Drinking water source resistance gene relative risk value space-time dimension calibration method

Abstract

The invention discloses a drinking water source resistance gene relative risk value space-time dimension calibration method. The method comprises the steps that resistance gene relative risk value data and water taking time data, water taking point location data and influence variable data corresponding to the resistance gene relative risk value data are acquired; constructing a spatial adjacency matrix to obtain a spatial information file; constructing a resistance gene relative risk value calibration model containing space and time random effects; adding a spatial information file, and performing Bayesian inference on the resistance gene relative risk value calibration model by using an integrated nested Laplacian approximation algorithm; according to a deviation information amount criterion, determining a final resistance gene relative risk value calibration model; and using the final resistance gene relative risk value calibration model to calibrate the target resistance gene relative risk value data. The calibration accuracy of the relative risk value of the antibiotic resistance gene of the drinking water source can be improved.

Description

Technical Field

[0001] The invention relates to the field of calibrating the relative risk of antibiotic resistance genes in drinking water source water bodies, and relates to a method for calibrating the time-space dimension of the relative risk value of resistance genes in drinking water sources. Background Art

[0002] With the rapid development of urbanization, human production activities have brought serious pollution to drinking water sources (rivers, reservoirs and lakes). In particular, the abuse of antibiotics in production and life has led to the induction of antibiotic resistance genes (ARGs) in some bacteria in open water sources, which in turn have resistance to the corresponding antibiotics, thus threatening human health. ARGs in water bodies bring potential risks to the ecological environment and human health that cannot be ignored. The multidimensional uncertainty in the ARGs-related water system (the uncertainty of gene abundance measurement caused by the uncertainty of DNA extraction in water sample processing, the uncertainty of the dynamic changes of various driving factors over time, the uncertainty of the influence between driving factors and ARGs, etc.) makes the relative risk value of ARGs evaluated in the system have uncertainty in the spatiotemporal dimension. In addition, most of the ARGs show an indirect relationship with environmental variables. Although the correlation is statistically significant, the correlation coefficient between ARGs and driving factors is small. The characteristics of multidimensional uncertainty and small correlation of multivariate variables bring great challenges to the calibration of the relative risk value of ARGs in the spatiotemporal dimension of drinking water sources. Therefore, an effective method is urgently needed to calibrate the relative risk values ​​of resistance genes in drinking water sources in the temporal and spatial dimensions. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for calibrating the spatiotemporal dimensions of the relative risk value of resistance genes in drinking water sources, which solves the problem of inaccurate measurement of antibiotic resistance genes in drinking water source water bodies in the prior art.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for calibrating the spatiotemporal dimension of the relative risk value of resistance genes in drinking water sources, comprising the following steps:

[0005] S1. Obtaining the relative risk value data of the resistance gene and the water collection time data, water collection point data and influencing variable data corresponding to the relative risk value data of the resistance gene, and preprocessing the relative risk value data of the resistance gene, the water collection time data, the water collection point data and the influencing variable data;

[0006] S2. Construct a spatial adjacency matrix based on the preprocessed water intake point data to obtain a spatial information file;

[0007] S3. Based on Poisson distribution, a relative risk value calibration model of resistance genes including spatial and temporal random effects was constructed;

[0008] S4. Based on the pre-processed resistance gene relative risk value data, water intake time data, water intake point data, influencing variable data and spatial information files, the integrated nested Laplace approximation algorithm is used to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain the trained resistance gene relative risk value calibration model;

[0009] S5. Determine the final resistance gene relative risk value calibration model according to the deviation information criterion;

[0010] S6. Use the final resistance gene relative risk value calibration model to calibrate the target resistance gene relative risk value data, and visualize the calibrated target resistance gene relative risk value data.

[0011] The beneficial effects of the above scheme are:

[0012] (1) The present invention uses the Bayesian spatiotemporal Poisson regression ARGs risk value calibration model, combined with easily measurable land use types and multidimensional influencing variables such as common types of bacteria, to significantly improve the spatial and temporal resolution of the calibration results, and well explain the spatiotemporal uncertainty of the ARGs relative risk value. This enables water source management departments to obtain more accurate ARGs relative risk values ​​and take corresponding control measures in a timely manner, thereby better protecting public health.

[0013] (2) The present invention fully considers the selection of model influencing variable combinations and spatiotemporal interactions, and proposes a comprehensive model selection method based on multi-stage screening and multidimensional weighted indicators. This method is different from the traditional single indicator selection. For the specific Bayesian spatiotemporal Poisson regression model selection scenario, it can not only improve the efficiency of model selection, but also avoid missing the best model under a single indicator, and enhance the reliability of the ARGs relative risk value calibration results. This method is not only applicable to the above-mentioned model selection, but can also be extended to similar model selection processes, and has a high reference value.

[0014] (3) By constructing a spatial adjacency matrix and using the INLA algorithm, the present invention can efficiently process large-scale spatiotemporal data, thereby improving the calculation speed and applicability of the model. This method is not only applicable to the above-mentioned ARGs relative risk value calibration, but can also be extended to other environmental pollution risk management with spatiotemporal uncertainty, and has broad application prospects.

[0015] Furthermore, step S1 specifically includes:

[0016] S11, obtaining resistance gene relative risk value data, water collection time data, water collection point data and influencing variable data from a resistance gene relative risk value database;

[0017] S12, converting the resistance gene relative risk value data into an integer form to obtain the pre-processed resistance gene relative risk value data;

[0018] S13, converting the water intake time data and the water intake point data into numerical data to obtain the pre-processed water intake time data and the pre-processed water intake point data;

[0019] S14. Use the K-nearest neighbor algorithm to perform weighted averaging on the influencing variable data to obtain preprocessed influencing variable data.

[0020] Further, step S2 specifically includes:

[0021] S21, marking the pre-processed water intake point data in the water intake area map, and obtaining the adjacent relationship between the water bodies corresponding to the pre-processed water intake point data according to the relevant positions of the pre-processed water intake point data in the water intake area map;

[0022] S22, constructing a spatial adjacency matrix according to the adjacent relationship;

[0023] S23. According to the spatial adjacency matrix, an ASCII file corresponding to the spatial adjacency matrix is ​​created to obtain a spatial information file.

[0024] The above further scheme has the following beneficial effects: it helps to reduce noise and errors in the modeling process, ensures the quality and accuracy of the basic data of the model, and facilitates the subsequent model construction and model inference. It captures the spatial correlation between different water bodies and effectively enhances the model's ability to grasp the spatial characteristics of drinking water source water bodies.

[0025] Further, step S3 specifically includes:

[0026] S31. Use the intrinsic conditional autoregressive specification to model the spatial structured effect. The model of the spatial structured effect is:

[0027]

[0028] Among them, u i represents the spatial structured effect, u -i represents the set of spatial structured effects except region i, n represents the number of regions, μ i represents the mean of region i, μ j represents the mean of region j, N i represents the number of neighbors in region i, is the variance of the same region, is the variance parameter, a ij If area i and area j are neighbors, a ij The value of is assigned 1, otherwise it is assigned 0;

[0029] S32. An independent and identically distributed Gaussian prior model is used to model the spatial unstructured effect. The model of the spatial unstructured effect is:

[0030]

[0031] Among them, v i represents the spatial unstructured effect, represents the variance parameter;

[0032] S33. Use first-order random walk and second-order random walk to model the time structured effect respectively. The first-order random walk model of the time structured effect is:

[0033]

[0034] Among them, γ j represents the time structured effect, γ j-1 represents the temporal structuring effect of the previous moment, represents the variance of the first-order random walk process;

[0035] The second-order random walk model of the time-structured effect is:

[0036]

[0037] Among them, γ j represents the temporal structured effect, γ j-1 represents the temporal structure effect of the previous moment, γ j-2 represents the time structured effect of the first two moments, represents the variance of the second-order random walk process;

[0038] S34. An independent and identically distributed Gaussian prior model is used to model the temporal unstructured effect. The model of the temporal unstructured effect is:

[0039]

[0040] in, represents the time unstructured effect, represents the variance parameter;

[0041] S35. Define the spatiotemporal interaction based on the different interactions of spatial structured effect, spatial unstructured effect, temporal structured effect and temporal unstructured effect, and based on the first-order random walk model of temporal structured effect and the second-order random walk model of temporal structured effect;

[0042] Among them, there are 6 types of space-time interactions;

[0043] S36. According to the model of spatial structured effect, the model of spatial unstructured effect, the first-order random walk model of temporal structured effect, the first-order random walk model of temporal structured effect, the model of temporal unstructured effect and the spatial-temporal interaction, and based on Poisson distribution, a resistance gene relative risk value calibration model is established. The resistance gene relative risk value calibration model is:

[0044] Y ij =[γy ij ]~Poisson(λ ij )

[0045]

[0046] Among them, Y ij represents the relative risk value of the resistance gene in region i at time j after scaling, y ij represents the relative risk value of the resistance gene corresponding to region i at time j, γ represents the coefficient used to scale the relative risk value, and λ ij represents the Poisson distribution parameter, b0 represents the bias term of the regression model, u i represents the spatial structuring effect, v i represents the spatial unstructured effect, γ j represents the time structuring effect, represents the time unstructured effect, δ ij Represents space-time interactions.

[0047] The beneficial effects of the above further scheme are: it can more comprehensively consider the impact of time and space effects on the relative risk of ARGs, further consider the uncertainty in the time and space dimensions of the relative risk of ARGs, and improve the accuracy and applicability of the model.

[0048] Further, in step S4, the integrated nested Laplace approximation algorithm is used to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain the trained resistance gene relative risk value calibration model, which specifically includes:

[0049] S41. Combining the model of spatial structured effect with the model of spatial unstructured effect to construct the Bessager-York-Molier model;

[0050] S42. Based on the Bessage-York-Molier model, the second-order random walk model of time-structured effects and the model of time-unstructured effects, the integrated nested Laplace approximation algorithm was used to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain the trained resistance gene relative risk value calibration model.

[0051] Furthermore, after step S4, the method further includes:

[0052] The model of spatial structured effects, the model of spatial unstructured effects, the first-order random walk model of temporal structured effects, the first-order random walk model of temporal structured effects, the model of temporal unstructured effects and space-time interaction are inferred. Based on the space-time interaction, the deviation information amount of the trained resistance gene relative risk value calibration model corresponding to the space-time interaction is obtained respectively.

[0053] The beneficial effects of the above further scheme are: when dealing with complex models, facing computational difficulties, the integrated nested Laplace approximation inference is used to efficiently estimate the regression model parameters, which facilitates the subsequent optimal model; six types of spatiotemporal interactions are further taken into account, and their modeling inference facilitates the subsequent model selection, making the final model more reliable and robust.

[0054] Furthermore, step S5 specifically includes:

[0055] S51. Use Pearson correlation analysis to perform correlation analysis on the influencing variable data, and screen the results of the correlation analysis according to a preset correlation threshold to obtain an initial correlation combination;

[0056] S52, respectively adding the initial correlation combination to the model of spatial structured effect, the model of spatial unstructured effect, the first-order random walk model of temporal structured effect, the first-order random walk model of temporal structured effect, the model of temporal unstructured effect and the space-time interaction, to obtain a set of calibration models of relative risk values ​​of candidate resistance genes;

[0057] S53, using the deviation information criterion to screen the candidate resistance gene relative risk value calibration model set;

[0058] S54, using the first weighted score to perform a second screening on the screened candidate resistance gene relative risk value calibration model set;

[0059] S55, using the second weighted score to perform sensitivity analysis on the relative risk value calibration model set of candidate resistance genes after the second screening, and perform a third screening;

[0060] S56, performing a fourth screening on the candidate resistance gene relative risk value calibration model set after the third screening according to a preset screening rule, and determining a final resistance gene relative risk value calibration model set;

[0061] S57. Determine the final resistance gene relative risk value calibration model based on the final resistance gene relative risk value calibration model set.

[0062] Furthermore, in step S57, the final resistance gene relative risk value calibration model is:

[0063] Y ij=[γy ij ]~Poisson(λ ij )

[0064]

[0065] Among them, Y ij represents the relative risk value of the resistance gene in region i at time j after scaling, y ij represents the relative risk value of the resistance gene corresponding to region i at time j, γ represents the coefficient used to scale the relative risk value, and λ ij represents the Poisson distribution parameter, b0 represents the bias term of the regression model, u i represents the spatial structuring effect, v i represents the spatial unstructured effect, γ j represents the time structuring effect, represents the time unstructured effect, δ ij represents the time-space interaction, variable 1 represents the first influencing variable in the selected influencing variable combination, and variable 2 represents the second influencing variable in the selected influencing variable combination.

[0066] The beneficial effect of the above further scheme is that it takes into account the impact of different combinations of influencing variables and spatiotemporal interactions on the model, ensures that the final model can best explain multidimensional uncertainties, and greatly improves the reliability of calibrated ARGs relative risk values.

[0067] Further, step S6 specifically includes:

[0068] S61. Calibrate the target resistance gene relative risk value data using the final resistance gene relative risk value calibration model to obtain a model operation result;

[0069] S62, extracting the posterior mean in the model running result as the relative risk calibration value of the relative risk value data of the target resistance gene;

[0070] S63. Extract the mean of the time random effect in the model running results, draw a time series graph, and visualize the calibrated relative risk value data of the target resistance gene.

[0071] Further, after step S62, the method further includes:

[0072] The relative risk calibration value was compared with the target resistance gene relative risk value data to determine the calibration accuracy of the final resistance gene relative risk value calibration model.

[0073] The beneficial effects of the above further scheme are: making the model results more interpretable and practical, and the visualization results provide decision makers with an intuitive environmental risk management tool. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 Schematic diagram of a method for calibrating the spatiotemporal dimensions of relative risk values ​​of resistance genes in drinking water sources. DETAILED DESCRIPTION

[0075] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0076] like Figure 1 As shown, a method for calibrating the spatiotemporal dimension of the relative risk value of resistance genes in drinking water sources comprises the following steps:

[0077] S1. Obtain the relative risk value data of resistance genes and the water collection time data, water collection point data and influencing variable data corresponding to the relative risk value data of resistance genes, and pre-process the relative risk value data of resistance genes, water collection time data, water collection point data and influencing variable data.

[0078] In this embodiment, step S1 specifically includes:

[0079] S11, obtaining resistance gene relative risk value data, water collection time data, water collection point data and influencing variable data from a resistance gene relative risk value database;

[0080] S12, converting the resistance gene relative risk value data into an integer form to obtain the pre-processed resistance gene relative risk value data;

[0081] S13, converting the water intake time data and the water intake point data into numerical data to obtain the pre-processed water intake time data and the pre-processed water intake point data;

[0082] S14. Use the K-nearest neighbor algorithm to perform weighted averaging on the influencing variable data to obtain preprocessed influencing variable data.

[0083] Exemplarily, the resistance gene relative risk value data is converted into an integer form by expanding the ARGs relative risk value in the drinking water source water body by 10n times, and the minimum value of n is determined by the fact that the expanded ARGs relative risk value is an integer.

[0084] Optionally, the influencing variable data may also include alternative influencing variable data, which may include variables related to land use in the water intake area and variables related to potential bacteria affecting water bodies. Variables related to land use in the water intake area may include the proportion of residential areas, urban areas, green areas, and agricultural areas in the drinking water source area. Different regional proportions imply the spatial information of the water intake area and the spatial correlation between water intake areas, which can explain the uncertainty in the relative risk of ARGs to a certain extent. Variables related to potential bacteria affecting water bodies may include the content of Escherichia coli, enterococci, and Pseudomonas aeruginosa in the drinking water source water body. For this type of bacterial infection, more potent antibiotics or a combination of multiple antibiotics can be selected for treatment. Therefore, it can be considered that their content in the water body can further correct the uncertainty of the relative risk of ARGs in the water body.

[0085] S2. Based on the preprocessed water intake point data, a spatial adjacency matrix is ​​constructed to obtain a spatial information file.

[0086] In this embodiment, step S2 specifically includes:

[0087] S21, marking the pre-processed water intake point data in the water intake area map, and obtaining the adjacent relationship between the water bodies corresponding to the pre-processed water intake point data according to the relevant positions of the pre-processed water intake point data in the water intake area map;

[0088] S22, constructing a spatial adjacency matrix according to the adjacent relationship;

[0089] S23. According to the spatial adjacency matrix, an ASCII file corresponding to the spatial adjacency matrix is ​​created to obtain a spatial information file.

[0090] Exemplarily, the preprocessed water intake point data are marked in the water intake area map. For the water bodies involved (catchment areas / tributaries), the adjacency relationship between each pair of water bodies is obtained according to their distribution location and coverage area. For adjacent water bodies, a value of 1 is assigned; for non-adjacent water bodies, a value of 0 is assigned, so that a spatial adjacency matrix can be constructed. The number of rows in the ASCII file is equal to the number of regions n plus 1. The first row is n, and each of the next n rows specifies: (i) the region label indexed 1, 2, ..., n, (ii) the number of its neighbors, and (iii) the labels of the neighboring regions.

[0091] S3. Based on Poisson distribution, a relative risk value calibration model of resistance genes was constructed, which included spatial and temporal random effects.

[0092] In this embodiment, step S3 specifically includes:

[0093] S31. Use the intrinsic conditional autoregressive specification to model the spatial structured effect. The model of the spatial structured effect is:

[0094]

[0095] Among them, u i represents the spatial structured effect, u -i represents the set of spatial structured effects except region i, n represents the number of regions, μ i represents the mean of region i, μ j represents the mean of region j, N i represents the number of neighbors in region i, is the variance of the same region, is the variance parameter, a ij If area i and area j are neighbors, a ij The value of is assigned 1, otherwise it is assigned 0;

[0096] S32. An independent and identically distributed Gaussian prior model is used to model the spatial unstructured effect. The model of the spatial unstructured effect is:

[0097]

[0098] Among them, v i represents the spatial unstructured effect, represents the variance parameter;

[0099] S33. Use first-order random walk and second-order random walk to model the time structured effect respectively. The first-order random walk model of the time structured effect is:

[0100]

[0101] Among them, γ j represents the temporal structured effect, γ j-1 represents the temporal structuring effect of the previous moment, represents the variance of the first-order random walk process;

[0102] The second-order random walk model of the time-structured effect is:

[0103]

[0104] Among them, γ j represents the temporal structured effect, γ j-1 represents the temporal structure effect of the previous moment, γ j-2 represents the time structured effect of the first two moments, represents the variance of the second-order random walk process;

[0105] S34. An independent and identically distributed Gaussian prior model is used to model the temporal unstructured effect. The model of the temporal unstructured effect is:

[0106]

[0107] in, represents the time unstructured effect, represents the variance parameter;

[0108] S35. Define the spatiotemporal interaction based on the different interactions of spatial structured effect, spatial unstructured effect, temporal structured effect and temporal unstructured effect, and based on the first-order random walk model of temporal structured effect and the second-order random walk model of temporal structured effect;

[0109] Among them, there are 6 types of space-time interactions;

[0110] S36. According to the model of spatial structured effect, the model of spatial unstructured effect, the first-order random walk model of temporal structured effect, the first-order random walk model of temporal structured effect, the model of temporal unstructured effect and the spatial-temporal interaction, and based on Poisson distribution, a resistance gene relative risk value calibration model is established. The resistance gene relative risk value calibration model is:

[0111] Y ij =[γy ij ]~Poisson(λ ij )

[0112]

[0113] Among them, Y ij represents the relative risk value of the resistance gene in region i at time j after scaling, y ij represents the relative risk value of the resistance gene corresponding to region i at time j, γ represents the coefficient used to scale the relative risk value, and λ ij represents the Poisson distribution parameter, b0 represents the bias term of the regression model, u i represents the spatial structuring effect, v i represents the spatial unstructured effect, γ j represents the time structuring effect, represents the time unstructured effect, δ ij Represents space-time interactions.

[0114] Exemplarily, when defining a spatiotemporal interaction, the spatiotemporal interaction can be decomposed into the Kronecker product of the spatiotemporal interaction of the corresponding main effects of the interaction. And according to the spatial effect u i 、v i and time effect γ j , There are four ways to define the spatial and temporal interactions, the temporal structured effect γ j By using first-order or second-order random walks in the interaction process, we can eventually obtain six types of spatiotemporal interactions.

[0115] S4. Based on the preprocessed resistance gene relative risk value data, water collection time data, water collection point data, influencing variable data and spatial information files, the integrated nested Laplace approximation algorithm is used to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain the trained resistance gene relative risk value calibration model.

[0116] In this embodiment, in step S4, the integrated nested Laplace approximation algorithm is used to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain the trained resistance gene relative risk value calibration model, which specifically includes:

[0117] S41. Combining the model of spatial structured effect with the model of spatial unstructured effect to construct the Bessager-York-Molier model;

[0118] S42. Based on the Bessage-York-Molier model, the second-order random walk model of time-structured effects and the model of time-unstructured effects, the integrated nested Laplace approximation algorithm was used to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain the trained resistance gene relative risk value calibration model.

[0119] Optionally, after step S4, the method further comprises:

[0120] The model of spatial structured effects, the model of spatial unstructured effects, the first-order random walk model of temporal structured effects, the first-order random walk model of temporal structured effects, the model of temporal unstructured effects and space-time interaction are inferred. Based on the space-time interaction, the deviation information amount of the trained resistance gene relative risk value calibration model corresponding to the space-time interaction is obtained respectively.

[0121] S5. Determine the final resistance gene relative risk value calibration model based on the deviation information criterion.

[0122] In this embodiment, step S5 specifically includes:

[0123] S51. Use Pearson correlation analysis to perform correlation analysis on the influencing variable data, and screen the results of the correlation analysis according to a preset correlation threshold to obtain an initial correlation combination;

[0124] S52, respectively adding the initial correlation combination to the model of spatial structured effect, the model of spatial unstructured effect, the first-order random walk model of temporal structured effect, the first-order random walk model of temporal structured effect, the model of temporal unstructured effect and the space-time interaction, to obtain a set of calibration models of relative risk values ​​of candidate resistance genes;

[0125] S53, using the deviation information criterion to screen the candidate resistance gene relative risk value calibration model set;

[0126] S54, using the first weighted score to perform a second screening on the screened candidate resistance gene relative risk value calibration model set;

[0127] S55, using the second weighted score to perform sensitivity analysis on the relative risk value calibration model set of candidate resistance genes after the second screening, and perform a third screening;

[0128] S56, performing a fourth screening on the candidate resistance gene relative risk value calibration model set after the third screening according to a preset screening rule, and determining a final resistance gene relative risk value calibration model set;

[0129] S57. Determine the final resistance gene relative risk value calibration model based on the final resistance gene relative risk value calibration model set.

[0130] In step S57 of this embodiment, the final resistance gene relative risk value calibration model is:

[0131] Y ij =[γy ij ]~Poisson(λ ij )

[0132]

[0133] Among them, Y ij represents the relative risk value of the resistance gene in region i at time j after scaling, y ij represents the relative risk value of the resistance gene corresponding to region i at time j, γ represents the coefficient used to scale the relative risk value, and λ ij represents the Poisson distribution parameter, b0 represents the bias term of the regression model, u i represents the spatial structuring effect, v i represents the spatial unstructured effect, γ j represents the time structuring effect, represents the time unstructured effect, δ ij represents the time-space interaction, variable 1 represents the first influencing variable in the selected influencing variable combination, and variable 2 represents the second influencing variable in the selected influencing variable combination.

[0134] Exemplarily, the influencing variable data may include 7 influencing variables, and the preset correlation threshold may be 0.4. Pearson correlation analysis is used to check the correlation of the 7 influencing variables in pairs. Strong correlation may lead to multicollinearity problems and affect the calibration ability of the model, so the combination of two variables with a correlation coefficient greater than 0.4 is not selected. In the total variable combination, the combination with the correlation coefficient that does not meet the requirements is screened out, and the remaining m1 combinations are the initial correlation combinations.

[0135] The remaining m1 available influencing variable combinations were added to the model (each combination combined with the 6 interactions would produce 6 different models), and the Deviation Information Criterion (DIC) was used for model screening: for the 6 models produced by each variable combination, the one with the smallest DIC value was selected. The models with DIC values ​​greater than the original model were screened out, and the m2 models with smaller DIC values ​​(ranked in the top 50%) were selected from the remaining models.

[0136] The remaining m2 models are selected using the first weighted scoring to select the m3 models with the top 1 / 3 scores: the models can be screened for the second time based on the five indicators of the model's DIC value, DIC saturation value, number of effective model parameters, marginal log-likelihood value, and total model time consumption. The weight allocation and scoring rules can be as follows:

[0137] DIC value (ω DIC =0.5), total model time (ω Time =0.15), the number of effective model parameters (ω p.eff =0.1) from low to high scores, the marginal log-likelihood value (ω likelihood =0.15) from high to low, the scores are m2, m2-1, ..., 1; DIC saturation value: DIC value / DIC saturation value greater than 0.5 is scored Otherwise, it is 1. The model score is pk is the score of the kth indicator.

[0138] The second weighted score was used to conduct a sensitivity analysis on the remaining m3 models for a third screening. The scoring rules can be the same as above, and the indicator weight change rules can be: (i) the DIC weight range is [0.2, 0.5], and the DIC saturation value weight is fixed at 0.2; (ii) the other weight ranges are [0.1, 0.4]; (iii) all weight definitions and changes are based on 0.1 as 1 unit.

[0139] Exemplarily, the preset screening rule may be that the score ranking is in the top 50% on average.

[0140] For example, m4 models with good stability to weight changes (average score ranking in the top 50%) were selected. Finally, considering the degree of DIC improvement and the introduction of more variables and types, the final model was selected from the m4 models, and the final resistance gene relative risk value calibration model was further determined.

[0141] For example, in one embodiment of the present invention, the ARGs relative risk values ​​of eight adjacent natural watersheds after evaluation were extracted, the real data of the actual drinking water environment was obtained, and the ARGs relative risk values ​​were accurately calibrated. The final resistance gene relative risk value calibration model is:

[0142] Y ij =[γy ij ]~Poisson(λ ij )

[0143]

[0144] S6. Use the final resistance gene relative risk value calibration model to calibrate the target resistance gene relative risk value data, and visualize the calibrated target resistance gene relative risk value data.

[0145] In this embodiment, step S6 specifically includes:

[0146] S61. Calibrate the target resistance gene relative risk value data using the final resistance gene relative risk value calibration model to obtain a model operation result;

[0147] S62, extracting the posterior mean in the model running result as the relative risk calibration value of the relative risk value data of the target resistance gene;

[0148] S63. Extract the mean of the time random effect in the model running results, draw a time series graph, and visualize the calibrated relative risk value data of the target resistance gene.

[0149] Optionally, after step S62, the method further includes:

[0150] The relative risk calibration value was compared with the target resistance gene relative risk value data to determine the calibration accuracy of the final resistance gene relative risk value calibration model.

[0151] For example, the mean of the time random effect in the extraction model running results can be μ j , draw a time series graph, the horizontal axis can be the time index 1,2,...,T, and the vertical axis can be the risk estimate R after adjustment for time j j and the actual risk value E j :

[0152] R j=exp(μ j )×S j

[0153]

[0154] Among them, S j Represents the calculation result of the exponential expectation value of the random effects in the model; RR raw.ji Represents the relative risk value of the original ARGs in region i at time j.

[0155] The present invention first extracts the ARGs relative risk value of the evaluated drinking water source water body, pre-processes the ARGs relative risk value, water collection time, water collection point information and influencing variable data, then constructs an adjacency matrix according to the adjacent relationship of each water body, further obtains a spatial information file, and then establishes a Bayesian spatiotemporal Poisson regression ARGs risk value calibration model containing spatial and temporal random effects, then adds the spatial information file, uses the INLA method to perform Bayesian inference, further selects the influencing variable combination and spatiotemporal interaction according to the DIC criterion, etc., to obtain the final model, and finally runs the final model to calibrate the ARGs spatiotemporal dimension relative risk value under uncertainty conditions, and visualizes it, so as to achieve accurate calibration of the relative risk value of antibiotic resistance genes in drinking water sources.

[0156] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the invention.

Claims

1. A method for calibrating the spatiotemporal dimension of the relative risk value of resistance genes in drinking water sources, characterized in that: The method comprises: S1. Obtaining the relative risk value data of the resistance gene and the water collection time data, water collection point data and influencing variable data corresponding to the relative risk value data of the resistance gene, and preprocessing the relative risk value data of the resistance gene, the water collection time data, the water collection point data and the influencing variable data; S2. Construct a spatial adjacency matrix based on the preprocessed water intake point data to obtain a spatial information file; S3. Based on Poisson distribution, a relative risk value calibration model of resistance genes including spatial and temporal random effects was constructed; S4, based on the pre-treated resistance gene relative risk value data, water intake time data, water intake point data, influencing variable data and the spatial information file, using an integrated nested Laplace approximation algorithm to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain a trained resistance gene relative risk value calibration model; S5. Determine the final resistance gene relative risk value calibration model according to the deviation information criterion; S6. Use the final resistance gene relative risk value calibration model to calibrate the target resistance gene relative risk value data, and visualize the calibrated target resistance gene relative risk value data.

2. The method according to claim 1, characterized in that The step S1 specifically includes: S11, acquiring the resistance gene relative risk value data, the water collection time data, the water collection point data and the influencing variable data from a resistance gene relative risk value database; S12, converting the resistance gene relative risk value data into an integer form to obtain pre-processed resistance gene relative risk value data; S13, converting the water intake time data and the water intake point data into numerical data to obtain pre-processed water intake time data and pre-processed water intake point data; S14. Use a K-nearest neighbor algorithm to perform weighted averaging on the influencing variable data to obtain preprocessed influencing variable data.

3. The method according to claim 1, characterized in that The step S2 specifically includes: S21, marking the pre-processed water intake point data in a water intake area map, and obtaining the neighboring relationship between the water bodies corresponding to the pre-processed water intake point data according to the relevant positions of the pre-processed water intake point data in the water intake area map; S22, constructing a spatial adjacency matrix according to the adjacent relationship; S23. Create an ASCII file corresponding to the spatial adjacency matrix according to the spatial adjacency matrix to obtain the spatial information file.

4. The method according to claim 1, characterized in that: The step S3 specifically includes: S31. Modeling the spatial structured effect using the intrinsic conditional autoregressive specification, the model of the spatial structured effect is: Among them, u i represents the spatial structured effect, u -i represents the set of spatial structured effects except region i, n represents the number of regions, μ i represents the mean of region i, μ j represents the mean of region j, N i represents the number of neighbors in region i, is the variance of the same region, is the variance parameter, a ij If area i and area j are neighbors, a ij The value of is assigned 1, otherwise it is assigned 0; S32, performing independent and identically distributed Gaussian prior modeling on the spatial unstructured effect, the model of the spatial unstructured effect is: Among them, v i represents the spatial unstructured effect, represents the variance parameter; S33, using the first-order random walk and the second-order random walk to model the time structured effect respectively, the first-order random walk model of the time structured effect is: Among them, γ j represents the temporal structured effect, γ j-1 represents the temporal structuring effect of the previous moment, represents the variance of the first-order random walk process; The second-order random walk model of the time structured effect is: Among them, γ j represents the temporal structured effect, γ j-1 represents the temporal structured effect of the previous moment, γ j-2 represents the time structured effect of the first two moments, represents the variance of the second-order random walk process; S34, performing independent and identically distributed Gaussian prior modeling on the temporal unstructured effect, wherein the model of the temporal unstructured effect is: in, represents the time unstructured effect, represents the variance parameter; S35, defining a spatiotemporal interaction according to different interactions of the spatial structured effect, the spatial unstructured effect, the temporal structured effect, and the temporal unstructured effect, and based on a first-order random walk model of the temporal structured effect and a second-order random walk model of the temporal structured effect; Among them, the space-time interactions are 6 types; S36. According to the model of the spatial structured effect, the model of the spatial unstructured effect, the first-order random walk model of the temporal structured effect, the first-order random walk model of the temporal structured effect, the model of the temporal unstructured effect and the spatiotemporal interaction, and based on Poisson distribution, a relative risk value calibration model for the resistance gene is established, and the relative risk value calibration model for the resistance gene is: Y ij =[γy ij ]~Poisson(λ ij ) Among them, Y ij represents the relative risk value of the resistance gene in region i at time j after scaling, y ij represents the relative risk value of the resistance gene corresponding to region i at time j, γ represents the coefficient used to scale the relative risk value, and λ ij represents the Poisson distribution parameter, b0 represents the bias term of the regression model, u i represents the spatial structuring effect, v i represents the spatial unstructured effect, γ j represents the time structuring effect, represents the time unstructured effect, δ ij Represents space-time interactions.

5. The method according to claim 4, characterized in that In step S4, the integrated nested Laplace approximation algorithm is used to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain a trained resistance gene relative risk value calibration model, which specifically includes: S41, combining the model of the spatial structured effect and the model of the spatial unstructured effect to construct a Bessage-York-Molière model; S42. Based on the Bessage-York-Molier model, the second-order random walk model of the time structured effect and the model of the time unstructured effect, the integrated nested Laplace approximation algorithm is used to perform Bayesian inference on the resistance gene relative risk value calibration model to obtain the trained resistance gene relative risk value calibration model.

6. The method according to claim 4, characterized in that After step S4, the method further includes: The model of the spatial structured effect, the model of the spatial unstructured effect, the first-order random walk model of the temporal structured effect, the first-order random walk model of the temporal structured effect, the model of the temporal unstructured effect and the space-time interaction are inferred, and based on the space-time interaction, the deviation information amount of the trained resistance gene relative risk value calibration model corresponding to the space-time interaction is obtained respectively.

7. The method according to claim 6, characterized in that The step S5 specifically includes: S51, performing a correlation analysis on the influencing variable data using Pearson correlation analysis, and screening the results of the correlation analysis according to a preset correlation threshold to obtain an initial correlation combination; S52, respectively adding the initial correlation combination to the model of the spatial structured effect, the model of the spatial unstructured effect, the first-order random walk model of the temporal structured effect, the first-order random walk model of the temporal structured effect, the model of the temporal unstructured effect and the spatiotemporal interaction to obtain a set of relative risk value calibration models for candidate resistance genes; S53, using the deviation information criterion to screen the candidate resistance gene relative risk value calibration model set; S54, using the first weighted score to perform a second screening on the screened candidate resistance gene relative risk value calibration model set; S55, using the second weighted score to perform sensitivity analysis on the relative risk value calibration model set of candidate resistance genes after the second screening, and perform a third screening; S56, performing a fourth screening on the candidate resistance gene relative risk value calibration model set after the third screening according to a preset screening rule, and determining a final resistance gene relative risk value calibration model set; S57. Determine the final resistance gene relative risk value calibration model based on the final resistance gene relative risk value calibration model set.

8. The method according to claim 7, characterized in that In step S57, the final resistance gene relative risk value calibration model is: Y ij =[γy ij ]~Poisson(λ ij ) Among them, Y ij represents the relative risk value of the resistance gene in region i at time j after scaling, y ij represents the relative risk value of the resistance gene corresponding to region i at time j, γ represents the coefficient used to scale the relative risk value, and λ ij represents the Poisson distribution parameter, b0 represents the bias term of the regression model, u i represents the spatial structuring effect, v i represents the spatial unstructured effect, γ j represents the time structuring effect, represents the time unstructured effect, δ ij represents the time-space interaction, variable 1 represents the first influencing variable in the selected influencing variable combination, and variable 2 represents the second influencing variable in the selected influencing variable combination.

9. The method according to claim 1, characterized in that: The step S6 specifically includes: S61, using the final resistance gene relative risk value calibration model to calibrate the target resistance gene relative risk value data to obtain a model operation result; S62, extracting the posterior mean in the model operation result as the relative risk calibration value of the relative risk value data of the target resistance gene; S63. Extract the mean of the time random effect in the model operation results, draw a time series graph, and visualize the calibrated target resistance gene relative risk value data.

10. The method according to claim 9, characterized in that After step S62, the method further includes: The relative risk calibration value is compared with the target resistance gene relative risk value data to determine the calibration accuracy of the final resistance gene relative risk value calibration model.

Images