;

[0024] Preferably, the variant sites are selected from the variant sites of the CDS region non-synonymous mutation subgroup and the variant sites of drug-resistant genes or related important functional genes in the upstream and downstream of the non-CDS region group;

[0025] In some embodiments, step 6) specifically includes: performing a second round of screening on the remaining features after group annotation and screening in step 5) using a LASSO regression model to obtain key variant sites that are highly correlated with the drug-resistant phenotype, and calculating weight coefficients;

[0026] Preferably, the analysis method of the LASSO regression model is: using the gene detection distribution matrix (preferably a 0-1 matrix, 0 represents no detection, 1 represents detection) and the antibiotic drug phenotype structure corresponding to the drug-resistant bacteria as input files, using the R package glmnet to associate the variation characteristics with the drug sensitivity phenotype, and performing cross-validation to screen for variation sites associated with the drug resistance phenotype;

[0027] In some embodiments, step 7) specifically includes: annotating important variant sites for drug resistance, comparing the protein sequence translated from the sequence of the region where the site is located in the CDS region and the protein sequence translated from the sequence within 5 kb upstream and downstream of the non-CDS region with the CARD database and the NCBI-nt database, and determining that the region where the screened variant site is located or the region upstream and downstream thereof is a drug resistance gene or an important related functional gene.

[0028] This application also provides a system for screening bacterial resistance variation characteristics based on the whole genome level, including the following modules:

[0029] Module 1): Acquisition of bacterial genome and drug sensitivity data;

[0030] Module 2): Quality control of bacterial genome and drug sensitivity data;

[0031] Module 3): Compare and obtain whole genome variation information;

[0032] Module 4): Association analysis between variant information and drug sensitivity phenotype;

[0033] Module 5): Variant feature annotation and grouping;

[0034] Module 6): LASSO regression model second round screening;

[0035] Module 7): Drug resistance annotation of important variant sites.

[0036] Furthermore, the modules are respectively used for the aforementioned steps 1)-7).

[0037] The present application also provides an electronic device, comprising: a processor and a memory; the processor and the memory are connected, wherein the memory is used to store a computer program, and the processor is used to call the computer program to execute any of the methods described above.

[0038] The present application also provides a computer storage medium, wherein the computer storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, any of the aforementioned methods is executed.

[0039] Beneficial technical effects of this application:

[0040] 1. This application uses a method that associates genome-wide variation detection with drug resistance characteristics to establish a machine learning prediction model, which takes into account the correlation between variation in coding region genes and drug resistance phenotypes, as well as the correlation between variation in non-coding regions and drug resistance phenotypes, making it more comprehensive. In addition, this method is also effective for regions with multiple copies of genes on the genome (for example, the membrane porin OmpK36 on the lung genome mostly has two copies, and the gene sequence identity of the two copies is not 100%, an advantage that existing methods do not have).

[0041] 2. When screening variant sites, this application sets thresholds from a mathematical perspective and also considers the biological significance of the comparison sites. First, select variant sites with a high consistency rate with the drug-resistant phenotype (default PPV>=0.9) and a high frequency of occurrence (default frequency of occurrence>=3); then, for gene coding regions, only retain non-synonymous SNP or InDel variant sites; for non-coding regions, only select variant sites near the upstream and downstream of drug-resistant genes or important functional genes (default 5Kbp region). Finally, use the Lasso regression model to rank the candidate drug-resistant variant features obtained from the above initial screening and select important drug-resistance-related variant sites.

[0042] 3. Finally, based on the presence or absence of drug resistance characteristics of genes known or obtained by other methods and the important variation characteristics screened by this method, this application can further construct a drug resistance phenotype prediction model based on strain WGS or clinical specimen metagenomic sequencing, especially a drug sensitivity prediction model based on clinical specimens. Compared with traditional drug sensitivity tests, it does not need to consider the culture positivity rate of the corresponding pathogens and can directly test the samples. It can detect the DNA sequences of all microorganisms in the samples, and is more timely than the culture drug sensitivity test. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 , Flowchart of this application for screening bacterial resistance variation characteristics based on the whole genome level;

[0044] Figure 2 , detection of characteristic combinations of 2967 strains of Klebsiella pneumoniae;

[0045] Figure 3 , the prediction results of cefoxitin susceptibility of Klebsiella pneumoniae. DETAILED DESCRIPTION

[0046] The embodiments of the present application will be described in detail below with reference to the examples, but it will be understood by those skilled in the art that the following examples are merely illustrative of the present application and should not be construed as limiting the scope of the present application. In the examples, if no specific conditions are specified, the conditions are carried out according to conventional conditions or manufacturer recommendations. The reagents or instruments used are not specified by the manufacturer and are conventional products that can be purchased on the market.

[0047] Definition of some terms

[0048] Unless otherwise defined below, the meaning of all technical terms and scientific terms used in the specific embodiments of the present application is intended to be the same as that generally understood by those skilled in the art. Although it is believed that the following terms are well understood by those skilled in the art, the following definitions are still set forth to better explain the present application.

[0049] As used in this application, an indefinite or definite article when referring to a singular noun eg "a" or "an", "the" includes a plural of that noun.

[0050] As used in this application, the terms "comprises," "comprising," "having," "containing," or "involving" are inclusive or open-ended and do not exclude other unrecited elements or method steps. The term "consisting of" is considered a preferred embodiment of the term "comprising." If a group is defined below as comprising at least a certain number of embodiments, this should also be understood to disclose a group that preferably consists only of these embodiments.

[0051] The term "about" in this application indicates an accuracy range that can be understood by those skilled in the art and still ensures the technical effect of the characteristics discussed. This term usually indicates a deviation from the indicated value by ±10%, preferably ±5%.

[0052] Furthermore, the terms first, second, third, (a), (b), (c), and the like, in the specification and claims, are used to distinguish between similar elements and are not necessarily intended to describe a sequential or chronological order. It is understood that the terms so used are interchangeable under appropriate circumstances, and that the embodiments described herein can be practiced in other sequences than described or illustrated herein.

[0053] The above terms or definitions are provided merely to aid understanding of the present application and should not be construed as having a scope less than that understood by a person skilled in the art.

[0054] The following are specific examples, which are used to illustrate the present application but not to limit it.

[0055] Example 1: Construction of the basic method of this application

[0056] According to the attached Figure 1 The technical roadmap of this application is completed, and each step is described in detail as follows:

[0057] 1. Acquisition of bacterial genomes and drug susceptibility data. Bacterial strain genomes are primarily sourced from the NCBIDARD database (http: / / www.nvbi.nlm.nih.gov / isolates) and the PATRIC database platform (http: / / patricbrc.org), which collects genome sequences and phenotypic data of strains with drug susceptibility test results. The number of target strains collected should be greater than 100, with a balanced ratio of resistant to sensitive strains. The strains should be sourced from a wide range of sources, with no more than 50% originating from the same region or project, to avoid monoclonal strains.

[0058] 2. Quality control of bacterial genome and drug susceptibility data. Use BLAST software to align the 16s gene with housekeeping genes to assess genome size. Remove strains that do not align with the 16s gene and housekeeping genes, as well as strains with abnormal genome sizes. Use FASTANI software to calculate the bacterial genome ANI, retaining strain genomes with a threshold above 95. For strains with only drug susceptibility test values but no drug susceptibility conclusions in the drug susceptibility data, determine the drug susceptibility conclusions based on the CISA drug susceptibility judgment criteria, and filter out strains without drug susceptibility data.

[0059] 3. Compare and obtain whole-genome variation information. Select the reference genome of the target pathogen and antibiotic-susceptible strain from the NBCI genome database as the reference sequence. Use BLAST software to align all downloaded genomes with the reference sequence, output in m0 format, and parse to obtain m6 format. Also output inDel and SNP information contained in each aligned sequence. Filter the sequence results and retain the best alignment.

[0060] 4. Perform association analysis on the variant information obtained for a particular antibiotic and the bacterial susceptibility phenotype. Calculate the association analysis between each variant information and the bacterial strain susceptibility phenotype, and calculate the positive consistency rate of each variant feature relative to the antibiotic resistance phenotype. (PPV = number of strains with the current gene detected and resistant / total number of strains detected for the current gene)

[0061] 5. Based on the annotation of mutation characteristics, they are divided into CDS regions and non-CDS regions; CDS regions are divided into synonymous mutation groups and non-synonymous mutation groups; thresholds are set to screen mutation sites with a PPV higher than 0.9 and a detection frequency of more than 3; mutation sites in the CDS region non-synonymous mutation group and non-CDS region group whose upstream and downstream are drug-resistant genes or related important functional genes are preferred.

[0062] 6. Correlate the drug-susceptibility phenotypes of drug-resistant bacteria with the variant signatures to identify key variant signatures that may be associated with drug resistance. Use the LASSO regression model to identify key variant sites that are highly correlated with the drug-resistant phenotype and calculate the weight coefficients.

[0063] The Lasso regression analysis method is: using the gene detection distribution matrix (0-1 matrix, 0 represents no detection, 1 represents detection) and the antibiotic drug phenotype structure corresponding to the drug-resistant bacteria as input files, using the R package glmnet to associate the variation characteristics with the drug sensitivity phenotype, and perform cross-validation to screen out important genes related to the drug resistance phenotype.

[0064] 7. Annotate important mutation sites, compare the CDS region where the site is located or its upstream and downstream genes with the CARD database and NCBI-nt database, and determine whether the screened mutation site or its upstream and downstream regions are drug-resistant genes or important functional genes.

[0065] 8. When the important weight variation features related to drug sensitivity screened out in 7 are determined, the ROC curve is drawn for this drug sensitivity model, and the performance of the model is determined by calculating the AUC value, Accuracy (consistency rate), PPV (positive consistency rate), and NPV (negative consistency rate).

[0066] 9. Select important features in the model and apply the model to clinical samples.

[0067] Example 2: Detection and analysis of whole genome variation in cefoxitin resistance in Klebsiella pneumoniae

[0068] This application uses the analysis of cefoxitin resistance of Klebsiella pneumoniae as an example to illustrate the specific details of this application, which specifically include the following aspects.

[0069] Step 1. Search and download the K. pneumoniae genome and corresponding cefoxitin susceptibility phenotypic data from public databases (NCBI NDARO and PATRIC). Download from the NCBI NDARO database: Open http: / / www.nvbi.nlm.nih.gov / isolates and search for "Klebsiella pneumoniae" to obtain K. pneumoniae information. In the subsequent Matched Isolates subwindow, click "Cose columns" and select "AST pheotypes." Enter "Laboratory Method" in the KEYWORDS column to obtain K. pneumoniae susceptibility information and download the genome.

[0070] Step 2: Combine the genomes downloaded from NCBI and PATRIC, and screen for genomes of qualified quality through housekeeping gene alignment, average nucleotide identity calculation, and genome size assessment. Perform quality control on drug sensitivity information for genomes with existing drug sensitivity data but no drug sensitivity conclusions according to CISA standards, and retain the drug sensitivity information and genomes that have passed the screening.

[0071] Step 3: Using the PATRIC database genome (573.18472) as the reference sequence, the downloaded K. pneumoniae strain genome was aligned with it (ncbi-blast-2.9.0+ / blastn-task blastn-outfmt 0 -evalue1e-5 -num_alignments 10000). The alignment results were in blast's m0 format, containing information on the variants (InDel\SNP) for each strain relative to the reference sequence. A custom script was used to convert the m0 format to m6 format, output the included variant information, and filter for the best alignment. The alignment results for all strains were statistically analyzed and associated with the bacterial resistance phenotype to obtain a correlation table of variant sites, resistance phenotypes, and PPV.

[0072] Step 4: Select candidate variants with a PPV threshold > 0.9 and a frequency greater than 6 and annotate them. Annotation information includes the variant type, CDS sequence ID number, and specific variant information. The filtered matrix is transposed and converted to the 0-1 matrix input format for the LASSO model, and further features are filtered using a machine learning model.

[0073] Step 5: Determine the selected drug-resistant variant sites, add the corresponding drug-resistant gene subtypes or newly discovered drug-resistant genes to the CARD database, and re-perform the model construction, performance evaluation and other processes to evaluate the selected features.

[0074] Detection of cefoxitin-resistant genomic variation characteristics in 2967 Klebsiella pneumoniae strains:

[0075] Table 1: Statistics of Klebsiella pneumoniae-cefoxitin resistance characteristics and resistance positive consistency rate

[0076]

[0077]

[0078]

[0079] Step 6: Use the above optimization model to verify the clinically isolated single bacterial strain. If the characteristics in Table 1 are detected, it is considered resistant. If none of the characteristics in Table 1 are detected, it is considered sensitive. Figure 2 and Figure 3The characteristic combination detection and drug sensitivity results of 2967 strains of Klebsiella pneumoniae were predicted, and a cefoxitin drug sensitivity prediction model for Klebsiella pneumoniae was constructed, with a model AUC value of 0.909.

[0080] The comprehensive prediction performance of cefoxitin resistance in Klebsiella pneumoniae is as follows:

[0081]

[0082] The results show that the model established with the above characteristics meets the needs and standards of practical applications.

[0083] Example 3: Testing of clinical samples of Klebsiella pneumoniae for cefoxitin resistance

[0084] 1. Data Collection: Clinical samples of bronchoalveolar lavage fluid and sputum were collected for metagenomic sequencing. Sequencing results identified 20 samples of Klebsiella pneumoniae. Clinical drug susceptibility culture results were also collected. These results revealed 10 susceptible strains, 9 resistant strains, and 1 intermediate strain.

[0085] 2. Detection of drug resistance signature genes and mutations in clinical metagenomics

[0086] Clinical bronchoalveolar lavage fluid and sputum samples were used to sequence fastq data for drug resistance analysis.

[0087] 1) KMA software (version 1.4.12a) was used to align the host-free fq sequence with the resistance signature combination database with the alignment parameters: -matrix-ef-nc-na-nf-a-sam to generate an alignment sam format file.

[0088] 2) Perform filtering statistics on the alignment results in sam format and annotate genes and variant sites.

[0089] Filtering conditions: Genes or variant sites of detected specific reads are annotated and then output; for reads without specificity, reads are ranked according to the gene polymorphism within their gene family or subfamily or the number of gene detection sites, and the top two genotypes are output.

[0090] 3) Determination of the identity of drug-resistant characteristic genes and prediction of cefoxitin phenotype

[0091] The applicant has previously published a patent (CN202310625501.8, a model construction method for predicting the species attribution of drug-resistant genes), which attributes the two detected drug-resistant features to species and compares the drug-resistant genes attributed to Klebsiella pneumoniae with the genes in the predicted cefoxitin-resistant feature combination of Klebsiella pneumoniae.

[0092] Strains that detect any of the characteristics in the combination are expected to be cefoxitin-resistant phenotypes. Klebsiella pneumoniae strains that do not detect any characteristics and whose genome coverage exceeds the threshold are predicted to be cefoxitin-susceptible phenotypes. The results of the cefoxitin susceptibility phenotype predictions for the 20 samples obtained above are shown in the following table through analysis of this example:

[0093]

[0094] Clinical sample evaluation statistical results:

[0095]

[0096] The clinical results show that the prediction of the drug-resistant phenotype and sensitive phenotype of Klebsiella pneumoniae in the clinical metagenome detected by using the feature combination of the present application is highly consistent with the actual results.

[0097] It can be seen that the features screened out by the method of screening bacterial resistance variation features based on the whole genome level of this application perform very well in clinical prediction. In addition, it can be seen from the specific processes of Examples 2 and 3 above that this basic method is not limited to the types of pathogens and antibiotics and has universal applicability.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for screening bacterial drug resistance variation characteristics based on the whole genome level, characterized in that: The method comprises the following steps: 1) Acquisition of bacterial genome and drug sensitivity data; 2) Quality control of bacterial genome and drug sensitivity data; 3) Compare and obtain whole genome variation information; 4) Association analysis between variant information and drug sensitivity phenotype; 5) Variant feature annotation and grouping; 6) LASSO regression model second round screening; 7) Drug resistance annotation of important variant sites.

2. The method according to claim 1, characterized in that The step 1) specifically includes: collecting the genome sequence and phenotypic data of the strains with drug sensitivity test results from the NCBIDARD and PATRIC databases; Preferably, during the collection process, the number of strains is greater than 100, the ratio of resistant strains to sensitive strains is balanced, the number of strains collected continuously from the same region, the same literature source, or the same team does not exceed 50%, and monoclonal strains are avoided; 3. The method according to any one of claims 1-2, characterized in that The step 2) specifically includes: evaluating the genome size based on sequence alignment of the 16s gene and / or housekeeping gene, and removing strains that do not have the 16s gene and / or housekeeping gene aligned; Preferably, the quality control further comprises: removing strains with abnormal genome size; calculating the average nucleotide identity (ANI) of the bacterial genome, and retaining the genomes of strains with an ANI threshold greater than 95%; More preferably, the quality control further comprises: for strains whose drug sensitivity data only contain drug sensitivity test values but no drug sensitivity conclusions, determining drug sensitivity conclusions based on the drug sensitivity values according to the CISA drug sensitivity determination criteria, and filtering out strains without drug sensitivity data; 4. The method according to any one of claims 1 to 3, characterized in that: The step 3) specifically includes: selecting a reference genome of a target pathogen target antibiotic-sensitive strain in the NBCIgenome database as a reference sequence; aligning all downloaded genomes with the reference sequence to obtain all variation information on the entire genome, including InDel and SNP information in CDS and non-CDS regions; filtering the alignment results and retaining the best alignment; 5. The method according to any one of claims 1 to 4, characterized in that: The step 4) specifically includes: performing an association analysis on all variant information contained in the pathogen genome corresponding to a certain antibiotic and the bacterial drug sensitivity phenotype, and calculating the positive consistency rate (PPV) of each variant feature relative to the drug resistance and drug sensitivity phenotype of the antibiotic; Preferably, the PPV=the number of strains in which the current gene is detected and is drug-resistant / the number of all strains in which the current gene is detected; 6. The method according to any one of claims 1 to 5, characterized in that: The step 5) specifically includes: locating and grouping the variant sites into CDS regions and non-CDS regions based on the CDS regions and non-CDS regions on the whole genome obtained by gene prediction; dividing the CDS regions into synonymous mutation subgroups and non-synonymous mutation subgroups; setting a threshold to screen variant sites with a PPV>0.9 and a detection frequency>3 in drug-resistant bacteria; Preferably, the variant sites are selected from the variant sites of the CDS region non-synonymous mutation subgroup and the variant sites of drug-resistant genes or related important functional genes in the upstream and downstream of the non-CDS region group; 7. The method according to any one of claims 1 to 6, characterized in that: The step 6) specifically includes: performing a second round of screening on the remaining features after group annotation and screening in step 5) using the LASSO regression model to obtain key variant sites that are highly correlated with the drug-resistant phenotype, and calculating weight coefficients; Preferably, the analysis method of the LASSO regression model is: using the gene detection distribution matrix and the antibiotic drug phenotype structure corresponding to the drug-resistant bacteria as input files, using the R package glmnet to associate the variation characteristics with the drug sensitivity phenotype, and performing cross-validation to screen the variation sites associated with the drug resistance phenotype; 8. The method according to any one of claims 1 to 7, characterized in that: The step 7) specifically includes: annotating the important variant sites for drug resistance, comparing the protein sequences translated from the sequences in the CDS region where the sites are located and the protein sequences translated from the sequences within 5 kb upstream and downstream of the non-CDS region with the CARD database and the NCBI-nt database, and determining whether the region where the screened variant sites are located or the regions upstream and downstream thereof are drug resistance genes or important related functional genes.

9. An electronic device, characterized in that: include: processor and memory; The processor is connected to a memory, wherein the memory is used to store a computer program, and the processor is used to call the computer program to execute the method according to any one of claims 1 to 8.

10. A computer storage medium, characterized in that The computer storage medium stores a computer program, wherein the computer program includes program instructions. When the program instructions are executed by a processor, the method according to any one of claims 1 to 8 is executed.

Images