Method for predicting relative abundance of antibiotic resistance genes in surface water and application

Abstract

The invention provides a method for predicting the relative abundance of antibiotic resistance genes in surface water and application. The method comprises the following steps: acquiring ARGs relative abundance, water quality physicochemical indexes and genus level bacteria relative abundance of a water sample to be detected, and judging potential bacterial hosts of ARGs by Spearman correlation; the method comprises the following steps: taking water quality physicochemical indexes and relative abundance of potential bacterial hosts as characteristic variables, and respectively taking relative abundance of multidrug ARGs and total ARGs as target variables; adopting automatic machine learning to screen an optimal algorithm; determining an optimal feature set through feature importance ranking and model comparison; and establishing an optimal prediction model to predict the relative abundance of the multi-drug ARGs and the total ARGs of the water sample to be detected. According to the method, the water environment ARGs are predicted based on environmental factors and bacterial indexes, the model accuracy is high, and main driving factors and potential bacterial hosts influencing propagation and proliferation of the ARGs can be identified through interpretable analysis between features and target variables.

Description

Technical Field

[0001] The present invention relates to the fields of environmental monitoring and intelligent environmental management, and in particular to a method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators. Background Art

[0002] Antibiotic resistance genes (ARGs) are mainly caused by the abuse of antibiotics, which leads to bacteria or other microorganisms becoming resistant to them and accelerate their widespread spread among bacteria in the environment, posing a serious threat to ecosystems and human health. It is generally believed that environmental factors such as antibiotics and nutritional status have a significant driving effect on ARGs in the water environment, and bacterial genetic transmission is the main mechanism leading to the proliferation of ARGs. However, the factors affecting the composition and transmission of ARGs in the water environment are still largely unknown, especially the complex interactions between biological and abiotic factors and their synergistic mechanisms on ARGs. As a result, there is a lack of ARGs prediction model methods based on transmission and proliferation factors, which in turn makes ARGs risk control and supervision very difficult.

[0003] At present, a few studies have attempted to use machine learning (ML) to build ARGs prediction models in the environment to explore their propagation and distribution patterns. For example, one study used random forests to predict the abundance of ARGs in activated sludge in sewage treatment plants based on different bacterial groups; another study used deep learning technology and long short-term memory methods to predict the abundance of four ARGs in recreational beaches based mainly on climate variables such as temperature, precipitation, and tides. ML models show strong self-learning and generalization capabilities in solving such complex multivariate nonlinear problems, but these ARGs prediction models are based on meteorological, hydrological, geographical and other influencing factors, rather than on the fundamental mechanisms and emission factors of ARGs propagation and proliferation. Therefore, the accuracy of the prediction models is limited, and it is impossible to effectively control the source of ARGs pollution. At present, there is no prediction model for ARGs in water environments based on both environmental factors and bacterial indicators. Summary of the invention

[0004] Therefore, the present invention aims to provide a method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators.

[0005] To achieve the above object, the present invention provides the following technical solutions: A method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators comprises the following steps: (1) Several surface water samples were collected to obtain the relative abundance of antibiotic resistance genes, water quality physical and chemical index data, and genus-level bacterial relative abundance data. Spearman correlation analysis was performed on bacterial variables to preliminarily screen bacterial variables for input into the prediction models of multidrug resistance genes and total antibiotic resistance genes.

[0006] Dataset construction of Multidrug ARGs prediction model: After the initial cleaning and screening of the original data, the remaining variables include the water quality physical and chemical indicators: ammonia nitrogen, nitrate nitrogen, nitrite nitrogen, total nitrogen, dissolved organic carbon, chemical oxygen demand, dissolved oxygen, chlorophyll a, suspended solids, temperature, pH, redox potential, conductivity, transparency, ozone, total chlorine, quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, tetracycline antibiotics; bacterial indicators: UBA、 Acinetobacter、Planktothrix、Microcystis、Planktophila、AcAMD、Unclassed、 Shewanella、Aquiluna、Pontimonas、Polaribacter、Mycolicibacterium、Vulcanococcus、 Fonsibacter、Limnohabitans、Thiosymbion、Flavobacterium、Pseudomonas、 Hydrogenophaga、HIMB、TMED、PCC、RBG、Rhodoluna ; Preprocess the relative abundance data of the target variable multidrug resistance gene (Multidrug), replace the 0 value of the target variable with 1 / 10 of the minimum value, and then uniformly logarithmize it to form a data set containing characteristic variables and target variables.

[0007] Dataset construction of the total ARGs prediction model: After the initial cleaning and screening of the original data, the remaining variables include the water quality physical and chemical indicators: ammonia nitrogen, nitrate nitrogen, nitrite nitrogen, total nitrogen, dissolved organic carbon, chemical oxygen demand, dissolved oxygen, chlorophyll a, suspended solids, temperature, pH, redox potential, conductivity, transparency, ozone, total chlorine, quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, tetracycline antibiotics; bacterial indicators: Pseudomonas、 Desulfobacca、RBG、Acinetobacter、Microcystis、0-14-0-80-60-11、UTPRO2、 Planococcus_B、Planktothrix、Haliscomenobacter、Idiomarina、Bacteriovorax、 Kuenenia、SG8-40、LD21、Desulfosarcina、PALSA-1003、Brevefilum、Sphaerospermopsis、 Desulforhopalus、Caldilinea、Thiosymbion、Nitrosomonas、Pararheinheimera、 Hyphomicrobium、GCF-002259525、Phenylobacterium ; Preprocess the relative abundance data of total antibiotic resistance genes of the target variable, replace the 0 value of the target variable with 1 / 10 of the minimum value, and then uniformly logarithmize it to form a data set containing characteristic variables and target variables.

[0008] (2) The above-mentioned water quality physical and chemical indicators and bacterial indicators were used as characteristic variables of the relative abundance prediction model of multidrug resistance genes (Multidrug) and total antibiotic resistance genes, respectively. The relative abundance of multidrug resistance genes (Multidrug) and total antibiotic resistance genes were used as target variables. The automatic machine learning model was used to screen out the optimal algorithm GBM. Based on GBM, the initial prediction model of the relative abundance of multidrug resistance genes (Multidrug) and total antibiotic resistance genes was constructed.

[0009] (3) Based on the above initial prediction model, feature importance analysis is performed, and the feature variables of the model are optimized based on the importance ranking of the feature variables. 2 The optimization results of the model were evaluated by using the MSE results, and finally the top 6 and top 10 feature variables ranked in importance in the initial prediction models of multidrug resistance genes and total antibiotic resistance genes were determined as the optimal feature variable sets, which were used to construct the optimal prediction models for the relative abundance of multidrug resistance genes and total antibiotic resistance genes, respectively. (4) Using the optimal prediction model for the relative abundance of multidrug resistance genes (Multidrug) and total antibiotic resistance genes determined above, the relative abundance of antibiotic resistance genes in the surface water samples to be tested was predicted.

[0010] Furthermore, in step (1), the relative abundance of antibiotic resistance genes in surface water samples is obtained by performing metagenomic sequencing on the samples.

[0011] Furthermore, the method for obtaining the relative abundance of ARGs in surface water samples includes: First, water samples were collected, metagenomic sequencing data of the samples were extracted, and the raw sequencing data (rawreads) were quality filtered to obtain high-quality reads (clean reads); Assemble the clean reads obtained from each sample to obtain contigs; Furthermore, according to the nucleic acid composition and abundance change pattern of contigs, the sequences were clustered and binned to obtain metagenomic assembled genomes (MAGs). All MAGs were evaluated according to their completeness and contamination rate, and high-quality representative MAGs that met the requirements were selected. The high-quality representative MAGs were de-redundanted to obtain a non-redundant MAGs set.

[0012] Open reading frames (ORFs) were predicted for all contigs of the non-redundant MAGs set, and ORFs with a length of 100 bp or longer were retrieved and translated into amino acid sequences using the NCBI translation table. A non-redundant gene catalog was constructed using CD-HIT v4.6.1 with 90% sequence identity and 90% coverage. SOAPaligner v 2.21 mapped these reads to the non-redundant gene catalog at 95% identity and assessed gene abundance in each sample. All analyses were performed on marker abundance normalized to reads per kilobase per million (RPKM). To identify ARGs, Diamond v 0.9.19 was used with an e-value cutoff of 10−5 ) The ORFs were matched to the nucleotide sequences of the Structured ARG (SARG) database.

[0013] The relative abundance of each ARG in each sample was calculated according to formula (1):

[0014] Mapped reads: is the number of mapped reads of the contig; Average read length: is the average length of sequenced reads; Total reads: The total number of all sequencing reads.

[0015] Calculation of the abundance of bacterial MAGs in samples: A mapping-based method was used to obtain the percentage of reads in each sample that could be successfully aligned with all MAGs in the non-redundant MAGs set to the total number of reads in the sample, as well as the coverage of each MAG in the non-redundant gene set in each sample. MAGs were classified at the genus level, and the abundance of bacteria at the genus level in each sample was calculated according to formula (2):

[0016] Furthermore, in step (1), the relative abundances of multidrug resistance genes (Multidrug) and total antibiotic resistance genes were correlated with the relative abundance of bacteria at the genus level using the Spearman correlation analysis in the psych package in the R software. Bacteria with r > 0.9 or r < -0.9 and p < 0.001 were considered to be potential bacterial hosts of multidrug resistance genes (Multidrug) and total antibiotic resistance genes and were used as the initial bacterial characteristic variables of the prediction model.

[0017] Furthermore, in step (2), the regression model is established by using the h2o package in the R software.

[0018] Furthermore, in step (3), the method for optimizing the relative abundance model of multidrug resistance genes (Multidrug) and total antibiotic resistance genes of the surface water sample to be tested includes: Multidrug ARGs: Obtain the water quality physical and chemical index data and genus-level bacterial relative abundance data of the surface water sample to be tested, using NH4 + -N, NO3 - -N, NO2 --N, TN, DOC, COD, DO, Chl-a, SS, Temp, pH, ORP, COND, SD, O3, ZCL, quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, tetracycline antibiotics, UBA、Acinetobacter、Planktothrix、Microcystis、Planktophila、AcAMD、 Unclassed、Shewanella、Aquiluna、Pontimonas、Polaribacter、Mycolicibacterium、 Vulcanococcus、Fonsibacter、Limnohabitans、Thiosymbion、Flavobacterium、 Pseudomonas、Hydrogenophaga、HIMB、TMED、PCC、RBG、Rhodoluna As initial feature variables, the feature variable data are substituted into the prediction model of the relative abundance of multidrug resistance genes (Multidrug), and the importance ranking of 45 initial feature variables is output.

[0019] Total antibiotic resistance genes: Obtain the water quality physical and chemical index data and genus-level bacterial relative abundance data of the surface water samples to be tested, using NH4 + -N, NO3 - -N, NO2 - -N, TN, DOC, COD, DO, Chl-a, SS, Temp, pH, ORP, COND, SD, O3, ZCL, quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, tetracycline antibiotics, Pseudomonas、Desulfobacca、RBG、Acinetobacter、Microcystis、0-14-0-80-60- 11、UTPRO2、Planococcus_B、Planktothrix、Haliscomenobacter、Idiomarina、 Bacteriovorax, Kuenenia, SG8-40, LD21, Desulfosarcina, PALSA-1003, Brevefilum, Sphaerospermopsis, Desulforhopalus, Caldilinea, Thiosymbion, Nitrosomonas, Pararheinheimera, Hyphomicrobium, GCF-002259525, Phenylobacterium As initial characteristic variables, the characteristic variable data are substituted into the prediction model of the relative abundance of total antibiotic resistance genes, and the importance ranking of 48 initial characteristic variables is output.

[0020] After eliminating 34 variables with a value of 0 by ranking the importance of the characteristic variables for predicting multidrug resistance genes (Multidrug), the top 11 variables were used as characteristic variables and the prediction model was constructed according to method (2) to complete the prediction results. Then, the model was gradually refined and the variables ranked in the top 10, top 9, top 8, top 7, top 6, and top 5 in importance were tried as characteristic variables. Six prediction models were constructed again according to method (2) to complete the prediction. The R values ​​output by each model were used as the prediction results. 2 The optimal model is evaluated by MSE to ensure the improvement of the accuracy and generalization ability of the prediction model.

[0021] After eliminating 27 variables with a value of 0 based on the importance of variables for predicting total antibiotic resistance genes, the top 21 variables were used as feature variables and the prediction model was constructed according to method (2) to complete the prediction results. The model was then gradually refined and optimized. Four prediction models were constructed using the top 20, top 15, top 10, and top 5 variables as feature variables and the prediction was completed according to method (2). The R values ​​of the models were calculated and the prediction results were compared. 2The optimal model interval is evaluated by using the MSE, and then the variables ranked in the top 14, top 13, top 12, top 11, and top 9 in importance are used as feature variables. Five prediction models are constructed again according to the method (2) and the prediction is completed. Finally, the R values ​​output by each model are used to calculate the prediction results. 2 The optimal model is evaluated by MSE to ensure the improvement of the accuracy and generalization ability of the prediction model.

[0022] Furthermore, in step (4), the method for predicting the relative abundance of antibiotic resistance genes in the surface water sample to be tested includes: The top six variables determined after the optimization of the multidrug resistance gene prediction model include: Planktothrix, Microcystis, Acinetobacter, Thiosymbion, COD, Pseudomonas The top 10 variables determined after the optimization of the total antibiotic resistance gene prediction model are the characteristic variables of the prediction model. Bacteriovorax, Acinetobacter, PALSA-1003, 0-14-0-80-60-11 、NO2 - -N, Pseudomonas, Microcystis, Nitrosomonas, Kuenenia , O3 are characteristic variables of the prediction model. The characteristic variable data are respectively substituted into the prediction models of the relative abundance of multidrug resistance genes (Multidrug) and total antibiotic resistance genes. The target variable prediction value output by the running model is the prediction result of the relative abundance of antibiotic resistance genes in the surface water sample to be tested.

[0023] Furthermore, the method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators also includes: randomly dividing the data set of surface water samples into a training set and a test set, using the training set to establish the prediction model, and using the test set to verify the prediction ability of the prediction model. Preferably, 75% of the data set is used as the training set and 25% of the data set is used as the test set.

[0024] Furthermore, the fitting coefficient R 2 Measure the predictive power of the forecast model: When R 2 When ≤0.3, the fitting effect between the predicted value and the observed value is weak, and the prediction ability of the prediction model is poor; When 0.3<R 2 When ≤0.4, the fitting effect between the predicted value and the observed value is weak, and the prediction ability of the prediction model is weak; When 0.4<R 2 When ≤0.6, the fitting effect between the predicted value and the observed value is moderate, and the prediction ability of the prediction model is moderate; When 0.6<R 2 When ≤1.0, the predicted value and the observed value have a strong fitting effect, and the prediction model has a strong prediction ability.

[0025] The present invention further discloses the application of a method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators in accurately predicting the relative abundance of antibiotic resistance genes (ARGs) in surface water environments. The relative abundance refers to the ratio of the number of antibiotic resistance genes (ARGs) in a sample to the total amount of all genes in the sample. This ratio can be used to estimate the abundance of the gene in the sample. The higher the relative abundance of ARGs, the greater its potential environmental hazard. Predicting the relative abundance of ARGs in a water environment is intended to monitor the spread and change of ARGs in the water environment. Experimental results show that the prediction model shows high accuracy and meets the strong fitting standard, and can be used to monitor the spread and change of ARGs in a surface water environment. Specifically, the relative abundance prediction model of multidrug resistance genes (Multidrug) had an R²=0.96, MSE=0.03 on the training set, and an R²=0.90, MSE=0.22 on the test set; the relative abundance prediction model of total antibiotic resistance genes had an R²=0.891, MSE=0.08 on the training set, and an R²=0.892, MSE=0.06 on the test set.

[0026] The method disclosed in the present invention for predicting the relative abundance of antibiotic resistance genes in surface water environment based on water quality physical and chemical indicators and bacterial indicators has the following technical effects compared with the prior art: The present invention innovatively combines water quality physicochemical indicators with bacterial indicators through automatic machine learning technology to construct an efficient prediction model that can accurately predict the relative abundance of antibiotic resistance genes (ARGs) in surface water environments. In this method, the physicochemical indicators in water quality detection and the relative abundance data of bacteria at the genus level are used as model characteristic variables, and the relative abundances of multidrug resistance genes (Multidrug) and total antibiotic resistance genes are logarithmically processed and used as target variables. After the data set is organized and formed, the optimal algorithm, Gradient boosting regression (GBM), is selected by building an automatic machine learning model as the algorithm core of the prediction model for the relative abundance of multidrug resistance genes (Multidrug) and total antibiotic resistance genes, and a prediction model is built. It has been verified that the model has the following significant advantages: (1) Combination of environmental factors and bacterial indicators: Taking into account both the physical and chemical indicators of water quality and the bacterial indicators, a model is constructed based on the key factors of the proliferation and spread of ARGs, providing a more solid guarantee for the accurate prediction of ARGs.

[0027] (2) High precision: The use of advanced automatic machine learning technology to determine the optimal algorithm for the model not only omits the tedious process of manual algorithm selection, but also ensures the accuracy and prediction efficiency of the model.

[0028] (3) Easy operation: After the model is successfully constructed, only the physical and chemical indicators in water quality testing and the relative abundance data of bacteria at the genus level are needed to accurately predict ARGs, which simplifies the complex ARGs detection process.

[0029] (4) Effective control: The model can identify key water quality factors and potential bacterial hosts that affect the spread of ARGs through interpretable analysis between characteristic variables and target variables, providing a basis for ARGs risk control, which is of great significance for reducing the spread of ARGs and protecting human and animal health.

[0030] In summary, this invention significantly improves the accuracy of antibiotic resistance gene (ARGs) prediction by using automatic machine learning technology and comprehensively considering environmental factors and bacterial indicators. At the same time, the interpretability analysis of the model further reveals the key factors affecting the spread of ARGs, providing a scientific basis for in-depth understanding and effective control of antibiotic resistance in aquatic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the specific implementation of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementations of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. Figure 1 It is the model training effect and test effect of using water quality physical and chemical indicators and bacterial indicators to predict the relative abundance of multidrug antibiotic resistance genes (Multidrug) in the embodiment of the present invention; Figure 2 It is a ranking diagram of the importance of each characteristic variable to the target variable in the prediction model of the relative abundance of multidrug antibiotic resistance genes (Multidrug); Figure 3 The model training effect and test effect of using water quality physical and chemical indicators and bacterial indicators to predict the relative abundance of total antibiotic resistance genes in the embodiment of the present invention; Figure 4 It is the ranking diagram of the importance of each characteristic variable to the target variable in the prediction model of the relative abundance of total antibiotic resistance genes; Figure 5 It is a flow chart of the method for constructing the relative abundance prediction model of antibiotic resistance genes described in the present invention. DETAILED DESCRIPTION

[0032] The following examples are provided for a better understanding of the present invention, but are not intended to limit the best mode of implementation, nor to limit the content and protection scope of the present invention. Any product identical or similar to the present invention obtained by anyone under the inspiration of the present invention or by combining the features of the present invention with other prior arts shall fall within the protection scope of the present invention.

[0033] If no specific experimental steps or conditions are specified in the examples, the conventional experimental steps or conditions described in the literature in the art can be used. The raw materials or instruments used are all conventional products that can be obtained commercially, including but not limited to the raw materials or instruments used in the examples of this application. Example

[0034] This embodiment provides a method for establishing a prediction model for the relative abundance of antibiotic resistance genes, and the specific steps are as follows: (1) Water sample collection The selected water sampling point is a water area in Binhai, Tianjin, located in Binhai New Area, Tianjin. Its core area, Jinghu, is a landscape lake supplied by recycled water, and is adjacent to the Eco-City Wastewater Treatment Plant to the west. The transit waters near Jinghu are Jiyun Canal and Gudao River. Jiyun Canal and Yongding New River merge into the Bohai Sea. Samples were collected in the summer, autumn and winter of 23 and 24. A total of 5 sampling points were set up in the Jinghu Lake area, two sampling points were set up in the surrounding rivers and river confluences, and 3 sampling points were set up in the offshore waters. There were a total of 10 water sampling points, and 25 water samples were collected. After the water sample was vacuum filtered through a 0.22 μm microporous filter membrane, the filter membrane was rolled up and placed in a 10 mL sterile centrifuge tube and stored in a refrigerator at -80 ℃. The water samples were subsequently sent to Shanghai Meiji Biological Co., Ltd. for metagenomic sequencing of the water body.

[0035] (2) Testing of water quality physical and chemical indicators The following conventional water quality physical and chemical indicators of water samples were tested: NH4 + -N, NO3 - -N, NO2 - -N, TN, DOC, COD, DO, Chl-a, SS, Temp, pH value, ORP, COND, SD, O3, ZCL, the detection methods, reference standards and main instruments and equipment of each index are shown in Table 1.

[0036]

[0037] The antibiotic indicators in the water samples were tested, including: quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, and tetracycline antibiotics.

[0038]

[0039] Antibiotic standard samples: Standard substances: 11 sulfonamide drug mixed standard solutions, 100μg / mL (sulfamethoxazole, sulfadimethoxine, sulfadiazine, sulfadimethoxine, sulfamethoxazole, sulfadimethoxine, sulfathiazole, sulfachloropyridazine, sulfamethoxazole, sulfaquinoxaline); 4 tetracycline drug solid mixed standards, 100μg (oxytetracycline, tetracycline hydrochloride, chlortetracycline hydrochloride, doxycycline monohydrochloride hemiethanol hemihydrate); 11 quinolone drugs Mixed standard solution of 8 kinds of macrolide antibiotics, 100μg / mL (sparfloxacin, difloxacin, flumequine, enrofloxacin, ciprofloxacin, danofloxacin, sarafloxacin, ofloxacin, norfloxacin, pefloxacin, lomefloxacin); mixed standard solution of 8 kinds of macrolide antibiotics, 100μg / mL (lincomycin hydrochloride, clindamycin hydrochloride, roxithromycin, josamycin, tylosin, kitasamycin, tilmicosin, clarithromycin); chloramphenicol solution, 100μg / mL, a total of 35 standards.

[0040] (3) Detection of relative abundance of bacteria Collect water samples, extract metagenomic sequencing data of the samples, and perform quality filtering on the raw sequencing data (rawreads) to obtain high-quality reads (clean reads); Assemble the clean reads obtained from each sample to obtain contigs; Further, according to the nucleic acid composition and abundance change pattern of contigs, the sequences were clustered and binned to obtain metagenomic assembled genomes (MAGs). All MAGs were evaluated according to completeness and contamination rate, and high-quality representative MAGs that met the requirements were selected. The high-quality representative MAGs were de-redundanted to obtain a non-redundant MAG set. After obtaining the non-redundant MAG set, GTDB-tk was used for annotation analysis.

[0041] (4) Detection of relative abundance of antibiotic resistance genes First, water samples were collected, metagenomic sequencing data of the samples were extracted, and the raw sequencing data (rawreads) were quality filtered to obtain high-quality reads (clean reads); Assemble the clean reads obtained from each sample to obtain contigs; Furthermore, according to the nucleic acid composition and abundance change patterns of contigs, the sequences were clustered and binned to obtain the bacterial genome draft (metagenomic assembled genomes, MAGs). All MAGs were evaluated according to their completeness and contamination rate, and high-quality representative MAGs that met the requirements were selected. The high-quality representative MAGs were de-redundanted to obtain a non-redundant MAGs set.

[0042] Open reading frames (ORFs) were predicted for all contigs of the non-redundant MAGs set, and ORFs with a length of 100 bp or longer were retrieved and translated into amino acid sequences using the NCBI translation table. A non-redundant gene catalog was constructed using CD-HIT v4.6.1 with 90% sequence identity and 90% coverage. SOAPaligner v 2.21 mapped these reads to the non-redundant gene catalog at 95% identity and assessed gene abundance in each sample. All analyses were performed on marker abundance normalized to reads per kilobase per million (RPKM). To identify ARGs, Diamond v0.9.19 was used with an e-value cutoff of 10 −5 ) The ORFs were matched to the nucleotide sequences of the Structured ARG (SARG) database.

[0043] The relative abundance of each ARG in each sample was calculated according to formula (1):

[0044] Calculation of the abundance of bacterial MAGs in samples: A mapping-based method was used to obtain the percentage of reads in each sample that could be successfully aligned with all MAGs in the non-redundant MAGs set, as well as the coverage of each MAG in the non-redundant gene set in each sample. MAGs were classified at the genus level, and the abundance of bacteria at the genus level in each sample was calculated according to formula (2):

[0045] (5) The Spearman correlation coefficient was used to determine the degree of association between antibiotic resistance genes and bacteria at the genus level, thereby identifying potential bacterial hosts of antibiotic resistance genes.

[0046] The psych package in R software performed correlation analysis on the relative abundance of multidrug resistance genes (Multidrug) and total antibiotic resistance genes obtained with the relative abundance of bacteria. Those with r>0.9 or r<-0.9 and p<0.001 were considered to be potential bacterial hosts of multidrug resistance genes (Multidrug) and total antibiotic resistance genes and were used as the initial bacterial characteristic variables of the prediction model.

[0047] (6) Use automated machine learning to build predictive models Based on the above 21 water quality physical and chemical indicators (NH4 + -N, NO3 - -N, NO2 - -N, TN, DOC, COD, DO, Chl-a, SS, Temp, pH, ORP, COND, SD, O3, ZCL, quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, tetracycline antibiotics) and 24 bacterial indicators (UBA, Acinetobacter, Planktothrix, Microcystis, Planktophila, AcAMD, Unclassed, Shewanella, Aquiluna, Pontimonas, Polaribacter, Mycolicibacterium, Vulcanococcus, Fonsibacter, Limnohabitans, Thiosymbion, Flavobacterium, Pseudomonas, Hydrogenophaga, HIMB, TMED, PCC, RBG, Rhodoluna ) as the characteristic variable, and the relative abundance of multidrug resistance genes (Multidrug) as the target variable.

[0048] Based on 21 water quality physical and chemical indicators (NH4 + -N, NO3 - -N, NO2 - -N, TN, DOC, COD, DO, Chl-a, SS, Temp, pH, ORP, COND, SD, O3, ZCL, quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, tetracycline antibiotics) and 27 bacterial indicators ( Pseudomonas, Desulfobacca, RBG, Acinetobacter, Microcystis, 0-14-0-80-60-11, UTPRO2, Planococcus_B, Planktothrix, Haliscomenobacter, Idiomarina, Bacteriovorax, Kuenenia, SG8-40, LD21, Desulfosarcina, PALSA-1003, Brevefilum, Sphaerospermopsis, Desulforhopalus, Caldilinea, Thiosymbion, Nitrosomonas, Pararheinheimera, Hyphomicrobium, GCF- 002259525, Phenylobacterium ) as the feature variable and the relative abundance of total antibiotic resistance genes as the target variable.

[0049] The relative abundance of antibiotic resistance genes was used as the target variable. When its value was 0, 1 / 10 of the minimum value was taken as a replacement, and the target variable was further logarithmically transformed with a base of 10. The parameters ntree and max_depth were determined when the calculation results reached stability. For each model, 75% of the data set was randomly divided into training sets, and the remaining 25% was used as a test set. The automatic machine learning model was established using the above data set, and the relative abundance prediction model of antibiotic resistance genes was further established based on the determined optimal algorithm gradient boosting regression (GBM). At the same time, the mean square error (MSE) and the fitting coefficient R based on the predictions based on the training set and the test set were used. 2 The two performance parameters are used to evaluate the model. Since different indicators have different numerical ranges, MSE, as an absolute numerical indicator, is only applicable to the comparison of prediction results between the same indicators. In addition, the smaller the MSE, the better the R 2 The larger the value, the larger the R2 As the main evaluation parameter. 2 When ≤0.3, the predicted value and the observed value have poor fitting effect; when 0.3<R 2 ≤0.4, the predicted value and the observed value have a weak fitting effect; when 0.4<R 2 When ≤0.6, the predicted value and the observed value have a moderate fitting effect; when 0.6<R 2 When ≤1.0, the predicted value and the observed value have a strong fit. The data set was randomly divided using the ratios parameter in the h2o package of R software (v.4.4.0). The model was constructed using the h2o.gbm function.

[0050] (7) Selection of important variables and model optimization When establishing the relative abundance prediction model of antibiotic resistance genes based on water quality indicators and bacterial indicators, the input variables were ranked by importance to select important input variables for further model optimization. The variable %IncMSE ranking was obtained by "importance" and "varImpPlot" in R software (v.4.3.3). Through comparative analysis, it was finally determined that the variables ranked in the top 6 and top 10 for predicting multidrug resistance genes (Multidrug) and total antibiotic resistance genes were used as input variables, respectively. Based on the determined GBM algorithm, the model was modeled and evaluated again according to the method in (6). Example

[0051] Model training effect based on training set The results of predicting the relative abundance of multidrug resistance genes (Multidrug) based on the training set using the top 6 feature variables ranked by importance are shown in Table 3. The fitting results of the predicted values ​​and observed values ​​are shown in Figure 1 Black part. It can be seen that the prediction model can achieve a strong fitting effect (R 2 =0.96), and the MSE is 0.03.

[0052]

[0053] The results of predicting the relative abundance of total antibiotic resistance genes using the top 10 feature variables ranked by importance based on the training set are shown in Table 4. The fitting results of the predicted values ​​and observed values ​​are shown in Figure 3 Black part. It can be seen that the prediction model can achieve a strong fitting effect (R 2 =0.892), and the MSE is 0.06.

[0054]

[0055] (2) Model prediction effect based on the test set The prediction model based on the training set is used in the test set to verify the generalization ability of the model, that is, the prediction ability for new data sets, using the mean square error MSE and the fitting coefficient R 2 Conduct an assessment.

[0056] The prediction results of the optimal prediction model for the relative abundance of total antibiotic resistance genes for the test set are shown in Table 5, and the fitting results of the predicted values ​​and observed values ​​are shown in Figure 1 In the gray part, it can be seen that the prediction model can achieve a strong fitting effect (R 2 =0.904), and the MSE is 0.227.

[0057]

[0058] The prediction results of the optimal prediction model for the relative abundance of total antibiotic resistance genes for the test set are shown in Table 6, and the fitting results of the predicted values ​​and observed values ​​are shown in Figure 3 In the gray part, it can be seen that the prediction model can still achieve a strong fitting effect (R 2 =0.892), and the MSE is 0.06.

[0059]

[0060] pass Figure 2 and Figure 4 In the importance ranking diagram, key factors affecting the spread and proliferation of multidrug resistance genes and total antibiotic resistance genes can be identified. Planktothrix ), Acinetobacter ( Acinetobacter ) and Microcystis ( Microcystis ) is the main factor affecting the spread of multidrug resistance genes. bacteriovorax )、Acinetobacter( Acinetobacter ), Microcystis spp. Microcystis )、 PLASA-1003 and NO2 - -N is the main factor in the spread of total antibiotic resistance genes; the feature importance analysis of the prediction model of multidrug resistance genes and total antibiotic resistance genes showed that bacteria have a more significant effect on the spread of ARGs than environmental stress, and Acinetobacter ( Acinetobacter ) and Microcystis ( Microcystis) showed a high importance for both multidrug resistance genes (Multidrug) and total antibiotic resistance genes, indicating that it may be of universal importance in the spread of antibiotic resistance genes.

[0061] It can be seen that the prediction model established by the method provided by the present invention can accurately predict the relative abundance of antibiotic resistance genes in surface water, and can be used to monitor the propagation and change patterns of antibiotic resistance genes.

[0062] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators, characterized in that: The following steps are involved: (1) Take several surface water samples, obtain the relative abundance of their antibiotic resistance genes, and obtain their water quality physical and chemical index data and genus-level bacterial relative abundance data as raw data, and clean and normalize the raw data; (2) After cleaning and preprocessing the raw data, the Spearman correlation coefficient was used to determine the degree of association between antibiotic resistance genes and bacteria at the genus level, thereby identifying potential bacterial hosts of antibiotic resistance genes; Identification of potential bacterial hosts of multidrug resistance genes (Multidrug) includes UBA , Acinetobacter Planktothrix, Microcystis, Planktophila, AcAMD, Unclassed, Shewanella, Aquiluna, Pontimonas, Polaribacter, Mycolicibacterium, Vulcanococcus, Fonsibacter, Limnohabitans, Thiosymbion, Flavobacterium, Pseudomonas, Hydrogenophaga, HIMB, TMED, PCC, RBG, Rhodoluna; Identification of potential bacterial hosts of total antibiotic resistance genes includes Pseudomonas, Desulfobacca, RBG, Acinetobacter,Microcystis,0-14-0-80-60-11,UTPRO2,Planococcus_B,Planktothrix, Haliscomenobacter, Idiomarina, Bacteriovorax, Kuenenia, SG8-40, LD21, Desulfosarcina, PALSA-1003, Brevefilum, Sphaerospermopsis, Desulforhopalus, Caldilinea, Thiosymbion, Nitrosomonas, Pararheinheimera, Hyphomicrobium, GCF- 002259525, Phenylobacterium; (3) The characteristic variables preliminarily determined as the multidrug resistance gene (Multidrug) relative abundance prediction model include the water quality physical and chemical indicators: ammonia nitrogen, nitrate nitrogen, nitrite nitrogen, total nitrogen, dissolved organic carbon, chemical oxygen demand, dissolved oxygen, chlorophyll a, suspended solids, temperature, pH, redox potential, conductivity, transparency, ozone, total chlorine, quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, tetracycline antibiotics; bacterial indicators: UBA, Acinetobacter, Planktothrix, Microcystis, Planktophila, AcAMD, Unclassed, Shewanella, Aquiluna, Pontimonas, Polaribacter, Mycolicibacterium, Vulcanococcus, Fonsibacter,Limnohabitans,Thiosymbion,Flavobacterium,Pseudomonas, Hydrogenophaga, HIMB, TMED, PCC, RBG, Rhodoluna ; The characteristic variables preliminarily determined as the prediction model of the relative abundance of total antibiotic resistance genes include the water quality physical and chemical indicators: ammonia nitrogen, nitrate nitrogen, nitrite nitrogen, total nitrogen, dissolved organic carbon, chemical oxygen demand, dissolved oxygen, chlorophyll a, suspended solids, temperature, pH, redox potential, conductivity, transparency, ozone, total chlorine, quinolone antibiotics, sulfonamide antibiotics, chloramphenicol antibiotics, macrolide antibiotics, tetracycline antibiotics; bacterial indicators: Pseudomonas, Desulfobacca, RBG, Acinetobacter, Microcystis, 0-14-0-80-60-11, UTPRO2, Planococcus_B, Planktothrix, Haliscomenobacter, Idiomarina, Bacteriovorax, Kuenenia, SG8-40, LD21, Desulfosarcina, PALSA-1003, Brevefilum, Sphaerospermopsis, Desulforhopalus, Caldilinea, Thiosymbion, Nitrosomonas, Pararheinheimera, Hyphomicrobium, GCF- 002259525, Phenylobacterium; (4) Preprocess the characteristic variables of the relative abundance prediction model of multidrug resistance genes and total antibiotic resistance genes determined above, replace the 0 value of the target variable with 1 / 10 of the minimum value and then perform a unified logarithmic processing; The relative abundance of multidrug resistance genes (Multidrug) and total antibiotic resistance genes after pretreatment and logarithmization was used as the target variable. The characteristic variables of each model mentioned in (3) were used to build an automatic machine learning model to determine the best model algorithm, which was gradient boosting regression (GBM), as the prediction model for the relative abundance of multidrug resistance genes (Multidrug) and total antibiotic resistance genes. (5) Finely optimize the above-established multidrug resistance gene (Multidrug) and total antibiotic resistance gene relative abundance prediction models to enable them to have good generalization ability and prediction accuracy; Based on the importance ranking of the 45 characteristic variables output by the multidrug resistance gene relative abundance prediction model, the characteristic variables with an importance of 0 were initially filtered out, and the remaining top 11 characteristic variables included Planktothrix,Microcystis,Acinetobacter,Thiosymbion 、NO2 - -N, RBG, UBA, AcAMD 、O3、COD、 Planktophila ,TN, Tetracycline ,SS, TMED To predict the relative abundance of multidrug resistance genes (Multidrug), we changed the important input variables and relied on the computational model R 2 The model accuracy was evaluated by using the MSE and the optimal prediction model for the relative abundance of multidrug resistance genes (Multidrug) was determined to be based on the top 6 characteristic variables including Planktothrix, Microcystis,Acinetobacter,Thiosymbion 、COD、 Pseudomonas Prediction model as input variables; Based on the importance ranking of the 48 characteristic variables output by the total antibiotic resistance gene relative abundance prediction model, the characteristic variables with importance of 0 were initially filtered out, and the remaining top 21 characteristic variables included Bacteriovorax, Acinetobacter, PALSA-1003, 0-14-0-80-60-11, NO2 - -N ,Pseudomonas,Microcystis, Nitrosomonas, Kuenenia, O3, Planktothrix, Haliscomenobacter, UTPRO2, Thiosymbion ,TN, Phenylobacterium 、NO3 - -N , Brevefilum , COD, pH, ORP to predict the relative abundance of total antibiotic resistance genes, changing the important input variables, and relying on the calculation model R 2 The model accuracy was evaluated by using the MSE and the optimal prediction model for the relative abundance of total antibiotic resistance genes was determined to be the top 10 characteristic variables including Bacteriovorax, Acinetobacter, PALSA-1003, 0-14-0-80-60-11 、NO2 - -N, Pseudomonas, Microcystis, Nitrosomonas, Kuenenia, Prediction model with O3 as input variable; (6) Predicting the relative abundance of antibiotic resistance genes in the surface water sample to be tested using the optimized prediction model for relative abundance of antibiotic resistance genes.

2. The method of claim 1 for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators, characterized in that In step (4), the target variable is logarithmically transformed when establishing a regression model based on the GBM algorithm; the regression model is established using the h2o package in the R software.

3. The method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators according to claim 1, characterized in that In step (1), the relative abundance of antibiotic resistance genes in surface water samples is obtained by performing metagenomic sequencing on the samples.

4. The method for predicting the relative abundance of antibiotic resistance genes based on water quality physical and chemical indicators and bacterial indicators according to claim 1, characterized in that Methods for obtaining the relative abundance of antibiotic resistance genes in surface water samples include: Open reading frames (ORFs) were predicted for all contigs in the non-redundant MAGs set, and ORFs with a length of 100 bp or longer were retrieved and translated into amino acid sequences using the NCBI translation table. Diamond v 0.9.19 (with an e-value cutoff of 10 −5 ) The ORFs were matched to the nucleotide sequences of the Structured ARG (SARG) database.

5. The method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physical and chemical indicators and bacterial indicators according to claim 1, characterized in that It also includes randomly dividing the data set of the surface water samples into a training set and a test set, using the training set to establish the prediction model, and using the test set to verify the prediction ability of the prediction model. Preferably, 75% of the data set is used as the training set and 25% of the data set is used as the test set.

6. The method for predicting the relative abundance of antibiotic resistance genes based on water quality physical and chemical indicators and bacterial indicators according to claim 1, characterized in that The fitting coefficient R 2 Measure the predictive power of the forecast model: When R 2 When ≤0.3, the fitting effect between the predicted value and the observed value is weak, and the prediction ability of the prediction model is poor; When 0.3<R 2 When ≤0.4, the fitting effect between the predicted value and the observed value is weak, and the prediction ability of the prediction model is weak; When 0.4<R 2 When ≤0.6, the fitting effect between the predicted value and the observed value is moderate, and the prediction ability of the prediction model is moderate; When 0.6<R 2 When ≤1.0, the predicted value and the observed value have a strong fitting effect, and the prediction model has a strong prediction ability.

7. Application of the method for predicting the relative abundance of antibiotic resistance genes in surface water based on water quality physicochemical indicators and bacterial indicators as described in claim 1 to accurately predict the relative abundance of antibiotic resistance genes (ARGs) in surface water environments; the relative abundance refers to: the ratio of the number of antibiotic resistance genes (ARGs) in a sample to the total amount of all genes in the sample, and this ratio is used to estimate the abundance of the gene in the sample; the higher the relative abundance of ARGs, the greater their potential environmental hazard; predicting the relative abundance of ARGs in surface water environments is intended to monitor the spread and changing patterns of ARGs in surface water environments.

Images