Mycobacterium tuberculosis drug resistance gene detection data analysis system

Abstract

The invention relates to the technical field of gene detection, and discloses a mycobacterium tuberculosis drug resistance gene detection data analysis system which comprises a data import module, a signal correction module, a feature extraction module, a multi-stage interpretation module, a result fusion module and a report generation module. The system performs baseline drift correction, abnormal point detection and nonlinear normalization on a fluorescence intensity value output by a detector, extracts multi-dimensional features such as peak intensity, a mutation ratio, a consistency coefficient and an in-batch difference index, and constructs a dynamic threshold function to realize wild type and mutation type layered judgment. A drug resistance risk conclusion is generated through drug resistance characteristic matrix mapping, a standardized detection report is output, and intelligent analysis of a detection result and automatic report generation are achieved. Through multi-stage signal correction, multi-dimensional feature extraction and dynamic threshold interpretation, intelligent analysis and standardized report output of a mycobacterium tuberculosis drug resistance gene detection result are realized, and data judgment accuracy and clinical application efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of gene detection technology, and more specifically, to a data analysis system for detecting drug resistance genes in Mycobacterium tuberculosis. Background Technology

[0002] Tuberculosis (TB) is a serious chronic infectious disease caused by Mycobacterium tuberculosis (MTB), posing a significant threat to human health and being one of the major global public health problems. Due to the irrational use of anti-tuberculosis drugs, drug resistance in TB is becoming increasingly serious. The prevalence of multidrug-resistant (MDR) and extensively drug-resistant (XDR) TB is considered one of the most challenging threats to TB control. Rapid and accurate detection of drug resistance mutations is crucial for clinical treatment.

[0003] Currently, there are various methods for detecting drug resistance genes in Mycobacterium tuberculosis, including quantitative real-time PCR and next-generation sequencing (NGS). Among them, the multiplex detection platform based on microsphere arrays has the following advantages: High throughput: It can detect multiple targets simultaneously, improving detection efficiency; High sensitivity: capable of detecting analytes at low abundance; High specificity: accurately identifies and detects targets; Easy to operate: The experimental procedure is relatively simple and easy to standardize.

[0004] However, the data generated by multi-detection platform technology based on microsphere arrays is complex, involving the analysis of fluorescence signals from multiple microspheres. Existing technologies have the following problems: 1) Data analysis is complex, relies on human experience, is highly subjective, prone to errors, and is difficult to analyze large batches of samples; 2) The software accompanying microsphere array-based multiplex analyzers typically only performs simple threshold calculations and preliminary positive / negative results, lacking dedicated analytical logic for detecting Mycobacterium tuberculosis drug resistance genes. For example, it cannot intelligently handle the relationship between "wild-type" and "mutant" types, nor can it integrate mutation results from multiple sites into a single test report that meets clinical needs. 3) Inconsistent result interpretation standards make it difficult to achieve standardized operations; 4) The report generation efficiency is low, making it difficult to meet the needs of rapid clinical diagnosis.

[0005] Therefore, the purpose of this invention is to solve the above problems and provide a system or method for analyzing drug resistance gene detection data of Mycobacterium tuberculosis. Summary of the Invention

[0006] This invention provides a data analysis system for detecting drug resistance genes in Mycobacterium tuberculosis to solve the above-mentioned problems.

[0007] On the one hand, the Mycobacterium tuberculosis drug resistance gene detection data analysis system provided by the present invention includes: The module includes a data import module, a signal correction module, a feature extraction module, a multi-level interpretation module, a result fusion module, and a report generation module. The signal correction module is used to perform multi-stage correction on the original fluorescence intensity value MFI, including baseline drift correction, outlier detection and nonlinear normalization; The feature extraction module is used to extract multi-dimensional features based on the MFI data after multi-stage correction. The multi-dimensional features include peak intensity, mutation ratio, probe consistency coefficient and intra-batch difference index. The multi-level interpretation module constructs a dynamic threshold function based on multi-dimensional features to perform stratified determination of whether the corresponding nucleic acid sequence of the detection target is wild-type or mutant. The result fusion module maps the results of each detection site to the drug category by combining the drug resistance feature matrix and outputs a drug resistance risk warning. The report generation module generates standardized clinical test reports based on the fusion results.

[0008] Furthermore, the signal correction module includes: The baseline drift correction unit is used to perform linear benchmark fitting on the MFI distribution of different detection batches; Anomaly detection unit is used to determine sudden intensity anomalies based on local statistical deviation ratio and perform local smoothing. The nonlinear normalization unit is used to unify the signal amplitude in each detection system to the 0-1 range using a piecewise Sigmoid function, so as to eliminate the difference in detection sensitivity between different batches.

[0009] Furthermore, in the feature extraction module, the mutation ratio is the ratio of the mutant signal to the corresponding wild-type signal, the consistency coefficient is the inverse ratio of the difference between adjacent probe signals under the same detection system, and the intra-batch difference index is the ratio of the standard deviation to the mean of all probe signals in the sample.

[0010] Furthermore, the multi-level interpretation module includes three operational phases: (1) In the signal layer determination stage, the basic threshold determination result is calculated based on the mutation ratio and peak intensity; (2) In the structural layer determination stage, the signal reliability is determined based on the consistency coefficient and the intra-batch difference index. When the reliability is lower than the set lower limit, it is marked as suspicious. (3) During the logic layer synthesis stage, conflict analysis is performed on the results of the signal layer and the structure layer. If the conflict ratio exceeds 20%, a second judgment is made and an anomaly report is generated.

[0011] Furthermore, the result fusion module performs drug category mapping based on the drug resistance feature matrix, where the rows of the matrix represent the detected genes, the columns represent the drug categories, and the matrix elements are the mutation influence weights. The comprehensive drug resistance score for each drug is calculated through matrix multiplication, and the drugs are divided into three categories: sensitive, suspected, and resistant, based on the score range.

[0012] Furthermore, it also includes: The abnormal sample correction module is used to identify samples whose deviation from the characteristics of historical samples exceeds a set threshold in the secondary judgment, and to perform numerical correction on the deviation site by signal re-estimation based on the neighborhood sample set. After correction, the data re-enters the signal layer judgment stage.

[0013] Furthermore, it also includes: The cross-system consistency verification module calculates the signal consistency rate between the systems and performs weight reallocation when the detection results of the same sample are inconsistent in the KIE system and the GR system.

[0014] Furthermore, the signal consistency rate between systems is calculated and weights are reallocated, including: (1) Match the fluorescence signals of the detection sites in the two systems one-to-one according to the detection site number; (2) Calculate the signal consistency rate of the two systems at each detection site. The signal consistency rate is 1 minus the ratio of the signal difference between the two systems to the maximum signal value. When the consistency rate is less than the set threshold, it is marked as an inconsistent site. (3) For inconsistent sites, calculate the weight value based on the signal-to-noise ratio and coefficient of variation of the two systems. The weight value is the signal-to-noise ratio divided by the sum of the signal-to-noise ratio and the coefficient of variation. (4) Recalculate the fused signal based on the weight values. The fused signal is the sum of the KIE system signal multiplied by the KIE weight and the GR system signal multiplied by the GR weight. (5) Input the fused signal as the correction result into the feature extraction module and the multi-level interpretation module.

[0015] On the other hand, the present invention also provides a method for analyzing Mycobacterium tuberculosis drug resistance gene detection data using the above-mentioned system, comprising the following steps: (1) Import the original fluorescence intensity value MFI and perform signal correction; (2) Extract peak intensity, mutation ratio, consistency coefficient and intra-batch difference index; (3) Output wild-type and mutant results based on the three-stage interpretation logic; (4) Construct a drug resistance feature matrix and calculate the comprehensive drug resistance score; (5) Generate standard test reports and drug resistance spectrum reasoning reports.

[0016] In another aspect, the present invention also protects a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the above-described method.

[0017] The present invention has the following technical effects: By constructing a comprehensive analysis system encompassing multi-stage signal correction, feature parameter extraction, multi-level interpretation, and result fusion, the system automates the processing of multiplex detection data based on microsphere arrays from raw signals to clinical reports. This system can perform baseline drift correction, anomaly identification, and nonlinear normalization of fluorescence signals, significantly reducing batch-to-batch variations and human interpretation errors. Through the joint extraction of multi-dimensional features such as peak intensity, mutation ratio, consistency coefficient, and intra-batch variation index, the system can comprehensively characterize signal significance and reliability. Utilizing a dynamic threshold function and hierarchical judgment logic, it can adaptively distinguish between wild-type and mutant signals. Based on a result fusion mechanism using a drug resistance feature matrix, multi-site mutation information can be mapped to drug resistance risk, enabling quantitative determination of multidrug resistance profiles. Combined with cross-system consistency verification and anomaly sample correction mechanisms, the system effectively improves the stability, repeatability, and clinical interpretability of detection results. Overall, this system can intelligently analyze and generate standardized reports of Mycobacterium tuberculosis drug resistance gene data without human intervention, improving detection efficiency and clinical application value. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a framework diagram of a Mycobacterium tuberculosis drug resistance gene detection data analysis system provided in an embodiment of the present invention; Figure 2 A flowchart illustrating a method for analyzing drug resistance gene detection data in Mycobacterium tuberculosis, provided in an embodiment of the present invention; Figure 3 This is the resistance result of sample No. 1 in the embodiments of the present invention to isoniazid, rifampin and quinolones; Figure 4 This is the resistance result of sample No. 2 to isoniazid and rifampin in the embodiment of the present invention; Figure 5 This is the resistance result of sample No. 3 to isoniazid and ethambutol in the embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] See Figure 1 As shown, this embodiment of the invention provides a data analysis system for detecting drug resistance genes in Mycobacterium tuberculosis, comprising: The module includes a data import module, a signal correction module, a feature extraction module, a multi-level interpretation module, a result fusion module, and a report generation module. The signal correction module is used to perform multi-stage correction on the original fluorescence intensity value MFI, including baseline drift correction, outlier detection and nonlinear normalization; The feature extraction module is used to extract multidimensional features based on the multi-stage corrected MFI data. The multidimensional features include peak intensity, mutation ratio, probe consistency coefficient and intra-batch difference index. The multi-level interpretation module constructs a dynamic threshold function based on multi-dimensional features to perform hierarchical determination of wild-type and mutant types; The results fusion module maps the results of each detection site to the drug category by combining the drug resistance feature matrix and outputs a conclusion on drug resistance risk. The report generation module generates standardized clinical test reports based on the fusion results.

[0023] Specifically, the data import module is used to read the raw detection files output by the device. These files include the fluorescence intensity value (MFI) of each microsphere probe, probe number, sample number, detection batch number, and channel information. During import, the module standardizes the data format and performs checks for field integrity, duplicate samples, and type validity to ensure the structural consistency and parsability of the input data, establishing a unified data foundation for subsequent analysis.

[0024] The signal correction module performs multi-stage correction on the raw MFI signal, including baseline drift correction, outlier detection, and nonlinear normalization. Baseline drift correction eliminates the overall offset caused by instrument drift by fitting a linear benchmark to the MFI distribution of different detection batches. Outlier detection identifies sudden intensity anomalies based on local statistical deviation ratios and performs smoothing repair, avoiding false signals caused by optical path interference or bubbles. Nonlinear normalization uses a piecewise sigmoid function to map the signal amplitude to the 0–1 range, achieving sensitivity uniformity across different batches and channels. The signal data after multi-stage correction maintains relative differences and is comparable.

[0025] The feature extraction module takes the corrected MFI data as input and extracts multidimensional features that reflect signal significance and stability, including peak intensity, mutation ratio, probe consistency coefficient, and intra-batch difference index. Peak intensity is the maximum effective signal value at the detection site, used to characterize the significance of the hybridization response; the mutation ratio is the ratio of mutant to wild-type signals, used to measure the intensity of mutation; the probe consistency coefficient reflects the stability of adjacent probe signals within the same system; and the intra-batch difference index is the ratio of the signal standard deviation to the mean, used to identify the degree of dispersion within the system. The feature vector output by this module serves as the input for subsequent interpretation logic.

[0026] The multi-level interpretation module constructs a dynamic threshold function based on the aforementioned multi-dimensional features to perform tiered judgment of wild-type and mutant types for each detection site. Its logic comprises three levels: at the signal level, a basic threshold judgment result is generated based on the peak intensity and mutation ratio; at the structural level, signal reliability is calculated based on the consistency coefficient and intra-batch difference index, and low-confidence sites are marked; at the logic level, conflict analysis is performed on the results of the first two levels, and a secondary judgment is automatically triggered when the conflict ratio exceeds 20%, recalculating features for suspicious sites and outputting anomaly alerts, thereby improving the accuracy and stability of mutation judgment.

[0027] The results fusion module integrates the site results from multi-level interpretation. A drug resistance feature matrix is ​​pre-established, where rows represent detected genes, columns represent drug categories, and matrix elements are the weights of mutations on drug resistance. The mutation result vector is multiplied by the matrix to obtain the comprehensive drug resistance score for each drug. Based on the score range, the drugs are categorized into three types: sensitive (<0.4), suspected (≥0.4 and <0.7), and resistant (≥0.7), forming a quantitative conclusion on drug resistance risk. It should be noted that the range divisions for sensitive, suspected, and resistant drugs can be adjusted according to the actual situation.

[0028] The report generation module automatically generates standardized test reports based on the fusion results. The report content includes sample information, test batch, point mutation status and confidence level, drug resistance classification results and comprehensive conclusions. The output format can be directly connected to the laboratory information management system to realize automatic archiving and clinical sharing of results.

[0029] In some embodiments of this application, the signal correction module includes: The baseline drift correction unit is used to perform linear benchmark fitting on the MFI distribution of different detection batches; Anomaly detection unit is used to determine sudden intensity anomalies based on local statistical deviation ratio and perform local smoothing. The nonlinear normalization unit is used to unify the signal amplitude in each detection system to the 0-1 range using a piecewise Sigmoid function, so as to eliminate the difference in detection sensitivity between different batches.

[0030] Specifically, the signal correction module is implemented as follows: First, after data import, the baseline drift correction unit is automatically invoked. This unit takes the original MFI signal of each batch as input, establishes a linear fitting model of the batch signal distribution using the least squares method, and selects the average signal of the standard reference sample as the global baseline. It calculates the average difference ΔB between each batch's signal and the baseline, and subtracts this offset from all signals in that batch, making the signal baselines of different batches numerically consistent. For example, if the average MFI value of a batch is 100 units higher than the baseline sample, all signals in that batch are shifted down by 100 units to eliminate the offset caused by instrument drift or reagent batch differences.

[0031] Secondly, the anomaly detection unit operates on the baseline-corrected data. Using the probe sequence as the analysis window, the difference between each probe signal and the mean of its two adjacent probe signals is calculated. Then, the standard deviation σ of this local signal set is determined. When a probe signal deviates from the local mean by more than 2σ, it is identified as a sudden anomaly. For detected anomalies, a sliding window smoothing method is used for repair. This involves replacing the anomaly with the weighted average of the two probe signals to the left and right of the point. The weights are determined based on the probe spacing and signal similarity. This process effectively eliminates spike signals caused by bubbles, optical path jitter, or uneven microsphere distribution without disrupting the overall signal trend.

[0032] Finally, the non-linear normalization unit performs piecewise Sigmoid function normalization on the smoothed signal. First, the signal intensity distribution range of each system is statistically analyzed, and three piecewise thresholds, namely the low interval, the middle interval, and the high interval, are set. In the low interval, a Sigmoid function with a larger amplification factor is used to enhance the ability to distinguish weak signals; in the middle interval, an approximate linear relationship is maintained to preserve the signal ratio; in the high interval, a smoothing compression function is used to prevent strong signal saturation. The normalized signal is mapped to the 0-1 interval to ensure the comparability of signal intensities between different detection systems. For example, low-intensity signals can still reflect subtle differences after normalization, while high-intensity signals will not mask the change trend due to saturation.

[0033] It should be noted that the setting of the three piecewise thresholds is a data-driven and dynamically correctable process in this application. It is not a fixed constant but is automatically determined based on the statistical results of batch signal distribution, the sensitivity characteristics of the detection system, and the sample quantity. The setting logic and implementation steps are as follows: At the beginning of each detection batch, the system first performs statistical analysis on the original MFI values of all detection probes to form a batch signal intensity distribution histogram. Based on this distribution, the system calculates the percentiles of the signal intensity, where the 10th percentile signal intensity P10, the 50th percentile signal intensity P50, and the 90th percentile signal intensity P90 are used as the initial statistical references. Subsequently, according to the instrument calibration curve and the response range of the detection system, the system automatically corrects these three percentile values to avoid the interference of abnormal samples. The correction coefficient is usually determined based on the relative magnitude of the signal variance. For example, when the variance in the high-signal region exceeds 1.5 times the global variance, the upper threshold P90 is increased by 5%; conversely, if the variance in the low-signal region is significantly amplified, the lower threshold P10 is decreased by 5%.

[0034] After determining the final thresholds, the system defines the low interval as MFI < P10, the middle interval as P10 ≤ MFI ≤ P90, and the high interval as MFI > P90. The low interval mainly contains background signals and low-abundance mutant probe signals. Here, an enhanced function with a larger Sigmoid function parameter β is used to expand the differences between weak signals, so that probes close to the background still have distinguishability; the middle interval is the part where the signal linearly changes most concentratedly, and the Sigmoid function parameter β is approximately equal to 1, and the output maintains an approximate linear relationship with the input, which is used to maintain the signal ratio relationship; the high interval is the saturation region. To prevent strong signals from overly occupying the ratio and causing numerical distortion, the Sigmoid function uses a smaller compression parameter β (such as 0.3–0.5) to achieve smoothing compression.

[0035] Furthermore, the system automatically updates these three thresholds across multiple consecutive testing batches. When the number of samples in a testing batch exceeds 50, the system will use a moving average method to dynamically and smoothly update P10, P50, and P90, avoiding sudden changes in the normalization interval caused by a single abnormal distribution. For extreme samples (such as all-positive or all-negative batches), the system automatically calls historical reference distribution templates for replacement, thereby ensuring that the division of low, medium, and high intervals is stable and repeatable.

[0036] Through this threshold setting mechanism based on statistical percentiles and dynamic correction, the piecewise sigmoid function can adapt to the differences in sensitivity between different detection batches and systems, achieving enhanced resolution of weak signals, suppressed compression of strong signals, and cross-batch comparability of the overall signal. The final output normalized signal can reflect subtle differences in the 0–1 range while avoiding high signal saturation, effectively improving the quantitative consistency among multiple systems and the reliability of mutation identification.

[0037] Through the collaborative processing of the three units mentioned above, the signal correction module can effectively correct batch offsets, eliminate local anomalies, and unify amplitude scales without altering the true physical meaning of the detected signal, generating signal data with high stability and statistical consistency. The result processed by this module is directly input into the feature extraction module. In some embodiments of this application, the mutation ratio in the feature extraction module is the ratio of the mutant signal to the corresponding wild-type signal, the consistency coefficient is the inverse ratio of the difference between adjacent probe signals under the same detection system, and the intra-batch difference index is the ratio of the standard deviation to the mean of all probe signals in the sample.

[0038] Specifically, the feature extraction module is used to perform multidimensional parameter calculations on the MFI data after signal correction processing, in order to achieve quantitative characterization of the significance, stability, and consistency of the detection signal. The implementation process of this module includes three stages: mutation ratio calculation, consistency coefficient calculation, and intra-batch difference index calculation.

[0039] First, in the mutation ratio calculation stage, signal data are categorized into mutant probes and wild-type probes based on probe type, and paired within the same sample. The normalized MFI signal intensities of mutant and wild-type probes are read separately, and their ratio is calculated. If the mutant signal is significantly higher than the wild-type signal, a mutation trend is considered possible at that site; if the wild-type signal is stronger than the mutant signal, the site is wild-type. To prevent bias caused by low or abnormal signals, data with signal intensities below the noise lower limit are automatically filtered out before calculation, and abnormally high values ​​are truncated. This step yields mutation ratio results that reflect the significance of mutation characteristics.

[0040] Secondly, in the consistency coefficient calculation stage, the signal differences between adjacent probes are statistically analyzed according to the probe arrangement order within the detection system, and the overall average difference is calculated. The overall smoothness of the probe signals is compared with the average signal intensity to obtain the channel consistency coefficient. If the coefficient is close to 1, it indicates that the channel signal is continuous and has strong response regularity, and the system detection is stable; if the coefficient drops significantly, it indicates that there are discontinuous or abnormal fluctuation areas in the signal. To avoid interference from probe signals at the channel edges with the overall judgment, probes located at both ends are assigned lower weights, and the final consistency coefficient is calculated through weighted averaging. This index is used to comprehensively evaluate the stability of the detection system and the integrity of local signals.

[0041] Secondly, during the intra-batch difference index calculation stage, statistical analysis is performed on all probe signals of the same sample. The average value and fluctuation range of the signals are calculated separately, and the ratio between the two is taken as the difference index. This index is used to reflect the signal dispersion of the sample within the same testing batch. When the intra-batch difference index is small (e.g., below 0.15), it indicates that the signal distribution of the batch is uniform and the system is stable; when the difference index is high (e.g., above 0.3), it indicates that the detection system fluctuates greatly, and the sample will be marked as a high-variability sample, and its calculation weight will be reduced or a review will be triggered in the subsequent interpretation stage.

[0042] After calculating the three types of features, the mutation ratio, consistency coefficient, and intra-batch difference index are normalized to map their values ​​uniformly to the range of zero to one, ensuring scale consistency among different parameters. Subsequently, these three types of features are combined into a feature vector according to the sample number and the detection site number, which serves as the input to the multi-level interpretation module for dynamic threshold construction and stratified judgment.

[0043] In some embodiments of this application, the multi-level interpretation module includes: (1) In the signal layer determination stage, the basic threshold determination result is calculated based on the mutation ratio and peak intensity; (2) In the structural layer determination stage, the signal reliability is determined based on the consistency coefficient and the intra-batch difference index. When the reliability is lower than the set lower limit, it is marked as suspicious. (3) During the logic layer synthesis stage, conflict analysis is performed on the results of the signal layer and the structure layer. If the conflict ratio exceeds 20%, a second judgment is made and an anomaly report is generated.

[0044] Specifically, the multi-level interpretation module is used to perform hierarchical analysis and comprehensive judgment on the detection data after signal correction and feature extraction, ensuring the accuracy and stability of Mycobacterium tuberculosis drug resistance gene mutation identification. The entire process includes a signal layer judgment stage, a structure layer judgment stage, and a logic layer synthesis stage, achieving multi-factor fusion judgment through step-by-step calculation, cross-validation, and feedback correction.

[0045] First, in the signal layer determination stage, each detection site is initially classified using mutation ratio and peak intensity as the core input parameters. A basic threshold range is pre-established based on historical sample statistics, and the signal difference between mutant and wild-type probes in the current sample is calculated. When the mutation ratio is significantly higher than the threshold and the peak intensity exceeds the minimum significance limit, the site is determined to be mutant; conversely, if the mutation ratio is lower than the threshold and the peak intensity is insufficient, it is determined to be wild-type. When the detection signal is in the critical range, it is marked as "pending" and verified in the next layer of determination. This stage is equivalent to quantitative screening of single-point signal intensity, ensuring the initial distinction between mutant and non-mutant signals.

[0046] Secondly, in the structural layer determination stage, a consistency coefficient and an intra-batch variation index are introduced to verify the reliability of the signal layer determination results. First, the continuity and fluctuation levels of all probe signals under the same detection system are calculated and compared with a standard range to obtain a signal reliability score. When the reliability is higher than a set threshold (e.g., 0.8), it indicates that the detection system signal is stable, and the signal layer determination result can be directly accepted. When the reliability is lower than the threshold, it indicates that there may be microsphere aggregation, localized uneven reaction, or batch fluctuations in the system. The sample is marked as "suspicious," and its weight is reduced or transferred to the logic layer for secondary analysis. The purpose of this stage is to eliminate misjudgments caused by system instability and ensure the statistical reliability of the input data.

[0047] Secondly, during the logic layer synthesis stage, conflict detection and consistency verification are performed on the judgment results of the signal layer and the structure layer. The proportion of inconsistent judgment results between the two layers at all detection points is statistically analyzed. When the conflict proportion is below 20%, the judgment is considered stable, and the final result is directly output. When the conflict proportion exceeds 20%, a secondary judgment procedure is triggered, recalculating the mutation ratio and signal reliability for the conflict points, and performing weighted adjustments based on the statistical distribution of historical samples. After the secondary judgment is completed, an anomaly report is generated, recording the conflict point number, signal strength, ratio parameters, and judgment results before and after correction, for manual verification or report traceability.

[0048] Through the combined operation of the above three stages, the signal layer ensures a sensitive response to individual sites, the structure layer guarantees the overall reliability of the system signal, and the logic layer realizes conflict coordination and dynamic correction of cross-layer results. This multi-level interpretation mechanism can automatically adjust the judgment logic under conditions of complex signals, system fluctuations, or uneven sample quality, thereby significantly improving the accuracy, repeatability, and robustness of mutation identification.

[0049] In some embodiments of this application, the result fusion module performs drug category mapping based on the drug resistance feature matrix, where the rows of the matrix represent the detected genes, the columns represent the drug categories, the matrix elements are the mutation influence weights, the comprehensive drug resistance score of each drug is calculated by matrix multiplication, and the drugs are divided into three categories: sensitive, suspected, and resistant according to the score range.

[0050] Specifically, the results fusion module integrates the site-level results output by the multi-level interpretation module into drug category-level drug resistance determination conclusions, realizing the transformation from "mutation detection" to "clinical drug risk analysis." This module maps the correspondence between detected genes and drug categories using a pre-established drug resistance feature matrix, and calculates and outputs a comprehensive drug resistance score for each drug using the matrix. Its implementation steps and logic are as follows: First, a drug resistance feature matrix is ​​constructed based on known drug resistance associations of Mycobacterium tuberculosis. The rows of this matrix correspond to the detected genes or mutation sites (e.g., katG315, inhA-15, rpoB531, etc.), and the columns correspond to different anti-tuberculosis drug classes (e.g., isoniazid, rifampin, ethambutol, quinolones, etc.). The matrix elements are "mutation impact weights," representing the contribution of each gene mutation to the resistance of the corresponding drug. The weight values ​​are derived from multiple layers of information: first, confirming the relationship between mutations and drug resistance in publicly available databases (e.g., TB-Profiler or the WHO mutation database); second, statistically analyzing the correlation between the frequency of mutations and drug resistance phenotypes in samples within the statistical system; and third, introducing a cross-correction coefficient to avoid duplicate scoring for the same mutation site across multiple drugs. All weights are normalized and limited to between 0 and 1, forming a computable matrix structure.

[0051] Secondly, based on the mutation determination results output by the multi-level interpretation module, a mutation state vector is generated. This vector is arranged in matrix row order; if a site is determined to be mutant, it is assigned a value of 1, and if it is determined to be wild-type, it is assigned a value of 0. This mutation state vector is then multiplied with the drug resistance feature matrix to obtain a comprehensive drug resistance score for each drug. This score represents the probability of drug resistance in the sample under a specific drug; the higher the score, the greater the risk of drug resistance.

[0052] Next, the calculated comprehensive drug resistance score is graded. Based on the distribution of long-term sample data, three threshold intervals are set: scores below 0.4 are considered "sensitive," scores between 0.4 and 0.7 are considered "suspected," and scores above 0.7 are considered "resistant." When a drug in a sample corresponds to multiple mutation sites, the scores are automatically accumulated and weighted averaged to reflect the comprehensive impact of the synergistic effect of mutations. If a key resistance site mutation exists in the sample (such as the decisive role of rpoB531 in rifampicin resistance), the weights are automatically adjusted to increase the contribution of this mutation to the total score, thus maintaining clinical relevance.

[0053] Then, the drug resistance scores of each drug are aggregated to generate a drug resistance spectrum. This table is structured with drug category as the row and drug resistance score, determination level, and key mutation sites as the column, forming a structured output. Based on this, confidence analysis is performed. If the scores of some drugs are close to the threshold boundary, they are marked as "drugs to be reviewed" to indicate the possible existence of boundary mutations or signal fluctuations.

[0054] Finally, the fusion module transmits the drug resistance spectrum results to the report generation module. The report will clearly list the determination level of each drug, the main influencing mutation sites, and the confidence value, allowing clinicians to interpret the information directly.

[0055] Through the above implementation steps, the results fusion module achieves a systematic mapping from site-level data to drug resistance conclusions. The row and column structure of the resistance feature matrix ensures the algorithm's universality and scalability, while the weight allocation mechanism ensures the scientific quantification of the contribution of different mutations to drug resistance. This module can not only quantify the resistance risk of a single drug but also achieve multi-drug linkage analysis through matrix expansion, supporting subsequent clinical medication decisions and epidemiological data accumulation, significantly improving the clinical interpretive value and decision-making reference significance of the test results.

[0056] In some embodiments of this application, it also includes: The abnormal sample correction module is used to identify samples whose deviation from the characteristics of historical samples exceeds a set threshold in the secondary judgment, and to perform numerical correction on the deviation site by signal re-estimation based on the neighborhood sample set. After correction, the data re-enters the signal layer judgment stage.

[0057] Specifically, the abnormal sample correction module is used to identify, analyze, and correct sample data that deviates from the normal statistical distribution during the detection process, ensuring that the interpretation can still output accurate and stable results under conditions of signal drift, batch-to-batch differences, or occasional anomalies. This module forms a closed-loop feedback mechanism with the multi-level interpretation module, achieving adaptive correction of abnormal signals through feature comparison of neighboring samples. Its implementation steps are as follows: I. Abnormal Sample Identification Stage (1) Extract the signal feature set of all detected samples from the secondary judgment results, including peak intensity, mutation ratio, structural consistency coefficient and noise baseline amplitude; (2) Calculate the feature deviation of each sample from the same category of samples in the historical sample library. The deviation is defined as the distance between the target sample and the mean vector of the reference sample in the multidimensional feature space. (3) When the deviation exceeds the set threshold (e.g., mean ± 2.5 times standard deviation), the sample is marked as an “abnormal sample”; if the deviation of a single site exceeds the threshold but the overall sample is still within the acceptable range, it is marked as a “local abnormal site”. (4) Output the abnormal sample and abnormal site information to the correction queue and wait for signal re-evaluation processing.

[0058] It should be noted that the feature deviation is used to measure the degree of difference between a sample and historical normal samples in multiple feature parameters (such as signal strength, ratio, morphology, etc.). The larger the value, the more obvious the difference between the sample and the normal distribution.

[0059] Multidimensional feature space refers to a mathematical space composed of multiple dimensions used to characterize signal features, such as a coordinate system jointly formed by signal strength, noise level, peak width, and symmetry. II. Neighborhood Sample Set Construction Stage (1) Select samples from the historical sample library that are similar to the abnormal sample detection conditions (e.g., the same batch, the same equipment, the same template area) to form a neighborhood sample set; (2) Calculate the mean μ of each site in the neighborhood sample. n and variance σ n This allows us to obtain the signal distribution range of this type of sample under normal conditions. (3) If the number of neighborhood samples is insufficient (e.g., less than 10), the time window will be automatically expanded or the similarity requirement will be relaxed to ensure the statistical reliability of the neighborhood sample set.

[0060] It should be noted that the neighborhood sample set refers to a group of historical samples that are most similar to the target sample in terms of detection conditions, sample source, and reaction system, and is used to provide normal reference signal characteristics.

[0061] Similarity requirements determine whether two samples are similar, such as identical testing batches, signal amplitude difference less than 20%, peak width difference less than 10%, etc.

[0062] III. Signal Reassessment Phase (1) Perform signal reestimation on the marked anomalous sites. The reestimation formula is: Correction value = Original signal value × (Neighborhood mean / Outlier sample mean) This operation adjusts the abnormal signal to the normal statistical range of the neighboring samples by scaling the signal. (2) For noise-interference type anomalies (i.e. signal jitter or peak drift), the moving average filtering algorithm is used to recalculate the local baseline and smooth the abnormal fluctuations; (3) When signal loss or oversaturation is detected, the signal is completed by neighborhood sample interpolation algorithm, such as weighted average or nearest neighbor interpolation. (4) The re-evaluated signal data is saved as a corrected version, along with a correction identifier and correction parameter record, for future reference.

[0063] It should be noted that signal re-estimation refers to recalculating or correcting the deviated signal based on the statistical characteristics of the neighboring normal samples when abnormal samples appear, so that it returns to a reasonable range.

[0064] Moving average filtering is a signal smoothing algorithm that reduces short-term fluctuations or noise by averaging over consecutive time points. It will not be elaborated on further here.

[0065] IV. Re-determination stage (1) The corrected signal data is re-input into the signal layer determination module to recalculate the mutation ratio, peak intensity and basic threshold determination results; (2) If the corrected signal is consistent with the normal distribution, update the original judgment result and replace the original abnormal record; (3) If significant deviations still exist after correction, a manual review will be automatically triggered, and the system will enter the quality control queue; (4) Finally, a correction report is generated, which records the original data, correction method, neighborhood sample parameters and correction results for subsequent quality tracking and method verification.

[0066] The abnormal sample correction module achieves dynamic self-correction of detection data through feature deviation detection, neighborhood sample comparison, and signal reestimation. This module is designed to address the "discard upon single abnormality" problem in traditional drug resistance gene detection, enabling the repair and reuse of abnormal samples. Through statistical constraints on neighborhood samples, it can automatically adjust abnormal signals without human intervention, giving the overall detection process self-correction, data consistency, and high repeatability. This mechanism significantly reduces the false positive rate and improves the stability and robustness of the detection system under complex sample and multi-batch conditions.

[0067] In some embodiments of this application, it also includes: The cross-system consistency verification module calculates the signal consistency rate between the systems and performs weight reallocation when the detection results of the same sample are inconsistent in the KIE system and the GR system.

[0068] Specifically, the KIE system is for detecting resistance-related genes of isoniazid and ethambutol, while the GR system is for detecting resistance-related genes of quinolones and rifampin. The cross-system consistency verification module is used to assess the consistency of signal data between different detection systems (such as the KIE and GR systems) and dynamically adjust weights when inconsistencies occur in the detection results of the same sample, ensuring the stability and reliability of the final mutation determination results. The implementation steps are as follows: I. Results Comparison Phase (1) After completing the KIE system and GR system detection on the same sample, the signal intensity and judgment label of the corresponding detection sites of the two systems were extracted. The judgment label included the mutant type that may lead to the drug resistance phenotype and the wild type. The wild type means that there is no risk of drug resistance. (2) Match the detection sites one by one to form a system result comparison table and record the dual system signal value and judgment result (mutant or wild type) for each site. (3) When the results of any site are inconsistent in the two systems (for example, the KIE system is mutant and the GR system is wild-type), the site is automatically marked as a "system conflict site".

[0069] The KIE and GR systems are two different detection reaction systems or reagent platforms, typically differing in probe structure, reaction temperature, and fluorescent labeling. Comparing the results of the two systems can effectively identify false positives or false negatives caused by system characteristics.

[0070] Import the original data file according to the software operation steps described in this invention, and perform three sets of data analysis (samples 1-3). The results of the sample fluorescence detection values ​​(MFI) are detailed in Table 1. The drug resistance detection results for samples 1-3 are shown in [Table 1]. Figure 1-3 .

[0071] Table 1 Summary of Sample Detection Data

[0072] II. Signal Consistency Calculation Stage (1) Extract the MFI (mean fluorescence intensity) data of all non-conflicting sites and calculate the linear correlation coefficient and mean deviation ratio between the two systems; (2) The signal consistency rate is defined as a comprehensive index of the correlation between the signals of the two systems and the consistency of the results. Its calculation form is as follows: Consistency rate = 0.5 × (signal strength correlation coefficient) + 0.5 × (result consistency ratio); (3) When the consistency rate is ≥0.85, the consistency between the systems is considered good; when the consistency rate is between 0.65 and 0.85, it is considered moderately consistent; when the consistency rate is <0.65, it is considered significantly inconsistent. (4) Automatically exclude low signal (MFI < background threshold) sites during the calculation process to prevent noise from interfering with the evaluation results.

[0073] It should be noted that the signal consistency rate is used to quantitatively evaluate the degree of consistency between two systems in terms of overall signal trend and mutation judgment results. Combining the two dimensions of intensity and result evaluation, it is more stable than a simple "result comparison".

[0074] III. Weight Redistribution Phase (1) When a decrease in the consistency rate between systems is detected (below 0.85), the weights of the results of the two systems are adjusted according to the consistency rate; (2) In the initial state, the weights of the KIE system and the GR system are each 0.5; when the consistency rate is between 0.65 and 0.85, the weight of the fluctuating system is reduced to 0.3 and the weight of the stable system is increased to 0.7. (3) If the consistency rate is lower than 0.65, further analyze the source of signal fluctuation at each point (such as signal-to-noise ratio, peak shape shift, etc.), determine the unstable system, automatically reduce its weight to 0.2 or below, and add "inconsistency mark" to the final result; (4) The results after weight correction are re-input into the result fusion module for drug resistance feature matrix calculation to ensure that the final drug resistance score is dominated by high confidence data.

[0075] It should be noted that the weight redistribution mechanism ensures that abnormal systems do not dominate the judgment results in parallel detection of multiple systems, thereby reducing the risk of false positives.

[0076] IV. Verification Closed-Loop Stage (1) Record the results of each system consistency verification, including the consistency rate, weight adjustment ratio and list of conflict sites; (2) If the consistency rate of the same system is lower than 0.7 in consecutive batches, the quality control alarm will be automatically triggered, prompting the testing platform to be recalibrated or the reagents to be replaced; (3) All verification results are written to the log to track the trend of detection stability.

[0077] The cross-system consistency verification module achieves automatic data fusion and reliability enhancement across multiple detection systems through signal comparison and dynamic weight correction mechanisms. When systematic deviations or local drifts exist between detection systems, this module can automatically correct the results through consistency rate quantification analysis, making the overall judgment more robust. This design effectively reduces errors caused by system differences, significantly improves the consistency of mutation detection and the accuracy of drug resistance determination, and ensures the consistency of results for the Mycobacterium tuberculosis drug resistance gene detection system in parallel application scenarios across multiple systems.

[0078] In some embodiments of this application, calculating the signal consistency rate between systems and performing weight reallocation includes: (1) Match the fluorescence signals of the detection sites in the two systems one-to-one according to the detection site number; (2) Calculate the signal consistency rate of the two systems at each detection site. The signal consistency rate is 1 minus the ratio of the signal difference between the two systems to the maximum signal value. When the consistency rate is less than the set threshold, it is marked as an inconsistent site. (3) For inconsistent sites, calculate the weight value based on the signal-to-noise ratio and coefficient of variation of the two systems. The weight value is the signal-to-noise ratio divided by the sum of the signal-to-noise ratio and the coefficient of variation. (4) Recalculate the fused signal based on the weight values. The fused signal is the sum of the KIE system signal multiplied by the KIE weight and the GR system signal multiplied by the GR weight. (5) Input the fused signal as the correction result into the feature extraction module and the multi-level interpretation module.

[0079] In this embodiment, the specific implementation steps are as follows: (1) First, the detection results of the KIE system and the GR system are matched at the site level according to the site number of the detection chip or probe array; (2) The corresponding rules for each point are automatically established by the initialization template to ensure that the same number represents the same gene locus or mutation probe; (3) After pairing is completed, a “system correspondence table” is generated. The table records the KIE signal value, GR signal value, background intensity and detection timestamp, which provides a basis for subsequent consistency rate calculation.

[0080] The purpose of this step is to establish a one-to-one correspondence between the two systems and prevent data mismatch. Since the probe placement locations or channel numbers may differ between different systems, a unified index is used through site numbers to ensure the accuracy of signal matching.

[0081] II. Signal Consistency Calculation Stage (1) Calculate the signal consistency rate between the two systems point by point. The consistency rate is defined as: Consistency rate = ; (2) In this formula, MFI(KIE) and MFI(GR) are the mean fluorescence intensity values ​​of the KIE system and the GR system, respectively, reflecting the signal brightness level; (3) When the consistency rate is lower than the set threshold (e.g., 0.85), the site is marked as an "inconsistent site" and requires subsequent weight correction processing; (4) At the same time, the overall average consistency rate is calculated to evaluate the global consistency of signals between systems.

[0082] The consistency rate reflects the degree of matching between the two systems in terms of signal amplitude. A higher consistency rate indicates that the two systems are stable and have good reproducibility.

[0083] III. Weight Calculation Stage (1) For the marked inconsistent sites, extract their signal-to-noise ratio (SNR) and coefficient of variation (CV). (2) The signal-to-noise ratio is defined as the ratio of signal strength to background noise strength, which reflects signal reliability; the coefficient of variation is the ratio of standard deviation to mean, which reflects signal stability. (3) The formula for calculating the weight value is: Weight = SNR / (SNR + CV); (4) Calculate the weight values ​​of the KIE system and the GR system respectively. When a certain system has a high signal-to-noise ratio and a low coefficient of variation, its weight will be automatically increased. (5) All weight values ​​are normalized so that the sum of the weights of the KIE system and the weights of the GR system equals 1.

[0084] This step dynamically adjusts the proportion of the two systems in the fusion result by simultaneously considering signal quality (SNR) and stability (CV), thereby preventing a single system from dominating the final judgment due to accidental fluctuations.

[0085] IV. Fusion Signal Generation Stage (1) Based on the calculated weight values, the signals of the two systems are weighted and fused using the following formula: Fusion signal = (KIE MFI ×weight KIE )+(GR MFI ×weight GR ); (2) The fused signal represents the unified expression of the two systems under weighted equilibrium, which retains the high-confidence signal and weakens the influence of the unstable system; (3) Record the source and calculation parameters of each weight during the fusion process and generate a "fusion log" to ensure data traceability; (4) For sites with significant drift (e.g., the fusion signal deviates from the average signal by more than 20%), they are automatically marked and entered into the abnormal sample correction module for review.

[0086] The calculation of the fused signal is the core step. It integrates signals from multiple systems by proportional superposition, giving higher weights to high-quality system data and automatically reducing the weights of low-quality system data, thus ensuring that the output signal is statistically the most stable.

[0087] V. Results Feedback and Reuse Stage (1) The fused signal, after correction, is transmitted as input data to the feature extraction module for recalculating peak intensity, abrupt change ratio and consistency coefficient; (2) At the same time, the fused signal also enters the multi-level interpretation module to update the hierarchical determination results of wild type and mutant type; (3) If the result after the fusion signal correction is consistent with the result of the original system, update the status to "correction passed"; if there is still a difference, retain both results and mark them as "requires manual review"; (4) All fusion data are displayed in the report generation module with the label “signal fusion” to ensure that clinicians can trace the source of correction at each step.

[0088] Example 1 See Figure 2 As shown, this embodiment of the invention provides a method for analyzing the detection data of drug resistance genes in Mycobacterium tuberculosis, comprising the following steps: (1) Import the original fluorescence intensity value MFI and perform signal correction; (2) Extract peak intensity, mutation ratio, consistency coefficient and intra-batch difference index; (3) Output wild-type and mutant results based on the three-stage interpretation logic; (4) Construct a drug resistance feature matrix and calculate the comprehensive drug resistance score; (5) Generate standard test reports and drug resistance spectrum reasoning reports.

[0089] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A data analysis system for detecting drug resistance genes in Mycobacterium tuberculosis, characterized in that, include: The module includes a data import module, a signal correction module, a feature extraction module, a multi-level interpretation module, a result fusion module, and a report generation module. The signal correction module is used to perform multi-stage correction on the original fluorescence intensity value MFI, including baseline drift correction, outlier detection and nonlinear normalization; The feature extraction module is used to extract multi-dimensional features based on the MFI data after multi-stage correction. The multi-dimensional features include peak intensity, mutation ratio, probe consistency coefficient and intra-batch difference index. The multi-level interpretation module constructs a dynamic threshold function based on multi-dimensional features to perform stratified determination of whether the corresponding nucleic acid sequence of the detection target is wild-type or mutant. The result fusion module maps the results of each detection site to the drug category by combining the drug resistance feature matrix and outputs a drug resistance risk warning. The report generation module generates standardized clinical test reports based on the fusion results.

2. The Mycobacterium tuberculosis drug resistance gene detection data analysis system according to claim 1, characterized in that, The signal correction module includes: The baseline drift correction unit is used to perform linear benchmark fitting on the MFI distribution of different detection batches; Anomaly detection unit is used to determine sudden intensity anomalies based on local statistical deviation ratio and perform local smoothing. The nonlinear normalization unit is used to unify the signal amplitude in each detection system to the 0-1 range using a piecewise Sigmoid function, so as to eliminate the difference in detection sensitivity between different batches.

3. The Mycobacterium tuberculosis drug resistance gene detection data analysis system according to claim 1, characterized in that, In the feature extraction module, the mutation ratio is the ratio of the mutant signal to the corresponding wild-type signal, the consistency coefficient is the inverse ratio of the difference between adjacent probe signals under the same detection system, and the intra-batch difference index is the ratio of the standard deviation to the mean of all probe signals in the sample.

4. The Mycobacterium tuberculosis drug resistance gene detection data analysis system according to claim 1, characterized in that, The multi-level interpretation module includes three operational phases: (1) In the signal layer determination stage, the basic threshold determination result is calculated based on the mutation ratio and peak intensity; (2) In the structural layer determination stage, the signal reliability is determined based on the consistency coefficient and the intra-batch difference index. When the reliability is lower than the set lower limit, it is marked as suspicious. (3) During the logic layer synthesis stage, conflict analysis is performed on the results of the signal layer and the structure layer. If the conflict ratio exceeds 20%, a second judgment is made and an anomaly report is generated.

5. The Mycobacterium tuberculosis drug resistance gene detection data analysis system according to claim 4, characterized in that, The result fusion module maps drug categories based on a drug resistance feature matrix, where rows of the matrix represent detected genes, columns represent drug categories, and matrix elements are mutation impact weights. The comprehensive drug resistance score for each drug is calculated through matrix multiplication, and the drugs are classified into three categories—sensitive, suspected, and resistant—based on the score range.

6. The Mycobacterium tuberculosis drug resistance gene detection data analysis system according to claim 5, characterized in that, Also includes: The abnormal sample correction module is used to identify samples in the secondary judgment whose deviation from the characteristics of historical samples exceeds a set threshold, and to perform numerical correction on the deviation site by signal re-estimation based on the neighborhood sample set. After correction, the data re-enters the signal layer judgment stage.

7. The Mycobacterium tuberculosis drug resistance gene detection data analysis system according to claim 6, characterized in that, Also includes: The cross-system consistency verification module calculates the signal consistency rate between the systems and performs weight reallocation when the detection results of the same sample are inconsistent in the KIE system and the GR system.

8. The Mycobacterium tuberculosis drug resistance gene detection data analysis system according to claim 7, characterized in that, Calculate the signal consistency rate between systems and perform weight redistribution, including: (1) Match the fluorescence signals of the detection sites in the two systems one-to-one according to the detection site number; (2) Calculate the signal consistency rate of the two systems at each detection site. The signal consistency rate is 1 minus the ratio of the signal difference between the two systems to the maximum signal value. When the consistency rate is less than the set threshold, it is marked as an inconsistent site. (3) For inconsistent sites, calculate the weight value based on the signal-to-noise ratio and coefficient of variation of the two systems. The weight value is the signal-to-noise ratio divided by the sum of the signal-to-noise ratio and the coefficient of variation. (4) Recalculate the fused signal based on the weight values. The fused signal is the sum of the KIE system signal multiplied by the KIE weight and the GR system signal multiplied by the GR weight. (5) Input the fused signal as the correction result into the feature extraction module and the multi-level interpretation module.

9. A method performed using the Mycobacterium tuberculosis drug resistance gene detection data analysis system according to any one of claims 1-8, characterized in that, Includes the following steps: (1) Import the original fluorescence intensity value MFI and perform signal correction; (2) Extract peak intensity, mutation ratio, consistency coefficient and intra-batch difference index; (3) Output wild-type and mutant results based on the three-stage interpretation logic; (4) Construct a drug resistance feature matrix and calculate the comprehensive drug resistance score; (5) Generate standard test reports and drug resistance spectrum reasoning reports.

10. A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, performs the method of claim 9.

Images