Methods for Establishing a Filtering Model for Tag Skipping Contamination in Targeted High-Throughput Sequencing Based on Nanopore Sequencing Platform

Abstract

This invention relates to a method for establishing a filtering model for tag skipping contamination in targeted high-throughput sequencing based on a nanopore sequencing platform, specifically relating to a filtering method, model, and application of tag skipping contamination in sequencing. This invention designs a tag skipping contamination filtering scheme based on statistical principles, employing targeted sequence clustering technology and a contamination rate estimation algorithm to fundamentally distinguish between real signals and skipping signals. It uses a probability distribution model, such as the Poisson probability distribution, to describe the inherent laws governing the random event of tag skipping and accurately calculates its probability of occurrence, integrating statistical evidence into a stable and interpretable discriminant conclusion. The entire process is based on rigorous statistical inference, and its results are transparent and traceable, significantly improving the reliability of the detection. It also possesses universality and can be applied to second-generation and third-generation sequencing technologies.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics technology and relates to a method for establishing a filtering model for targeted high-throughput sequencing tag skipping contamination based on a nanopore sequencing platform. Background Technology

[0002] Driven by the demand for precise pathogen detection and genomic analysis, the combined application of multiplex PCR amplification technology and third-generation sequencing technology has become a core solution to break through the limitations of traditional detection. Multiplex PCR improves detection sensitivity by targeting and enriching target sequences, while third-generation sequencing solves the problem of complex sequence analysis with its ultra-long read length. The two technologies have been applied in clinical infection diagnosis, pathogen drug resistance gene analysis, and source tracing of emerging infectious diseases.

[0003] Multiplex PCR is a technique that adds multiple pairs of specific primers to a single reaction system to simultaneously amplify multiple target gene fragments in a sample. It boasts core advantages such as high targeted enrichment efficiency, outstanding sensitivity, and adaptability to multiple samples in parallel processing. It addresses the pain points of third-generation sequencing, namely "high sequencing cost and high proportion of non-target sequences," and is well-suited to the application requirements of third-generation sequencing. Third-generation sequencing technologies, such as PacBio single-molecule real-time sequencing (SMRT) and Oxford Nanopore nanopore sequencing, have core advantages in resolving complex sequences with long reads, eliminating PCR bias, and enabling real-time detection. Nanopore technology supports "sequencing while analyzing," requiring only 1-2 hours from sample processing to obtaining results. Combined with the targeted enrichment of multiplex PCR, it can meet the rapid response needs of emergency detection (such as sudden bacterial infection outbreaks).

[0004] The combination of multiplex PCR and third-generation sequencing offers the dual advantages of "high-sensitivity targeted capture + real-time sequencing and rapid analysis," making it a preferred solution for complex pathogen detection and genomic analysis. However, this technology combination faces the "index hopping" problem in parallel multi-sample testing, which severely limits the accuracy of the test results. 1) The formation mechanism of tag skipping In the combined workflow of multiplex PCR amplification and third-generation sequencing, tag skipping is closely related to the characteristics of both technologies and mainly occurs in two key stages: The library construction stage, especially during multiplex PCR amplification and adapter ligation, is a high-risk period for tag skipping. The root cause lies in complex intermolecular malconnections within the reaction system: such as "index-template mismatch" caused by primer dimers or non-specific amplification during PCR; non-specific binding of free indexes to heterologous adapters in PacBio SMRTbell adapter ligation; and "barcode-heterologous fragment ligation" caused by abnormal ligase efficiency in Nanopore barcode ligation. These errors all lead to the formation of chimeric molecules carrying incorrect sample tags, becoming the direct source of tag skipping. Third-generation sequencing detection stage: The "single-molecule detection" characteristic of third-generation sequencing is more sensitive to contamination - even a single tag-jumping nucleic acid molecule may be captured and identified as a valid sequence by the sequencer; while Nanopore's "nanopore channel reuse" design may cause the tag sequence of the previous sample to remain in the channel, contaminating the sequencing data of subsequent samples. 2) Problems caused by tag skipping Tag skipping under this combination of technologies will directly negate the core advantages of "multiplex PCR targeting" and "third-generation sequencing high fidelity", leading to multiple risks such as sequence analysis errors, false positive species, and failure of sample traceability; 3) Existing methods for reducing index homing and their limitations Currently, the industry's handling of tag skipping in the "multiplex PCR amplification + third-generation sequencing" combination largely follows solutions from traditional sequencing technologies, and no dedicated method adapted to the characteristics of this combination has been developed, resulting in significant limitations. Experimental design optimization methods, such as dual-label design, high Hamming distance design of label sequences, and use of high-fidelity polymerase, can reduce the probability of label skipping from the source, but cannot completely eliminate non-specific binding during PCR amplification and adapter ligation. They are especially sensitive to single-molecule contamination in third-generation sequencing and still have residual risks. Sequencing platforms have built-in filtering functions: PacBio's SMRT Link software can filter reads with "abnormal tag-fragment binding time", and Nanopore's Guppy software supports tag signal separation. However, these functions are only designed for the characteristics of the platform itself and do not take into account the problem of "tag skipping signal amplification after target fragment enrichment" caused by multiplex PCR amplification, resulting in insufficient filtering accuracy. Traditional data analysis methods, such as threshold filtering based on tag frequency (e.g., removing tag sequences with a proportion of <1%) and reference sequence alignment filtering (removing sequences without a corresponding reference genome), cannot distinguish between "low-proportion tag-jumping sequences" and "true low-abundance target sequences" (e.g., low-load target sequences after multiplex PCR amplification), which can easily lead to false negatives or false positives. Furthermore, they do not combine the primer specificity of multiplex PCR and the long read characteristics of third-generation sequencing, resulting in weak tag-jumping recognition capabilities for chimeric fragments and structurally variable sequences.

[0005] Predictive models based on second-generation sequencing technology (such as neural networks) are difficult to apply to third-generation sequencing technology, which has a higher random error rate and a different sequencing principle, and its "black box" nature is not conducive to the verification of clinical diagnostic results.

[0006] In summary, developing a solution specifically designed to address tag skipping contamination in multiplex PCR and targeted third-generation sequencing is one of the most pressing issues that needs to be resolved. Summary of the Invention

[0007] To address the shortcomings of existing technologies and practical needs, this invention provides a method for establishing a filtering model for tag skipping contamination in targeted high-throughput sequencing based on a nanopore sequencing platform, aiming to accurately quantify tag skipping contamination in sequencing and accurately exclude false positive results.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for filtering tag skipping contamination during sequencing, the filtering method comprising the following steps: (1) For all samples detected in the same sequencing batch, clustering is performed based on sequence similarity to obtain different cluster groups; (2) For each cluster group, traverse all the sequences contained therein, identify and filter out candidate label skip sequences; (3) Calculate the proportion of the total number of candidate tag skipping sequences in each cluster group to the total abundance of the cluster group, and record it as the candidate contamination rate f_cluster of the cluster group. Use the median of the candidate contamination rates f_cluster of all cluster groups as the estimated value f of the global tag skipping contamination rate of this sequencing batch. (4) For a single target detection event that needs to be determined, the statistical probability caused by label skipping is calculated by a statistical prediction model. P ; (5) Set a confidence level threshold α for decision-making, and perform decision-making and correction: a) If P If the value is greater than or equal to α, the target detection event is determined to be a label skipping event, and the count of the corresponding sequence of the target detection event is corrected in subsequent analysis. b) If P If the value is less than α, the detected event is determined to be a non-labeled skip event and is retained.

[0009] This invention designs a statistically based filtering scheme for tag skipping contamination in sequencing. First, it distinguishes between true signals and skipping signals at the source using targeted sequence clustering technology and a contamination rate estimation algorithm. Then, it utilizes a probability distribution model, such as the Poisson probability distribution, to characterize the inherent patterns of the random event of tag skipping and accurately calculate its probability of occurrence, thus fusing statistical evidence into a stable and interpretable judgment. The entire process is based on rigorous statistical inference, and its results are transparent and traceable, greatly enhancing the reliability of the detection. It also possesses universality and can be extended to second- and third-generation sequencing technologies.

[0010] Optionally, the criteria for clustering based on sequence similarity in step (1) include: sequences with a similarity of ≥97% are divided into one cluster group.

[0011] In this invention, the similarity between sequences refers to the edit distance between two sequences. The method for determining / calculating the similarity between sequences can be to use algorithms such as BLAST and Needleman-Wunsch to calculate the similarity.

[0012] Optionally, the criteria for selecting candidate label skip sequences in step (2) include: The sequence has an extremely low count c in its sample, satisfying 0 < c ≤ s×(N_cluster / n), where N_cluster is the total abundance of the cluster to which the sequence belongs in all samples, n is the total number of samples in this batch of sequencing, and s is the threshold used to define low abundance; and the sequence appears in at least 3 samples.

[0013] Optionally, s is selected from 0.05 to 0.2.

[0014] Optionally, the method for constructing the statistical prediction model in step (4) includes constructing it based on a probability distribution model.

[0015] Optionally, the probability distribution model includes a Poisson distribution model or a negative binomial distribution model.

[0016] Optionally, the method for constructing the statistical prediction model in step (4) includes: (4-1) Set the null hypothesis H0 as: the target detection event is contamination caused by label skipping; (4-2) Using the global tag skipping contamination rate estimate f obtained in step (3), calculate the expected contamination value z = f×N_cluster / n for the target detection event, where N_cluster is the total abundance of the cluster group to which the sequence corresponding to the target detection event belongs in all samples, and n is the total number of samples sequenced in this batch. (4-3) Calculate the statistical probability based on the Poisson distribution model. P The calculation formula is: ; Alternatively, the statistical probability can be calculated using the negative binomial distribution model. P The calculation formula is: P = P (X ≥ c | H0) = 1 - F_NB(c-1; mean=z, dispersion=r); Where F_NB is the cumulative distribution function of the negative binomial distribution, z is the expected pollution value, and r is the discrete parameter estimated from the data.

[0017] Optionally, the confidence level threshold α in step (5) is selected from 0.01 to 0.2.

[0018] Optionally, the correction in step (5) includes setting the count of the sequence corresponding to the target detection event to 0.

[0019] The tag skipping contamination filtering method in sequencing proposed in this invention has broad application prospects. It can be applied to clinical diagnosis as well as non-disease diagnosis purposes. Specifically, it can be applied to the processing of targeted sequencing data in research related to targeted pathogen sequencing (tNGS), environmental microbiology, agricultural diseases, and food hygiene. It is compatible with different types of sequencing library construction schemes and provides reliable contamination filtering support for various sequencing studies.

[0020] Secondly, the present invention provides a filtering model for tag skipping contamination in sequencing, the filtering model including a clustering module, a candidate tag skipping sequence screening module, a global tag skipping contamination rate calculation module, and a statistical probability module. P Calculation module and decision and correction module; The clustering and grouping module is used to perform the following: clustering the target sequences detected in all samples within the same sequencing batch according to sequence similarity to obtain different clustering groups; The candidate label skip sequence filtering module is used to perform the following: for each cluster group, traverse all sequences contained therein, identify and filter out candidate label skip sequences; The global tag skip contamination rate calculation module is used to perform the following: calculate the proportion of the total count of candidate tag skip sequences in each cluster group to the total abundance of the cluster group, denoted as the candidate contamination rate f_cluster of the cluster group, and use the median of the candidate contamination rates f_cluster of all cluster groups as the estimated value f of the global tag skip contamination rate of this sequencing batch. The statistical probability P The calculation module is used to perform the following: for a single target detection event that needs to be determined, calculate the statistical probability of its occurrence due to label skipping using a statistical prediction model. P ; The decision-making and correction module is used to perform the following: setting a confidence level threshold α for decision-making, and then making a decision and making corrections. a) If P If the value is greater than or equal to α, the target detection event is determined to be a label skipping event, and the count of the target detection event is corrected in subsequent analysis. b) If P If the value is less than α, the detected event is determined to be a non-labeled skip event and is retained.

[0021] Optionally, the filtering model for tag skipping contamination in sequencing is used to perform the steps in the filtering method for tag skipping contamination in sequencing described in the first aspect.

[0022] Thirdly, the present invention provides a targeted sequencing and data analysis method, the targeted sequencing and data analysis method comprising: Nucleic acid is extracted from the sample to be tested, the extracted nucleic acid is targeted for enrichment and a library is constructed, the constructed library is sequenced, the sequencing data is filtered using the label skipping contamination filtering method in the first aspect or the label skipping contamination filtering model in the second aspect, and the filtered data is subjected to bioinformatics analysis.

[0023] Optionally, the sequencing includes second-generation sequencing or third-generation sequencing.

[0024] Fourthly, the present invention provides an electronic device comprising one or more processors and a memory for storing executable instructions, wherein the one or more processors are configured to invoke the executable instructions stored in the memory to implement the function of the tag skipping contamination filtering model in sequencing described in the second aspect.

[0025] Fifthly, the present invention provides a computer-readable storage medium having computer program instructions stored thereon, characterized in that, when the computer program instructions are executed by a processor, they implement the function of the filtering model for tag skipping contamination in sequencing as described in the second aspect.

[0026] Compared with the prior art, the present invention has at least the following beneficial effects: This invention designs a label skipping contamination filtering scheme in sequencing based on statistical principles. First, targeted sequence clustering technology and a contamination rate estimation algorithm are used to distinguish between true signals and skipping signals at the source. Second, a probability distribution model, such as the Poisson probability distribution, is used to describe the inherent laws governing the random event of label skipping and to accurately calculate its probability of occurrence, integrating statistical evidence into a stable and interpretable judgment. The entire process is based on rigorous statistical inference, and its results are transparent and traceable, significantly improving the reliability of the detection. It also has universality and can be applied to second- and third-generation sequencing technologies. Detailed Implementation

[0027] The technical solution of the present invention will be further illustrated below through specific embodiments. However, the following examples are merely simplified examples of the present invention and do not represent or limit the scope of protection of the present invention. The scope of protection of the present invention is determined by the claims.

[0028] Where specific techniques or conditions are not specified in the examples, they shall be performed in accordance with the techniques or conditions described in the literature in this field, or in accordance with the product instructions. Reagents or instruments whose manufacturers are not specified are all conventional products that can be purchased from legitimate channels.

[0029] This invention addresses the tag skipping contamination problem in the combination of multiplex PCR and third-generation sequencing technologies. It completely abandons "black box" machine learning models (such as neural networks) and instead proposes an innovative hybrid model based on statistical principles, starting from first principles: First, targeted sequence clustering technology and a contamination rate estimation algorithm are used to fundamentally distinguish between real and skipping signals; then, the Poisson probability distribution is used to characterize the inherent patterns of the random event of tag skipping and to accurately calculate its probability of occurrence, thus fusing statistical evidence into a stable and interpretable discriminant conclusion. The entire process is based on rigorous statistical inference, and its results are transparent and traceable, greatly enhancing the reliability of the detection.

[0030] Specifically, this invention provides a highly interpretable tag skipping contamination filtering scheme based on sequence clustering and probabilistic statistics to accurately quantify tag skipping contamination in third-generation sequencing and reliably exclude false positives. The scheme mainly includes the following steps: (1) Cluster analysis of target sequences For all samples detected in the same sequencing batch, the target sequences are clustered according to their sequence similarity (sequences with similarity ≥ 0.97 are grouped together) to form different clusters.

[0031] Objective: To group similar sequences originating from the same target species but arising from sequencing errors or microvariations into the same operational taxonomic unit, providing a stable data foundation for subsequent analysis. Each cluster is considered an independent analytical unit.

[0032] (2) Identification and screening of candidate label skipping sequences For each cluster group, iterate through all the sequences contained within it, identify and filter out "candidate label skipping sequences".

[0033] Selection criteria: A sequence is considered a candidate label skip sequence if it simultaneously meets the following two conditions: (a) Low abundance characteristic: The sequence count c in its sample is extremely low, specifically, it must satisfy 0 < c ≤ s × (N_cluster / n), where: N_cluster is the total abundance of this cluster across all samples; n is the total number of samples sequenced in this batch; s is an adjustable sensitivity parameter (the default value can be set to 0.1) used to define the threshold for "low abundance"; (b) Distributive characteristics: The sequence must appear in a certain number of samples (by default S_min ≥ 3, where S_min is the minimum number of samples, i.e. the sequence appears in ≥3 samples) to exclude the interference of random errors or sequencing noise and ensure that the pattern is systematic.

[0034] (3) Robust estimation of label skipping contamination rate within a batch Calculate the proportion of the total count of candidate label skip sequences in each cluster group to the total abundance of that cluster, and denote it as the candidate contamination rate f_cluster of that cluster.

[0035] To avoid outliers in individual clusters having an excessive impact on the overall estimate, the median f of the contamination rates of all candidate clusters is used as the global tag skipping contamination rate estimate for this sequencing batch. This step ensures the robustness of the contamination rate estimate to outlier data.

[0036] (4) Probability test of a single detection event based on a statistical model 1) For a single target detection event that needs to be determined (i.e., a low abundance count c of a certain cluster in a certain sample), we construct a statistical prediction model to calculate its statistical probability of being "purely caused by label skipping". P The statistical prediction model can be constructed based on the Poisson distribution.

[0037] 2) Null hypothesis (H0): The detection event was caused by contamination resulting from label skipping.

[0038] 3) Test statistic: We use the global contamination rate f obtained in step (3) to calculate the expected contamination value z = f×N_cluster / n for the detected event.

[0039] 4) Probability Calculation: A monotonically decreasing function (e.g., a heuristic scoring function t = 2 / (1 + exp(c / z)), or a probability calculation based on a Poisson distribution) is used to quantify the goodness of fit between the observed count c and the null hypothesis. This function maps c to a probability. P The smaller the value of c, P The larger the value, the stronger the evidence supporting the null hypothesis.

[0040] (5) Hypothesis testing and automatic correction based on confidence level 1) Set a confidence level threshold α for decision-making (e.g., α = 0.1).

[0041] 2) Decision-making and correction: a) If P If the value is greater than or equal to α, then we cannot reject the null hypothesis, determine that the detected event is a label skipping event, and correct the count in subsequent analysis (e.g., set it to zero).

[0042] b) If P If < α, then we reject the null hypothesis, determine that the detected event is a non-labeled skipping event (i.e., it may be a real low-abundance target sequence signal), and retain it.

[0043] Example 1 This embodiment is based on a probability calculation model using the Poisson distribution.

[0044] A statistical prediction model is constructed based on the Poisson distribution to calculate the statistical probability that it is "purely caused by label skipping". P The Poisson distribution is a probability distribution widely used to describe the number of times a random event occurs within a unit of time or space.

[0045] (1) Model building principle Under the null hypothesis (H0) of label skipping contamination, we consider the count c of low-abundance sequences belonging to a certain cluster group observed in a specific sample to be a random event that follows a Poisson distribution with an average occurrence rate of z. The expected contamination value z is calculated as follows: z = f×N_cluster / n; In the formula, f: global label skipping contamination rate within the batch; N_cluster: total abundance of the cluster group in all samples; n: total number of samples within the batch.

[0046] 2. Probability Calculation Formula According to the definition of the Poisson distribution, given an average of z occurrences, the probability of actually observing exactly k events is: P (X = k) = (e^{-z}×z^k) / k!; Where e is the natural constant and k! is the factorial of k.

[0047] In this invention, we focus on cumulative probability. This is to calculate the probability that the observation count c is purely due to random contamination. P We calculate its right-tail probability, which is the total probability that the Poisson distribution, under the null hypothesis, produces an observed value c and a value more extreme (larger) than c. The formula is as follows: P = 1 - F(c-1; z); Where F(c-1; z) is the cumulative distribution function value of a Poisson distribution with mean z at point c-1, that is, the sum of all probabilities from 0 to c-1. This formula is equivalent to: P = 1 - [ P (X=0) + P (X=1) + ... + P (X=c-1) ]; in, P (X=k) = (e^{-z}×z^k) / k!, which is the probability mass function of the Poisson distribution.

[0048] Those skilled in the art should understand that any monotonically decreasing probability function that can be calculated based on the observed value c and the expected value z to measure the likelihood of the null hypothesis being true falls within the protection scope of this invention.

[0049] Example 2 This embodiment performs tag skipping contamination identification and correction based on two-step multiplex PCR and third-generation sequencing.

[0050] (1) Nucleic acid extraction DNA and RNA were extracted using nucleic acid extraction or purification reagents (Guangzhou Jinqirui Biotechnology Co., Ltd., KS118-BYTQ-96) or (Guangzhou Jinqirui Biotechnology Co., Ltd., KS132-TNGSTQA96 / KS132-TNGSTQM96).

[0051] (2) Targeted enrichment Targeted enrichment and library construction were performed using Respiratory 100 premixed solution (Guangzhou Jinqirui Biotechnology Co., Ltd., KS608-100HXD96) and nucleic acid extraction or purification reagents (Guangzhou Jinqirui Biotechnology Co., Ltd., KS138-TNGSTQ-96).

[0052] 3. Sequencing Library construction and sequencing were performed using the Rapid Barcode Library Construction Kit (Chengdu Qitan Technology Co., Ltd., C-001-0008) and the Rapid Barcode Library Construction Kit (Chengdu Qitan Technology Co., Ltd., C-002-0005).

[0053] (3) Sequencing process and data integration Hybrid sequencing: The final libraries of different samples are mixed and sequenced in real time on a third-generation sequencer.

[0054] Data output and splitting: The sequences generated by sequencing directly contain sample index information, and the samples are split using bioinformatics software.

[0055] (4) Integration and application with the statistical methods of this invention After sequencing, data processing is seamlessly integrated with the statistical scheme described in this invention: Step 1: Sequence Clustering: Utilizing the read length advantage of third-generation sequencing, the split sequences are clustered with high precision to obtain more accurate analysis units.

[0056] Steps 2 and 3: Contamination rate estimation: Identify candidate label skipping sequences within cluster groups and robustly estimate the global label skipping contamination rate f.

[0057] Step 4: Probability Calculation and Verification: The probability is calculated using the Poisson distribution model preferred in this embodiment. P The formula is: ; Step 5: Automatic Judgment: Based on P The value is compared with the threshold α, and the judgment result is automatically output.

[0058] Based on the above process, the following beneficial effects are achieved: 1. Standardized and Flexible Process: The clearly defined two-step process aligns better with standardized practices in molecular biology, making it easier to implement in routine laboratories. Furthermore, separating index addition from adapter ligation allows the targeted sequencing kit to be more flexibly adapted to different versions or brands of third-generation sequencing platforms, requiring only a change of the second-round primers.

[0059] 2. Higher index quality: The sample index is added in the first round, avoiding the inefficiency and primer dimer problems that may be caused by multiple amplification in the presence of complex adapter sequences. This ensures the high fidelity of the index sequence and improves the accuracy of subsequent sample splitting from the source, providing a cleaner data starting point for the statistical model of this invention.

[0060] 3. The Synergistic Advantages of Technology: Third-generation sequencing technology offers significant real-time sequencing capabilities, enabling rapid data output. However, its inherent index skipping problem in pooled testing limits the specificity and reliability of test results. The statistical scheme provided in this invention forms a synergistic solution with the third-generation sequencing platform: the platform ensures the timeliness of testing, while the scheme of this invention ensures the accuracy of results by accurately identifying and filtering false positive signals caused by index skipping. The combination of these two approaches satisfies the dual requirements of speed and accuracy in testing.

[0061] 4. Forming a complete technical solution: This embodiment constitutes a complete, closed-loop technical solution that can withstand practical testing, from four levels: special reagents (two-step primer system) -> standard procedures (two-step PCR) -> cutting-edge platform (third-generation sequencing) -> core algorithm (statistical model).

[0062] Example 3 This embodiment verifies the filtration effect of the present invention on a culture with a known composition.

[0063] 1. Sample and Experimental Design 1) Sample composition: a) Culture A: Contains the pathogen Streptococcus pneumoniae; b) Culture B: Contains the pathogen Streptococcus agalactiae; c) Culture C: Contains the pathogen Mycobacterium tuberculosis; d) NTC (Negative Control without Template): Water without nucleic acid.

[0064] 2) Experimental Design: The above-mentioned standards and negative control samples were co-sequencing on the same sequencing chip. Since each culture has its own specific and unique expected pathogen spectrum, the authenticity of any pathogen detected in any sample can be clearly determined. A. True positive signal: Streptococcus pneumoniae detected in culture A, Streptococcus agalactiae detected in culture B, and Mycobacterium tuberculosis detected in culture C; B. Label skipping contamination (false positives): a) Agalactococcus and Mycobacterium tuberculosis detected in culture A (originating from culture B or culture C).

[0065] b) Streptococcus pneumoniae and Mycobacterium tuberculosis detected in culture B (derived from culture A or culture C).

[0066] c) Streptococcus pneumoniae and Streptococcus agalactiae detected in culture C (from culture A or culture B).

[0067] d) Any sequence detected in NTC (Streptococcus pneumoniae, Streptococcus agalactiae, Mycobacterium tuberculosis).

[0068] 2. Method Implementation Process The following analysis was performed on the offline data: 1) Sequence processing and clustering: The sequencing sequences of all samples (3 groups of cultures and negative control) were quality controlled and split, and clustered according to sequence similarity to form cluster groups.

[0069] 2) Gold standard label definition: True positives as defined above are labeled as “expected positive clusters”; while the four types of false positives as defined above are labeled as “label skipping contamination”.

[0070] 3) Applying the method of this invention: Input all data as a whole batch into the statistical model described in this invention, and perform the following steps: Calculate the global label skipping contamination rate f within the batch; For each low abundance detection in the negative control samples, calculate the probability that it originated from label skipping. P ; The nature of the detected substance is automatically determined based on a preset confidence level α (e.g., α=0.1).

[0071] 3. Effect Evaluation and Result Analysis We will evaluate the model's performance across all possible contamination pathways.

[0072] Sensitivity: The proportion of all known contamination events (such as the four types of false positives defined above) that are correctly identified as "contamination" by the model.

[0073] Specificity: The proportion of all known true signals (such as true positives as defined above) that are correctly identified as "true" by the model.

[0074] Accuracy: The proportion of events that are actually contaminated out of all events identified as "contaminated" by the model.

[0075] The performance evaluation results of known cultures and NTC are summarized in Table 1. The global contamination rate f estimated by this invention is in high agreement with the contamination rate calculated directly from the culture samples (estimated value f = 1.62%, measured value 1.77%), which proves the reliability of the contamination rate estimation; the sensitivity reaches 100%, indicating that all contamination signals are effectively captured and removed.

[0076] Table 1 The pollution path analysis and filtration effect are shown in Table 2. The results clearly reveal the network-like characteristics of the label jumping.

[0077] Table 2 The improved cleanliness of each sample data is shown in Table 3. No false positive sequences were detected in any culture or NTC after filtration.

[0078] Table 3 This embodiment demonstrates, through a multi-culture cross-contamination model, that tag skipping is a prevalent, network-like contamination phenomenon in pooled sequencing. As shown in Table 2, the method of this invention can accurately trace the source and path of contamination and exhibits stable and efficient filtering capabilities for contamination events in all directions. Ultimately, as shown in Tables 1 and 3, the method of this invention completely filters the total number of contaminated sequences in the entire batch, significantly improving the reporting cleanliness and accuracy of individual samples within the batch, and providing a crucial quality control tool for applying sequencing technology to demanding clinical differential diagnoses.

[0079] Example 4 This embodiment was validated using real clinical samples.

[0080] This embodiment aims to verify the effectiveness and practicality of the method of the present invention in complex real-world application environments using real clinical samples.

[0081] 1. Sample and Experimental Design Seven clinical bronchoalveolar lavage fluid samples with clear clinical significance were selected. These samples contained high-load pathogens, medium- and low-load pathogens, mixed-infection pathogens, and pathogens that were difficult to culture.

[0082] Each sample has a clinical verification result as the gold standard, which is determined by traditional microbial culture, qPCR or clinical targeted genomics (tNGS based on multiplex PCR amplification, target NGS) and combined with clinical manifestations.

[0083] 2. Method Implementation and Evaluation These 7 clinical samples and 1 NTC (template-free negative control) were tested on the same third-generation sequencing chip to ensure that all samples were in the same sequencing environment, and that tag skipping events during the process could be effectively acquired and analyzed.

[0084] The data were analyzed using both the "unfiltered raw analysis process" and the "analysis process integrating the filtering method of this invention".

[0085] Based on the clinical review results, the performance indicators of the two processes were calculated and compared: Sensitivity and specificity; Positive predictive value and negative predictive value; Overall agreement rate with the gold standard results.

[0086] 3. Results and Data Analysis The results verified by this embodiment are shown in Table 4 below.

[0087] Table 4 The performance metrics for both unfiltered and filtered processes are shown in Table 5.

[0088] Table 5 In 7 clinical samples, the unfiltered process generated 15 false positive reports due to label skipping, resulting in a specificity of only 17%. Applying the method of this invention successfully identified and corrected these 15 cases, significantly improving the specificity to 100%. However, this method introduced 2 false negatives, with a sensitivity of 86%. The results indicate that this invention can greatly improve the specificity and reliability of clinical reports while maintaining high sensitivity.

[0089] This demonstrates that the method of the present invention can effectively reduce the false positive rate and improve the clinical decision support value of the test results in the context of complex real-world samples, thus avoiding overdiagnosis and unnecessary treatment caused by false positives.

[0090] Example 5 This embodiment constructs an alternative statistical model based on the negative binomial distribution.

[0091] This embodiment aims to demonstrate the flexibility of the core statistical testing framework of the present invention and to provide an optimized solution when the data is excessively discrete.

[0092] 1. Model Building Principles In some sequencing data, due to biological and technical variations, the variance of sequence counts may be much greater than the mean; this phenomenon is called "overdispersion." The Poisson distribution assumption may no longer apply in this case.

[0093] This embodiment uses the negative binomial distribution as the probability calculation model. By introducing a discrete parameter k, this distribution can better fit overly discrete count data.

[0094] 2. Probability Calculation Formula The probability calculation formula is as follows: P = P (X ≥ c | H0) = 1 - F_NB(c-1; mean=z,dispersion=r) Where F_NB is the cumulative distribution function of the negative binomial distribution, z is the expected pollution value, and r is the discrete parameter estimated from the data.

[0095] 3. Implementation and Verification Using the same dataset as in Example 4, the Poisson distribution model (Example 1) and the negative binomial distribution model (this example) were applied respectively.

[0096] Comparing the performance of the two models, the negative binomial distribution may exhibit a lower misclassification rate due to its better fit to data fluctuations.

[0097] Table 6 The results, shown in Table 6, demonstrate that the hypothesis testing framework of this invention does not rely on a single statistical distribution. By introducing a negative binomial distribution, the applicability and robustness of the method are broadened, ensuring high performance under different data characteristics.

[0098] Example 6 This embodiment determines and optimizes the key discrimination threshold (α).

[0099] 1. Optimization Methods and Design Choose a dataset that contains known label jump events (such as the data in Example 4).

[0100] On this dataset, a series of α values ​​(such as 0.01, 0.05, 0.1, 0.2, 0.3) are systematically traversed.

[0101] For each α value, calculate its corresponding sensitivity and specificity.

[0102] 2. Results and Decisions Plot the receiver operating characteristic (ROC) curve and calculate the area under the curve to assess the model’s overall discriminative ability.

[0103] Select the optimal α value based on the clinical application scenario: In screening scenarios, high sensitivity should be prioritized by selecting a threshold with a high α value (such as 0.2).

[0104] In the case of confirmed diagnosis, high specificity should be prioritized, and a threshold with a low α value (such as 0.05) should be selected.

[0105] It is recommended to use α=0.1 as the default threshold that balances sensitivity and specificity.

[0106] This embodiment uses a data-driven approach to determine the core parameters of the method, proving that the present invention is not a "black box," but a mature solution that has been fully optimized and verified and can be adapted to different application needs, thus enhancing the scientific validity and credibility of the method.

[0107] Example 7 This embodiment applies a tag skipping contamination filtering scheme on a second-generation sequencing platform.

[0108] This embodiment aims to emphasize the platform universality of the method of the present invention. It is designed for third-generation sequencing technology, but can also be extended to the more mainstream second-generation sequencing platform.

[0109] 1. Method Adaptability Description The core input to the label skipping contamination filtering scheme is the counting matrix after sequence clustering, which is independent of a specific sequencing platform.

[0110] For second-generation sequencing platforms such as Illumina, their data preprocessing steps (sequence denoising, generation of ASV / OTU tables and other clustering matrix tables, etc.) can be seamlessly integrated with the method of this invention, and the method of this invention can be directly applied to the final generated clustering matrix table.

[0111] 2. Implementation and Verification We used a dataset that was sequenced on the Illumina MiSeq platform and included the Mock community and NTC.

[0112] Applying the method of this invention involves performing the exact same steps: pollution rate estimation, probability calculation (Poisson distribution can be used), and hypothesis testing.

[0113] The results show that the method of the present invention can effectively identify and filter false positive reports generated by Index Hopping in second-generation sequencing data, significantly improving data quality.

[0114] In summary, this invention addresses the tag skipping contamination problem in the combined multiplex PCR and third-generation sequencing technologies. It completely abandons "black box" machine learning models (such as neural networks) and designs a hybrid model based on statistical principles: First, targeted sequence clustering technology and contamination rate estimation algorithms are used to fundamentally distinguish between real signals and skipping signals; then, Poisson probability distribution and other methods are used to characterize the inherent patterns of the random event of tag skipping and accurately calculate its probability of occurrence, fusing statistical evidence into a stable and interpretable discriminant conclusion. The entire process is based on rigorous statistical inference, and its results are transparent and traceable, greatly enhancing the reliability of the detection. It also possesses universality and can be extended to second-generation sequencing technologies.

[0115] The applicant declares that the above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention fall within the protection and disclosure scope of the present invention.

Claims

1. A method for filtering tag skipping contamination in sequencing, characterized in that, The filtering method includes the following steps: (1) For all samples detected in the same sequencing batch, clustering is performed based on sequence similarity to obtain different cluster groups; (2) For each cluster group, traverse all the sequences contained therein, identify and filter out candidate label skip sequences; (3) Calculate the proportion of the total number of candidate tag skipping sequences in each cluster group to the total abundance of the cluster group, and record it as the candidate contamination rate f_cluster of the cluster group. Use the median of the candidate contamination rates f_cluster of all cluster groups as the estimated value f of the global tag skipping contamination rate of this sequencing batch. (4) For a single target detection event that needs to be determined, the statistical probability caused by label skipping is calculated by a statistical prediction model. P ; (5) Set a confidence level threshold α for decision-making, and perform decision-making and correction: a) If P If the value is greater than or equal to α, the target detection event is determined to be a label skipping event, and the count of the target detection event is corrected in subsequent analysis. b) If P If the value is less than α, the detected event is determined to be a non-labeled skip event and is retained.

2. The method for filtering tag skipping contamination in sequencing according to claim 1, characterized in that, The criteria for clustering based on sequence similarity in step (1) include: sequences with a similarity of ≥97% are grouped into one cluster.

3. The method for filtering tag skipping contamination in sequencing according to claim 1 or 2, characterized in that, The criteria for selecting candidate label skip sequences in step (2) include: The sequence has an extremely low abundance count (c) in its sample, satisfying 0 < c ≤ s×(N_cluster / n), where N_cluster is the total abundance of the cluster to which the sequence belongs in all samples, n is the total number of samples in this batch of sequencing, and s is the threshold used to define low abundance; furthermore, the sequence appears in at least 3 samples. Optionally, s is selected from 0.05 to 0.

2.

4. The method for filtering tag skipping contamination in sequencing according to any one of claims 1-3, characterized in that, The statistical prediction model construction method in step (4) includes construction based on probability distribution models; Optionally, the probability distribution model includes a Poisson distribution model or a negative binomial distribution model.

5. The method for filtering tag skipping contamination in sequencing according to claim 4, characterized in that, The methods for constructing the statistical prediction model in step (4) include: (4-1) Set the null hypothesis H0 as: the target detection event is contamination caused by label skipping; (4-2) Using the global tag skipping contamination rate estimate f obtained in step (3), calculate the expected contamination value z = f×N_cluster / n for the target detection event, where N_cluster is the total abundance of the cluster group to which the sequence corresponding to the target detection event belongs in all samples, and n is the total number of samples sequenced in this batch. (4-3) Calculate the statistical probability based on the Poisson distribution model. P The calculation formula is: ； Alternatively, the statistical probability can be calculated using the negative binomial distribution model. P The calculation formula is: P = P (X ≥ c | H0) = 1 - F_NB(c-1; mean=z, dispersion=r); Where F_NB is the cumulative distribution function of the negative binomial distribution, z is the expected pollution value, and r is the discrete parameter estimated from the data.

6. The method for filtering tag skipping contamination in sequencing according to any one of claims 1-5, characterized in that, In step (5), the confidence level threshold α is selected from 0.01 to 0.2; Optionally, the correction in step (5) includes setting the count of the sequence corresponding to the target detection event to 0.

7. A filtering model for tag skipping contamination in sequencing, characterized in that, The filtering model includes a clustering module, a candidate label skip sequence filtering module, a global label skip contamination rate calculation module, and a statistical probability module. P Calculation module and decision and correction module; The clustering and grouping module is used to perform the following: clustering the target sequences detected in all samples within the same sequencing batch according to sequence similarity to obtain different clustering groups; The candidate label skip sequence filtering module is used to perform the following: for each cluster group, traverse all sequences contained therein, identify and filter out candidate label skip sequences; The global tag skip contamination rate calculation module is used to perform the following: calculate the proportion of the total count of candidate tag skip sequences in each cluster group to the total abundance of the cluster group, denoted as the candidate contamination rate f_cluster of the cluster group, and use the median of the candidate contamination rates f_cluster of all cluster groups as the estimated value f of the global tag skip contamination rate of this sequencing batch. The statistical probability P The calculation module is used to perform the following: for a single target detection event that needs to be determined, calculate the statistical probability of its occurrence due to label skipping using a statistical prediction model. P ; The decision-making and correction module is used to perform the following: setting a confidence level threshold α for decision-making, and then making a decision and making corrections. a) If P If the value is greater than or equal to α, the target detection event is determined to be a label skipping event, and the count of the target detection event is corrected in subsequent analysis. b) If P If the value is less than α, the detected event is determined to be a non-labeled skip event and is retained.

8. A targeted sequencing and data analysis method, characterized in that, The targeted sequencing and data analysis methods include: Nucleic acid is extracted from the sample to be tested, the extracted nucleic acid is targeted for enrichment and a library is constructed, the constructed library is sequenced, the sequencing data is filtered using the label skipping contamination filtering method in sequencing as described in any one of claims 1-6 or the label skipping contamination filtering model in sequencing as described in claim 7, and the filtered data is subjected to bioinformatics analysis.

9. An electronic device comprising one or more processors and a memory for storing executable instructions, characterized in that, The one or more processors are configured to invoke executable instructions stored in the memory to implement the filtering model for tag skipping contamination in sequencing as described in claim 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the function of the tag skipping contamination filtering model in sequencing as described in claim 7.

Images