Data processing method and system for sequencing mutation sites

Abstract

The present application relates to the field of bioinformatics data processing, and provides a sequencing mutation site data processing method and device and a storage medium, comprising: obtaining sequencing mutation site data and extracting multi-dimensional feature data; identifying sample type identification and determining an evaluation threshold based thereon; mapping attribute labels to corresponding sites based on the multi-dimensional feature data, wherein part of the label mapping is based on the evaluation threshold; and comprehensively ruling on the attribute labels based on a preset label ruling rule to generate a filtering state identification, the rule including priority ordering and conflict resolution rules when multiple labels coexist. The present application effectively balances sensitivity and specificity, and improves the accuracy of sequencing data analysis.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics data processing, specifically to a data processing method, apparatus, and storage medium for sequencing mutation sites. Background Technology

[0002] With the widespread application of next-generation sequencing technology, variant detection in sequencing data has become an important tool in precision medicine research. However, in existing sequencing data analysis workflows, to eliminate false-positive variants introduced by sequencing errors, sample preparation artifacts, or alignment errors, hard filtering strategies based on fixed thresholds or rule-based filtering strategies with a single order are typically used. This approach often lacks a comprehensive understanding of multidimensional contextual features, making it difficult to effectively balance sensitivity and specificity. On the one hand, fixed filtering thresholds cannot adapt to the inherent signal-to-noise ratio differences between different sample types (such as paraffin-embedded tissue samples and circulating tumor DNA samples), easily leading to the misfiltering of low-abundance true signals or the retention of background noise. On the other hand, when the same mutation site simultaneously carries both clinically significant preservation characteristics and technical artifact-representing filtering characteristics, existing linear filtering logic lacks an effective conflict resolution mechanism, easily causing critically valuable variants to be incorrectly removed, or introducing systematic artifacts due to overprotection, thereby reducing the accuracy and reliability of sequencing data analysis results. Summary of the Invention

[0003] To address the problem of low data processing accuracy caused by the lack of context awareness and sample adaptability in existing sequencing mutation site filtering strategies, this invention provides the following technical solution: A data processing method for sequencing mutation sites, the method comprising: acquiring sequencing mutation site data and extracting multi-dimensional feature data of the sequencing mutation site data; identifying a sample type identifier corresponding to the sequencing mutation site data and determining an evaluation threshold for the sequencing mutation site based on the sample type identifier; mapping at least one attribute label to the corresponding sequencing mutation site based on the multi-dimensional feature data; wherein at least some of the attribute labels are mapped based on the evaluation threshold; and comprehensively adjudicating the attribute labels carried by the sequencing mutation site based on a preset label adjudication rule to generate a filtering status identifier for the sequencing mutation site; wherein the label adjudication rule includes priority ranking and conflict resolution rules for multiple attribute labels coexisting.

[0004] The above scheme decouples complex multi-dimensional feature data into independent attribute labels and dynamically determines the evaluation threshold by combining sample type identifiers, so that the label mapping process can adapt to the signal-to-noise ratio characteristics of different samples. At the same time, it uses label adjudication rules that include priority ranking and conflict resolution rules to comprehensively adjudicate the coexistence state of multiple labels, which overcomes the limitations of traditional linear filtering logic in handling feature conflicts and achieves the maximum retention of real variation signals while strictly controlling background artifacts.

[0005] As one implementation, the multi-dimensional feature data includes at least one of technical artifact features, clinical significance features, data quality features, and population frequency features; the step of determining the evaluation threshold of the sequencing mutation site based on the sample type identifier includes: determining the allele frequency threshold and the mutation depth threshold corresponding to the sample type identifier based on a preset mapping relationship between sample type and threshold, as the evaluation threshold.

[0006] In one implementation, the attribute tags include a first type of tag that tends to retain sites and a second type of tag that tends to filter sites; the comprehensive adjudication of the attribute tags carried by the sequencing mutation sites based on preset tag adjudication rules includes: when the sequencing mutation site carries both the first type of tag and the second type of tag, adjudication is made based on the coverage rule or exemption rule in the tag adjudication rules; wherein, the coverage rule indicates that the second type of tag has a higher priority than the first type of tag to perform a filtering operation, and the exemption rule indicates that the first type of tag has a higher priority than the second type of tag to perform a retention operation.

[0007] This implementation defines a clear conflict resolution mechanism, namely, handling the game-theoretic relationship between the first and second categories of labels through coverage and exemption rules. This mechanism enables the system to make decisions that conform to biological logic and technical specifications when facing complex scenarios. For example, it can strictly enforce filtering in the face of confirmed technical artifacts, while appropriately relaxing quality standards in the face of high-confidence clinical significance, thereby improving the rationality of the adjudication results.

[0008] In one implementation, the second type of label includes: background noise labels determined by matching a normal control sample library, and short tandem repeat region labels determined by genomic region features; the first type of label includes: hotspot mutation labels determined by matching a known hotspot mutation database, and carcinogenicity labels determined by matching a tumor knowledge base.

[0009] This implementation concretizes abstract label categories into executable data matching logic. By introducing a normal control sample library and genomic region features to identify technical artifacts, systematic background noise can be effectively eliminated; by integrating a hotspot mutation database and a tumor knowledge base to confirm clinical significance, the integrity of key variant information is ensured. This concretization enhances the operability of the scheme in actual bioinformatics analysis workflows.

[0010] As one implementation, the adjudication based on the coverage rule or exemption rule in the label adjudication rule includes: when the sequencing mutation site carries both the hotspot mutation label and the background noise label, triggering the coverage rule to make the background noise label cover the hotspot mutation label, generating the filtering status identifier representing the filtering; when the sequencing mutation site carries both the hotspot mutation label and the low-frequency label, the low-frequency label represents data quality lower than the evaluation threshold, triggering the exemption rule to make the hotspot mutation label exempt the low-frequency label, generating the filtering status identifier representing the retention.

[0011] This implementation provides two typical conflict resolution scenarios: for high-risk technical artifacts such as background noise, even if the site is a hotspot, it is filtered out to prevent systematic errors from interfering with the results; for simple low-frequency or low-quality signals, if the site has clear hotspot attributes, it is retained to avoid losing important information due to fluctuations in sequencing depth. This differentiated adjudication strategy reflects a refined balance between data quality and clinical value.

[0012] In one implementation, the sample type identifier includes a paraffin-embedded tissue sample identifier and a circulating tumor DNA sample identifier; the allele frequency threshold and the mutation depth threshold corresponding to the circulating tumor DNA sample identifier are both less than the allele frequency threshold and the mutation depth threshold corresponding to the paraffin-embedded tissue sample identifier.

[0013] As one implementation method, the step of comprehensively adjudicating the attribute tags carried by the sequencing mutation sites based on preset tag adjudication rules to generate a filtering status identifier for the sequencing mutation sites includes: determining the filtering status identifiers of the sequencing mutation sites layer by layer according to a preset priority order; the priority order from high to low includes: first priority, when the attribute tag contains a blacklist tag or the sequencing mutation site is not in the gene list of the current detection project, generating the filtering status identifier indicating filtering; second priority, when the attribute tag contains a hotspot mutation tag or a carcinogenic tag and does not contain a background noise tag, generating the filtering status identifier indicating retention; when the gene identifier corresponding to the sequencing mutation site belongs to a preset special gene list, the limitation of the evaluation threshold is ignored, and the filtering status identifier indicating retention is generated; third priority, when the attribute tag contains a clinically benign tag, conditional adjudication is performed according to the gene panel type of the current detection project; default priority: when the attribute tag does not trigger any of the above priority adjudication conditions, the filtering status identifier indicating retention is generated based on the default retention rule.

[0014] The hierarchical architecture of this implementation avoids logical deadlocks between rules, improving the efficiency and stability of automated processing.

[0015] As one implementation, mapping at least one attribute label to the corresponding sequencing mutation site based on the multi-dimensional feature data includes: obtaining a preset blacklist and historical abundance data recorded in the preset blacklist; mapping a blacklist label to the sequencing mutation site when the sequencing mutation site matches the preset blacklist and the current abundance data of the sequencing mutation site is lower than a preset fold of the historical abundance data; and exempting the matching evaluation of the preset blacklist from the sequencing mutation site when the sequencing mutation site carries the first type of label.

[0016] In addition, the present invention provides a data processing system for sequencing mutation sites, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements a data processing method for sequencing mutation sites.

[0017] Beneficial effects: The technical solution provided by this invention decouples the multi-dimensional features of sequencing mutation sites into independent attribute labels and combines them with sample type identifiers to dynamically adapt and evaluate thresholds. This fundamentally changes the rigid mode of traditional fixed threshold filtering, enabling the data processing process to adapt to the signal-to-noise ratio characteristics of different samples and effectively reducing the risk of false negatives and false positives caused by threshold mismatch. Attached Figure Description

[0018] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a schematic flowchart of the data processing method for sequencing mutation sites according to an embodiment of the present invention. Detailed Implementation

[0019] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] Method Implementation Examples like Figure 1 As shown, this embodiment provides a data processing method for sequencing mutation sites. This method achieves high-precision filtering of mutation sites by performing multi-dimensional feature decoupling and intelligent adjudication on sequencing data. Specifically, the method includes the following steps: Step S100: Obtain sequencing mutation site data and extract multi-dimensional feature data from the sequencing mutation site data. In this embodiment, obtaining sequencing mutation site data does not refer to the original sequencing signal acquisition process, but rather to reading standardized variant annotation files from the output of the variant detection process as the processing object.

[0022] Specifically, the input variant annotation file can be a VCF (Variant Call Format) 4.2 / 4.3 format file, a MAF (Mutation Annotation Format) file, or a TSV / CSV structured table file. The raw fields parsed from this file include at least: chromosome number (CHROM), genomic physical coordinates (POS), reference genomic bases (REF), mutant bases (ALT), variant quality value (QUAL), filter status (FILTER), INFO attribute set (such as DP, AF, AO, RO, FS, SOR, etc.), and FORMAT sample format information. After parsing, the system transforms the data into a unified data structure, facilitating subsequent feature vector extraction and analysis.

[0023] When extracting multi-dimensional feature data, the original fields can be parsed and reorganized to transform them into feature vectors that can be used for subsequent logical judgments.

[0024] For example, multidimensional feature data can include at least one of the following: technical artifact features, clinical significance features, data quality features, and population frequency features. Technical artifact features are used to characterize systematic biases introduced by sequencing errors or sample preparation, such as normal control library matching information or short tandem repeat region markers; clinical significance features are used to characterize the biological value of variants, such as hotspot mutation levels or knowledge base carcinogenicity ratings; data quality features are used to characterize the reliability of sequencing signals, such as reference matching variant depth or chain bias indices; and population frequency features are used to characterize the distribution of variants in healthy individuals. This multidimensional feature extraction approach transforms complex biological and technical information into independent evaluation dimensions, providing a data foundation for subsequent refined decision-making.

[0025] Examples of various features are given below.

[0026] 1. Characteristics of technical illusions: PoN: Matching frequency in the normal control sample library, denoted as N. PoN This is used to measure the frequency of this variant in tumor-free healthy individuals or in the same batch of control samples; Region markers: indicate whether the variation falls on a single nucleotide repeat (such as 5 or more consecutive A / T / C / G bases) or a short tandem repeat region of 2-6 bp; Strand Bias: Measured using Fisher Strand (FS) or Strand Odds Ratio (SOR), a high strand bias is indicated when FS > 60 or SOR > 3. Alignment and sequencing quality distribution: mean alignment quality (MQ), mean base quality (BQ), and mismatch rate of paired-end sequencing overlap regions; Trinucleotide context: Extract one base each upstream and downstream of the mutation site and compare the inherent sequencing mismatch rate of a specific sequencing library in this context.

[0027] 2. Clinical significance and characteristics: Known hotspot mutation database: Matches the hotspot level (such as Tier 1 / Tier 2) in the Cancer Hotspots, CIVIC, and COSMIC (v90 and above) databases. Oncogenic / Knowledge Base: Matches oncogenicity ratings (such as Oncogenic, Likely Oncogenic) in OncoKB, MyOncoDB, or NCCN guideline mutation libraries. Clinical pathogenicity rating (ClinVar / InterVar): Matching clinical significance ratings (such as Pathogenic, Likely Pathogenic, VUS, Benign) to the ClinVar database.

[0028] 3. Data quality characteristics: Allele frequency (AF): Calculated using the formula AF = AO / DP, where AO is the number of reads supporting the variant and DP is the total sequencing depth at that location; Number of mutations (Reads); Reference genome depth (RO); Support ratio of Reads before and after PCR deduplication.

[0029] 4. Population frequency characteristics: Healthy Population Database (gnomAD): Extracts the highest allele frequency (AF) from the population in gnomAD v2.1.1 / v3.1.2. popmax Or East Asian population frequency AF eas ; Local / Chinese population-specific high-frequency polymorphism database (ChinaAF).

[0030] Step S200: Identify the sample type identifier corresponding to the sequencing mutation site data, and determine the evaluation threshold for the sequencing mutation site based on the sample type identifier. Existing technologies typically use a fixed hard threshold to uniformly filter all samples, ignoring the inherent signal-to-noise ratio differences between different sample types. To address this issue, this embodiment establishes a dynamic mapping mechanism between sample type and evaluation threshold. Specifically, the system first parses the sample type identifier from metadata or filenames. This identifier can be a paraffin-embedded tissue sample identifier, a circulating tumor DNA sample identifier, a leukocyte sample identifier, or other type identifiers. Subsequently, the system queries a preset threshold configuration table to determine the allele frequency threshold and mutation depth threshold corresponding to the current sample type identifier, which are used as evaluation thresholds. For example, for circulating tumor DNA samples, due to their high degree of DNA fragmentation and low tumor release, true mutations often exist at extremely low abundance; therefore, the corresponding allele frequency threshold is set to 0.1%, and the mutation depth threshold is correspondingly lowered. For paraffin-embedded tissue samples, due to the potential for deamination artifacts caused by formalin fixation, the background noise is high; therefore, the allele frequency threshold is typically set to 1% or higher. This differentiated threshold setting allows the system to adapt to the signal-to-noise ratio characteristics of different samples, maximizing the detection capability of low-frequency real mutations in trace samples while ensuring specificity.

[0031] The system combines sequencing platforms (such as Illumina NovaSeq / MGI DGISEQ / Ion Torrent), sequencing library types (such as Targeted Panel capture, WES whole exome, WGS whole genome), and sample type identifiers to establish a detailed dynamic evaluation threshold mapping table, as shown in Table 1: Table 1. Dynamic evaluation threshold mapping matrix between sample type and sequencing configuration

[0032] When the actual AF, DP, AO of the sequencing mutation site does not meet the restrictions corresponding to the current sample type in Table 1, the system will label it with the corresponding quality defect label (such as LowAF, LowDP, LowAO) in the subsequent mapping steps.

[0033] Step S300: Based on multi-dimensional feature data, at least one attribute label is mapped to the corresponding sequencing mutation site; wherein, at least some attribute labels are mapped based on an evaluation threshold. This step is a crucial step in transforming abstract features into discrete decision labels. The system traverses each sequencing mutation site and assigns a corresponding attribute label based on the comparison results of its feature value with the evaluation threshold or the database matching results. Attribute labels include a first type of label that tends to preserve sites and a second type of label that tends to filter sites. The first type of label may include hotspot mutation labels determined based on matching with a known hotspot mutation database, and oncogenic labels determined based on matching with a tumor knowledge base; the second type of label may include background noise labels determined based on matching with a normal control sample library, and short tandem repeat region labels determined based on genomic region features.

[0034] During the mapping process, a relative abundance assessment mechanism is used to determine the blacklist labels, rather than simple coordinate matching. Specifically, the system obtains a preset blacklist and historical abundance data recorded in the preset blacklist. When a sequencing mutation site matches the preset blacklist and the current abundance data of the sequencing mutation site is lower than a preset fold of the historical abundance data, a blacklist label is mapped to the sequencing mutation site. When the sequencing mutation site carries a first-class label, it is exempt from the matching assessment of the preset blacklist. For example, if a site is located at the coordinate position recorded in the blacklist, but its current allele frequency is significantly higher than twice the artifact abundance at that position in the historical record, the system does not assign a blacklist label, thus avoiding misclassification of real high-frequency variants as known artifacts. At the same time, if the site has been identified as a drug-related hotspot (i.e., carrying a first-class label), the blacklist assessment process is skipped directly, preventing key clinical variants from being incorrectly labeled due to the limitations of historical data. This anti-false filtering mechanism effectively improves the robustness of the system in complex scenarios.

[0035] Specifically, the pre-defined blacklist database records the coordinates of historically high-incidence false positives introduced by systematic hybridization capture enrichment artifacts, PCR primer dimer artifacts, or specific sequencing gel mismatches, as well as the historical average abundance of these artifacts (AF). bg With the highest abundance AF max .

[0036] The determination logic is as follows: If the chromosomal location and base change of the sequencing mutation site match the blacklist database, the system extracts the AF at that location. bg : When the current sequencing abundance AF curr <N AF bg (The preset multiple N can be selected as 1.5, 2.0 or 3.0, and N=2.0 is preferred in this example.) When the site is mapped to a blacklist tag (Blacklist_Tag), the site is mapped to a blacklist tag. When AFcurr ≥N AF bg At that time, it was determined that the abundance of the mutation was significantly higher than the historical background noise, and therefore it was not mapped to the blacklist label.

[0037] Meanwhile, the specific rules for system mapping attribute tags are summarized in Table 2: Table 2 Attribute Tag Classification and Mapping Conditions Table

[0038] When a site carries any of the first-class tags Hotspot_Tag, Oncogenic_Tag, or Pathogenic_Tag, the system directly bypasses the blacklist matching process and does not assign a Blacklist_Tag, thereby achieving strong protection for important clinical variants.

[0039] Step S400: Based on preset tag adjudication rules, the system comprehensively adjudicates the attribute tags carried by the sequencing mutation sites to generate a filtering status identifier for the sequencing mutation sites. The tag adjudication rules include priority ranking and conflict resolution rules for multiple coexisting attribute tags. This is the core step in achieving intelligent filtering in this embodiment. When the same sequencing mutation site carries both a first-type tag and a second-type tag, the system adjudicates based on the coverage rule or exemption rule in the tag adjudication rules.

[0040] Specifically, the coverage rule indicates that the second type of label has higher priority than the first type of label for performing filtering operations. For example, when a sequencing mutation site carries both a hotspot mutation label and a background noise label, the coverage rule is triggered, causing the background noise label to cover the hotspot mutation label, generating a filtering status identifier that represents the filtering process. This is because background noise labels usually originate from repeated validation in normal control sample libraries and represent a systematic artifact with extremely high confidence. Even if the site is a known hotspot, it is very likely a sequencing artifact rather than a real biological signal. Therefore, strict filtering must be implemented to ensure data purity.

[0041] Conversely, the exemption rule prioritizes the first type of tag over the second type for retention. For example, when a sequencing mutation site carries both a hotspot mutation tag and a low-frequency tag, the low-frequency tag indicates data quality below the evaluation threshold. In this case, the exemption rule is triggered, exempting the hotspot mutation tag from the low-frequency tag and generating a filtering status identifier indicating retention. This is because the low-frequency tag only reflects that the sequencing depth or abundance has not reached the general standard, but given that the hotspot mutation has clear biological significance, moderate quality fluctuations should not be a reason to remove key information. Through these two diametrically opposed conflict resolution strategies, the system achieves context-based awareness, avoiding both false positives and the loss of true positives due to over-filtering.

[0042] In the comprehensive adjudication process, the Boolean logic operations for conflict resolution are as follows: 1. Overwrite rules: Filter_Status = FILTER If and only if This holds true even if the site simultaneously satisfies... The second type of high signal-to-noise ratio spoof label still forcibly covers the first type of label.

[0043] Exemption rules: Filter_Status=PASS If and only if When PoN_Tag = FALSE, even if the site carries LowQual_Tag (e.g., AF = 0.08% in ctDNA samples, which is less than the 0.1% threshold), the first type of tag is exempt from the restrictions of quality defect tags.

[0044] At the specific priority execution level, the system determines the filtering status identifier of sequencing mutation sites layer by layer according to the preset priority order.

[0045] The first priority is the mandatory filtering layer. When the attribute label contains a blacklist label or the sequencing mutation site is not in the gene list of the current detection project, a filtering status identifier representing the filtering is directly generated to ensure the compliance and cleanliness of the basic data.

[0046] The second priority is the mandatory retention layer. When the attribute tag contains hotspot mutation tags or oncogenic tags but not background noise tags, a filter status identifier representing retention is generated. Specifically, when the gene identifier corresponding to the sequencing mutation site belongs to a pre-defined special gene list, the restriction of the evaluation threshold is ignored, and a filter status identifier representing retention is generated. For example, germline variations of the BIM gene (BCL2L11) have special significance in certain hereditary tumor screenings, and should be retained even if its abundance does not meet the conventional germline determination criteria. It should be understood that this special gene list can be not only a pre-set static list, but also a maintainable list dynamically loaded through an external configuration file, so as to flexibly expand the scope of protection as medical research progresses.

[0047] The third priority is the conditional adjudication layer. When the attribute label contains a clinically benign label, a conditional adjudication is made based on the gene panel type of the current test item. For example, specific benign polymorphic sites are retained in the BRCA-specific panel, while they are filtered out in the general cancer panel.

[0048] Default priority: If none of the above priority decision conditions are triggered, a filter status identifier representing retention is generated based on the default retention rule to avoid accidental data loss due to missing rules.

[0049] In the hierarchical adjudication architecture, the specific execution branches are detailed as follows: First priority: If Blacklist_Tag = TRUE (and is not exempted by the first type of tag), output FILTER: Blacklist; If the mutated coordinates are not within the Target Bed capture range of the current detection item, output FILTER:OffTarget; If PopFreq_Tag=TRUE (crowd frequency) Furthermore, there is no clear evidence of a strong pathogenic hotspot, so the output is FILTER: HighPopFreq.

[0050] Second priority (mandatory preservation and special gene protection layer): like And PoN_Tag = FALSE, so the output is PASS: Hotspot_Retained; Special Gene List: This includes tumor-targeting driver genes (such as EGFR, KRAS, BRAF, ALK, ROS1, MET, RET, HER2 / ERBB2), tumor suppressor genes (such as TP53, BRCA1, BRCA2, ATM, PALB2), and specific genes with deep structural polymorphism (such as BIM / BCL2L11 intron deletion variants). When a locus falls within the above special gene list, or belongs to a specific variant type (such as frameshift imprinting (InDel) or splice-site variants), the AF and DP restrictions are waived, and PASS: SpecialGene_Protection is directly output.

[0051] Third priority: When the site carries a ClinVar-annotated Benign or Likely_Benign tag: If the current project panel is a BRCA1 / 2 genetic breast cancer panel or an HRR homologous recombination defect panel, retain these benign polymorphic sites and output PASS: Polymorphism_QC for haplotype analysis and internal laboratory quality control. If the current project panel is a pan-cancer panel or a whole-exome WES project, then filter is performed and the output is FILTER: Benign_Variant.

[0052] Default priority: If the site does not trigger any of the above explicit filtering rules and there is no evidence of verification artifacts, output PASS:Default_Retained.

[0053] As an alternative implementation, the rule-based label adjudication logic described above can also be implemented using a weighted scoring model or a machine learning classifier. As long as the model's input is multi-dimensional attribute labels or their corresponding original features, and the output is a state identifier representing retention or filtering, and its internal decision boundary can be trained or configured to simulate the aforementioned priority ranking and conflict resolution mechanisms, it should be considered within the scope of this invention. For example, positive weights can be assigned to the first type of label, negative weights to the second type of label, and the final state can be determined by a total score threshold; or a random forest classifier can be trained using labeled true and false positive datasets to automatically learn the nonlinear game relationship between labels. This alternative provides greater flexibility and scalability while maintaining the core technical concept.

[0054] When using a weighted scoring model to adjudicate conflicts, the weight settings for each label and the formula for calculating the overall score S are as follows: in Indicates whether the i-th attribute tag is included. Assign the corresponding weights, for example: , , (Strong positive pull); , , , (Negative punishment).

[0055] The decision rule is as follows: when the overall score S > 0, a PASS state is generated; when S < 0, a FILTER state is generated. This mathematical model completely and equivalently maps the hierarchical game logic of the aforementioned coverage and exemption rules.

[0056] The data processing method provided in this embodiment decouples complex multi-source heterogeneous features into independent labels and, combined with a dynamic threshold adaptation evaluation method for sample types and a multi-label conflict resolution mechanism, fundamentally changes the rigid mode of traditional fixed threshold filtering. This method can maximize the retention of potentially valuable real variant signals while strictly controlling background artifacts, significantly improving the accuracy and reliability of sequencing data analysis and providing a high-quality data foundation for downstream bioinformatics interpretation.

[0057] Based on the above embodiments, the following specific application examples are given.

[0058] Application Example 1: Preservation of Ultra-Low Frequency Hotspot Mutations in ctDNA Liquid Biopsy Samples This embodiment provides a specific data processing scenario for circulating tumor DNA (ctDNA) liquid biopsy samples. In ctDNA detection, due to the extremely low proportion and severe fragmentation of tumor DNA in peripheral blood, true mutations manifest as ultra-low allele frequencies (AFs), making traditional hard filtering prone to missed detections.

[0059] The specific processing procedure is as follows: Data Acquisition and Feature Extraction: The system read the VCF file exported from peripheral blood ctDNA sequencing of a lung cancer patient and extracted a C>T mutation (corresponding to the p.L858R protein change) at the EGFR gene coordinate (chr7:55249071). The data quality characteristics were calculated as follows: mutation depth. Supporting the number of variant Reads Allele frequency %.

[0060] Dynamic threshold determination: The system identifies the sample's metadata identifier as ctDNA_Liquid_Biopsy, queries the preset mapping table (as shown in Table 1), and determines the allele frequency threshold for this sample type. %, mutation depth threshold .

[0061] Attribute tag mapping: because %, the system maps the quality defect label (LowAF_Tag) in the second type of label to this site; Meanwhile, the system queries the Cancer Hotspots and CIVIC databases and finds that the p.L858R mutation precisely hits a known first-order oncogenic hotspot. Therefore, it is mapped to the hotspot mutation tag (Hotspot_Tag) in the first category of tags. The normal control library (PoN) was queried, and this site did not appear in the PoN. ).

[0062] Integrated adjudication and conflict resolution: This site carries both a Hotspot_Tag (Type 1 tag) indicating a tendency to retain it and a LowAF_Tag (Type 2 tag) indicating a tendency to filter it. The system triggers an exemption rule: high-confidence clinical hotspot attributes exempt low-frequency quality defect restrictions. The integrated adjudication engine ultimately generates a PASS status identifier indicating retention, ensuring that this critical targeted drug use site is not incorrectly excluded.

[0063] Application Example 2: Background Noise Overlay Filtering for FFPE Tissue Samples This embodiment provides a specific data processing scenario for formalin-fixed paraffin-embedded (FFPE) tissue samples. FFPE samples are prone to generating numerous deamination artifacts (C>T / G>A) due to DNA damage, and are also prone to generating recurring systematic background noise during sequencing library construction.

[0064] The specific processing procedure is as follows: Data Acquisition and Feature Extraction: The system read the variation data of a colorectal cancer FFPE tissue sample and extracted the KRAS gene G12D mutation site. Sequencing features are as follows: , , %.

[0065] Dynamic threshold determination: The sample type is identified as FFPE_Tissue, and its evaluation threshold is determined by matching. %, The quality indicators of this site are all above the evaluation threshold.

[0066] Attribute tag mapping: This site matches a common hotspot in the COSMIC database and is mapped to the first-class label Hotspot_Tag; Meanwhile, the system compared the same batch and historically accumulated normal control sample library (PoN) and found that this specific coordinate and base change appeared 5 times in normal human blood samples and blank controls. This indicates that it belongs to the typical library construction or hybridization capture of systematic artifacts, and therefore maps to the background noise label (PoN_Tag) in the second type of label.

[0067] Integrated adjudication and conflict resolution: This site carries both Hotspot_Tag and PoN_Tag. The system triggers a coverage rule: confirmed technical artifacts (PoN_Tag) have higher priority than clinical hotspot attributes, thus covering the Hotspot_Tag. The integrated adjudication engine generates a FILTER for characterization filtering: PoN_Noise status flag, successfully eliminating false positive interference.

[0068] Application Example 3: Blacklist Relative Abundance Assessment and High-Frequency Real Mutation Protection Scenario This embodiment provides a specific data processing scenario based on dynamic assessment of a preset blacklist based on relative abundance to protect high-frequency real mutations.

[0069] The specific processing procedure is as follows: Data Acquisition and Feature Extraction: The system analyzed a lung adenocarcinoma sample and extracted a missense mutation site in the TP53 gene. Current sequencing abundance... .

[0070] Blacklist comparison and relative abundance assessment: The system queries the preset blacklist database and finds that the gene coordinates match a blacklist record (this location is recorded in the historical artifact capture set, with a historical average artifact abundance). %); The system executes the relative abundance determination logic: calculates the preset multiple threshold (set to...). ), %. Due to the current actual abundance If the signal is significantly higher than the historical background noise level, the system determines that the current signal is not a technical artifact and will not map it to the blacklist tag.

[0071] Further exemption based on the first-category tag: After comparison with the knowledge base, this TP53 variant was also assessed as a carcinogenic variant by OncoKB (mapped to the Oncogenic_Tag). According to the rules, sites carrying the first-category tag are directly exempt from blacklist matching assessment.

[0072] Overall decision: The site successfully passed the first and second priority assessments, and a PASS status identifier with preserved characteristics was ultimately generated, avoiding the false impression of misclassifying high-frequency real driver mutations as known blacklist mutations.

[0073] Application Example 4: Preset Special Genes (such as BIM Genes) Forced Retention Scenario This embodiment provides a forced protection data processing scenario for key variants in a preset special gene list.

[0074] The specific processing procedure is as follows: Data Acquisition and Feature Extraction: The system analyzed a hereditary tumor screening sample and extracted a 298bp deletion variant from the intron region of the BIM (BCL2L11) gene. Sequencing features showed that the mutation depth was low. (and it only supports a maximum of 3 Reads).

[0075] Dynamic threshold comparison: This identifies the sample type as a germline control. The standard germline threshold requires DP ≥ 50X. According to conventional standards, this site will be tagged with LowDP_Tag due to insufficient depth and will face the risk of being filtered out.

[0076] The hierarchical logic in comprehensive adjudication: The system enters the second priority (forced preservation and special gene protection layer) judgment; The system retrieves a pre-defined list of special genes and identifies the BIM gene as a pre-defined gene associated with genetic susceptibility and drug resistance (BIM deletion polymorphism is closely related to EGFR-TKI targeted drug resistance). Triggering the mandatory retention rule: The system ignores it. If the value falls below the evaluation threshold, the subsequent filtering logic is immediately interrupted.

[0077] Final output: Generate a PASS: SpecialGene_BIM status identifier with preserved characteristics, ensuring that gene variants with special clinical and genetic significance are not mistakenly rejected by the quality hard threshold.

[0078] Application Example 5: Conditional Triage Scenario for Clinically Benign Labels in Different Panels This embodiment provides a data processing scenario for conditionally triaging clinically benign labels based on the Panel gene list type of the current testing item.

[0079] The specific processing procedure is as follows: Data acquisition and feature extraction: The system reads the variation data and extracts a synonymous mutation site (c.2612C>T, p.P871P) on the BRCA1 gene.

[0080] Attribute tag mapping: The system queries the ClinVar and InterVar databases. This site is clearly marked as benign / likely benign polymorphism (Benign / Likely Benign), so it is mapped to the clinical benign tag (Benign_Tag) in the second category of tags.

[0081] Conditional decision logic (third priority execution): Scenario A (Hereditary Breast Cancer Panel): If the currently entered test item is identified as BRCA1 / 2_Hereditary_Panel, the system recognizes that this panel belongs to the genetic susceptibility screening project. The adjudication engine triggers the special retention rule, marks the benign site as PASS: Polymorphism_QC and retains it for haplotype linkage analysis and laboratory-internal linkage disequilibrium (LD) quality control.

[0082] Scenario B (Pan-Cancer Somatic Panel): If the currently input test item is identified as Pan_Cancer_500_Panel, the system recognizes that this item is tumor plasmid mutation analysis. The adjudication engine triggers a filtering rule, generates a FILTER:Benign_Polymorphism status identifier, and removes it to reduce redundancy in downstream clinical report interpretation.

[0083] Application Example 6: Forced Filtering Scenario Based on High-Frequency Population Database This embodiment provides a data processing scenario that performs forced filtering on high-frequency germline polymorphisms based on a healthy population genome database (such as gnomAD).

[0084] The specific processing procedure is as follows: Data acquisition and feature extraction: The system read the variation data of a solid tumor FFPE sample and extracted a missense mutation site of unknown significance. Sequencing quality was good. , %).

[0085] Attribute tag mapping: The system searched Cancer Hotspots and OncoKB and found that this site did not match any hotspots or cancer ratings (no Category 1 label). The system queried the gnomAD population database and found the allele frequency of this locus in East Asian populations. The percentage is much higher than the preset high-frequency threshold (1.0%), so the system maps it to a high-frequency population tag (PopFreq_Tag).

[0086] Comprehensive ruling (first priority execution): The system evaluates this site with the highest priority (forced filtering layer); Since this site carries the PopFreq_Tag and does not have any strong evidence of pathogenic hotspots (no Hotspot_Tag or Oncogenic_Tag), the system determines that it is a germline single nucleotide polymorphism (SNP) common in the general healthy population, rather than a tumor-specific somatic driver mutation.

[0087] Final output: Directly generates the FILTER: High_Population_Frequency status flag for characterization filtering, effectively preventing high-frequency germline background contamination of the tumor somatic mutation spectrum.

[0088] System Implementation Examples This embodiment also provides a data processing system for sequencing mutation sites, including a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the data processing method for sequencing mutation sites as described in the method embodiment. The storage medium can be a tangible, non-volatile memory, such as flash memory, a hard disk, or an optical disk, or a medium capable of carrying program code through signal transmission. When the processor loads and executes the program, it can drive hardware resources to complete the entire process from data acquisition, dynamic threshold adaptation, multi-dimensional label mapping to priority comprehensive decision-making, thereby achieving the same technical effect as in Embodiment 1. Through this combination of hardware and software, the technical solution of this invention can be flexibly deployed and reused in different computing environments, significantly improving the standardization and processing efficiency of the sequencing data analysis process.

[0089] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A data processing method for sequencing mutation sites, characterized in that, The method includes: Acquire sequencing mutation site data and extract multi-dimensional feature data from the sequencing mutation site data; Identify the sample type identifier corresponding to the sequencing mutation site data, and determine the evaluation threshold of the sequencing mutation site based on the sample type identifier; Based on the multi-dimensional feature data, at least one attribute label is mapped to the corresponding sequencing mutation site; wherein, at least some of the attribute labels are mapped based on the evaluation threshold. Based on preset tag adjudication rules, the attribute tags carried by the sequencing mutation sites are comprehensively adjudicated to generate a filtering status identifier for the sequencing mutation sites; wherein, the tag adjudication rules include priority ranking and conflict resolution rules for multiple attribute tags coexisting.

2. The method according to claim 1, characterized in that, The multidimensional feature data includes at least one of the following: technical artifact features, clinical significance features, data quality features, and population frequency features; The evaluation threshold for determining the sequencing mutation site based on the sample type identifier includes: Based on the preset mapping relationship between sample type and threshold, the allele frequency threshold and variation depth threshold corresponding to the sample type identifier are determined as the evaluation threshold.

3. The method according to claim 2, characterized in that, The attribute labels include a first type of label that tends to preserve sites and a second type of label that tends to filter sites; The method of comprehensively adjudicating the attribute tags carried by the sequencing mutation sites based on preset tag adjudication rules includes: When the sequencing mutation site carries both the first type of tag and the second type of tag, the decision is made based on the coverage rule or exemption rule in the tag decision rules. The coverage rule indicates that the second type of label has a higher priority than the first type of label to perform a filtering operation, and the exemption rule indicates that the first type of label has a higher priority than the second type of label to perform a retention operation.

4. The method according to claim 3, characterized in that, The second type of labels includes: background noise labels determined based on matching with a normal control sample library, and short tandem repeat region labels determined based on genomic region features; The first type of label includes: hotspot mutation labels determined by matching based on a known hotspot mutation database, and carcinogenicity labels determined by matching based on a tumor knowledge base.

5. The method according to claim 4, characterized in that, The adjudication based on the coverage rule or exemption rule in the label adjudication rules includes: When the sequencing mutation site carries both the hot spot mutation tag and the background noise tag, the coverage rule is triggered, causing the background noise tag to cover the hot spot mutation tag, and generating the filtering status identifier that represents the filtering. When the sequencing mutation site carries both the hot spot mutation tag and the low frequency tag, the low frequency tag indicates that the data quality is lower than the evaluation threshold, triggering the exemption rule, which exempts the hot spot mutation tag from the low frequency tag, and generates the filter status identifier that indicates retention.

6. The method according to claim 2, characterized in that, The sample type identifiers include paraffin-embedded tissue sample identifiers and circulating tumor DNA sample identifiers; The allele frequency threshold and the mutation depth threshold corresponding to the circulating tumor DNA sample identifier are both less than the allele frequency threshold and the mutation depth threshold corresponding to the paraffin-embedded tissue sample identifier.

7. The method according to claim 1, characterized in that, The method of comprehensively judging the attribute tags carried by the sequencing mutation sites based on preset tag adjudication rules to generate a filtering status identifier for the sequencing mutation sites includes: judging the filtering status identifier of the sequencing mutation sites layer by layer according to a preset priority order; the priority order from high to low includes: First priority: When the attribute tag contains a blacklist tag or the sequencing mutation site is not in the gene list of the current detection project, generate the filtering status identifier that represents the filtering. In the second priority, when the attribute tag contains a hotspot mutation tag or a carcinogenic tag but does not contain a background noise tag, the filter status identifier with retained characteristics is generated; when the gene identifier corresponding to the sequencing mutation site belongs to a preset special gene list, the filter status identifier with retained characteristics is generated regardless of the limitation of the evaluation threshold. The third priority is to make a conditional decision based on the gene panel type of the current test item when the attribute label contains a clinically benign label. Default priority: When the attribute tag does not trigger any of the above priority adjudication conditions, the filter status identifier representing retention is generated based on the default retention rule.

8. The method according to claim 3, characterized in that, The mapping of at least one attribute label to the corresponding sequencing mutation site based on the multi-dimensional feature data includes: Obtain a preset blacklist and the historical abundance data recorded in the preset blacklist; When the sequencing mutation site hits the preset blacklist, and the current abundance data of the sequencing mutation site is lower than the preset fold of the historical abundance data, the sequencing mutation site is mapped to a blacklist label. When the sequencing mutation site carries the first type of tag, the matching evaluation of the preset blacklist is exempted.

9. A data processing system for sequencing mutation sites, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by a processor, it implements the data processing method for sequencing mutation sites as described in any one of claims 1 to 8.

Images