Method for predicting prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data

Abstract

The application is suitable for the technical field of medical health prediction and big data analysis, and provides a prediction method for allergic skin disease prevalence based on machine learning and multi-source remote sensing data, which comprises the following steps: acquiring multi-source remote sensing environment data and prevalence data, performing time series analysis by using segmented regression, and selecting environment factors according to the contribution degree of the environment factors evaluated by using an information entropy model; constructing multiple machine learning models to select an optimal prediction model, and quantifying the contribution degree and interaction of the environment factors in combination with a geographic detector; performing metagenomic sequencing on fecal samples to obtain the composition and functional characteristics of intestinal flora; performing spatial metabolomics analysis on skin tissue by using mass spectrometry imaging to identify differential metabolites and spatial distribution; and taking the environment factors as independent variables, the differential microorganisms as intermediate variables and the differential metabolites as dependent variables, and constructing a ternary correlation network through intermediate effect analysis.The application realizes the systematic integration from macro-environment prediction to micro-molecular mechanism.

Description

Technical Field

[0001] This invention relates to the field of medical and health prediction and big data analysis technology, specifically a method for predicting the prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data. Background Technology

[0002] Allergic skin diseases (ASD), especially atopic dermatitis (AD), have become a global public health problem. Currently, the incidence of AD in children is approximately 7%-30%, and in adults approximately 1%-10%, and it continues to rise globally, placing a heavy burden on patients' quality of life, mental health, and socioeconomic status. AD is a complex, chronic, relapsing, inflammatory skin disease whose pathogenesis involves the interaction of multiple factors, including genetic susceptibility, impaired skin physical barrier function, immune dysregulation, and environmental exposure.

[0003] In recent years, the impact of climate change on allergic diseases has received increasing attention. Environmental factors such as temperature, humidity, aerosols, and ultraviolet radiation affect the occurrence and development of Alzheimer's disease (AD) through multiple pathways. Rising global temperatures and carbon dioxide concentrations not only affect pollen concentration, the duration of the pollen season, and pollen allergenicity, but also influence the spread of allergens by altering precipitation patterns and wind conditions. However, traditional epidemiological survey methods largely rely on patient subjective reports and clinician diagnostic differences, making it difficult to dynamically and accurately describe the complex relationship between disease and environmental factors. Traditional statistical models (such as linear regression) struggle to capture the nonlinear, lag-related, and interactive characteristics of the relationship between environmental factors and disease.

[0004] In the study of AD pathological mechanisms, existing technologies have the following main shortcomings: First, in terms of the association analysis between environmental exposure and disease, existing studies mostly use traditional statistical regression models, which are difficult to handle the complex relationship between high-dimensional, nonlinear environmental variables and disease outcomes, and lack technical solutions to combine multi-source remote sensing environmental data with machine learning to predict regional prevalence and extrapolate future scenarios. Second, in terms of gut microbiota research, most existing microbiome studies rely on 16S rRNA gene amplicon sequencing technology, which has limited resolution at the species level, making it difficult to directly reveal the functional activity and metabolic potential of the microbial community, and lacks systematic analysis of the topological properties of the microbial symbiotic network and key ecological nodes. Third, in the study of skin tissue metabolism, traditional metabolomics analysis usually requires homogenization of tissue samples, which results in the loss of spatial distribution information of metabolites in different compartments of the tissue (such as the epidermis, dermis, and inflammatory cell infiltration areas). The metabolic state in AD lesions has significant spatial heterogeneity, and existing technologies are unable to reveal this compartmentalized metabolic interaction and spatial reprogramming characteristics at the in situ level of the tissue. Fourth, at the level of multi-omics integrated research, there is still a lack of systematic empirical research and quantitative analysis on how environmental factors, gut microbiota and host skin metabolism constitute a coherent pathogenic pathway; existing technologies usually study changes at a single omics level in isolation, without establishing a complete research framework that connects macro-environmental exposure with micro-molecular mechanisms.

[0005] Therefore, in view of the above situation, there is an urgent need to provide a method for predicting the prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data, so as to overcome the shortcomings in current practical applications. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting the prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data, effectively solving the problems mentioned in the background art.

[0007] This invention is implemented as follows: a method for predicting the prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data, comprising the following steps: Step S1: Obtain multi-source remote sensing environmental data and allergic skin disease prevalence data for the study area to construct an environment-disease spatiotemporal dataset; use a piecewise regression model to analyze the temporal variation characteristics of the prevalence and identify inflection points; use an information entropy model to evaluate the contribution of each environmental factor to the prediction of the prevalence, and select environmental factors accordingly. Step S2: Construct multiple machine learning regression models, and use the environmental factors selected after evaluation by the information entropy model in Step S1 to train and validate the prevalence of allergic skin diseases; optimize the model hyperparameters through cross-validation and grid search algorithms; evaluate the performance of each model based on mean squared error, mean absolute error and correlation coefficient, and select the optimal prediction model; use a geospatial detector to quantify the independent explanatory power of each environmental factor on the prevalence and the contribution of the interaction between factors. Step S3: Obtain fecal samples from patients with allergic skin diseases and healthy controls, extract sample DNA, and perform metagenomic paired-end sequencing; perform quality control, sequence assembly, and gene prediction on the sequencing data; perform species annotation and abundance quantification on the assembled sequences to obtain a microbial community composition profile; perform α-diversity analysis and β-diversity analysis on the microbial community based on this composition profile to compare differences in community structure; use linear discriminant analysis effect size method to screen for differential marker microorganisms between groups; construct a microbial symbiotic network, calculate network topological properties, and identify key nodes; align the sequencing sequences with functional databases to perform functional annotation and pathway enrichment analysis. Step S4: Obtain skin tissue samples from patients with allergic dermatitis and healthy controls, prepare frozen sections, and acquire data from the tissue sections using matrix-assisted laser desorption / ionization mass spectrometry (MALDI-MS). Perform peak detection, metabolite identification, and spatial distribution analysis on the primary mass spectrometry data. Use partial least squares discriminant analysis to differentiate between metabolographic profiles, and screen differentially expressed metabolites using variable projection importance indices and hypothesis testing. Perform KEGG pathway enrichment analysis on the screened differentially expressed metabolites to determine key metabolic pathways and their spatial distribution characteristics. Step S5: Integrate the data analysis results from steps S2 to S4, using environmental factors as independent variables, differentially expressed microorganisms screened in step S3 as mediating variables, and differentially expressed metabolites screened in step S4 as dependent variables. Quantify the direct effects of environmental factors on host metabolites and the indirect effects mediated by gut microbiota through mediation effect analysis, and calculate the proportion of mediation effects. Incorporate the mediation effect results into the association network to construct a ternary association network of environmental factors—gut microbiota—host metabolites. Evaluate the importance of each environmental factor using a random forest model. Based on the mediation effect analysis results, determine the cascade pathway by which environmental factors affect host skin metabolism through gut microbiota, and output the multi-omics association analysis results for allergic skin diseases.

[0008] As a further aspect of the present invention: in step S1, the multi-source remote sensing environmental data includes environmental element data, hydrological element data, and meteorological element data; The environmental data include aerosol optical thickness, ozone concentration, and ultraviolet index. The hydrological data include the monthly maximum runoff and the monthly minimum runoff; The meteorological data include daily maximum temperature, daily minimum temperature, daily average temperature, daily maximum precipitation, daily minimum precipitation, easterly wind speed at 10 meters height, and northerly wind speed at 10 meters height.

[0009] As a further aspect of the present invention, step S6 is also included: The optimal prediction model obtained in step S2 is combined with the multi-scenario climate prediction data provided by the CMIP6 Coupled Model Comparison Program to extract the predicted values ​​of environmental factors under different shared socio-economic pathway scenarios in the future. These values ​​are then input into the optimal prediction model to predict the trend of changes in the prevalence of allergic skin diseases under different scenarios in the future. The shared socioeconomic path scenarios include the SSP126 low-compulsion scenario, the SSP370 medium-compulsion scenario, and the SSP585 high-compulsion scenario. For environmental factors that are not directly output by the CMIP6 model, a generalized linear model is constructed to establish the mapping relationship between the factor and other output factors for estimation. In the generalized linear model, aerosol optical thickness and ozone are used as predictor variables to estimate the ultraviolet index.

[0010] As a further aspect of the present invention: in step S2, the machine learning regression model includes a decision tree regression model, an elastic network regression model, a random forest regression model, a ridge regression model, and a support vector regression model; The cross-validation is 10-fold cross-validation.

[0011] As a further aspect of the present invention: in step S2, the geographic detector includes a factor detector and an interaction detector; The factor detector passes through The values ​​quantify the independent explanatory power of each environmental factor on the morbidity rate. A higher value indicates a stronger explanatory power of the factor for the prevalence of disease. The formula for calculating the value is:

[0012] Where h = 1, 2, 3, …, represents a specific type; L represents the stratification number of the independent or dependent variable; N and h represent the number of cells in layer h and the entire region, respectively. and These are the variances of category h and the variance of the entire region, respectively. The interaction detector is used to assess the change in the explanatory power of two environmental factors acting together relative to their independent effects.

[0013] As a further aspect of the present invention: in step S3, the metagenomic paired-end sequencing is performed using the Illumina NovaSeq platform; The functional database is the KEGG database; the identification of the key nodes adopts the Zi-Pi thresholding method; The network topology properties include average path length, clustering coefficient, and modularity; the key nodes include module hub nodes and connecting nodes.

[0014] As a further aspect of the present invention: in step S4, the metabolic pathways that the KEGG pathway enrichment analysis focuses on include: Glucose metabolism pathway, tricarboxylic acid cycle pathway, ceramide synthesis pathway, arachidonic acid metabolism pathway, arginine metabolism pathway, and tryptophan metabolism pathway.

[0015] As a further aspect of the present invention: in step S1, in the piecewise regression model, it is assumed that the dataset is segmented into... For each paragraph Fitting a linear regression model ,in and They are the first The intercept and slope of the segment, It is an error term; The function values ​​of adjacent segments must be continuous at the breakpoint; The optimal breakpoint location and number of segments are determined using either the Akaike information criterion or the Bayesian information criterion.

[0016] As a further aspect of the present invention: in step S3, the results of the functional annotation and pathway enrichment analysis are used to identify the functional remodeling characteristics of the gut microbiota in patients with allergic dermatitis, the functional remodeling characteristics including: It leads to a decline in basal nutritional metabolic capacity, as well as an enhancement of environmental stress response pathways, substance transport pathways, and host inflammatory response-related pathways.

[0017] As a further aspect of the present invention: in step S4, the spatial distribution characteristics include the differential enrichment or absence of each metabolite in the epidermis, dermis and inflammatory infiltration areas of the skin.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention is the first to integrate multi-source remote sensing environmental data, machine learning prediction models, metagenomics, and spatial metabolomics into a unified analytical framework, realizing a complete research system from macro-environmental prediction to micro-tissue metabolic mechanisms, and from single omics to systems biology, filling the gap in the current research field of allergic skin diseases that lacks multi-omics system integration methods.

[0019] In terms of disease incidence prediction, this invention overcomes the technical bottleneck of traditional statistical models in handling the nonlinear relationship between environment and disease by integrating multi-source remote sensing data (12 factors including environmental, hydrological, and meteorological elements) with five machine learning models. The geographic detector further quantifies the independent contributions and interactive enhancement effects of each factor, significantly improving the interpretability of the prediction model. Combined with CMIP6 multi-scenario prediction, this invention enables quantitative extrapolation of disease burden under different future emission pathways.

[0020] In the field of gut microbiota analysis, this invention uses metagenomic sequencing technology to replace traditional 16S rRNA amplicon sequencing, achieving a resolution leap from the genus level to the species level of microbial communities. For the first time, neutral community models, symbiotic network topology analysis, and Zi-Pi key node identification are introduced into the study of gut microbiota in allergic dermatitis, systematically revealing the functional remodeling characteristics of AD microbiota from a steady-state fermentation type to an inflammatory adaptive type.

[0021] In the field of skin metabolism analysis, this invention is the first to apply matrix-assisted laser desorption / ionization mass spectrometry imaging technology to the spatial metabolomics study of skin lesions in allergic dermatitis. It overcomes the inherent defect of traditional homogenized metabolomics in losing spatial information and realizes in-situ visualization of the heterogeneous partitions of energy metabolism, lipid metabolism and amino acid metabolism in AD skin lesions.

[0022] In terms of multi-omics integration, this invention proposes a differential trend association analysis method, effectively solving the technical challenge of integrating and analyzing unpaired multi-omics data. More importantly, this invention, for the first time, quantitatively reveals the cascade pathway of environmental factors → gut microbiota → host metabolites through mediation effect analysis, verifying the core pivotal role of gut microbiota in the relationship between environmental exposure and host skin metabolism (mediation effect accounting for over 80%). This provides crucial empirical evidence for the gut-skin axis theory and offers a new multi-level strategy for the precise prevention and treatment of Alzheimer's disease, ranging from macro-environmental regulation to micro-microbiota intervention. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is an overview map of the study area in an embodiment of the present invention; Figure 2 This is a technical roadmap for the method of predicting the prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data, as described in this invention. Figure 3 This is a graph showing the monthly ASD prevalence trend and piecewise regression results from 2014 to 2023 in an embodiment of the present invention. Figure 4 This is a trend chart of the average monthly ASD prevalence over many years in an embodiment of the present invention; Figure 5 This is a monthly trend chart of 12 environmental factors from 2014 to 2023 in an embodiment of the present invention; Figure 6 This is a heatmap showing the correlation between 12 environmental factors and the prevalence of ASD in an embodiment of the present invention. Figure 7 This is a graph showing the information gain values ​​of various environmental factors on the prevalence of ASD in an embodiment of the present invention; Figure 8 This is a performance evaluation comparison chart of five machine learning models in an embodiment of the present invention; Figure 9 This is a diagram showing the prediction performance of the optimal model in the training and test sets in this embodiment of the invention. Figure 10 This is a graph showing the changing trends of 12 environmental factors under three SSPs scenarios from 2024 to 2099 in this embodiment of the invention. Figure 11 This is a predicted trend chart of ASD prevalence under three SSP scenarios from 2024 to 2099 in an embodiment of the present invention; Figure 12 This is a diagram illustrating the independent explanatory power of each environmental factor on the prevalence of ASD and the explanatory power of interactions between factors in embodiments of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The present invention will be further explained below with reference to specific embodiments.

[0027] The method for predicting the prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data provided in this invention includes the following steps: Step S1: Obtain multi-source remote sensing environmental data and allergic skin disease prevalence data for the study area to construct an environment-disease spatiotemporal dataset; use a piecewise regression model to analyze the temporal variation characteristics of the prevalence and identify inflection points; use an information entropy model to evaluate the contribution of each environmental factor to the prevalence prediction and select environmental factors accordingly; the multi-source remote sensing environmental data includes environmental element data, hydrological element data, and meteorological element data. The environmental data include aerosol optical thickness, ozone concentration, and ultraviolet index. The hydrological data include the monthly maximum runoff and the monthly minimum runoff; The meteorological data include daily maximum temperature, daily minimum temperature, daily average temperature, daily maximum precipitation, daily minimum precipitation, easterly wind speed at 10 meters height, and northerly wind speed at 10 meters height.

[0028] Step S2: Construct multiple machine learning regression models, using the environmental factors selected through the information entropy model evaluation in Step S1 to train and validate the prevalence of allergic skin diseases; optimize model hyperparameters through cross-validation and grid search algorithms; evaluate the performance of each model based on mean squared error, mean absolute error, and correlation coefficient, and select the optimal prediction model; use a geographic detector to quantify the independent explanatory power of each environmental factor on the prevalence and the contribution of interactions between factors; the machine learning regression models include decision tree regression model, elastic network regression model, random forest regression model, ridge regression model, and support vector regression model; The cross-validation is 10-fold cross-validation; The geographic detector includes a factor detector and an interaction detector; The factor detector passes through The values ​​quantify the independent explanatory power of each environmental factor on the morbidity rate. A higher value indicates a stronger explanatory power of the factor for the prevalence of disease. The formula for calculating the value is:

[0029] Where h = 1, 2, 3, …, represents a specific type; L represents the stratification number of the independent or dependent variable; N and h represent the number of cells in layer h and the entire region, respectively. and These are the variances of category h and the variance of the entire region, respectively. The interaction detector is used to assess the change in the explanatory power of two environmental factors acting together relative to their independent effects.

[0030] Step S3: Obtain fecal samples from patients with allergic skin diseases and healthy controls, extract sample DNA, and perform metagenomic paired-end sequencing; perform quality control, sequence assembly, and gene prediction on the sequencing data; perform species annotation and abundance quantification on the assembled sequences to obtain a microbial community composition profile; perform α-diversity analysis and β-diversity analysis on the microbial community based on this composition profile to compare differences in community structure; use linear discriminant analysis effect size method to screen for differential marker microorganisms between groups; construct a microbial symbiotic network, calculate network topological properties, and identify key nodes; align the sequencing sequences with functional databases, and perform functional annotation and pathway enrichment analysis; the metagenomic paired-end sequencing was performed using the Illumina NovaSeq platform; The functional database is the KEGG database; the identification of the key nodes adopts the Zi-Pi thresholding method; The network topology properties include average path length, clustering coefficient, and modularity; the key nodes include module hub nodes and connecting nodes.

[0031] Step S4: Obtain skin tissue samples from patients with allergic dermatitis and healthy controls, prepare frozen sections, and acquire data from the tissue sections using matrix-assisted laser desorption / ionization mass spectrometry (MADS). Perform peak detection, metabolite identification, and spatial distribution analysis on the primary mass spectrometry data. Use partial least squares discriminant analysis to differentiate between metabolographic profiles, and screen differentially expressed metabolites using variable projection importance indices and hypothesis testing. Perform KEGG pathway enrichment analysis on the screened differentially expressed metabolites to determine key metabolic pathways and their spatial distribution characteristics. The metabolic pathways of focus in the KEGG pathway enrichment analysis include: Glucose metabolism pathway, tricarboxylic acid cycle pathway, ceramide synthesis pathway, arachidonic acid metabolism pathway, arginine metabolism pathway, and tryptophan metabolism pathway.

[0032] Step S5: Integrate the data analysis results from steps S2 to S4, using environmental factors as independent variables, differentially expressed microorganisms screened in step S3 as mediating variables, and differentially expressed metabolites screened in step S4 as dependent variables. Quantify the direct effects of environmental factors on host metabolites and the indirect effects mediated by gut microbiota through mediation effect analysis, and calculate the proportion of mediation effects. Incorporate the mediation effect results into the association network to construct a ternary association network of environmental factors—gut microbiota—host metabolites. Evaluate the importance of each environmental factor using a random forest model. Based on the mediation effect analysis results, determine the cascade pathway by which environmental factors affect host skin metabolism through gut microbiota, and output the multi-omics association analysis results for allergic skin diseases.

[0033] Step S6: Combine the optimal prediction model obtained in step S2 with the multi-scenario climate prediction data provided by the CMIP6 Coupled Model Comparison Program, extract the predicted values ​​of environmental factors under different shared socio-economic pathway scenarios in the future, input them into the optimal prediction model, and predict the trend of changes in the prevalence of allergic skin diseases under different scenarios in the future. The shared socioeconomic path scenarios include the SSP126 low-compulsion scenario, the SSP370 medium-compulsion scenario, and the SSP585 high-compulsion scenario. For environmental factors that are not directly output by the CMIP6 model, a generalized linear model is constructed to establish the mapping relationship between the factor and other output factors for estimation. In the generalized linear model, aerosol optical thickness and ozone are used as predictor variables to estimate the ultraviolet index.

[0034] In a more specific example, in step S1, the piecewise regression model assumes the dataset is segmented into... For each paragraph Fitting a linear regression model ,in and They are the first The intercept and slope of the segment, It is an error term; The function values ​​of adjacent segments must be continuous at the breakpoint; The optimal breakpoint location and number of segments are determined using either the Akaike information criterion or the Bayesian information criterion.

[0035] In a more specific example, in step S3, the results of the functional annotation and pathway enrichment analysis are used to identify functional remodeling features of the gut microbiota in patients with allergic dermatitis, the functional remodeling features including: It leads to a decline in basal nutritional metabolic capacity, as well as an enhancement of environmental stress response pathways, substance transport pathways, and host inflammatory response-related pathways.

[0036] In a more specific example, in step S4, the spatial distribution characteristics include the differential enrichment or absence of each metabolite in the epidermis, dermis, and inflammatory infiltration areas of the skin.

[0037] Please see Figures 1-12 Using Urumqi, Xinjiang Uygur Autonomous Region, China as the research area, this invention provides a detailed description of the method for predicting the prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data, as follows: 1. Research Content and Methods 1.1 Overview of the Study Area Urumqi is located in the central part of Xinjiang Uygur Autonomous Region, China, in the heart of Eurasia. It covers an area of ​​approximately 13,800 square kilometers and has a resident population of about 4.08 million. Figure 1 This region has a typical continental arid climate with extremely significant seasonal temperature differences. The average annual temperature is about 6.7℃, and the average annual precipitation is only about 280 mm. Evaporation is intense, and the climate is dry. This arid and low-humidity environment has a significant impact on the spread, concentration, and duration of local allergens (such as pollen and dust mites): low humidity may prolong the suspension time of airborne pollen, while drought and high evaporation easily lead to the lifting of surface particulate matter, increasing the concentration of suspended particulate matter in the air, thereby exacerbating allergic reactions. Topographically, Urumqi exhibits a significant differentiation: "mountains in the east and plains in the west." The east consists of mountains and hills on the northern slope of the Tianshan Mountains, while the west belongs to the plains area on the southern edge of the Junggar Basin. This geomorphological pattern profoundly affects the local wind field, temperature distribution, and precipitation patterns, forming diverse microclimate environments, which in turn leads to complex differences in the spatiotemporal distribution of allergens. Therefore, Urumqi's unique geographical location, distinct arid climate, and complex topography together constitute a typical natural setting for studying how environmental factors (such as climate and landforms) affect the distribution and prevalence of allergic diseases.

[0038] 1.2 Data Sources 1.2.1 Data on allergic skin diseases First, data on allergic dermatitis in Urumqi from 2014 to 2023 were collected. Then, annual population data for Urumqi from 2014 to 2023 were downloaded from the National Bureau of Statistics website (https: / / www.stats.gov.cn / ). Finally, the monthly incidence rate for each year was divided by the corresponding total population to obtain the monthly prevalence rate of ASD.

[0039] 1.2.2 Environmental Impact Factors Based on the principles of scientific rigor, completeness, and independence, 12 evaluation indicators were selected as explanatory factors from three levels: environmental, hydrological, and meteorological elements, to construct predictive models for two allergic diseases in the study area. The selected influencing factors included aerosol optical depth (AOD), ozone (O3), ultraviolet index (UVI), and monthly maximum runoff (Q). max ), monthly minimum runoff (Q) min ), daily maximum temperature (T) max ), daily minimum temperature (T) min ), daily average temperature (T) mean ), maximum daily precipitation (P) max ), minimum daily precipitation (P) min ), Eastward wind speed at a height of 10 meters (W) U ), northerly wind speed at a height of 10 meters (W) V ).

[0040] (1) Environmental factors: AOD, O3, and UVI are the three main indicators used for characterization. AOD reflects the concentration of particulate matter in the air, which may trigger inflammatory reactions in the skin. O3 is a potent respiratory irritant and also has an important impact on allergic skin diseases. High UVI levels may damage skin cells and alter immune responses, thereby triggering the onset of allergic skin diseases. Therefore, we selected these three factors to establish a predictive model for allergic diseases. AOD data comes from MOD08_M3 V6.1 with a spatial resolution of 0.1° and a daily temporal resolution. AOD data was extracted using the Google Earth Engine platform. O3 data comes from the ground O3 dataset in the China High Air Pollutants (CHAP) dataset. This dataset uses artificial intelligence technology, with surface solar radiation intensity and air temperature as the main predictive factors, and combines ground-based observations, atmospheric reanalysis, and emission inventories to produce the maximum 8-hour moving average O3 data of the seamless ground surface from 2000 to the present. The data is cross-validated with a coefficient of determination R. 2 The value was 0.89, and the root mean square error (RMSE) was 15.77 µg / m².3 The main coverage area is the entire China region, with a spatial resolution of 1 km and a temporal resolution of day, month, and year, with units of µg / m³. 3 UVI data comes from NASA's The Clouds and the Earth's Radiant Energy System (CERES) project. Climate is controlled by the amount of sunlight absorbed by the Earth and the infrared energy emitted into space. These quantities and their differences define the Earth's radiation budget (ERB). The CERES project provides satellite observations of the ERB and clouds. It uses measurements from CERES instruments flying on multiple satellites, along with data from many other instruments, to generate a comprehensive suite of ERB data products for climate, weather, and applied science research. This includes UVI data with a spatial resolution of 1° and a temporal resolution on a monthly scale.

[0041] (2) Hydrological elements: mainly using monthly total surface runoff (Q) sum ) and maximum monthly runoff (Q max The runoff data is characterized using the ERA5-Land dataset. ERA5 is a comprehensive reanalysis dataset spanning from 1979 to near real-time, assimilating a large amount of observational data in the upper atmosphere and near the surface as comprehensively as possible. The ERA5 atmospheric model is coupled with land surface and wave models. ERA5-Land is a land surface dataset spanning from 1981 to the present, providing higher resolution (9 km) and driven by ERA5 atmospheric parameters corrected for temperature gradients. This invention uses Google Earth Engine to extract runoff data from ERA5-Land on a daily timescale, and calculates the Q on a monthly scale based on this. max and Q min。

[0042] (3) Meteorological elements: mainly using T max T min T mean P max P min W U W V Seven metrics were used to characterize the dataset, all of which were derived from the ERA5-Land dataset and extracted and post-processed using GEE.

[0043] 1.2.3 CMIP6 Prediction Output To systematically address the multidimensional challenges posed by global climate change, the World Climate Research Programme (WCRP) has launched the Sixth Coupled Model Intercomparison Project (CMIP6). This project aims to provide a crucial data foundation for understanding the physical processes of the climate system and assessing future climate change risks through multi-model ensemble simulations. It also provides important support for achieving the scientific objectives set forth in WCRP's "Major Challenges" program. To predict the future evolution of regional allergic skin diseases, this invention selects three representative shared socioeconomic pathways (SSPs) and representative concentration pathways (RCPs) within the CMIP6 framework to encompass different possibilities under the interaction of future socioeconomic development and climate policy: SSP126 (low forcing scenario) assumes a global transition to a sustainable development path and the implementation of strong emission reduction policies, with radiation forcing stabilizing at approximately 2.6 W / m² by 2100, potentially limiting global average temperature rise to around 1.5°C; SSP370 (medium forcing scenario) reflects a development path characterized by intensified regional competition and relatively limited climate policy, with radiation forcing of approximately 7.0 W / m², corresponding to a global temperature rise of approximately 3°C by the end of the century; while SSP585 (high forcing scenario) characterizes a development model highly dependent on fossil fuels, driven by economic growth, and with weak climate action, with radiation forcing reaching 8.5 W / m², potentially leading to a global average temperature increase of 4.5°C or more. These scenarios integrate socioeconomic drivers and climate system responses, providing more comprehensive and reliable climate change predictions, thus laying a scientific foundation for the quantitative analysis of health effects under different climate forcings in this invention.

[0044] 1.3 Research Methods 1.3.1 Piecewise Regression Model In some cases, the relationships between variables in a dataset are not simple linear relationships, but rather exhibit different trends in different segments. Piecewise regression (PR), also known as regression at discontinuities, is used to model data when multiple distinct linear relationships exist. The basic idea of ​​piecewise regression is to select different linear regression models to fit different data segments based on the characteristics of the data, thereby better describing the overall trend and changes in the data. Furthermore, by dividing the dataset into different segments, we can more accurately capture the non-linear characteristics of the data.

[0045] In piecewise regression, choosing appropriate breakpoints is crucial. Breakpoints can be determined through prior knowledge or estimated using data-driven methods. After determining the breakpoints, the data is divided into several segments, and a linear regression model is fitted to each segment. Assuming the dataset is divided into k segments, for each segment i, the following linear regression model can be fitted: (1-1) in, and These are the intercept and slope of the i-th segment, respectively. This is the error term.

[0046] To make the model smoother, it is generally required that the function values ​​of adjacent segments are continuous at the breakpoint, that is, the regression lines of adjacent segments intersect at the breakpoint. Assume the location of the first breakpoint is... The second breakpoint is located at The following conditions must be met at these breakpoints. (1-2) In addition, the least squares (LS) method is used to estimate the model parameters. For each data segment, an objective function is constructed and the sum of squared errors is minimized.

[0047] (1-3) By solving this optimization problem, the regression parameters for each curve segment can be estimated. Furthermore, the AIC or BIC evaluation metrics can be used to determine the optimal breakpoint location and number of segments.

[0048] 1.3.2 Information Entropy Model The preceding text only explained the rationale for selecting 12 factors in principle. To quantitatively demonstrate the rationality of the factors selected in this invention, the Information Gain (IG) model and multicollinearity diagnostic method are used for evaluation. IG is highly popular in feature selection and for measuring the amount of information contained in a factor; a larger value indicates that the explanatory factor contains more information and has stronger predictive power. Let D be a dataset consisting of n input samples, and F be the explanatory factor dataset. It belongs to the label in dataset D. The number of samples (at flooded or non-flooded locations). The formula for calculating the IG value is as follows:

[0049] in,

[0050]

[0051]

[0052] In the formula, |D| is the sample size. Let j represent the j-th input sample.

[0053] 1.3.3 Machine Learning Model This invention selects five classic machine learning regression models to predict the prevalence of ASD in the study area: Decision Tree Regression (DTR), Elastic Net Regression (ENR), Random Forest Regression (RFR), Ridge Regression (RR), and Support Vector Regression (SVR). All five models are implemented in Python, and their parameters are calibrated using a grid search algorithm and 10-fold cross-validation. The basic principles of these five machine learning models are described in detail below.

[0054] (1) Model Introduction Decision Tree Regression (DTR) is a supervised machine learning algorithm based on a tree structure, suitable for predictive modeling of continuous target variables. Its core mechanism involves recursively binary partitioning the feature space into a series of non-overlapping rectangular regions (corresponding to leaf nodes of the tree), and using the average value of the target variable as the prediction output within each region. The tree construction follows a top-down inductive process: at each step, the algorithm evaluates all features and their possible split points based on an impurity metric (such as the reduction in mean squared error or Gini impurity) to select the optimal splitting attribute and threshold that minimizes the variability of target values ​​within child nodes, aiming to make the sample response values ​​within each subset as homogeneous as possible. After model training, prediction of new samples is achieved by traversing downwards along the decision path from the root node according to its attribute values ​​until a leaf node is reached, and the stored constant value is used as the final prediction result. The significant advantages of this method are its intuitive model structure, ease of visualization and interpretation, low requirements for data preprocessing, and natural ability to handle both numerical and categorical features. However, standard decision trees are prone to overfitting due to excessive pursuit of fitting the training set, resulting in complex tree structures, high variance, and weak generalization ability. To overcome this limitation, regularization techniques such as pre-pruning and post-pruning are often introduced in practice, or they are integrated as base learners into more robust ensemble methods such as random forests and gradient boosting trees to improve prediction accuracy and stability.

[0055] Ridge regression (RR) is a biased estimating regression method specifically designed for handling multicollinear data. Its core principle is to constrain model complexity by introducing an L2 norm penalty term into the Ordinary Least Squares (OLS) loss function, thereby transforming an ill-posed problem into a well-posed one. Specifically, when the feature matrix X is not full column rank or there is a high correlation between features, the L2 norm penalty term in the OLS estimation (X...) becomes more complex. X) - The calculation of ¹ tends to be unstable, leading to excessively large variance in coefficient estimates. Ridge regression, through the corrected solution β_RR = (X X + λI) - ¹X The algorithm Y addresses this problem, where λ is the regularization parameter and I is the identity matrix. As λ increases from 0 to ∞, the regression coefficients continuously shrink towards zero but will not become completely zero, forming a continuous "ridge trace," seeking the optimal trade-off between bias and variance. Although this method sacrifices some unbiasedness, it significantly improves the model's numerical stability and generalization ability, especially showing strong fitting advantages in high-dimensional or ill-conditioned data scenarios. However, ridge regression cannot perform feature selection, and its performance is highly dependent on the selection of the regularization parameter λ, which usually needs to be determined through methods such as cross-validation, ridge trace method, or generalized cross-validation.

[0056] Elastic Network Regression (ENR) further integrates the characteristics of Ridge Regression (L2 regularization) and Lasso Regression (L1 regularization). Its loss function includes both L1 norm (∑|β_j|) and L2 norm (∑β_j²) penalty terms. The core principle is to introduce a composite penalty term, consisting of both Ridge Regression (L2 regularization) and Lasso Regression (L1 regularization) penalty terms, into the loss function of ordinary least squares. Its objective function can be expressed as minimizing... (1-8) in, Controlling the overall penalty intensity, mixed parameters Used to adjust the ratio between L1 and L2 penalties. This structure gives the model a dual mechanism of variable selection and coefficient shrinkage: L1 penalty achieves feature selection and model sparsity by causing some coefficients to be exactly zero, which is especially suitable for high-dimensional data; while L2 penalty can effectively handle multicollinearity problems, compressing the coefficients of highly correlated features to a similar degree, thus improving model stability. Compared to using Lasso or Ridge Regression alone, Elastic Networks show significant advantages in scenarios where the feature dimension is much higher than the sample size or where there is significant correlation between features. It avoids the limitation of Lasso, which may randomly select a single representative variable and ignore other important related variables, and overcomes the shortcomings of Ridge Regression, such as its inability to generate sparse models and its weak interpretability. Its main advantages include: stronger predictive stability in high-dimensional data; automatic feature selection and preservation of related variable groups; and the ability to adjust the ratio between L1 and L2 penalties. It can flexibly adapt to different data structure requirements. However, the limitation of this method lies in the introduction of... and The optimal values ​​of the two hyperparameters are usually determined through grid search using cross-validation, which is computationally expensive. Furthermore, when the signal-to-noise ratio of the data is low or the correlation between features is weak, its performance may not be significantly better than simple Lasso or Ridge regression. Therefore, elastic networks are often used in modeling complex datasets with high dimensionality and multicollinearity, such as those in genomics, financial econometrics, and text mining, becoming an important tool for balancing model explanatory power and prediction accuracy.

[0057] Random Forest Regression (RFR) is a nonlinear regression model based on the bagging ensemble learning concept. Its core principle is to construct a large number of independent regression decision trees (CART) and average their outputs to form the final prediction. This model introduces double randomness to ensure the diversity of the base learners and improve generalization ability. First, multiple subsets of samples are randomly drawn from the original training set using Bootstrap resampling to train each decision tree; Secondly, when splitting at each node of each tree, instead of evaluating all features, the optimal splitting feature is selected from a randomly chosen subset of features. This mechanism effectively reduces model variance and alleviates the problem of overfitting in a single decision tree. The main advantages of random forest regression include: 1) High accuracy and strong robustness, which typically performs well when processing high-dimensional data by integrating the average effective smoothing of noise and outliers; 2) The inherent verification mechanism can use out-of-bag (OOB) samples that are not involved in the construction of each tree to perform unbiased error estimation and feature importance assessment without the need for additional validation set partitioning; 3) It has relaxed requirements on data assumptions, does not require strict pre-normalization, and has a certain degree of adaptability to multicollinearity and interactions. However, this model also has significant limitations: Compared to linear models, it has poor interpretability and is considered a "black box" model; The computational and storage costs are high, and training a large number of deep decision trees requires more computing resources and memory. Furthermore, when the amount of data is limited or there is a smooth global functional relationship between the features and the target variable, its performance may be inferior to simpler parametric models. Therefore, random forest regression is more suitable for scenarios where prediction accuracy is the primary goal, the features have high dimensionality and complex relationships, and the requirements for model interpretability are not strict.

[0058] Support Vector Machines (SVR), a significant achievement in statistical learning theory, extends the classification concept of SVR to the field of regression estimation. Its basic principle is to transform the regression problem into finding an optimal linear hyperplane in a high-dimensional feature space by introducing a predefined ε-insensitive loss function. This process aims to ensure that the deviation between the model's predicted values ​​and the true values ​​does not exceed a preset tolerance ε, while adhering to the principle of minimizing structural risk and maximizing a margin defined as the prediction band (i.e., the "ε-insensitive band") to obtain a regression function that is as "flat" as possible, thereby improving the model's generalization ability. To achieve complex nonlinear fitting, SVR implicitly maps the original input space to a high-dimensional feature space through kernel function techniques (such as radial basis function kernels, multinomial kernels, etc.), and efficiently performs the aforementioned linear regression process in this space. This cleverly avoids the curse of dimensionality problem caused by directly performing high-dimensional calculations. The main characteristics of SVR (Sequential Dynamic Range) are its robustness to outliers (due to the ε-insensitive loss function penalizing only samples outside the error band), its ability to flexibly handle nonlinear relationships through kernel methods, and the global optimality and sparsity of its solutions. However, the performance of this method is highly dependent on the regularization parameter, the ε value, and the choice of kernel function type and parameters, making parameter tuning relatively complex and computationally expensive. Furthermore, its training efficiency drops significantly when the training sample size is extremely large, and its interpretability is generally weaker compared to models such as tree ensembles. Therefore, SVR is particularly suitable for small to medium-sized datasets with moderate sample sizes, complex nonlinear patterns, and high requirements for prediction robustness.

[0059] (2) Model evaluation To comprehensively evaluate the overall performance of the model, three metrics—Mean Squared Error (MSE), Mean Absolute Error (MAE), and Correlation Coefficient (r)—are used to evaluate the model's performance on both the training and validation sets. The formulas for calculating these three metrics are shown below.

[0060] (1-9) (1-10) (1-11) In the formula, It is the true value of the i-th sample. Let m represent the predicted value of the i-th sample, where m is the total number of samples.

[0061] 1.3.4 Geographic Detector To investigate the contribution of various environmental factors, such as temperature, humidity, and air pollution, to the incidence of allergic skin diseases, this invention employs a geographic detector for analysis. This method, initially applied to research on the risk of endemic diseases and related geographical influencing factors, is a novel spatial heterogeneity detection method, comprising four types of detectors: factor detectors, interaction detectors, risk zone detectors, and ecological detectors. Among them, the factor detector is used to quantify the impact of different environmental indicators on the prevalence of allergic skin diseases; the calculation formula is as follows:

[0062] In the formula, h = 1, 2, 3, …, represents a specific type; L is the number of strata for the independent or dependent variable, i.e., classification or partitioning; N and h represent the number of cells in layer h and the entire region, respectively. and These represent the variance of category h and the variance of the entire region, respectively. The value of Q ranges from [0,1]. A larger value indicates a higher contribution of the influencing factor to the occurrence of allergic skin diseases, and vice versa.

[0063] Interaction detectors can be used to assess whether the interaction between two influencing factors increases or decreases the explanatory power of the corresponding variable. In geodetectors, the interaction relationship between two factors is mainly classified into the following five categories: nonlinear attenuation, single-factor nonlinear attenuation, two-factor enhancement, independent interaction, and linear enhancement. Furthermore, this invention mainly selects 12 explanatory factors, including AMDP, FHR, DEM, and CN. To meet the input data requirements of the geodetector, an optimal parameter discretization method is first used for processing. Both the geodetector and the optimal parameter discretization method are implemented using the GD package in R language.

[0064] 1.4 Technology Roadmap This invention constructs a predictive framework for the prevalence of typical allergic skin diseases based on multi-source remote sensing data and multiple machine learning models. Figure 2 First, considering environmental, hydrological, and meteorological factors, 12 influencing factors were selected, and index screening was conducted using IG values ​​and correlation analysis. Next, five typical machine learning models were selected, and model parameters were calibrated using 10-fold cross-validation and a grid search algorithm. The overall performance of the five machine learning models was evaluated based on three assessment indicators: MSE, MAE, and r, and the optimal model for predicting the prevalence of allergic skin diseases was selected. Based on CMIP6 data, 12 influencing factors under three SSPs scenarios were extracted and input into the optimal prediction model to predict the future trend of allergic skin disease prevalence under different SSPs scenarios. Finally, the influence mechanism of different influencing factors on the prevalence of allergic skin diseases was explained using a geographic detector, enhancing the interpretability of the machine learning model.

[0065] 2. Results 2.1 Characteristics of changes in ASD prevalence Figure 3 The results of piecewise linear regression of ASD prevalence from 2014 to 2023 are presented, along with the slope and duration (start time, ST, and end time, ET) of each segment. The graph shows that the overall prevalence of ASD increased from 2014 to 2023. The lowest prevalence of ASD occurred in January 2014 at 0.006%, and the highest prevalence occurred in July 2018 at 0.079%. The piecewise regression results show a clear inflection point in March 2017; before this point, the prevalence of ASD increased, and after this point, it decreased. The multi-year average monthly ASD prevalence trend is also shown. Figure 4 The prevalence of ASD shows a clear seasonal trend. It increases from January to July and decreases from July to December, with a multi-year average monthly maximum of 0.030% and a minimum of 0.055%.

[0066] 2.2 Characteristics of Environmental Factor Changes Figure 5 The figure shows the time-series trends of 12 environmental factors from 2014 to 2023. As can be seen from the figure, AOD, Q... max Q min P max P sum W V The ozone concentration showed a downward trend, while all other factors showed an upward trend. Statistical analysis of the long-term trends of various environmental factors revealed significant heterogeneity. Ozone (O3) concentration exhibited a statistically significant upward trend, indicating that as an atmospheric chemical component with a clear direction of change, it may be becoming a continuously intensifying environmental stressor in the region. In contrast, physical climate parameters such as aerosol optical depth (AOD), temperature series indicators, and surface runoff did not show statistically significant directional trends during the analysis period. This pattern of "significantly rising ozone while most physical parameters show no clear trend" not only reflects the decoupling between photochemical processes and primary pollution and climate responses in regional complex environmental pollution, but also suggests that when constructing environmental driving models for allergic diseases, it is necessary to differentiate the mechanisms of action of different factors: the deterministic long-term trend of ozone makes it a key dynamic variable for predicting future changes in disease risk; while other physical factors, although showing no significant trend, exhibit seasonal and interannual fluctuations, mainly affect the intensity of disease incidence as key background fluctuation factors through their inherent high threshold values ​​or specific combination patterns. This analysis provides important insights into the temporal variation dimensions of the selection and interpretation of model input features.

[0067] Figure 6Correlation analysis showed that, except for the northerly wind speed at a height of 10 meters (W... V Apart from the lowest temperature (T), all other factors were positively correlated with the prevalence of ASD. min The correlation between aerosol optical density (AOD) and ASD was the most significant (correlation coefficient 0.63, p < 0.001). This is likely due to the dry and cold climate of Urumqi, where low temperatures are often accompanied by even lower air humidity, which easily leads to impaired skin barrier function and increased moisture loss from the stratum corneum, thereby increasing skin sensitivity and the risk of inflammatory reactions. In addition, low winter temperatures also cause residents to spend long periods in enclosed, heated environments, which may increase the exposure concentration and cumulative effect of indoor allergens (such as dust mites and pet dander). In contrast, the correlation between AOD and ASD was the weakest (correlation coefficient 0.29, p < 0.05). AOD mainly reflects the overall load of aerosol particles in the atmosphere. In inland arid areas like Urumqi, aerosol components are mainly non-biological particles such as dust and secondary inorganic salts, which are not completely synchronized with the sources and seasonal distribution of biological allergens such as pollen and mold spores that usually induce allergic reactions. This may be an important reason for the low correlation between the two. The results suggest that localized allergy risk assessment should focus more on monitoring specific allergens rather than relying solely on generalized particulate matter indicators.

[0068] Furthermore, considering the importance of prediction, Figure 7 The information gain values ​​of various environmental factors on the risk of ASD are concentrated in a narrow range of 6.6–7.0, indicating that in the machine learning model, these variables contribute relatively evenly to reducing prediction uncertainty, and there is no single dominant factor. This is especially true for maximum runoff (Q... max The information gain value of [the data] is relatively low, mainly because Urumqi has low annual precipitation and limited surface runoff impact. Therefore, its influence as a humidity-related indicator on the overall pattern of allergic diseases throughout the year is less than that of continuous factors such as temperature and sunshine. This pattern of similar contributions from various factors reflects the complexity of the synergistic and intertwined influence of multiple environmental factors in the pathogenesis of allergic diseases. Based on the above analysis, this invention ultimately includes [data] including the ultraviolet index (UVI) and cumulative precipitation (P). sum ) and minimum temperature (T) min A predictive model was constructed based on 12 environmental characteristics, including climate conditions in the Urumqi area, in order to comprehensively capture the combined effects of climate conditions on allergic skin diseases.

[0069] 2.3 Model Performance Evaluation from Figure 8As can be seen, for predicting the prevalence of ASD, all five machine learning models achieved r-values ​​exceeding 0.6 on the training set, demonstrating good performance. The r-values, from highest to lowest, are: Decision Tree Regression (DTR), Ridge Regression (RR), Support Vector Regression (SVR), Random Forest Regression (RFR), and Elastic Network Regression (ENR). DTR also performed well on the test set, with MSE, MAE, and r-values ​​of 0.49, 0.59, and 0.67, respectively, making it the best among the five models. In conclusion, the DTR model outperforms the other four models in predicting the prevalence of ASD; therefore, this model will be used for predicting the prevalence of ASD in different scenarios in the future.

[0070] Figure 9 The results show the consistency between the actual prevalence of ASD and the prevalence predicted by the optimal model. Overall, the DTR model effectively captures the peaks and troughs in the trend of ASD prevalence. For ASD prediction, the DTR model tends to overestimate high values ​​and underestimate low values ​​in the training dataset. This underestimation is more pronounced in the test dataset. Overall, the DTR model performs well in predicting ASD prevalence.

[0071] 2.4 Characteristics of Future ASD Prevalence Changes under Three Types of SSPs This invention utilizes multi-scenario climate prediction data provided by the CMIP6 model to provide an important foundation for exploring the impact of future environmental changes on allergic diseases. Among all environmental factors, the ultraviolet index (UVI) has been proven to play a crucial role in accurately predicting the prevalence of allergic dermatitis (ASD) due to its ability to regulate skin immune responses and influence individual sensitivity to allergens (Table 1). However, the CMIP6 model itself does not directly output UVI data; therefore, this paper constructs a generalized linear model (GLM) to establish the relationship between UVI and 11 other environmental factors (Table 1). The results show that using aerosol optical thickness (AOD) and ozone (O3) as predictor variables can accurately estimate UVI (R² = 0.78, p < 0.001). This is because AOD reflects the extinction effect of aerosols on ultraviolet light, while O3 has strong absorption characteristics for the ultraviolet band, especially UVB. Based on this relationship, combined with future scenario data of AOD and O3 from CMIP6, this invention predicts the trend of UVI changes under three SSP scenarios from 2024 to 2099 (Table 1). Figure 10 ).

[0072] Table 1. Parameters of the generalized linear model used to estimate UVI

[0073] Note: "***" indicates a significance level of 0.001; "**" indicates a significance level of 0.01; "*" indicates a significance level of 0.05.

[0074] Based on the optimized prediction model obtained through training and the CMIP6 meteorological dataset Figure 11 This study presents the annual trends of ASD prevalence in Urumqi from 2024 to 2099 under scenarios SSP126, SSP370, and SSP585. Except for scenario SSP126, the prevalence of ASD shows a significant upward trend in the other two scenarios, with SSP585 showing the largest increase. The prevalence of ASD in all three scenarios exhibits clear seasonal fluctuations, generally showing a cyclical pattern of gradual increase, peak, and then gradual decrease. If the current environment-disease relationship remains unchanged, the prevalence of ASD is projected to reach 0.043%, 0.047%, and 0.044% in scenarios SSP126, SSP370, and SSP585, respectively, by December 2099. The highest prevalence of ASD is expected to occur in summer months, such as July 2089 under scenario SSP126, January 2077 under scenario SSP370, and July 2096 under scenario SSP585, with peak values ​​of 0.079%, 0.079%, and 0.071%, respectively; while the lowest prevalence is expected to occur in early winter months, such as October 2041, September 2065, and October 2062, all at 0.036%.

[0075] 3. Discussion The results of the geographic detector analysis show that ( Figure 12 Except for total runoff, all other factors had a significant impact on the prevalence of allergic dermatitis (ASD) (P<0.01). Among them, the lowest temperature (T) min The greatest impact comes from ultraviolet radiation (UVI, q=0.392), followed by ozone (O3, q=0.379) and average temperature (T). mean The effect of aerosol optical thickness (q=0.364) was the weakest (q=0.120). This suggests that in the dry and cold climate of Urumqi, low temperature-induced skin barrier dysfunction is a key environmental driver of ASD pathogenesis.

[0076] Further analysis using interactive detectors revealed a complex synergistic effect of environmental factors on the prevalence of ASD. AOD and T min Q max Q sum P max P sumThe non-linear reinforcing relationship among these factors indicates that their interactions are not simply additive but dynamically change with factor levels. Most other factors exhibit a bi-factor reinforcing effect, meaning the combined effect is greater than the sum of the individual effects of each factor. Among all interaction combinations, the interaction between low temperature and total runoff had the strongest impact on ASD prevalence (q=0.556), while the interaction between AOD and maximum runoff had the weakest (q=0.340). In conclusion, environmental factors not only act individually on ASD prevalence but also exhibit significant reinforcing interactions, indirectly confirming the feasibility and necessity of using a multi-factor comprehensive prediction model to assess the risk of allergic skin diseases in this region.

[0077] This invention uses a machine learning model to predict the future prevalence of allergic skin diseases, based on the assumption that "the mechanistic relationship between environmental factors observed in historical periods and diseases will remain unchanged in the future." However, this assumption may be challenged by several future evolutionary factors: First, advances in medical technology (such as the widespread availability of new therapies or preventative vaccines) may reduce the actual disease burden, leading to an overestimation of future prevalence by the model; second, increased public health awareness and behavioral adaptations (such as enhanced indoor protection and the use of air purification equipment) can reduce individual allergen exposure, similarly causing predicted values ​​to be higher than actual future figures; third, the allergen ecosystem itself may evolve due to climate change (such as rising CO2 concentrations and temperatures), including increased pollen production, enhanced allergenicity, and the invasion of new allergenic plant species. These changes may exacerbate environmental allergen pressures, causing the model's predictions based on historical data to underestimate the true future risk. Therefore, the prediction results of this invention should be strictly considered as scenario extrapolation under the specific condition that "current relationships remain unchanged," rather than as a definite future trend. These dynamic factors highlight the necessity and importance of combining health prediction models with dynamic system models such as social adaptation and ecological evolution in future research.

[0078] 4. Summary This invention constructs a predictive framework for the prevalence of allergic dermatitis (ASD) in the Urumqi area based on a machine learning model.

[0079] 4.1. The study revealed the changing characteristics of ASD prevalence: From 2014 to 2023, the overall prevalence of allergic dermatitis (ASD) in Urumqi showed an upward trend, with a significant inflection point in March 2017. The prevalence of ASD exhibited obvious seasonal fluctuations, rising from January to July and declining from July to December.

[0080] 4.2. Key environmental influencing factors were identified: The study systematically analyzed the relationship between the typical arid climate and complex topography of the region and the distribution of environmental allergens, clarifying that climate factors are key environmental drivers of ASD. Correlation and information entropy analysis showed that the minimum temperature (T)min The correlation between environmental factors and ASD prevalence was the strongest, while the correlation with aerosol optical thickness (AOD) was the weakest. The importance of each environmental factor in prediction was relatively balanced.

[0081] 4.3 The optimal prediction model was constructed and validated: By integrating multi-source remote sensing and environmental data, 12 key influencing factors, including minimum temperature, UV index, and ozone concentration, were selected. Five machine learning models, including decision tree regression, were used for training and validation. Among the five machine learning models compared, the decision tree regression (DTR) model performed best in predicting the prevalence of ASD, with its test set evaluation metrics (MSE: 0.49, MAE: 0.59, r: 0.67) outperforming other models (ridge regression, support vector regression, random forest regression, and elastic network regression). It was selected as the model for future predictions.

[0082] 4.4. Predicted Prevalence Trends Under Different Scenarios: Based on three climate scenarios (SSP126, SSP370, SSP585) of CMIP6, predictions show that, except for the SSP126 scenario, the prevalence of ASD will show a significant upward trend in the future (2024-2099), with the largest increase under the SSP585 (high compulsion) scenario. The prevalence under all scenarios maintains a seasonal fluctuation pattern of first rising and then falling.

[0083] 4.5. The driving mechanisms and interactions of environmental factors were quantified: Through analysis of geospatial detectors, the independent contributions of various environmental factors to ASD and their complex nonlinear interactions were further revealed, finding that the lowest temperature (T... min The factor that had the strongest independent explanatory power for the prevalence of ASD was q=0.409. There were significant interaction-enhancing effects among multiple factors, especially the interaction between low temperature and total runoff (q=0.556), indicating that the incidence of ASD is the result of complex synergistic effects of multiple environmental factors.

[0084] This invention demonstrates the feasibility of using machine learning models to integrate environmental variables for predicting the risk of allergic diseases, providing a methodological basis and scientific reference for understanding environmental exposure mechanisms and developing regional public health intervention strategies.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the prevalence of allergic skin diseases based on machine learning and multi-source remote sensing data, characterized in that, Includes the following steps: Step S1: Obtain multi-source remote sensing environmental data and allergic skin disease prevalence data for the study area to construct an environment-disease spatiotemporal dataset; use a piecewise regression model to analyze the temporal variation characteristics of the prevalence and identify inflection points; use an information entropy model to evaluate the contribution of each environmental factor to the prediction of the prevalence, and select environmental factors accordingly. Step S2: Construct multiple machine learning regression models, and use the environmental factors selected after evaluation by the information entropy model in Step S1 to train and validate the prevalence of allergic skin diseases; optimize the model hyperparameters through cross-validation and grid search algorithms; evaluate the performance of each model based on mean squared error, mean absolute error and correlation coefficient, and select the optimal prediction model; use a geospatial detector to quantify the independent explanatory power of each environmental factor on the prevalence and the contribution of the interaction between factors. Step S3: Obtain fecal samples from patients with allergic skin diseases and healthy controls, extract DNA from the samples, and perform metagenomic paired-end sequencing; Quality control, sequence assembly, and gene prediction of sequencing data; Species annotation and abundance quantification were performed on the assembled sequences to obtain the microbial community composition profile. Based on this composition profile, α-diversity analysis and β-diversity analysis were performed on the microbial community to compare differences in community structure. We used linear discriminant analysis effect size method to screen differential marker microorganisms between groups; constructed a microbial symbiotic network, calculated the network topology properties and identified key nodes; and compared the sequencing sequences with functional databases to perform functional annotation and pathway enrichment analysis. Step S4: Obtain skin tissue samples from patients with allergic dermatitis and healthy controls, prepare frozen sections, and acquire data from the tissue sections using matrix-assisted laser desorption / ionization mass spectrometry (MALDI-MS). Perform peak detection, metabolite identification, and spatial distribution analysis on the primary mass spectrometry data. Use partial least squares discriminant analysis to differentiate between metabolographic profiles, and screen differentially expressed metabolites using variable projection importance indices and hypothesis testing. Perform KEGG pathway enrichment analysis on the screened differentially expressed metabolites to determine key metabolic pathways and their spatial distribution characteristics. Step S5: Integrate the data analysis results from steps S2 to S4, using environmental factors as independent variables, differentially expressed microorganisms screened in step S3 as mediating variables, and differentially expressed metabolites screened in step S4 as dependent variables. Quantify the direct effects of environmental factors on host metabolites and the indirect effects mediated by gut microbiota through mediation effect analysis, and calculate the proportion of mediation effects. Incorporate the mediation effect results into the association network to construct a ternary association network of environmental factors—gut microbiota—host metabolites. Evaluate the importance of each environmental factor using a random forest model. Based on the mediation effect analysis results, determine the cascade pathway by which environmental factors affect host skin metabolism through gut microbiota, and output the multi-omics association analysis results for allergic skin diseases.

2. The method according to claim 1, characterized in that, In step S1, the multi-source remote sensing environmental data includes environmental element data, hydrological element data, and meteorological element data. The environmental data include aerosol optical thickness, ozone concentration, and ultraviolet index. The hydrological data include the monthly maximum runoff and the monthly minimum runoff; The meteorological data include daily maximum temperature, daily minimum temperature, daily average temperature, daily maximum precipitation, daily minimum precipitation, easterly wind speed at 10 meters height, and northerly wind speed at 10 meters height.

3. The method according to claim 1, characterized in that, It also includes step S6: The optimal prediction model obtained in step S2 is combined with the multi-scenario climate prediction data provided by the CMIP6 Coupled Model Comparison Program to extract the predicted values ​​of environmental factors under different shared socio-economic pathway scenarios in the future. These values ​​are then input into the optimal prediction model to predict the trend of changes in the prevalence of allergic skin diseases under different scenarios in the future. The shared socioeconomic path scenarios include the SSP126 low-compulsion scenario, the SSP370 medium-compulsion scenario, and the SSP585 high-compulsion scenario. For environmental factors that are not directly output by the CMIP6 model, a generalized linear model is constructed to establish the mapping relationship between the factor and other output factors for estimation. In the generalized linear model, aerosol optical thickness and ozone are used as predictor variables to estimate the ultraviolet index.

4. The method according to claim 1, characterized in that, In step S2, the machine learning regression model includes decision tree regression model, elastic network regression model, random forest regression model, ridge regression model and support vector regression model; The cross-validation is 10-fold cross-validation.

5. The method according to claim 1, characterized in that, In step S2, the geographic detector includes a factor detector and an interaction detector. The factor detector passes through The values ​​quantify the independent explanatory power of each environmental factor on the morbidity rate. A higher value indicates a stronger explanatory power of the factor for the prevalence of disease. The formula for calculating the value is: Where h = 1, 2, 3, …, represents a specific type; L represents the stratification number of the independent or dependent variable; N and h represent the number of cells in layer h and the entire region, respectively. and These are the variances of category h and the entire region, respectively. The interaction detector is used to assess the change in the explanatory power of two environmental factors acting together relative to their independent effects.

6. The method according to claim 1, characterized in that, In step S3, the metagenomic paired-end sequencing is performed using the Illumina NovaSeq platform; The functional database is the KEGG database; the identification of the key nodes adopts the Zi-Pi thresholding method; The network topology properties include average path length, clustering coefficient, and modularity; the key nodes include module hub nodes and connecting nodes.

7. The method according to claim 1, characterized in that, In step S4, the KEGG pathway enrichment analysis focuses on the following metabolic pathways: Glucose metabolism pathway, tricarboxylic acid cycle pathway, ceramide synthesis pathway, arachidonic acid metabolism pathway, arginine metabolism pathway, and tryptophan metabolism pathway.

8. The method according to claim 1, characterized in that, In step S1, the piecewise regression model assumes the dataset is segmented into... For each paragraph Fitting a linear regression model ,in and They are the first The intercept and slope of the segment, It is an error term; The function values ​​of adjacent segments must be continuous at the breakpoint; The optimal breakpoint location and number of segments are determined using either the Akaike information criterion or the Bayesian information criterion.

9. The method according to claim 1, characterized in that, In step S3, the results of the functional annotation and pathway enrichment analysis are used to identify functional remodeling features of the gut microbiota in patients with allergic dermatitis. These functional remodeling features include: It leads to a decline in basal nutritional metabolic capacity, as well as an enhancement of environmental stress response pathways, substance transport pathways, and host inflammatory response-related pathways.

10. The method according to claim 1, characterized in that, In step S4, the spatial distribution characteristics include the differential enrichment or absence of each metabolite in the epidermis, dermis and inflammatory infiltration areas of the skin.

Images