Tumor genome variation detection method, device and equipment and storage medium

Abstract

The invention provides a tumor genome variation detection method, device and equipment and a storage medium, and relates to the technical field of bioinformatics. The detection method comprises the following steps: acquiring sequencing data and comparing the sequencing data to a reference genome to obtain a comparison sequence; and performing somatic mutation detection based on the comparison sequence. According to the method, the problem of noise interference in clinical FFPE sample detection is effectively solved through a dual mechanism of combining statistical model preliminary screening and chain orientation preference scoring. According to the method, the low-frequency mutation detection sensitivity is ensured, meanwhile, the chain preference false positive is accurately eliminated, and the mutation detection specificity and the result reliability are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of bioinformatics technology, and in particular to a method, apparatus, device, and storage medium for detecting tumor genome variations. Background Technology

[0002] Precision oncology relies heavily on next-generation sequencing (NGS) technology for comprehensive detection of genomic variations in patients. With the continuous accumulation of clinical testing data, real-world data (RWD) has become a crucial source of evidence for drug development, treatment optimization, and scientific translation. Sequencing tumor tissue or blood samples and analyzing somatic mutations, copy number variations, and genomic rearrangements can provide clinicians with critical treatment guidance. To realize the value of real-world data, ensuring the accuracy and standardization of sequencing data analysis is essential; this is a prerequisite for building high-quality clinical databases.

[0003] Currently, genomic data analysis primarily relies on alignment and variant detection workflows based on reference genomes. In research, high-depth data generated by whole-genome sequencing (WGS) or whole-exome sequencing (WES) are typically utilized, combined with public databases such as TCGA and ICGC for analysis. Most existing open-source analysis tools are designed for high-quality, fresh, frozen tissue samples. Their core algorithms usually assume high DNA quality and homogeneity in the sequenced samples, identifying biological variations through standard steps such as alignment, deduplication, and variant recall.

[0004] However, in actual clinical testing scenarios in hospitals, most samples are formalin-fixed paraffin-embedded (FFPE) tissue. Unlike fresh tissue used in research, FFPE samples undergo severe DNA damage during fixation and preservation, resulting in significant deamination (such as C>T conversion) and fragmentation, which introduces substantial background noise into sequencing data. Existing analytical workflows designed for research samples often lack targeted damage correction mechanisms when processing such clinical samples, making it difficult to effectively distinguish between extremely low-frequency (e.g., ≤2%) true somatic mutations and sequencing false positives or artifacts introduced by sample preparation. Furthermore, for the detection of genomic rearrangements (such as fusion genes), current technologies lack unified judgment criteria, leading to discrepancies in the description of variant structures and annotation of clinical significance between different platforms.

[0005] In summary, due to the lack of a unified standard operating procedure optimized for low-quality clinical samples, existing analytical methods face numerous challenges in processing real-world clinical data. Inconsistent parameter settings and reference standards between different testing batches or platforms weaken the comparability and reproducibility of test results. Furthermore, insufficient ability to analyze low-frequency mutations and complex structural variations makes it difficult to transform raw sequencing data into accurate and interpretable clinical evidence, thus limiting the effective integration and in-depth analysis of multi-source clinical data. Summary of the Invention

[0006] In a first aspect, the present invention provides a method for detecting tumor genomic variations, comprising: Sequencing data of tumor samples are obtained and aligned to a reference genome to obtain aligned sequences; Somatic mutation detection based on the alignment sequence includes: The alignment sequences are traversed to identify candidate mutation sites, and the mutation probability scores of the candidate mutation sites are calculated based on a statistical model for initial screening. For candidate mutation sites after initial screening, the number of mutation reads and reference reads on the forward and reverse strands are counted respectively, and the strand orientation preference score is calculated based on the statistical results. By combining the mutation probability score with the chain orientation preference score, false positive mutations with significant chain orientation preference are filtered out to determine the final somatic mutation.

[0007] In an optional implementation, calculating the mutation probability score of the candidate mutation site based on a statistical model includes: Obtain the total sequencing depth and number of mutated base reads for this candidate mutation site; The probability value of the number of reads of the mutated base at the candidate mutation site is calculated based on the Poisson distribution model, and the mutation probability score is calculated as follows: ;

[0008] Where p represents the probability value; Poisson represents the Poisson distribution function; k represents the number of mutated base reads; λ represents the expected value, calculated as λ = D × ε; D represents the total sequencing depth; ε represents the preset single-base sequencing error rate; S mut represents the mutation probability score; lg represents the logarithm to the base 10.

[0009] In an optional implementation, the calculation of the chain orientation preference score based on statistical results includes: Construct a 2×2 contingency table; the four elements of the contingency table are: the number of mutation reads on the forward chain, the number of reference reads on the forward chain, the number of mutation reads on the reverse chain, and the number of reference reads on the reverse chain; Fisher's exact test was performed on the contingency table to obtain two-tailed test values; The chain orientation preference score is calculated using the following formula: ; Among them, S bias p represents the chain orientation preference score; Fisher represents the two-tailed test value of the Fisher exact probability test; lg represents the logarithm to the base 10.

[0010] In an optional implementation, filtering out false-positive mutations with significant strand orientation bias to determine the final somatic mutation includes: Sequencing data of normal samples matched with the tumor sample are obtained, and mutation probability scores of the normal samples at candidate mutation sites are calculated. The final score will be used to determine the outcome; the formula for calculating the final score is as follows: ; Among them, S final This represents the final score used to determine whether a mutation is a somatic mutation; S mut S represents the mutation probability score of the candidate mutation site in the tumor sample; norm S represents the mutation probability score of the candidate mutation site in the normal sample; bias This represents the chain orientation preference score; If and only if the S final If the score exceeds the preset threshold, the candidate mutation site is retained.

[0011] In an optional implementation, when a matching normal sample is unavailable, the somatic mutation detection further includes: Extract features from the candidate mutation sites, wherein the features include at least mutation frequency and population frequency; The extracted features are input into a pre-trained XGBoost machine learning model for classification to identify somatic mutations.

[0012] In an optional implementation, the method further includes a genome rearrangement detection and annotation step: The breakpoints of genomic rearrangement events are identified based on the aligned sequences; Analyze the transcription direction and gene structure preservation of the genes on both sides of the breakpoint; When the transcription directions of the genes on both sides of the breakpoint are consistent, and the gene upstream of the breakpoint retains at least one complete coding region and the gene downstream of the breakpoint retains a complete start codon, the genome rearrangement event is determined to be a fusion gene.

[0013] In an optional implementation, the genome rearrangement detection and annotation step further includes: When the transcriptional directions of genes on both sides of the breakpoint conflict, it is marked as a chain conflict; When the breakpoint is located inside a single gene, it is marked as a break within a single gene; Extract sequences of preset lengths upstream and downstream of the breakpoint, and adjust the case representation of the sequences according to the positive and negative strand directions of the gene.

[0014] In an optional implementation, the method further includes a copy number variation detection step: Calculate the tumor purity, ploidy, and allele-specific copy number of the sample; Identifying copy number variation regions based on the depth ratio of aligned sequences; When a matching normal sample is unavailable, copy number variation analysis is performed using a pre-defined baseline library.

[0015] In an optional implementation, the method further includes a step of calculating a microsatellite instability score: Obtain sequencing data of normal samples that match the tumor sample; The microsatellite instability score is calculated using the following formula: ; Where MSI% represents the microsatellite instability score; N total N represents the total number of valid microsatellite loci located within the detection area and whose sequencing depth meets the preset standard; modified This represents the number of unstable microsatellite loci after correction; and the corrected unstable microsatellite loci must simultaneously meet the following two conditions: A. Chi-square test shows that the distribution difference between tumor samples and normal samples is significant (FDR ≤ 0.05), B. The entropy value of tumor samples is greater than that of normal samples.

[0016] In an optional implementation, the method further includes a step of calculating the tumor mutational burden based on determined somatic mutations, using the following formula: ; Wherein, TMB represents tumor mutation burden; L CDS N represents the total length of the protein-coding region within the target area; somaticThis represents the total number of effective somatic nonsynonymous mutations. The effective somatic nonsynonymous mutations are those obtained by filtering out false positive mutations with significant strand orientation bias in the previous step, and must simultaneously meet the following exclusion conditions: A. Exclude mutations with a mutation frequency less than a preset threshold; B. Exclude known tumor primary hotspot mutations; C. Exclude mutations located on HLA genes.

[0017] In an optional implementation, the method further includes a clinical annotation step: The identified somatic mutations or fusion genes are matched with a pre-defined clinical knowledge base; The clinical knowledge base contains the correspondence between gene mutations and drug treatment regimens, and outputs the clinical evidence level of the mutation based on the matching results.

[0018] In a second aspect, the present invention provides a tumor genome mutation detection device, comprising: The acquisition module is used to acquire sequencing data of tumor samples and align the sequencing data to a reference genome to obtain the aligned sequence. A detection module, used for somatic mutation detection based on the alignment sequence, includes: The initial screening unit is used to traverse the alignment sequence to identify candidate mutation sites and calculate the mutation probability score of the candidate mutation sites based on a statistical model for initial screening. The calculation unit is used to count the number of mutation reads and reference reads on the forward and reverse strands for candidate mutation sites after initial screening, and calculate the strand orientation preference score based on the statistical results. A filtering unit is used to combine the mutation probability score with the chain orientation preference score to filter out false positive mutations with significant chain orientation preference and determine the final somatic mutation.

[0019] Thirdly, the present invention provides a computer device including a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the tumor genome variation detection method described in any of the foregoing embodiments.

[0020] Fourthly, the present invention provides a computer storage medium storing a computer program, which, when executed on a processor, implements the tumor genome variation detection method according to any one of the foregoing embodiments.

[0021] The embodiments of this application have the following beneficial effects: To address the common problems of DNA damage and sequencing background noise in FFPE samples in hospital clinical testing scenarios, this invention proposes a targeted method for detecting tumor genomic mutations. By introducing a preliminary screening mechanism based on a statistical model, potential candidate mutation sites can be sensitively identified from massive amounts of alignment data. Even low-frequency mutation signals can be effectively captured, avoiding missed detections caused by rigid filtering and thus ensuring the sensitivity of the detection.

[0022] More importantly, this method innovatively employs a forward and reverse strand statistical strategy to address strand orientation-specific damage (such as C>T deamination) that is easily introduced during formalin fixation and other treatments. By separately counting mutations and reference reads on the forward and reverse strands and calculating a strand orientation preference score, the degree of strand preference for each candidate site can be accurately quantified. Combining this score with a mutation probability score enables specific filtering of false-positive mutations. This "double-checking" mechanism effectively solves the problem of traditional procedures struggling to distinguish between true low-frequency mutations and sequencing artifacts when processing low-quality clinical samples, significantly improving the specificity and accuracy of somatic mutation detection and ensuring the reliability and reproducibility of clinical test results. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and therefore should not be considered as a limitation on the scope of protection of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the hardware operating environment involved in an embodiment of the tumor genome variation detection method of the present invention; Figure 2 This is a schematic flowchart of Example 1 of the tumor genome variation detection method of the present invention; Figure 3 This is a detailed flowchart of step S200 in Example 3 of the tumor genome variation detection method of the present invention; Figure 4 This is a detailed flowchart of step S300 in Example 4 of the tumor genome variation detection method of the present invention; Figure 5 This is a detailed flowchart of step S700 in Example 4 of the tumor genome variation detection method of the present invention; Figure 6 This is a schematic diagram of the module connections of the tumor genome variation detection device of the present invention. Detailed Implementation

[0025] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0026] The components of the embodiments of this application described and illustrated in the accompanying drawings can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0027] In the following, the terms “comprising,” “having,” and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as excluding, firstly, the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more features, numbers, steps, operations, elements, components, or combinations thereof.

[0028] Furthermore, the terms "first," "second," and "third" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0029] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0030] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0031] like Figure 1 The diagram shown is a structural schematic of the hardware operating environment of the terminal involved in an embodiment of the present invention.

[0032] The tumor genome mutation detection system (device) of this invention can be a PC, or a mobile terminal device such as a smartphone, tablet, or laptop. The tumor genome mutation detection system may include: a processor 1001, such as a CPU; a network interface 1004; a user interface 1003; a memory 1005; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen, an input unit such as a keyboard, or a remote control; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM memory or a stable memory, such as a disk storage device. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001. Optionally, the tumor genome mutation detection system may also include RF (Radio Frequency) circuitry, audio circuitry, a Wi-Fi module, etc. In addition, this tumor genome mutation detection system can also be equipped with other sensors such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, which will not be described in detail here.

[0033] Those skilled in the art will understand that Figure 1 The system (device) shown is not intended to limit it and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Figure 1 As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a data interface control program, a network connection program, and a tumor genome mutation detection program.

[0034] In summary, the method provided by this invention effectively solves the problems of high detection noise and high false positive rate in clinical FFPE samples due to DNA damage. Through full-scan initial screening based on a statistical model, high sensitivity capture of low-frequency mutation signals is ensured; simultaneously, by combining statistical analysis of forward and reverse strand read counts with strand orientation preference scoring, sequencing artifacts with significant strand preferences are accurately identified and eliminated. This dual filtering mechanism significantly improves the specificity and accuracy of somatic mutation detection, providing reliable technical support for the standardized analysis of real-world clinical data.

[0035] Example 1 Reference Figure 2 This embodiment provides a method for detecting tumor genomic variations, including: Step S100: Obtain sequencing data of the tumor sample and align the sequencing data to a reference genome to obtain the aligned sequence.

[0036] This step is a fundamental prerequisite for variant detection. It involves obtaining raw DNA sequence fragments (Reads) from a high-throughput sequencer (such as the Illumina platform) and using bioinformatics algorithms to locate these short, jumbled sequence fragments to specific chromosomal positions on a standard human genome (such as hg19 or hg38). The principle is based on the complementary pairing principle of nucleic acid sequences. The Reads in the sequencing data have a high degree of similarity to the reference genome sequence; the optimal matching position is found through alignment algorithms.

[0037] Specifically, the input can be raw data in FASTQ format. After alignment processing, the output is a BAM or SAM file containing alignment location, alignment quality, and CIGAR value (indicating alignment status such as insertion or deletion). This step transforms disordered data into ordered genomic coordinate data, which is the basis for subsequent variant identification.

[0038] For example, industry-standard algorithms such as BWA-MEM can be used for comparison. This process typically includes preprocessing steps such as removing PCR duplicates to reduce bias in subsequent analyses.

[0039] Specifically, the FASTP software can be used to detect sequencing quality and remove low-quality sequences, the BWA-MEM algorithm can be used to align the sequences to the reference genome, and samblaster can be used to remove sequence duplications.

[0040] Step S200: Somatic mutation detection based on the aligned sequence. Specifically, this may include: Step S210: Traverse the alignment sequence to identify candidate mutation sites, and calculate the mutation probability score of the candidate mutation sites based on a statistical model for initial screening.

[0041] This step is the first stage of variant detection (high-sensitivity scanning).

[0042] The term "traversal" refers to checking every base position covered by the sequencing without pre-setting hotspots. For sites where non-reference bases (i.e., mutated bases) are found, the determination is not based on simple frequency thresholds (such as only looking at >5%), but rather on a statistical method to calculate the "confidence" score of the mutation event.

[0043] The principle behind this is the separation of signal and noise. Sequencing inherently involves a sequencing error rate. Statistical models are used to assess whether the currently observed number of mutated bases stems from random sequencing errors or from genuine biological variations.

[0044] Specifically, a preferred approach may employ a Poisson distribution model. A probability score is calculated, and the higher the score, the lower the probability that the mutation was caused by a sequencing error, i.e., the greater the likelihood of it being genuine.

[0045] Compared to traditional hard threshold filtering, this statistical model-based initial screening can retain low-frequency mutations to the greatest extent, significantly improve detection sensitivity, and avoid missed detections.

[0046] In addition to the Poisson distribution, the binomial distribution or Bayesian inference model can also be used to calculate this probability score.

[0047] Step S220: For the candidate mutation sites after initial screening, the number of mutation reads and reference reads on the forward and reverse strands are counted respectively, and the strand orientation preference score is calculated based on the statistical results.

[0048] This step is for feature extraction targeting sequencing artifacts. NGS sequencing constructs libraries by breaking down double-stranded DNA. True biological mutations should appear in similar proportions on both the forward and reverse strands. If a mutation is almost entirely distributed on the sense strand or entirely on the antisense strand, it is highly likely a false positive.

[0049] Its principle is based on the principle of Strand Orientation Bias (SOB).

[0050] It should be noted that formalin fixation leads to cytosine deamination (C>T), a chemical damage that often occurs on single-stranded DNA, resulting in significant strand specificity during sequencing. Polymerase errors during amplification also frequently manifest as strand bias.

[0051] Specifically, a contingency table can be constructed, creating a 2x2 table: Table 1, Contingency Table

[0052] For example, Fisher's exact test can be used to calculate a two-sided p-value, quantifying the change or shift at each mutation site. A higher score indicates a more severe chain bias and a greater likelihood of false positives.

[0053] Step S230: Combine the mutation probability score with the strand orientation preference score to filter out false positive mutations with significant strand orientation preference and determine the final somatic mutation.

[0054] This step is the final decision-making step. It does not look at the strength of mutation (probability score) or the size of preference (preference score) in isolation, but combines the two through mathematical operations or logic to achieve precise filtering.

[0055] It should be noted that relying solely on the "mutation probability score" will introduce a large number of false positives caused by FFPE damage (because damage also generates many mutant reads, resulting in a high probability score); relying solely on the "chain preference score" may mistakenly eliminate some real mutations in low sequencing depth or certain special sequence contexts; in this embodiment, a "penalty mechanism" is used, that is, the "bias score" is deducted from the "confidence score" of the mutation.

[0056] This method significantly improves specificity while preserving extremely low-frequency mutations (high sensitivity). It solves the problem of "noise looking very much like real signals" in clinical samples, ensuring that the final reported somatic mutations are highly reliable, reducing the workload of subsequent validation and the risk of clinical misdiagnosis.

[0057] In addition, if a matched normal control sample is available, the background noise score (S) of the normal sample can also be incorporated into this step. norm This will further improve accuracy.

[0058] Example 2 This embodiment provides a method for detecting tumor genomic variations. Step S210, which calculates the mutation probability score of the candidate mutation sites based on a statistical model, includes: Step S211: Obtain the total sequencing depth and number of mutated base reads for the candidate mutation site.

[0059] Step S212: Calculate the probability value of the number of reads of the mutated base at the candidate mutation site based on the Poisson distribution model, and calculate the mutation probability score. The calculation formula is as follows: (Formula 1); (Formula 2); Where p represents the probability value; Poisson represents the Poisson distribution function; k represents the number of mutant base reads; λ represents the expected value, calculated as λ = D × ε (Formula 3); D represents the total sequencing depth; ε represents the preset single-base sequencing error rate; S mut represents the mutation probability score; lg represents the logarithm to the base 10.

[0060] This embodiment aims to quantify the likelihood that a candidate mutation site is caused by random sequencing errors. It does not directly rely on the mutation frequency (VAF), but rather calculates the probability of the observed mutation support number occurring under the assumption of "pure background noise" using a statistical model.

[0061] The principle behind this is based on the statistical principle of rare events. Sequencing errors are typically low-probability events that occur randomly. Equation 3 defines the expected value of the background noise. Assuming a sequencing depth D of 1000 and a single-base error rate ε of 0.001 (i.e., Q30), the expected number of erroneous bases to be seen λ = 1. Equation 1 calculates the probability of actually observing k (e.g., 10) mutated bases when only one erroneous base is expected to be seen. Equation 2 converts the probability into a score similar to the Phred quality score. The smaller the probability p (i.e., the less likely it is to be a random error), the higher the score S. mut The higher.

[0062] For example, scene A (similar to noise) has a depth of 1000 and two abrupt changes. λ=1, and we calculate p ≈0.18, S mut ≈7.35. The low score indicates that it is likely just background noise.

[0063] Scene B (similar to a signal) has a depth of 1000 and 10 abrupt changes. λ=1, therefore p≈10 is calculated. -7 S mut ≈70. A high score indicates that this is highly unlikely to be a random error and is more likely a genuine mutation.

[0064] Compared to fixed VAF thresholds (e.g., requiring >1%), this method adapts to sequencing depth. At ultra-high depths, even with low VAF (e.g., 0.1%), as long as S... mut It is high enough that it can still be recognized, thus greatly improving the sensitivity.

[0065] Furthermore, the error rate ε can be dynamically adjusted based on the quality report of the actual sequencing data, rather than being fixed at 0.001.

[0066] In some implementations, step S220, calculating the chain orientation preference score based on statistical results, includes: Step S221: Construct a 2×2 contingency table; the four elements of the contingency table are: mutation read count on the forward chain, reference read count on the forward chain, mutation read count on the reverse chain, and reference read count on the reverse chain.

[0067] Step S222: Perform Fisher's exact test on the contingency table to obtain two-sided test values.

[0068] Step S223, calculate the chain orientation preference score, using the following formula: (Formula 4); Among them, S bias p represents the chain orientation preference score; Fisherrepresents the two-tailed test value of the Fisher exact probability test; lg represents the logarithm to the base 10.

[0069] The processing method described in this embodiment is used to detect whether a mutation has significant "strand specificity". If a mutation appears only on the forward or reverse strand, but is completely missing on the other strand, this is usually a false positive in the sequencing process.

[0070] The principle behind this is the independence test. In an ideal real mutation, the distribution ratio of the mutated bases on the forward and reverse strands should be approximately the same as the distribution ratio of the reference bases on the forward and reverse strands (i.e., the two are independent).

[0071] The Fisher test described above is used to accurately calculate the significance of this distributional difference (P-value) when the sample size is small. In Equation 4, the P-value is converted into a bias score. The smaller the P-value (the more extreme the distributional difference), the higher the score S. bias The higher the value, the more severe the chain preference.

[0072] For example, a high-bias (typical false positive) case shows mutations all in the positive strand (10 bases) and none in the reverse strand; the reference bases are 495 in each strand. The calculated P ≈ 0.0019, S bias ≈27.27. High score, indicating strong bias. Unbiased (typical of real mutations) example, mutated 5 positive strands and 5 anti strands. Calculated P=1.0, S bias =0. Low score, indicating a uniform distribution.

[0073] The method provided in this embodiment can accurately identify common deamination damage (C>T often only occurs on a single strand) in FFPE samples. Traditional filters usually only look for "whether there is a single strand with a value of 0", while the Fisher test can handle more complex situations (such as 100 positive strands and 1 anti-strand strand), providing a quantitative risk assessment.

[0074] Furthermore, step S230, filtering out false-positive mutations with significant strand orientation bias to determine the final somatic mutation, includes: Step S231: Obtain sequencing data of normal samples matched with the tumor sample, and calculate the mutation probability score of the normal sample at the candidate mutation site; Step S232: Make a judgment based on the final score; the formula for calculating the final score is as follows: (Formula 5); Among them, S final This represents the final score used to determine whether a mutation is a somatic mutation; S mut S represents the mutation probability score of the candidate mutation site in the tumor sample; normS represents the mutation probability score of the candidate mutation site in the normal sample; bias This represents the chain orientation preference score; Step S233, if and only if the S final If the score exceeds the preset threshold, the candidate mutation site is retained.

[0075] This embodiment presents the core decision-making mechanism of the entire algorithm. It performs two steps: "background removal" and "preference removal." Only when a mutation is strong enough (S... mut High), and not present in normal samples (S norm Low), and no chain preference (S) bias Only when the value is low will it be retained.

[0076] Its synergistic effect and mechanism lie in, firstly, the first subtraction, (S mut -S norm This is a somatic cell-specific filter. If the mutation is also present in the normal control (e.g., germline mutation or systematic sequencing error), S norm The value will be very high, causing the difference to become smaller, thus resulting in filtering. Secondly, the second subtraction (...-S) bias This is for quality-specific filtration. Even S mut Very high (looks like a strong signal), if it is due to chain damage (S bias (High), ultimately S final They will also be dragged down and thus eliminated.

[0077] Through the above calculations, a comprehensive score S can be calculated. final .

[0078] For example, assuming a threshold of 20, then the real somatic cell mutation is S. mut =70, S norm =0, S bias =0→S final =70 (retained). Germline mutation: S mut =70, S norm =70, S bias =0→S final =0 (filtered). Also, FFPE false positive, S mut =60 (signal strength), S norm =0, S bias =50 (preference for large) → S final =10 (filtered).

[0079] This weighted subtraction mechanism is more scientific than the traditional "multi-step hard filtering". It allows a mutation with a slight chain bias but a very strong signal to be preserved (because S...). mut Large enough to offset S biasAlternatively, a slightly weaker but extremely clean and unbiased mutation is preserved, thus achieving the best balance between sensitivity and specificity.

[0080] Specifically, this can be achieved by writing a script (such as Python or R) to read the BAM file, extract the depth and chain information for each site, and then substituting these into the formula for calculation. A preset threshold (such as S) is required. final ≥20).

[0081] Example 3 Reference Figure 3 This embodiment provides a method for detecting tumor genomic mutations. When a matching normal sample is unavailable, step S200, somatic mutation detection, further includes: Step S240: Extract features of the candidate mutation sites, the features including at least mutation frequency and population frequency.

[0082] In this embodiment, a particular dilemma commonly encountered in clinical testing is addressed: the Tumor-Only mode.

[0083] It's important to note that in standard procedures, distinguishing between somatic mutations (present only in tumors) and germline mutations (present from birth) primarily relies on comparison with the patient's own normal tissue (such as blood) (i.e., a subtraction strategy). This direct subtraction is impossible when a normal control is unavailable. Therefore, this step introduces a machine learning algorithm to attempt to predict whether a mutation is a somatic mutation based on its characteristic patterns.

[0084] The principle of data distribution differences: Somatic mutations and germline mutations differ significantly in certain statistical characteristics.

[0085] The above-mentioned population frequencies indicate that germline mutations usually have a certain frequency in population databases (such as 1000 Genomes, ExAC) (i.e., polymorphic SNPs), while somatic mutations are usually extremely rare or non-existent in healthy individuals.

[0086] The mutation frequency (VAF) mentioned above is typically close to 100% or 50% for homozygous germline mutations, while somatic mutations are affected by tumor purity and heterogeneity, resulting in a wider and generally lower VAF distribution.

[0087] In step S250, the extracted features are input into a pre-trained XGBoost machine learning model for classification to identify somatic mutations.

[0088] XGBoost, as described above, is a gradient boosting algorithm based on decision tree ensembles. It can learn from a large amount of existing data on "paired samples (tumor + normal)" to construct a combination of decision rules that can distinguish between somatic and germline mutations.

[0089] Specifically, the process can be as follows: First, feature extraction is performed, that is, for each detected candidate mutation, a multidimensional feature vector is extracted. Mutation frequency (VAF) refers to the proportion of reads with that mutation site out of the total reads. Population frequency refers to the frequency value of that mutation in the general population obtained by searching public databases (such as 1000 Genomes, gnomAD). Other features may include base substitution type (e.g., C>T), sequence context information, etc. Then, the above feature vectors are input into a pre-trained XGBoost model. The Genorui model outputs a probability value or classification label.

[0090] Through the above processing, even without a normal control, somatic mutations can still be separated from a large number of background germline variations with high precision. This method greatly expands the applicability of the detection, enabling effective gene testing for patients who cannot collect blood controls (such as those with only archived slides). Compared to simple filtering (such as filtering only loci with a population frequency >1%), XGBoost can use non-linear relationships to handle loci in the "fuzzy zone," significantly reducing false positives and false negatives.

[0091] In terms of specific algorithms, during the training phase (offline), a large number (e.g., thousands) of samples with "tumor-normal" pairings can be collected. The gold standard label can be "presence in normal samples": absence in normal samples = somatic cells (1), presence in normal samples = germline (0); features (VAF, gnomAD_AF, 1000g_AF, etc.) are selected to train the XGBoost classifier. During the prediction phase (online), the trained model file can be loaded, and the features of new samples can be input in real time to obtain prediction results.

[0092] In addition to XGBoost, random forests, support vector machines (SVMs), or deep neural networks (DNNs) can also be used to achieve the same classification purpose.

[0093] In addition, in terms of feature engineering, functional prediction scores of mutation sites (such as SIFT, PolyPhen) or inclusion information in the COSMIC database can be added as input features to further improve model performance.

[0094] Example 4 Reference Figure 4 This embodiment provides a method for detecting tumor genomic variations, the method further including step S300, genomic rearrangement detection and annotation step: Step S310: Identify the breakpoint of the genome rearrangement event based on the aligned sequence.

[0095] This step is the initial step in structural variation (SV) detection. By analyzing anomalous alignments (such as Split Reads or Discordant Pairs) in the BAM file, the specific coordinates of genome breaks and reconnections are precisely located.

[0096] As a concrete example, tools such as Delly2, manta, and lumpy can be used to detect gene rearrangement events and filter them in conjunction with the biological rules described in this application.

[0097] Step S320: Analyze the transcription direction and gene structure preservation of the genes on both sides of the breakpoint.

[0098] This step is a key analysis that transforms a simple "physical breakpoint" into a "biological function".

[0099] Specifically, you can consult genome annotation databases (MANE is the standard in clinical practice, while RefSeq, GENCODE, and other databases are also acceptable in research) to determine which gene, exon / intron, and whether the gene itself is transcribed in the positive or negative strand.

[0100] Step S330: When the transcription directions of the genes on both sides of the breakpoint are consistent, and the gene upstream of the breakpoint retains at least one complete coding region and the gene downstream of the breakpoint retains a complete start codon, the genome rearrangement event is determined to be a fusion gene.

[0101] The principle underlying this step is the formation of functional fusion proteins.

[0102] It should be noted that a gene rearrangement capable of translating a functional fusion protein must satisfy specific biological rules: First, the direction must be consistent, meaning that the two gene segments must be transcribed on the same mRNA strand. If the direction is opposite (such as head to head or tail to tail), an effective transcript usually cannot be formed.

[0103] Secondly, the reading frame must be preserved. The upstream gene (5' end) must provide a promoter and at least one coding sequence (CDS), and the reading frame cannot be shifted or terminated prematurely at the breakpoint. The downstream gene (3' end) must preserve the key functional domain after the start codon (ATG) and provide a stop codon.

[0104] Specifically, the algorithm performs logical checks on each detected SV. Only events that fully conform to the above "translatable rules" will be labeled with the gene_fusion tag.

[0105] This method addresses the limitation of traditional SV detection software that only reports location and not function. Through rigorous biological rule filtering, it eliminates a large number of non-coding rearrangements with no clinical significance, directly outputting functional fusion candidates with carcinogenic potential, greatly reducing the difficulty of clinical interpretation.

[0106] Furthermore, step S300, the genome rearrangement detection and annotation step, also includes: Step S340: When the transcriptional directions of the genes on both sides of the breakpoint conflict, it is marked as a chain conflict.

[0107] Step S350: When the breakpoint is located inside a single gene, it is marked as a breakpoint inside a single gene.

[0108] The above steps are for classifying and annotating other rearrangement events that do not meet the criteria for "fusion gene".

[0109] The aforementioned chain conflict refers to two gene segments being transcribed in opposite directions, making it physically impossible to form a continuous transcript.

[0110] Internal breakage of a single gene refers to breakpoints located within the same gene (such as exon deletion or duplication). Such variations do not form new genes, but may lead to loss or alteration of the gene's function.

[0111] This step provides a comprehensive overview of the variants. While these are not classic fusion genes, they are still valuable for understanding the instability of the tumor genome.

[0112] The above steps represent two directions of logical judgment; they can be performed simultaneously or decisions can be made based on logic.

[0113] Step S360: Extract sequences of preset lengths upstream and downstream of the breakpoint, and adjust the case representation of the sequences according to the positive and negative strand directions of the gene.

[0114] It should be noted that, for ease of subsequent validation (such as designing PCR primers), the sequence near the breakpoint needs to be output. Since genes have positive and negative strands, simply extracting the positive strand of the genome often does not reflect biological reality.

[0115] The mechanism is that if the gene is located on the negative strand, the algorithm will automatically take the reverse complement of the extracted sequence and may use lowercase letters to distinguish it.

[0116] This method standardizes the sequence output, allowing for direct integration with subsequent Sanger sequencing validation experiments.

[0117] In some embodiments, the method further includes step S400, a copy number variation detection step: Step S410: Calculate the tumor purity, ploidy, and allele-specific copy number of the sample.

[0118] Step S420: Identify copy number variation regions based on the depth ratio of the aligned sequences; In this step, we perform a process to infer gene copy number (CNV) by utilizing changes in sequencing depth.

[0119] It should be noted that if the copy number of a certain region is amplified, its sequencing depth will be significantly higher than the average level; conversely, deletion will result in decreased depth. The algorithm needs to be combined with tumor purity and ploidy correction baseline because the depth variation in low-purity samples will be diluted by normal cells.

[0120] Specifically, for paired samples, the cnv_facets algorithm can be used to calculate tumor purity and ploidy; for unpaired samples, the cnvkit tool can be used for analysis.

[0121] Step S430: When a matching normal sample is lacking, copy number variation analysis is performed using a pre-defined baseline library.

[0122] This step addresses the challenge of CNV detection in Tumor-Only scenarios.

[0123] Specifically, a "baseline library" can be pre-constructed, containing the average depth distribution of a large number of healthy samples in each exon. The depth of the tumor sample to be tested is compared with this baseline library (normalized) to identify abnormal regions.

[0124] This step expands the applicability of CNV detection, allowing testing to be performed even without a paired blood sample.

[0125] In some embodiments, the method further includes step S500, the calculation step of the microsatellite instability score: Step S510: Obtain sequencing data of normal samples that match the tumor sample.

[0126] Step S520: Calculate the microsatellite instability score using the following formula: (Formula 6); Where MSI% represents the microsatellite instability score; N total N represents the total number of valid microsatellite loci located within the detection area and whose sequencing depth meets the preset standard; modified This represents the number of unstable microsatellite loci after correction; and the corrected unstable microsatellite loci must simultaneously meet the following two conditions: A. Chi-square test shows that the distribution difference between tumor samples and normal samples is significant (FDR ≤ 0.05), B. The entropy value of tumor samples is greater than that of normal samples.

[0127] The above calculation method is a specific optimization of the microsatellite instability (MSI) scoring algorithm.

[0128] The aforementioned microsatellites refer to short repetitive sequences in the genome that are prone to length changes (slippage) when there are defects in MMR genes.

[0129] The entropy values ​​mentioned above are used to measure the degree of disorder in the distribution of sequence lengths. Normal samples have fixed microsatellite lengths and low entropy values; MSI samples have varying lengths and high entropy values.

[0130] In the aforementioned dual filtering mechanism, condition A (FDR≤0.05) uses a chi-square test to ensure that the length distributions of tumors and normal samples are indeed significantly different, excluding random errors. Condition B (entropy comparison) ensures that this difference is due to "tumors becoming more disordered" (MSI feature), rather than due to "normal samples being inherently disordered" (e.g., germline polymorphism).

[0131] Compared to traditional methods that only compare length distributions, introducing entropy comparison can effectively eliminate the interference of innate polymorphic sites and significantly reduce the false positive rate of MSI detection.

[0132] In some embodiments, the method further includes step S600, which involves calculating the tumor mutation burden based on determined somatic mutations, using the following formula: (Formula 7); Wherein, TMB represents tumor mutation burden; L CDS N represents the total length of the protein-coding region within the target area; somatic This represents the total number of effective somatic nonsynonymous mutations. The effective somatic nonsynonymous mutations are those obtained by filtering out false positive mutations with significant strand orientation bias in the previous step, and must simultaneously meet the following exclusion conditions: A. Exclude mutations with a mutation frequency less than a preset threshold; B. Exclude known tumor primary hotspot mutations; C. Exclude mutations located on HLA genes.

[0133] The aforementioned TMB is an important biomarker for immunotherapy, and its calculation must be based on an accurate "non-functional somatic mutation" load.

[0134] The principle behind this approach lies in molecular screening, excluding primary hotspots. Driver mutations (such as EGFR L858R) typically differ from neoantigen production mechanisms and may lead to resistance to immunotherapy; some algorithms suggest excluding them to reflect pure "mutation burden." Additionally, HLA genes are excluded because HLA regions are highly polymorphic and homologous, leading to numerous alignment errors and the introduction of noise; furthermore, HLA mutations themselves may affect antigen presentation, interfering with the predictive value of TMB.

[0135] Based on the high-precision denoised mutation results in the aforementioned implementation methods, and further superimposed with strict biological exclusion rules, the calculated TMB value can more accurately reflect the potential of tumors to produce new antigens, thus improving its accuracy as a predictive indicator of immunotherapy efficacy.

[0136] Example 5 Reference Figure 5 This embodiment provides a method for detecting tumor genomic variations, the method further including step S700, clinical annotation step: Step S710: Match the identified somatic mutations or fusion genes with a pre-defined clinical knowledge base.

[0137] Step S720: The clinical knowledge base contains the correspondence between gene mutations and drug treatment regimens, and outputs the clinical evidence level of the mutation based on the matching results.

[0138] In this embodiment, the method provided belongs to the final stage of the bioinformatics analysis process: the "clinical translation" stage. The method provided in the previous embodiment solved the problems of "accurate measurement" and "correct calculation," while this embodiment aims to solve the problem of "understandable interpretation." It transforms abstract gene variation information (such as EGFR L858R, EML4-ALK Fusion) into specific clinical treatment recommendations (such as "recommended use of osimertinib, evidence level 1").

[0139] The method can be based on evidence-based medicine principles. Clinical knowledge bases (such as OncoKB and CIViC) store a large number of expert-reviewed "variant-drug-evidence level" triple relationships. Furthermore, the algorithm uses the standardized naming of detected variants (such as the HGVS format) as the query key to perform precise retrieval in the knowledge base. Once a match is found, the associated drug and evidence level can be extracted.

[0140] Specifically, identified somatic mutations and fusion genes can be organized into a standard format (e.g., HUGO GeneSymbol + AA Change, such as BRAF V600E). These variant entries are then compared to a pre-defined localized database (e.g., the local version of OncoKB). Based on the matching results, the level of evidence is automatically assigned. This typically follows the grading standards of FDA or NCCN guidelines, for example: Level 1: FDA-approved standard treatment for this type of tumor.

[0141] Level 2: Standard treatments for other indications of this tumor.

[0142] Level 3 / 4: Supported by clinical or biological evidence.

[0143] Level R1: Clear evidence of drug resistance.

[0144] Using the methods described above, a structured clinical report containing "variant sites," "recommended drugs," and "level of evidence" can be generated. This replaces the tedious process of manually reviewing literature, greatly shortening the report generation cycle; it avoids subjective biases or omissions that may occur with manual interpretation, ensuring that all patient reports are based on the latest authoritative guidelines; and by regularly updating the backend knowledge base, it ensures that the basis for clinical decision-making is always up-to-date.

[0145] For example, a structured database (such as SQL or JSON format) can be built or imported, containing fields such as Gene, Alteration, Cancer_Type, Drug, and Level. A script (Python / Perl) can be written to iterate through the list of test results, performing a database query operation (Select) for each item. from KB where Gene=... and Alteration=...).

[0146] In addition to matching OncoKB, it can also match multiple databases such as ClinVar (pathogenicity rating) and COSMIC (frequency statistics) to provide more comprehensive annotations.

[0147] In addition to matching marketed drugs, it can also automatically match recruiting clinical trials based on mutation information to find new treatment opportunities for advanced patients.

[0148] Furthermore, to achieve comprehensive analysis of multimodal data, the device described in this application is also configured to process pathological image data. Specifically, it receives whole-slice scan images (WSI), converts them to a standard format (such as SVS), and removes patient-sensitive label information attached to the images. Simultaneously, based on the pathological diagnosis results, it uses the OncoTree tumor classification system to standardize and encode the samples to facilitate subsequent multi-dimensional correlation retrieval in the database.

[0149] Example 6 To verify the effectiveness and accuracy of the tumor genomic variation detection method described in this application, the inventors of this application conducted a systematic verification based on large-scale real-world clinical sequencing data.

[0150] Experimental methods: 1. Dataset Construction: Tumor tissue samples fixed in formalin and embedded in paraffin (FFPE) were collected, totaling 6,935 samples from 6,858 patients, covering 54 cancer types. Hybridization capture and sequencing were performed using a targeted sequencing panel covering 520 cancer-related genes.

[0151] 2. Performance verification of the core mutation detection algorithm (LFvars): To verify the performance of the algorithm described in this application, which combines statistical model-based initial screening with chain orientation preference scoring, the following comparative experiment was designed in this embodiment: (1) Proportional settings: Comparative Example 1 (Traditional Hard Filtering): The same BWA alignment results were used, but only the conventional filtering criteria (VAF>1%, depth>50) were used. Poisson distribution scoring was not performed, and only variants with a single chain read count of 0 were removed (simple chain preference filtering).

[0152] Comparative Example 2 (Mainstream Open Source Software): Analysis was performed using the industry-standard Mutect2 workflow.

[0153] (2) Method of this application: Perform the complete process described in Example 1 (Formula 5, threshold 20).

[0154] (3) Test dataset: 50 high-confidence mutation sites (covering 0.5%-5% frequency) verified by digital PCR (dPCR) and 50 known FFPE false positive sites were selected.

[0155] (4) Experimental results: Specificity (denoising ability): Among 50 false positive sites, Comparative Example 1 falsely reported 38 (false positive rate 76%), Comparative Example 2 falsely reported 12 (false positive rate 24%), while the method in this application only falsely reported 2 (false positive rate 4%). Analysis revealed that the false positive sites successfully removed by this method generally had high S... bias Score.

[0156] Sensitivity (low-frequency detection): For ultra-low frequency mutations in VAF in the range of 0.5%-2%, Comparative Example 2 missed 40% due to overly strict filtering, while the detection rate of the method in this application reached 92%.

[0157] In summary, the dual scoring mechanism proposed in this application effectively suppresses FFPE chain preference noise while significantly preserving the true low-frequency mutation signal, which is superior to existing mainstream methods.

[0158] 3. Validation of the mutation identification model without control samples: The XGBoost machine learning model described in Embodiment 3 of this application was trained and evaluated using paired samples (containing tumor tissue and normal controls) from the aforementioned cohort. The evaluation results showed that the model achieved an area under the ROC curve (AUC) of 0.99 on the task of distinguishing between somatic mutations and germline mutations. This indicates that the method described in this application can still maintain extremely high detection accuracy even in clinical scenarios lacking normal controls.

[0159] 4. Cross-platform consistency verification: To verify the accuracy of the bioinformatics workflow described in this application (especially the detection of fusion genes and copy number variations), the NGS test results were compared with the "gold standard" method in clinical pathology: ALK rearrangement verification: ALK fusion-positive samples were verified using immunohistochemistry (IHC, VENTANA-D5F3 antibody). In non-small cell lung cancer (NSCLC) patients, after retesting and confirmation of inconsistent cases, the positive concordance rate between NGS and IHC reached 85.7% (6 / 7). This demonstrates the reliability of the fusion gene determination rule in Example 4 of this application.

[0160] ERBB2 (HER2) amplification verification: The detected ERBB2 amplification samples were verified by IHC and fluorescence in situ hybridization (FISH). The results showed that the focal amplification (≤10 Mb) detected by NGS was highly consistent with the IHC 3+ / FISH+ results, with a concordance rate of 95.4%. This demonstrates the accuracy of the copy number variation detection method in Example 4 of this application.

[0161] 5. Clinical application value: Analysis of this cohort using the methods described in this application revealed that the EGFR mutation rate in Chinese lung adenocarcinoma patients (48.5%) was significantly higher than that in the Caucasian population (23.7%, p<10). -72 The KRAS mutation rate (11.1%) was significantly lower. This further demonstrates that the method described in this application is suitable for characterization of large-scale populations and has stable data output capabilities.

[0162] refer to Figure 6 The present invention provides a tumor genome variation detection device, comprising: The acquisition module 10 is used to acquire sequencing data of tumor samples and align the sequencing data to a reference genome to obtain the aligned sequence. Detection module 20, used for somatic mutation detection based on the alignment sequence, includes: The initial screening unit 21 is used to traverse the alignment sequence to identify candidate mutation sites and calculate the mutation probability score of the candidate mutation sites based on a statistical model for initial screening. The calculation unit 22 is used to count the number of mutation reads and reference reads on the forward and reverse strands respectively for the candidate mutation sites after initial screening, and calculate the strand orientation preference score based on the statistical results. The filtering unit 23 is used to combine the mutation probability score with the chain orientation preference score to filter out false positive mutations with significant chain orientation preference and determine the final somatic mutation.

[0163] It is understood that the device in this embodiment corresponds to the tumor genome variation detection method in the above embodiments, and the options in the above embodiments are also applicable to this embodiment, so they will not be described again here.

[0164] This application also provides a computer device, the computer device including a processor and a memory, the memory storing a computer program, and the processor being used to execute the computer program to implement the tumor genomic variation detection method described in any of the foregoing embodiments.

[0165] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0166] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving execution instructions.

[0167] This application also provides a computer storage medium storing a computer program, which, when executed on a processor, implements the tumor genome variation detection method according to any one of the foregoing embodiments.

[0168] The computer storage medium can be a readable storage medium, a non-volatile storage medium, or a volatile storage medium. For example, the computer storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0169] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0170] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0171] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0172] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for detecting tumor genomic variations, characterized in that, include: Sequencing data of tumor samples are obtained and aligned to a reference genome to obtain aligned sequences; Somatic mutation detection based on the alignment sequence includes: The alignment sequences are traversed to identify candidate mutation sites, and the mutation probability scores of the candidate mutation sites are calculated based on a statistical model for initial screening. For candidate mutation sites after initial screening, the number of mutation reads and reference reads on the forward and reverse strands are counted respectively, and the strand orientation preference score is calculated based on the statistical results. By combining the mutation probability score with the chain orientation preference score, false positive mutations with significant chain orientation preference are filtered out to determine the final somatic mutation.

2. The tumor genomic variation detection method as described in claim 1, characterized in that, The calculation of the mutation probability score of the candidate mutation sites based on the statistical model includes: Obtain the total sequencing depth and number of mutated base reads for this candidate mutation site; The probability value of the number of reads of the mutated base at the candidate mutation site is calculated based on the Poisson distribution model, and the mutation probability score is calculated as follows: ； ； Where p represents the probability value; Poisson represents the Poisson distribution function; k represents the number of mutated base reads; λ represents the expected value, calculated as λ = D × ε; D represents the total sequencing depth; ε represents the preset single-base sequencing error rate; S mut represents the mutation probability score; lg represents the logarithm to the base 10.

3. The tumor genomic variation detection method as described in claim 1, characterized in that, The calculation of chain orientation preference score based on statistical results includes: Construct a 2×2 contingency table; the four elements of the contingency table are: the number of mutation reads on the forward chain, the number of reference reads on the forward chain, the number of mutation reads on the reverse chain, and the number of reference reads on the reverse chain; Fisher's exact test was performed on the contingency table to obtain two-tailed test values; The chain orientation preference score is calculated using the following formula: ； Among them, S bias p represents the chain orientation preference score; Fisher represents the two-tailed test value of the Fisher exact probability test; lg represents the logarithm to the base 10.

4. The tumor genomic variation detection method according to claim 3, characterized in that, The process of filtering out false positive mutations with significant strand orientation bias to determine the final somatic mutation includes: Sequencing data of normal samples matched with the tumor sample are obtained, and mutation probability scores of the normal samples at candidate mutation sites are calculated. The final score will be used to determine the outcome; the formula for calculating the final score is as follows: ； Among them, S final This represents the final score used to determine whether a mutation is a somatic mutation; S mut S represents the mutation probability score of the candidate mutation site in the tumor sample; norm S represents the mutation probability score of the candidate mutation site in the normal sample; bias This represents the chain orientation preference score; If and only if the S final If the score exceeds the preset threshold, the candidate mutation site is retained.

5. The tumor genomic variation detection method as described in claim 1, characterized in that, When a matching normal sample is unavailable, the somatic mutation detection also includes: Extract features from the candidate mutation sites, wherein the features include at least mutation frequency and population frequency; The extracted features are input into a pre-trained XGBoost machine learning model for classification to identify somatic mutations.

6. The tumor genomic variation detection method as described in claim 1, characterized in that, The method also includes genome rearrangement detection and annotation steps: Breakpoints in genomic rearrangement events are identified based on the aligned sequences; Analyze the transcription direction and gene structure preservation of the genes on both sides of the breakpoint; When the transcription directions of the genes on both sides of the breakpoint are consistent, and the gene upstream of the breakpoint retains at least one complete coding region and the gene downstream of the breakpoint retains a complete start codon, the genome rearrangement event is determined to be a fusion gene.

7. The tumor genomic variation detection method as described in claim 6, characterized in that, The genome rearrangement detection and annotation steps also include: When the transcriptional directions of genes on both sides of the breakpoint conflict, it is marked as a chain conflict; When the breakpoint is located inside a single gene, it is marked as a break within a single gene; Extract sequences of preset lengths upstream and downstream of the breakpoint, and adjust the case representation of the sequences according to the positive and negative strand directions of the gene.

8. The tumor genomic variation detection method as described in claim 1, characterized in that, The method further includes a copy number variation detection step: Calculate the tumor purity, ploidy, and allele-specific copy number of the sample; Identifying copy number variation regions based on the depth ratio of aligned sequences; When a matching normal sample is unavailable, copy number variation analysis is performed using a pre-defined baseline library.

9. The tumor genomic variation detection method as described in claim 1, characterized in that, The method also includes a step for calculating the microsatellite instability score: Obtain sequencing data of normal samples that match the tumor sample; The microsatellite instability score is calculated using the following formula: ； Where MSI% represents the microsatellite instability score; N total N represents the total number of valid microsatellite loci located within the detection area and whose sequencing depth meets the preset standard; modified This represents the number of unstable microsatellite loci after correction; and the corrected unstable microsatellite loci must simultaneously meet the following two conditions: A. Chi-square test shows that the distribution difference between tumor samples and normal samples is significant (FDR ≤ 0.05), B. The entropy value of tumor samples is greater than that of normal samples.

10. The tumor genomic variation detection method as described in claim 1, characterized in that, The method further includes a step of calculating tumor mutation burden based on identified somatic mutations, using the following formula: ； Wherein, TMB represents tumor mutation burden; L CDS N represents the total length of the protein-coding region within the target area; somatic This represents the total number of effective somatic nonsynonymous mutations. The effective somatic nonsynonymous mutations are those obtained by filtering out false positive mutations with significant strand orientation bias in the steps, and must simultaneously meet the following exclusion conditions: A. Exclude mutations with a mutation frequency less than a preset threshold; B. Exclude known tumor primary hotspot mutations; C. Exclude mutations located on HLA genes.

11. The tumor genomic variation detection method as described in claim 1, characterized in that, The method also includes a clinical annotation step: The identified somatic mutations or fusion genes are matched with a pre-defined clinical knowledge base; The clinical knowledge base contains the correspondence between gene variants and drug treatment regimens, and outputs the clinical evidence level of the variant based on the matching results.

12. A tumor genome variation detection device, characterized in that, include: The acquisition module is used to acquire sequencing data of tumor samples and align the sequencing data to a reference genome to obtain the aligned sequence. A detection module, used for somatic mutation detection based on the alignment sequence, includes: The initial screening unit is used to traverse the alignment sequence to identify candidate mutation sites and calculate the mutation probability score of the candidate mutation sites based on a statistical model for initial screening. The calculation unit is used to count the number of mutation reads and reference reads on the forward and reverse strands for candidate mutation sites after initial screening, and calculate the strand orientation preference score based on the statistical results. A filtering unit is used to combine the mutation probability score with the strand orientation preference score to filter out false positive mutations with significant strand orientation preference and determine the final somatic mutation.

13. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the tumor genomic variation detection method according to any one of claims 1-11.

14. A computer storage medium, characterized in that, It stores a computer program, which, when executed on a processor, implements the tumor genome variation detection method according to any one of claims 1-11.

Images