Method and system for infectious disease pathogen monitoring analysis based on targeted high-throughput sequencing

Abstract

The application belongs to the technical field of bioinformatics, and discloses an infectious disease pathogen monitoring analysis method and system based on targeted high-throughput sequencing, which comprises the following steps: processing unstructured clinical text of a sample according to a preset medical keyword mapping rule and a negative logic rule to obtain a standard medical phenotype label, wherein the negative logic rule is used to avoid misjudging exclusion diagnosis information as a positive phenotype label; constructing a sample feature vector, wherein the sample feature vector at least contains a targeted high-throughput sequencing result of the sample and the standard medical phenotype label; and adopting bidirectional consistency checking based on environmental noise decoupling for the sample feature vector to eliminate pseudo-correlation signals introduced by environmental confounding factors and obtain a pathogen co-infection determination result. The application can realize accurate discrimination of pathogen interaction relationships.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics technology, specifically relating to a method and system for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing. Background Technology

[0002] Targeted next-generation sequencing (tNGS) technology combines the high sensitivity of PCR with the high throughput of traditional next-generation sequencing (NGS), enabling the simultaneous detection of over 153 pathogens and more than 370 drug resistance genes in a single reaction system. However, in actual clinical monitoring and data analysis, tNGS data exhibits characteristics of high dimensionality, sparseness, heterogeneity, and massive volume: a single detection can generate binary state information of hundreds of target dimensions, making it difficult for traditional data analysis methods to process and deeply mine the data efficiently. Furthermore, the high sensitivity of tNGS easily detects multiple co-existing pathogen signals simultaneously. Many of these co-existence phenomena are caused by confounding factors such as seasonal epidemics and environmental exposures, representing statistically significant co-detection and spurious correlations, rather than genuine pathogen biological co-infection and interaction. Traditional monitoring and analysis methods struggle to effectively distinguish between spurious correlations and genuine co-infections, easily leading to insufficient accuracy in pathogen interaction discrimination, misconceptions about co-infection networks, and misjudgments of epidemic trends, thus affecting the reliability of clinical diagnosis and epidemic early warning. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing, which can accurately identify pathogen interactions.

[0004] The first aspect of this invention discloses a method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing, comprising:

[0005] According to the preset medical keyword mapping rules and negation logic rules, the unstructured clinical text of the sample is processed to obtain standard medical phenotype labels. The negation logic rules are used to avoid misjudging exclusionary diagnostic information as positive phenotype labels.

[0006] Construct a sample feature vector, wherein the sample feature vector includes at least the targeted high-throughput sequencing results of the sample and the standard medical phenotypic label;

[0007] For the sample feature vector, a bidirectional consistency check based on environmental noise decoupling is used to eliminate spurious correlation signals introduced by environmental confounding factors and obtain the pathogen co-infection determination result.

[0008] In some implementations, after obtaining the co-infection determination result of pathogens, time series data of the detection rate of the target pathogens are also obtained, adaptive piecewise linear fitting is performed on the time series data, trend turning points are identified, trend change indicators of the latest time interval are calculated, and an early warning is issued when the trend change indicators meet the preset early warning conditions.

[0009] In some implementations, the step of adaptively segmenting and linearly fitting the time series data, identifying trend turning points, calculating a trend change index for the latest time interval, and issuing an early warning when the trend change index meets preset early warning conditions includes:

[0010] Traverse all possible breakpoints in the time series data and combine them to obtain several combinations of breakpoint positions;

[0011] Construct piecewise linear regression models corresponding to each combination of the aforementioned breakpoint positions;

[0012] Calculate the evaluation index for each piecewise linear regression model, determine the optimal piecewise linear regression model based on the evaluation index, and determine the trend inflection point based on the optimal piecewise linear regression model;

[0013] Based on the trend turning point, the time series data is divided into multiple time intervals;

[0014] Calculate the annual percentage change and slope for the latest time interval;

[0015] An early warning is issued when the slope exceeds a preset threshold and the annual change percentage shows an accelerating trend.

[0016] In some implementations, the step of processing the unstructured clinical text of the sample according to preset medical keyword mapping rules and negation logic rules to obtain standard medical phenotypic labels includes:

[0017] Pre-construct medical keyword mapping rules that include clinical symptoms;

[0018] Pre-construct negation logic rules that include negation words and distance constraints;

[0019] The unstructured clinical text is segmented and scanned. When a keyword is identified according to the medical keyword mapping rule, the characters within the distance constraint range are traced back according to the negation logic rule. If a negative word exists in the characters, the keyword is discarded; otherwise, the keyword is mapped to the standard medical phenotype label.

[0020] In some implementations, after constructing the sample feature vector, the method further includes performing a merging operation on multiple sample feature vectors of the same patient identifier within a preset time window.

[0021] In some implementations, the step of using a bidirectional consistency check based on environmental noise decoupling to remove spurious correlation signals introduced by environmental confounding factors to obtain the pathogen co-infection determination result for the sample feature vector includes:

[0022] For the pathogen pair to be analyzed, including the first pathogen and the second pathogen, a first logistic regression model is constructed with the first pathogen as the dependent variable and the second pathogen as the independent variable, and a second logistic regression model is constructed with the second pathogen as the dependent variable and the first pathogen as the independent variable. In both the first logistic regression model and the second logistic regression model, the environmental confounding factor is used as a covariate.

[0023] When the statistical significance of the core independent variables in both the first and second logistic regression models is less than the preset significance threshold, and the signs of the regression coefficients in the first and second logistic regression models are consistent, the first pathogen and the second pathogen are determined to have a true co-infection relationship and the result is saved to the pathogen co-infection determination result.

[0024] In some implementations, after obtaining the pathogen co-infection determination result, topological relationship data of pathogen co-infection is generated based on the pathogen co-infection determination result, and a co-infection network map that is different from environmental pseudo-correlation is displayed.

[0025] A second aspect of this invention discloses an infectious disease pathogen monitoring and analysis system based on targeted high-throughput sequencing, comprising:

[0026] The data standardization module is used to process the unstructured clinical text of the sample according to the preset medical keyword mapping rules and negation logic rules to obtain standard medical phenotype labels. The negation logic rules are used to avoid misjudging exclusionary diagnostic information as positive phenotype labels.

[0027] A multidimensional feature construction module is used to construct a sample feature vector, wherein the sample feature vector contains at least the targeted high-throughput sequencing results of the sample and the standard medical phenotypic label;

[0028] The co-infection determination module is used to perform a bidirectional consistency check based on environmental noise decoupling on the sample feature vector, eliminate spurious correlation signals introduced by environmental confounding factors, and obtain the pathogen co-infection determination result.

[0029] A third aspect of the present invention discloses an electronic device, including a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the infectious disease pathogen monitoring and analysis method based on targeted high-throughput sequencing disclosed in the first aspect.

[0030] The fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the infectious disease pathogen monitoring and analysis method based on targeted high-throughput sequencing disclosed in the first aspect.

[0031] The beneficial effects of this invention are that, by using preset medical keyword mapping rules and negation logic rules, unstructured text is converted into standard medical phenotypic labels, generating sample feature vectors that are compatible with unstructured clinical text and high-dimensional tNGS detection data, providing reliable data support for the accurate determination of pathogen co-infection; then, under the premise of eliminating the interference of confounding factors, bidirectional consistency verification is used to identify the co-infection patterns between pathogens based on the sample feature vectors, ensuring that the analysis results reflect the real biological synergistic or antagonistic relationships, and achieving accurate identification of pathogen interaction relationships. Attached Figure Description

[0032] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0033] Unless otherwise specified or defined, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0034] Figure 1 This is a flowchart of an embodiment of a method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing according to the present invention;

[0035] Figure 2 This is a flowchart illustrating the determination of true co-infection in an embodiment of the present invention;

[0036] Figure 3 This is a schematic diagram of the bidirectional logistic regression co-infection discrimination execution process according to an embodiment of the present invention;

[0037] Figure 4 This is a flowchart of trend inflection point identification and early warning based on adaptive piecewise linear fitting according to an embodiment of the present invention;

[0038] Figure 5 This is a schematic diagram of the structure of the infectious disease pathogen monitoring and analysis system based on targeted high-throughput sequencing according to an embodiment of the present invention;

[0039] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0040] Unless otherwise specified or defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. When combined with the technical solutions of the invention in a real-world scenario, all technical and scientific terms used herein may also have meanings corresponding to the purpose of achieving the technical solutions of the invention. The terms "first," "second," etc., used herein are merely for distinguishing names and do not represent a specific number or order. The term "and / or," as used herein, includes any and all combinations of one or more of the associated listed items.

[0041] It should be noted that when a component is considered "fixed" to another component, it can be directly fixed to the other component or there can be an intervening component; when a component is considered "connected" to another component, it can be directly connected to the other component or there can be an intervening component; when a component is considered "mounted" on another component, it can be directly mounted on the other component or there can be an intervening component; when a component is considered "placed" on another component, it can be directly placed on the other component or there can be an intervening component.

[0042] Unless otherwise specified or defined, the terms "described" or "the" as used herein refer to the technical features or technical content mentioned or described prior to the relevant section, which may be the same as or similar to the technical features or technical content mentioned herein. Furthermore, the terms "comprising" and "having," and any variations thereof, as used herein, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0043] To facilitate understanding of the present invention, specific embodiments of the present invention will be described in more detail below with reference to the accompanying drawings.

[0044] Addressing the technical challenges of existing infectious disease pathogen monitoring systems in handling massive amounts of heterogeneous data and accurately identifying pathogen interactions, this invention employs a unique "keyword mapping + negation filtering" technique. This technique cleanses unstructured text into standard medical terminology, generating sample feature vectors compatible with both unstructured clinical text and high-dimensional tNGS detection data. Then, after eliminating environmental confounding factors (such as season, age, and region), a bidirectionally validated mathematical model is established. Based on this bidirectional consistency verification, the true co-infection patterns between pathogens are accurately identified. Furthermore, based on a self-developed adaptive piecewise linear fitting algorithm, inflection points in epidemic trends are identified, enabling "early warning."

[0045] This invention provides a detailed description using the monitoring, analysis, and early warning of pathogens corresponding to respiratory infectious diseases as an example, but its scope of protection is not limited thereto. The technical solutions provided by this invention are also applicable to pathogen-related analysis, monitoring, and early warning scenarios for other common infectious diseases such as intestinal infectious diseases, vector-borne diseases, and bloodborne infectious diseases, and can be widely used in the fields of etiological research, clinical diagnostic assistance, and public health prevention and control of various infectious diseases.

[0046] like Figure 1 As shown, the embodiments of the present invention specifically include the following steps:

[0047] Step S100: Based on the preset medical keyword mapping rules and negation logic rules, process the unstructured clinical text of the sample to obtain standard medical phenotype labels. The negation logic rules are used to avoid misjudging exclusionary diagnostic information as positive phenotype labels.

[0048] The samples included unstructured clinical text and targeted high-throughput sequencing results. Unstructured clinical text refers to clinically relevant text information, the most common data format in clinical settings, used to record patient condition descriptions, diagnostic information, etc. Examples include doctor-written medical records, clinical notes in laboratory reports, as well as unformatted text content such as progress notes, doctor's orders, and symptom descriptions.

[0049] Specifically, firstly, two types of rules are pre-constructed: medical keyword mapping rules and negation logic rules. These two types of rules work together to ensure the accuracy of medical phenotypic label extraction. The medical keyword mapping rules are used to establish a correspondence between key information in clinical texts and standard medical phenotypic labels. These rules include three main categories of keywords: clinical symptoms, anatomical locations, and pathological descriptions. Each category of keywords can correspond to a pre-defined standard medical phenotypic label. For example, the clinical symptom keywords "lung inflammation, lung infection" correspond to the phenotypic label "pneumonia." This mapping rule can quickly transform key information in the text into standardized medical phenotypic labels that can be used for subsequent analysis.

[0050] Negation logic rules are used to filter negative expressions in text, preventing negative diagnostic information from being misclassified as positive phenotype labels. They mainly consist of two parts: a set of negation words and distance constraints. The set of negation words covers commonly used negative expressions in clinical scenarios, including "none," "not," "deny," "not seen," "exclude," and "does not exist." Distance constraints limit the scope of negation words, ensuring that they only take effect within the backward distance constraint range from the keyword's location, avoiding false negations across sentences or paragraphs, and ensuring the accuracy of negation logic validation. This prevents situations where, in long sentences like "influenza is ruled out, but the test result is pneumonia," the label "pneumonia" is mistakenly deleted because of the word "excluded."

[0051] Then, word segmentation and scanning operations are performed sequentially on the acquired unstructured clinical text. The word segmentation operation uses a clinically specific algorithm to break down continuous text content into independent lexical units, eliminating meaningless interjections, punctuation marks, and other interfering information. The scanning operation, based on pre-constructed medical keyword mapping rules, matches each segmented lexical unit individually or uses a multi-pattern matching algorithm (such as the AC automaton) to identify keywords related to clinical symptoms, anatomical locations, and pathological descriptions within the text. When a keyword matching the medical keyword mapping rules is identified, a negation logic verification process is triggered: based on the preset distance constraints in the negation logic rules, the distance constraint characters at the keyword's location are traced back, and negative word matching is performed on the characters within the traceback range. If any negative word from the set of negative words exists within the traceback range, the description corresponding to the keyword is determined to be a negative diagnosis, the keyword is discarded, and no phenotypic label mapping is performed; if no negative word is detected within the traceback range, the description corresponding to the keyword is determined to be a positive or definitive diagnosis, and the keyword is mapped to the corresponding standard medical phenotypic label according to the medical keyword mapping rules.

[0052] In this embodiment, the unstructured clinical test text is sourced from the LIS (Laboratory Information System). The unstructured clinical test text contains terms such as "suspected severe pneumonia" and "bilateral lung infection with respiratory failure." The medical keyword mapping rules include: keywords containing "lung inflammation" or "lung infection" are mapped to the standard label "pneumonia"; keywords containing "severe," "respiratory failure," or "white lung" and also containing "pneumonia" are mapped to "severe pneumonia"; keywords containing "septicemia" or "septicemia" are mapped to "septicemia"; keywords containing "tuberculosis" or "TB" are mapped to "pulmonary tuberculosis"; and keywords containing "tumor," "space-occupying lesion," and "lung" are mapped to "lung tumor."

[0053] By employing a unique "keyword mapping + negation filtering" technique, this method matches negative words and calculates the character distance between negative words and keywords. This not only effectively standardizes unstructured clinical text but also eliminates negative interference, ensuring the authenticity and accuracy of the extracted standard medical phenotypic labels. Compared to unstable deep learning models, this avoids the risk of misjudgment (such as "exclude" or "none"), and offers extremely fast processing speed.

[0054] In some implementations, suspected word rules can also be set, and the suspected word rules and negation logic rules form a logic rule base. The application process of suspected word rules is the same as that of negation logic rules, except that suspected labels are added on the basis of standard medical phenotypic labels. Suspected labels can also be used to construct sample feature vectors.

[0055] Step S200: Construct a sample feature vector, wherein the sample feature vector contains at least the targeted high-throughput sequencing results and standard medical phenotypic labels of the sample;

[0056] The data used to construct the sample feature vectors comes from high-quality targeted high-throughput sequencing results generated by tNGS technology and standard medical phenotypic labels converted from unstructured clinical text. This allows the sample feature vectors to include not only the detection results but also the clinical phenotype, greatly enriching the dimensions of data mining and providing reliable data support for the accurate determination of pathogen co-infection and the analysis of epidemic trends.

[0057] Compared to traditional pathogen detection methods (including conventional culture methods and ordinary PCR technology), targeted high-throughput sequencing (tNGS) significantly improves the positive detection rate of respiratory pathogens, effectively reducing the occurrence of missed diagnoses in clinical practice. Furthermore, thanks to its broad coverage of 153 pathogen targets, it can sensitively capture complex co-infection scenarios in clinical settings, obtaining realistic and comprehensive infection-related data. This allows subsequent logistic regression models used for pathogen co-infection determination to obtain sufficient and realistic sample data for model training and validation, thereby accurately revealing pathogen interaction patterns that are difficult to detect using traditional single-target detection methods.

[0058] Specifically, the sample feature vector V is defined as:

[0059] in: The k standard medical phenotype labels obtained by processing according to medical keyword mapping rules and negation logic rules are represented by 0 / 1 binary encoding (1 represents the existence of the phenotype, and 0 represents the non-existence of the phenotype). Represents the detection status of 153 target pathogens (covering viruses, bacteria, and fungi), using 0 / 1 binary coding (1 represents detection, 0 represents non-detection). This represents the detection status of 370 drug resistance genes; This represents the discretized age group code. For example, according to medical standards, continuous age is divided into 7 intervals: newborn [0,28] days, infancy (28,365] days, early childhood (1,3] years, preschool age (3,6] years, school age (6,18) years, adult [18,60) years, and elderly ≥60 years. Represents administrative division code (based on GB / T 2260 standard); Represents the specimen type code (e.g., 1=bronchoalveolar lavage fluid, 2=nasopharyngeal swab, 3=sputum, 4=pleural effusion and peritoneal effusion, 5=tissue, 6=other); Represents the seasonal code.

[0060] It is important to emphasize that the feature vector is constructed based on samples, not patients. Existing respiratory infectious disease surveillance systems (such as NDRS, SARI, NREVSS, etc.) often use patients as the statistical unit when faced with the massive data deluge from tNGS. In clinical practice, the same patient may submit multiple different types of samples (such as bronchoalveolar lavage fluid and nasopharyngeal swabs) during the same course of the disease. Their microecological environment and pathogen load vary significantly. Simply merging and statistically analyzing these samples can mask the pathogen spectrum characteristics of different sample types, affecting the determination of co-infection.

[0061] Furthermore, considering that critically ill patients often undergo multiple tests within a short period (e.g., one test upon admission, followed by a repeat test 3 days later), directly calculating the sample size would lead to an inflated prevalence rate of pathogens affecting severe cases (such as Acinetobacter baumannii). Therefore, an aggregation process is introduced: after constructing the sample feature vector, all samples with the same patient identifier, the same admission serial number, and within a preset window period are considered as a single treatment event. The union of the detected results is then used to generate a unique sample feature vector, ensuring the objectivity of epidemiological statistics.

[0062] Step S300: For the sample feature vector, a bidirectional consistency check based on environmental noise decoupling is used to remove spurious correlation signals introduced by environmental confounding factors and obtain the pathogen co-infection determination result.

[0063] To address the "co-detection" noise introduced by the high sensitivity of tNGS (i.e., two pathogens are detected simultaneously simply because they both tend to occur in winter, when there is actually no biological association), traditional statistical methods (such as Pearson correlation) easily misjudge "seasonal co-occurrence" as "co-infection," failing to distinguish between "environmental coexistence" and "biological synergy." Research has found that genuine pathogen interactions (such as bacteria using virally damaged mucous membranes for invasion) should statistically exhibit a strong bidirectional association, remaining significant even after removing environmental covariates such as temperature and season. If only one direction is significant, or the signs are contradictory, it is likely a spurious association dominated by environmental factors. Therefore, this invention employs bidirectional logistic regression and a sign consistency check mechanism to effectively eliminate false positive associations caused by confounding factors (such as co-occurrence solely due to the same season), ensuring that the analysis results reflect genuine biological synergistic or antagonistic relationships.

[0064] like Figure 2 As shown, the specific steps include:

[0065] Step S310: For each pair of pathogens to be analyzed, including the first pathogen and the second pathogen, construct a first logistic regression model with the first pathogen as the dependent variable and the second pathogen as the independent variable, and a second logistic regression model with the second pathogen as the dependent variable and the first pathogen as the independent variable, and use environmental confounding factors as covariates in both the first logistic regression model and the second logistic regression model.

[0066] Specifically, for each pair of pathogens (A, B) to be analyzed, the detection status (0 / 1) of the first pathogen A is used as the dependent variable Y, and the detection status of the second pathogen B is used as the core independent variable X. An environmental feature vector is incorporated as a noise filter to remove spurious correlation signals Z caused by non-biological factors such as seasonal overlap, thus constructing the first logistic regression model:

[0067] Where P(A=1) represents the probability of the first pathogen being positive; B represents the detection status of the second pathogen; This represents the i-th environmental confounding factor or interference characteristic; , , These are the model fitting coefficients. The first logistic regression model improves the accuracy of pathogen co-infection determination by correcting for environmental noise and confounding factors and eliminating spurious correlation signals.

[0068] Using the detection status (0 / 1) of the second pathogen B as the dependent variable Y, and the detection status of the first pathogen A as the core independent variable X, and incorporating the same covariates, a second logistic regression model is constructed:

[0069] Where P(B=1) represents the probability of the second pathogen being positive; A represents the detection status of the first pathogen; This represents the i-th environmental confounding factor or interference characteristic; , , These are the model fitting coefficients.

[0070] In this embodiment, the environmental confounding factors are temperature, season, and age group.

[0071] Step S320: When the statistical significance of the core independent variables in both the first logistic regression model and the second logistic regression model is less than the preset significance threshold, and the signs of the regression coefficients in the first logistic regression model and the second logistic regression model are consistent, the first pathogen and the second pathogen are determined to have a true co-infection relationship and saved to the pathogen co-infection determination results.

[0072] The sample feature vectors are input into the first logistic regression model and the second logistic regression model to achieve bidirectional cross-validation. A significant interaction between the first pathogen A and the second pathogen B is determined only if the following dual conditions are met: the statistical significance (P-value) of the core independent variables in both the first and second logistic regression models is less than a preset significance threshold, and the regression coefficient of the first logistic regression model is... Regression coefficients of the second logistic regression model The signs are in the same direction. At this time, if... : Determined to be a positive interaction (cooperative); if The interaction is determined to be negative (competitive / antagonistic). The actual co-infection relationship between pathogen A and pathogen B is then saved to the pathogen co-infection determination results.

[0073] The bidirectional logistic regression co-infection discrimination process in this embodiment can be referred to... Figure 3 As shown.

[0074] After obtaining the pathogen co-infection determination results, this embodiment further extracts information on the actual co-infecting pathogen pairs based on these results, generating topological relationship data for pathogen co-infection. Based on this topological relationship data, a pathogen co-infection network map, distinct from environmental spurious correlations, is drawn. This map visualization clearly presents the true pathogen interactions, effectively eliminating statistical spurious correlations caused solely by environmental confounding factors such as seasonal overlap. In practical implementation, the strength of co-infection associations between pathogens can be quantified by calculating the odds ratio (OR) and corresponding confidence intervals. This provides a quantitative basis for drawing the co-infection network map and eliminating spurious correlations, ensuring that the map accurately reflects the true biological synergistic infection relationships between pathogens, providing an intuitive and reliable reference for clinical diagnostic assistance and public health control.

[0075] Existing respiratory infectious disease surveillance systems often use a "fixed threshold method" for early warning (e.g., an alarm is triggered when the influenza detection rate exceeds 10%). This method is a "post-event response" and cannot detect subtle changes in the slope of the epidemic curve in the early stages (such as a sudden increase in the annual percentage change (APC)), making it difficult to issue an alert in the early stages of an outbreak.

[0076] To detect epidemic inflection points earlier than traditional threshold methods and achieve "early warning," this invention also designs a trend inflection point identification and early warning method based on adaptive piecewise linear fitting. Specifically, after obtaining the pathogen co-infection determination result, time series data of the target pathogen detection rate are acquired. The time series data is collected at preset time intervals (such as daily, weekly, and monthly), covering the detection rate change information of the target pathogen within the monitoring period. Then, adaptive piecewise linear fitting processing is performed on the time series data. The fitting algorithm automatically identifies the changing characteristics of the data trend, accurately locates the trend inflection point, and divides the entire time series into multiple time intervals with stable trends. Each time interval corresponds to a continuous and consistent detection rate change pattern. For the latest time interval obtained after fitting (i.e., the latest time interval and the time interval closest to the current monitoring node),... The system calculates the corresponding trend change indicators, including but not limited to annual percentage change (APC) and fitting slope. These indicators quantify the increase or decrease and rate of change of the detection rate of the target pathogen within the latest time interval. When the calculated trend change indicators meet the preset warning conditions (such as the slope exceeding the preset threshold and showing an accelerating upward trend, or the APC reaching the preset risk threshold), an early warning signal is automatically triggered. At the same time, relevant warning information (including warning time, trend change indicator value, corresponding time interval, etc.) is recorded, enabling timely monitoring and risk warning of the epidemic trend of the target pathogen, and providing timely and reliable technical support for public health prevention and control and clinical diagnosis and treatment.

[0077] The specific steps of this embodiment are as follows: Figure 4 As shown, it includes:

[0078] Step S11: Traverse all possible breakpoints in the time series data and combine them to obtain several breakpoint position combinations;

[0079] First, input the time series data of the target pathogen detection rate, and then perform smoothing preprocessing on the time series data. This time series data is denoted as... ,in (i=1,2,…,n) represent each monitoring time point. This represents the detection rate of the target pathogen at the corresponding time point. The data collection cycle can be set daily, weekly, or monthly according to actual monitoring needs. A maximum allowed number of breakpoints can be preset. (It can be flexibly adjusted according to the epidemic pattern of respiratory pathogens and the required monitoring accuracy, for example, setting...) =3). Adaptive piecewise linear fitting algorithms (such as top-down (TD), bottom-up (BU), and multivariate adaptive regression spline (MARS) algorithms) are used to process the time series data, automatically traversing all possible breakpoints. (k takes values ​​from 0 to 0) ≤ ), and all possible combinations of breakpoint positions for each breakpoint number.

[0080] Step S12: Construct piecewise linear regression models corresponding to the combinations of each breakpoint location;

[0081] Step S13: Calculate the evaluation index for each piecewise linear regression model, determine the optimal piecewise linear regression model based on the evaluation index, and determine the trend inflection point based on the optimal piecewise linear regression model;

[0082] The evaluation index can be the Akaike Information Content Criterion (ABIC) or the coefficient of determination. The ABIC value takes into account both model fit and complexity. It reflects the model's fit to time series data.

[0083] For each combination of breakpoint locations, a corresponding piecewise linear regression model is constructed. For each piecewise linear regression model, its Akaike Information Criterion (ABIC) value is calculated. By comparing the ABIC values ​​of all piecewise linear regression models, the piecewise linear regression model with the smallest ABIC value is selected as the optimal piecewise linear regression model. This optimal piecewise linear regression model is used to accurately determine the inflection point of the target pathogen detection rate trend. This ensures the accuracy of trend analysis and effectively avoids misjudgments caused by the fitting bias of a single model.

[0084] By employing an adaptive piecewise linear fitting algorithm and introducing information content criteria (such as ABIC) as an evaluation function for model optimization, the optimal structural breakpoints in time series can be automatically identified.

[0085] Step S14: Divide the time series data into multiple time intervals based on trend turning points;

[0086] Step S15: Calculate the annual percentage change and slope for the latest time interval;

[0087] Step S16: Issue an alert when the slope exceeds a preset threshold and the annual change percentage shows an accelerating trend.

[0088] Based on the temporal structure formed by trend inflection points, the time series data is divided into multiple time intervals. The slope and annual percentage change (APC) of the fitted line for the target pathogen detection rate corresponding to the latest time interval (i.e., the current time window) are calculated. If the fitted slope of the current time window is greater than a preset threshold, it indicates a significant upward trend in the pathogen detection rate; if the annual percentage change of the current time window shows a significant accelerating trend compared to the annual percentage change of the previous time window (i.e., the previous annual percentage change is less than the current annual percentage change), it indicates that the pathogen detection rate is increasing at a faster rate. When both of these conditions are met, there is a risk of an epidemic, and timely warnings are issued to remind relevant personnel to implement prevention and control measures to prevent the spread of the epidemic.

[0089] In summary, this embodiment cleanses unstructured text into standard medical terminology, generating sample feature vectors compatible with both unstructured clinical text and high-dimensional tNGS detection data. Under the premise of eliminating confounding factors, a two-way validated mathematical model is established to accurately identify the true co-infection patterns between pathogens based on two-way consistency verification. Furthermore, an adaptive piecewise linear fitting algorithm is used to identify inflection points in epidemic trends, achieving "early warning."

[0090] like Figure 5 As shown, based on the above-mentioned method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing, this embodiment of the invention discloses an infectious disease pathogen monitoring and analysis system based on targeted high-throughput sequencing, comprising:

[0091] The data standardization module 600 is used to process the unstructured clinical text of the sample according to the preset medical keyword mapping rules and negation logic rules to obtain standard medical phenotype labels. The negation logic rules are used to avoid misjudging exclusionary diagnostic information as positive phenotype labels.

[0092] The multidimensional feature construction module 610 is used to construct a sample feature vector, wherein the sample feature vector includes at least the targeted high-throughput sequencing results of the sample and the standard medical phenotypic label;

[0093] The co-infection determination module 620 is used to perform a bidirectional consistency check based on environmental noise decoupling on the sample feature vector, remove spurious correlation signals introduced by environmental confounding factors, and obtain the pathogen co-infection determination result.

[0094] The infectious disease pathogen monitoring and analysis system in this embodiment is deployed on the respiratory infection pathogen monitoring and analysis platform. The respiratory infection pathogen monitoring and analysis platform also includes a distributed data scheduling module. The distributed data scheduling module slices the cross-regional multi-source heterogeneous data capture task according to time window and region, distributes it to multiple computing nodes for parallel execution, and synchronizes the cleaned data to the cache database in real time.

[0095] By adopting a distributed task scheduling architecture, automatic data collection from LIS systems in 31 provinces across the country was achieved. When processing more than 2 million tNGS data points nationwide, a near real-time update of "T+3" from sampling to visualization of analysis results was realized, solving the data latency problem under the traditional single-point computing architecture and meeting the timeliness requirements of national-level public health monitoring.

[0096] The respiratory infection pathogen monitoring and analysis platform adopts a front-end and back-end separated B / S architecture design, with the underlying distributed computing cluster ensuring timeliness. It includes: a front-end interaction layer, a back-end service layer, a data storage and caching layer, and a core scheduling layer. The front-end interaction layer is a single-page application built on a progressive framework of front-end interaction modules, combined with Vue Router for routing management. The UI component library uses a visual interactive interface to ensure the standardization of medical-grade interfaces. The data visualization module deeply integrates ECharts to achieve dynamic rendering of geographic heatmaps, Sankey diagrams, and co-occurrence network diagrams. Network requests interact with the back-end RESTful API through the Axios library. The back-end service layer is developed based on a microservice framework for business logic processing servers, responsible for business logic distribution; it utilizes MyBatis / MyBatis-Plus for efficient data persistence operations; it integrates Shiro and JWT to achieve stateless distributed identity authentication and access control; and it introduces the EasyExcel component to handle streaming read and write of millions of monitoring data points. The data storage and caching layer uses structured data storage units to store structured sample information, test results, and metadata; a high-speed caching unit is introduced to cache frequently queried statistical indicators (such as the national positivity rate and the list of the top 10 pathogens) to improve the loading speed of large screens. The core scheduling layer deploys a distributed data scheduling module, abandoning the traditional single-point polling mechanism. Instead, it slices the data extraction tasks from 31 provinces across the country according to region and time window, and distributes them to multiple computing nodes to execute ETL jobs in parallel.

[0097] The respiratory infection pathogen monitoring and analysis platform operates as follows: it retrieves data from the LIS (Limited Infectious Disease System) over the past 24 hours via the CDC or a scheduled interface, performs parallel cleaning and vectorization transformation, writes the data to a relational database, and simultaneously updates the cache unit. This mechanism ensures that the entire process, from sample sequencing to visualization on a large screen, is strictly controlled within 3 days.

[0098] For detailed introductions to each module in the infectious disease pathogen monitoring and analysis system, please refer to the relevant introductions in the Infectious Disease Pathogen Monitoring and Analysis Methods section, which will not be repeated here.

[0099] To verify the effectiveness of the monitoring and analysis system and algorithm proposed in this invention, we designed the following comparative experiment based on the platform's accumulated data of over 2 million respiratory tNGS tests covering 31 provinces and cities across the country.

[0100] Example 1: Verification of the accuracy of co-infection determination

[0101] Verification scenario: In response to the "co-infection" phenomenon that is prone to occur during the peak of influenza season, the system selected a sample set containing the detection of influenza A virus (Flu A) from the historical database, and used the traditional correlation analysis method (control group) and the bidirectional logistic regression model of the present invention (experimental group) to determine co-infection.

[0102] Control group (traditional method): Analysis was performed solely based on Pearson correlation coefficient. Results showed that influenza A virus exhibited a high "statistically positive correlation" with multiple pathogens, including respiratory syncytial virus (RSV) and adenovirus. However, after review by epidemiological experts, these "correlation" findings were largely due to "co-detection" resulting from the fact that both viruses are prevalent in winter (seasonal overlap), rather than genuine biological synergy. This resulted in a high false-positive rate and made it difficult to guide precise clinical medication.

[0103] Experimental group (method of this invention): The logistic regression model of this system was applied, and "temperature," "season," and "age" were included as covariates in the model for two-way validation. The system successfully identified and eliminated spurious correlation combinations caused by seasonal factors, ultimately selecting significant interaction combinations (such as the synergistic relationship between Haemophilus influenzae and Streptococcus pneumoniae). This demonstrates that the "two-way validation mechanism" of this invention can effectively isolate environmental confounding factors, elevating the analytical dimension from simple "co-detection statistics" to the level of "biological attribution," thereby improving the specificity and clinical reference value of co-infection early warning.

[0104] Example 2: Timeliness Verification of Trend Turning Point Early Warning

[0105] Taking the previous "Mycoplasma pneumoniae" epidemic as an example, we compare the early warning timeliness of the present invention with that of traditional methods.

[0106] Control group (traditional threshold method): The warning rule is set as "absolute value of weekly detection rate > 20%" or "number of positive cases per week > specific threshold".

[0107] Traditional methods typically trigger a red alert only when the epidemic curve has already reached a high level (mid-to-late stage of the outbreak). At this point, hospital outpatient clinics are often already under heavy load, resulting in a significant lag in the alert signal.

[0108] Experimental group (method of this invention): Employing the system's built-in adaptive piecewise linear fitting engine. Review results show that this algorithm automatically captures early structural abrupt changes in the epidemic curve using the ABIC criterion. It identifies significant abrupt changes in the slope and triggers signals during the early, rapid rise of the epidemic curve. Compared to traditional methods, this invention can significantly detect early signs of an outbreak, achieving a leap from "post-event response" to "pre-event warning."

[0109] Example 3: Data Governance Efficiency Verification

[0110] To verify the ability of this invention to process massive amounts of heterogeneous clinical data, 10,000 raw sample records from LIS systems of different primary hospitals were randomly selected. These records contained a large number of non-standard detection texts (such as "increased lung markings, suspected lung infection," "TB screening," etc.). Based on a distributed scheduling engine, the system achieved a second-level response time for batch data cleaning and vectorization, meeting the "T+3" real-time data entry requirement for massive datasets. Utilizing the built-in "keyword mapping dictionary" and "negation logic rule base," the system successfully identified negative prefixes such as "exclude...", avoiding misclassification of "exclude pulmonary tuberculosis" as "pulmonary tuberculosis." Compared to traditional manual rules or simple fuzzy matching, the accuracy and coverage of converting unstructured text into standard medical terms (such as normalizing "lung inflammation" to "pneumonia") were significantly improved, ensuring the purity of the data for subsequent statistical analysis.

[0111] like Figure 6 As shown, an embodiment of the present invention discloses an electronic device, including a memory 401 storing executable program code and a processor 402 coupled to the memory 401;

[0112] Specifically, the processor 402 calls the executable program code stored in the memory 401 to execute the infectious disease pathogen monitoring and analysis method based on targeted high-throughput sequencing described in the above embodiments.

[0113] This invention also discloses a computer-readable storage medium storing a computer program that causes a computer to execute the infectious disease pathogen monitoring and analysis method based on targeted high-throughput sequencing described in the above embodiments.

[0114] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0115] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing, characterized in that, include: According to the preset medical keyword mapping rules and negation logic rules, the unstructured clinical text of the sample is processed to obtain standard medical phenotype labels. The negation logic rules are used to avoid misjudging exclusionary diagnostic information as positive phenotype labels. Construct a sample feature vector, wherein the sample feature vector includes at least the targeted high-throughput sequencing results of the sample and the standard medical phenotypic label; For the sample feature vector, a bidirectional consistency check based on environmental noise decoupling is used to eliminate spurious correlation signals introduced by environmental confounding factors and obtain the pathogen co-infection determination result.

2. The method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing as described in claim 1, characterized in that, After obtaining the result of pathogen co-infection determination, time series data of the detection rate of the target pathogen are also obtained. Adaptive piecewise linear fitting is performed on the time series data to identify trend turning points and calculate the trend change index of the latest time interval. When the trend change index meets the preset warning conditions, an early warning is issued.

3. The method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing as described in claim 2, characterized in that, The process of adaptively segmenting and linearly fitting the time series data, identifying trend turning points, calculating the trend change index for the latest time interval, and issuing an early warning when the trend change index meets preset early warning conditions includes: Traverse all possible breakpoints in the time series data and combine them to obtain several combinations of breakpoint positions; Construct piecewise linear regression models corresponding to each combination of the aforementioned breakpoint positions; Calculate the evaluation index for each piecewise linear regression model, determine the optimal piecewise linear regression model based on the evaluation index, and determine the trend inflection point based on the optimal piecewise linear regression model; Based on the trend turning point, the time series data is divided into multiple time intervals; Calculate the annual percentage change and slope for the latest time interval; An early warning is issued when the slope exceeds a preset threshold and the annual change percentage shows an accelerating trend.

4. The method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing as described in claim 1, characterized in that, The process of processing the unstructured clinical text of the sample according to preset medical keyword mapping rules and negation logic rules to obtain standard medical phenotypic labels includes: Pre-construct medical keyword mapping rules that include clinical symptoms; Pre-construct negation logic rules that include negation words and distance constraints; The unstructured clinical text is segmented and scanned. When a keyword is identified according to the medical keyword mapping rule, the characters within the distance constraint range are traced back according to the negation logic rule. If a negative word exists in the characters, the keyword is discarded; otherwise, the keyword is mapped to the standard medical phenotype label.

5. The method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing as described in claim 1, characterized in that, After constructing the sample feature vector, the process also includes: performing a merging operation on multiple sample feature vectors of the same patient identifier within a preset time window.

6. The method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing as described in claim 1, characterized in that, The process involves using a bidirectional consistency check based on environmental noise decoupling for the sample feature vector to remove spurious correlation signals introduced by environmental confounding factors, thereby obtaining the pathogen co-infection determination result, including: For the pathogen pair to be analyzed, including the first pathogen and the second pathogen, a first logistic regression model is constructed with the first pathogen as the dependent variable and the second pathogen as the independent variable, and a second logistic regression model is constructed with the second pathogen as the dependent variable and the first pathogen as the independent variable. In both the first logistic regression model and the second logistic regression model, the environmental confounding factor is used as a covariate. When the statistical significance of the core independent variables in both the first and second logistic regression models is less than the preset significance threshold, and the signs of the regression coefficients in the first and second logistic regression models are consistent, the first pathogen and the second pathogen are determined to have a true co-infection relationship and the result is saved to the pathogen co-infection determination result.

7. The method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing as described in claim 6, characterized in that, After obtaining the pathogen co-infection determination result, the topological relationship data of pathogen co-infection is generated based on the pathogen co-infection determination result, and a co-infection network map that is different from the pseudo-correlation of the environment is displayed.

8. A system for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing, characterized in that, include: The data standardization module is used to process the unstructured clinical text of the sample according to the preset medical keyword mapping rules and negation logic rules to obtain standard medical phenotype labels. The negation logic rules are used to avoid misjudging exclusionary diagnostic information as positive phenotype labels. A multidimensional feature construction module is used to construct a sample feature vector, wherein the sample feature vector contains at least the targeted high-throughput sequencing results of the sample and the standard medical phenotypic label; The co-infection determination module is used to perform a bidirectional consistency check based on environmental noise decoupling on the sample feature vector, eliminate spurious correlation signals introduced by environmental confounding factors, and obtain the pathogen co-infection determination result.

9. An electronic device, characterized in that, It includes a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the infectious disease pathogen monitoring and analysis method based on targeted high-throughput sequencing as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program causes a computer to execute the method for monitoring and analyzing infectious disease pathogens based on targeted high-throughput sequencing as described in any one of claims 1-7.

Images