A copy number variation detection method based on semi-supervised learning

Abstract

The application relates to the technical field of gene variation detection, in particular to a copy number variation detection method based on semi-supervised learning. The method comprises the following steps: obtaining read depth signals and mapping quality signals of each normal window of a reference genome sequence from alignment information of sequencing reads, correcting the read depth signals of all normal windows in terms of GC content bias, adopting a cyclic binary segmentation algorithm to divide all normal windows into segmented regions with uniform read depth signals, identifying copy number variation breakpoint positions in combination with a split read strategy, performing normalization processing on the mapping quality signals, performing smoothing and noise reduction processing on the read depth signals, labeling pseudo labels for corresponding segmented regions, performing clustering analysis on all segmented regions through an improved density clustering algorithm, integrating and determining the variation types of abnormal segmented regions, and outputting copy number variation detection results, so that efficient detection of copy number variation is realized, and the accuracy and reliability of the detection results are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of gene variation detection technology, specifically a copy number variation detection method based on semi-supervised learning. Background Technology

[0002] Genomic structural variations refer to significant changes in the length or position of DNA sequences, including insertions, deletions, inversions, translocations, and copy number variations (CNVs). Among these, copy number variations are the most common and have the highest research value, referring to duplications or deletions of genomic segments ranging in length from thousands to millions of base pairs. Studies have shown that copy number variations play an important role in gene expression regulation, phenotypic differences, and disease development, and are one of the main sources of human genomic diversity, closely related to various tumors and neurological diseases.

[0003] In recent years, with the rapid development of high-throughput sequencing technology and computational methods, copy number variant detection has evolved from methods based on simple statistical features to a detection framework that combines various advanced algorithms such as machine learning, deep learning, and graph modeling. Based on whether it relies on manually labeled data and how it is used, existing copy number variant detection methods are mainly divided into unsupervised detection methods and fully supervised detection methods.

[0004] Unsupervised detection methods primarily rely on the statistical characteristics of the sequencing data itself, including sequencing depth distribution, GC content bias, and pairwise end distance. Representative tools in unsupervised detection include CNVkit, GROM-RD, Matchclips2, and OTSUCV. CNVkit integrates targeted and extra-target sequencing data to detect copy number variations across the entire genome. CNVkit exhibits relatively consistent detection performance at different sequencing depths, but its relatively conservative variation identification strategy results in a relatively small number of detected copy number variations, affecting its overall performance metrics. GROM-RD detects copy number variations by analyzing sequencing depth bias in genomic data. It improves breakpoint detection resolution by using overlapping windows of varying sizes for copy number variation searching. However, GROM-RD's detection performance decreases at lower tumor purity. Matchclips2 utilizes read position information across structural variation regions and alignment features to identify breakpoints, achieving high breakpoint resolution. However, its detection performance is significantly limited in low-sequencing-depth data samples due to insufficient read position information. OTSUCNV employs an adaptive sequence segmentation method that does not rely on prior assumptions about data distribution and constructs a variant recognition framework based on image thresholding principles. However, due to the lack of fine-grained localization of variant boundaries, OTSUCNV still has room for improvement in breakpoint accuracy.

[0005] Fully supervised detection methods utilize a large number of known true positive copy number variants (CNVs) as training sets, enabling the model to learn the mapping relationship between sequencing signals and CNVs. By combining multimodal deep learning, graph modeling, and neural networks, they achieve automatic identification of CNV regions. Fully supervised detection methods typically outperform in terms of variant detection accuracy and boundary localization accuracy, but require higher computational resources and more complex model training processes, and are highly dependent on large-scale, high-quality labeled data. However, in real-world detection scenarios, obtaining sufficient true positive labels is often difficult, which limits the model's generalization ability. Summary of the Invention

[0006] This invention provides a copy number variation detection method based on semi-supervised learning, which effectively solves the problems of insufficient detection accuracy, limited boundary recognition accuracy, and excessive dependence on labeled data in complex sequencing environments. In the sequencing data sample to be detected, by utilizing a small amount of labeled feature information, the potential feature information in the unlabeled data is fully explored to achieve efficient detection of copy number variations and significantly improve the accuracy and reliability of the detection results.

[0007] The present invention provides a copy number mutation detection method based on semi-supervised learning, characterized by comprising the following steps:

[0008] Step 1: Input the reference genome sequence, the sequencing data sample to be tested, and the seed file. Align and sort the sequencing data sample with the reference genome sequence to obtain the sorted BAM file.

[0009] Step 2: Under the reference genome coordinate system, the reference genome sequence is divided into fixed-size, non-overlapping windows. Sequencing reads are obtained from the sorted BAM file. Based on the alignment information of the sequencing reads, the read count of all windows in the reference genome sequence is counted. Abnormal windows containing N bases in the reference genome sequence are identified and removed, while normal windows that do not contain N bases are retained. For missing values ​​in the alignment information of sequencing reads, zero-padding is used to replace the missing values.

[0010] Step 3: Obtain the read depth signal and mapping quality signal of each normal window of the reference genome sequence from the alignment information of the sequencing reads, and save the characteristics of all normal windows of the reference genome sequence; wherein, the read depth signal is the RD signal, which is used to represent the sequencing depth of bases in the sequencing data sample; the mapping quality signal is the MQ signal, which is used to represent the quality of the alignment between the sequencing data sample and the reference genome sequence.

[0011] Step 4: Correct the GC content deviation for the read segment depth signal of all normal windows;

[0012] Step 5: Based on the corrected read segment depth signal, a cyclic binary segmentation algorithm is used to divide all normal windows into segmented regions with uniform read segment depth signals. Combined with the segment splitting strategy, potential copy number variation breakpoint locations are identified.

[0013] Step 6: Normalize the mapping quality signals of all segmented regions;

[0014] Step 7: Perform smoothing and noise reduction processing on the reading depth signals of all segmented regions;

[0015] Step 8: Based on the start and end positions of copy number mutations in the seed file, label the corresponding segment regions with pseudo-labels;

[0016] Step 9: The segmented regions labeled with pseudo-labels are used as the clustering core of the semi-supervised learning clustering process. The reading depth signals of all segmented regions are used as clustering features. An improved density clustering algorithm is used to perform clustering analysis on all segmented regions. The obtained clustering analysis results are marked as normal segmented regions or abnormal segmented regions by clustering labels.

[0017] Step 10: Integrate the abnormal segmented regions and determine the mutation type, then output the copy number mutation detection results.

[0018] Furthermore, in step one, the sequencing data sample is aligned with the reference genome sequence using BWA software to generate an initial BAM file. Based on the position of the reference genome sequence, the initial BAM file is sorted using SAMtools software to obtain a sorted BAM file containing specific sequencing reads.

[0019] Furthermore, in step two, the positions containing N bases in the reference genome sequence are identified and processed. The reference genome sequence is divided into fixed-size, non-overlapping windows. Windows containing N bases are detected and defined as abnormal windows. The read count of abnormal windows is marked with a specific negative value for distinction. The read count refers to the number of times the sequencing read in the BAM file appears within the reference genome sequence window, which is used to reflect the sequencing coverage of the window. If the read count of a window is negative, the window is determined to be an abnormal window and deleted; otherwise, it is determined to be a normal window, and the start and end positions of the normal window are recorded. For missing values ​​in the alignment information of sequencing reads, zero-padding is used to replace the missing values.

[0020] Furthermore, in step three, the alignment information of the sequencing reads is obtained. Based on the start and end positions of each normal window in the reference genome sequence, the read depth signal of each base within the normal window is calculated, and the average read depth signal of all bases within the normal window is used as the read depth signal of the normal window.

[0021]

[0022] in, Indicates the first The read segment depth signal of a normal window. Indicates the length of a normal window. Indicates the first The first normal window The read depth signal of each base;

[0023] The alignment information of the sequencing reads is obtained. Based on the start and end positions of each normal window in the reference genome sequence, the mapping quality signal of each base within the normal window is calculated. The average mapping quality signal of all bases within the normal window is used as the mapping quality signal of the normal window.

[0024]

[0025] in, Indicates the first The mapping quality signal of a normal window, Indicates the length of a normal window. Indicates the first The first normal window The mapping quality signal of each base.

[0026] Furthermore, in step four, GC content deviation correction is performed on the read depth signals of all normal windows, i.e.

[0027]

[0028] in, Indicates the first correction The read segment depth signal of a normal window. Indicates the first time before correction The read segment depth signal of a normal window. This represents the average read depth signal of all normal windows before correction. Indicates the difference between before and after correction. The average read depth signal of a normal window with highly consistent GC content.

[0029] Furthermore, in step five, a cyclic binary segmentation algorithm is used to segment all normal windows, dividing them into segmented regions with uniform read depth signals. During the segmentation process, for discontinuous windows containing gaps, the discontinuous windows are cut and merged according to the fixed length of the segmented regions to restore the continuity of the reference genome sequence coordinates. For wide windows whose window length exceeds a preset length threshold, a fixed-size sliding window is used, starting from the beginning of the wide window and sliding sequentially according to a set step size to further subdivide the wide window into segmented regions of uniform length.

[0030] A CIGAR string-based segmentation strategy was employed to identify breakpoints in copy number variations. Breakpoints were inferred by comparing the relative positions of the truncated portion of the CIGAR string with the matching region of the reference genome sequence. If the truncated portion preceded the matching region, the breakpoint was located at the end of the truncated portion; if the truncated portion followed the matching region, the breakpoint was located at the beginning of the truncated portion.

[0031]

[0032] in, This indicates a contiguous region where the sequencing read matches the reference genome sequence. This indicates a soft truncation in the CIGAR string. This indicates a hard truncation in the CIGAR string. This indicates that there are sequencing reads. Each base matches the reference genome sequence. This indicates that the length of the sequencing read is... A single base did not align to the reference genome sequence but remains in the sequencing read. This indicates that the length of the sequencing read is... One base did not align to the reference genome sequence and was discarded in the sequencing read; and This indicates that the truncated portion is located after the matched region. and This indicates that the truncated portion is located before the matched region.

[0033] Furthermore, in step six, the mapping quality signals of all segmented regions are normalized to eliminate the systematic bias introduced by differences in sequencing depth, so that the mapping quality signals more accurately reflect the mapping quality.

[0034]

[0035] in, Indicates the number of digits after normalization. Mapping quality signal of each segmented region Indicates the first digit before normalization. The average value of the mapped quality signal of each segmented region, Indicates the first digit before normalization. The average value of the read depth signal of each segmented region.

[0036] Furthermore, in step seven, a total variational model is used to smooth and denoise the read depth signals of all segmented regions. A fidelity term is constructed by calculating the difference between the original read depth signal and the smoothed and denoised read depth signal to constrain the smoothing and denoising process, ensuring that the main characteristics of the original read depth signal are preserved during the process. A total variational regularization term is used by calculating the sum of the absolute values ​​of the differences between adjacent segmented regions in the read signal, quantifying the changes in read depth signals between adjacent segmented regions. Different optimization weights are assigned to different segmented regions based on the magnitude of the differences in read depth signals between adjacent segmented regions. That is,

[0037]

[0038] in This represents the read depth signal of the segmented region after smoothing and noise reduction. This represents the read depth signal of the segmented region before smoothing and noise reduction, i.e., the original read depth signal. This represents the regularization parameter, used to balance the fidelity and smoothness of the read segment depth signal. Indicates the number of segmented regions. It is the fidelity term of the total variation model, preventing the read depth signal after smoothing and denoising from deviating excessively from the original read depth signal. Indicates the first The segmented region and the first The degree of deviation of the read depth signal in each segmented region. This represents the total variation regularization term, which penalizes sharp fluctuations in the read depth signal, smooths the read depth signal, and preserves true inflection points or abrupt changes.

[0039] Furthermore, in step eight, based on the start and end positions of copy number mutations in the seed file, pseudo-labels are assigned to the corresponding segment regions. The seed file is constructed from the intersection of detection results from multiple existing copy number mutation detection tools. A segment region with a pseudo-label is selected if it meets the following two criteria: the pseudo-labeled segment region belongs to the intersection of detection results from existing copy number mutation detection tools, and the read depth signal corresponding to the pseudo-labeled segment region exceeds [a certain threshold]. The principle is that if the read depth signal of a segmented region labeled with a false label exceeds three standard deviations of the average read depth signal of all segmented regions, it is considered a high-confidence false-labeled region.

[0040]

[0041] in Indicates the first The read depth signal of each segmented region This represents the average value of the read depth signal across all segmented regions. This represents the standard deviation of the reading depth signal for all segmented regions.

[0042] Furthermore, in step nine, high-confidence pseudo-label regions are used as clustering cores, and the read depth signals of all segmented regions are used as clustering features. Clustering analysis is performed on all segmented regions based on the clustering cores and an improved density clustering algorithm. The improved density clustering algorithm, based on the traditional density clustering algorithm, introduces z-score as a third constraint, in addition to the original neighborhood radius and minimum neighbor number parameters. This z-score quantifies the deviation of the read depth signal of a certain segment from the average read depth signal of all segmented regions. Under the condition of satisfying the neighborhood radius and minimum neighbor number, segmented regions with z-scores higher than the deviation threshold are marked as abnormal segmented regions, while those with z-scores lower than the threshold are marked as normal segmented regions.

[0043]

[0044] For the first z-score of each segmented region Indicates the first The read depth signal of each segmented region This represents the average value of the read depth signal across all segmented regions. This represents the standard deviation of the reading depth signal for all segmented regions.

[0045] Furthermore, in step ten, the abnormal segment regions selected from the cluster analysis results are sorted according to their starting positions. If the interval between two adjacent abnormal segment regions is very small, the two abnormal segment regions are merged, and the average values ​​of the read depth signal, GC content, and mapping quality signal are taken as the new features of the merged segment region. If the interval is large, the abnormal segment region is recorded as an independent variant region. For the merged segment region, if the read depth signal of the abnormal segment region is higher than the average read depth signal of all merged segment regions, the variant type of the abnormal segment region is defined as duplicate; otherwise, the variant type is defined as missing. The starting position, ending position, read depth signal, and variant type of the merged abnormal segment region are integrated and saved as the detection result to a TXT file for output.

[0046] This invention provides a copy number variation detection method based on semi-supervised learning, applicable to the identification and localization of copy number variations in the human genome. For sequencing data samples generated by next-generation sequencing technology, it can utilize a small amount of high-confidence labeled data to detect a large amount of unlabeled data, effectively identifying potential copy number variation regions, and has the following technical advantages:

[0047] (1) This invention innovatively uses the results of existing copy number variation detection tools for semi-supervised learning methods, breaking through the previous limitation of only using them for benchmark verification, and significantly improving the utilization efficiency of detection results.

[0048] (2) Based on the traditional density clustering algorithm framework, this invention introduces z-score as the third constraint condition of the improved density clustering algorithm, which is used to quantify the deviation of the reading depth signal of a certain segment region from the average reading depth signal of all segment regions, thereby improving the identification accuracy of abnormal segment regions in the clustering process.

[0049] (3) The present invention embeds the high-confidence pseudo-label region as the clustering core into the clustering process, effectively guiding the determination of the clustering core and the clustering expansion process, and realizing more targeted detection of copy number variation regions. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the implementation of the present invention;

[0051] Figure 2 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 10X and the tumor purity is 0.4.

[0052] Figure 3 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 10X and the tumor purity is 0.6.

[0053] Figure 4 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 10X and the tumor purity is 0.8.

[0054] Figure 5The contour plot shows the comparison results of the present invention with existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 20X and the tumor purity is 0.4.

[0055] Figure 6 The contour plot shows the comparison results of the present invention with existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 20X and the tumor purity is 0.6.

[0056] Figure 7 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 20X and the tumor purity is 0.8.

[0057] Figure 8 Contour plot showing the comparison of detection results between the present invention and existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under simulated sequencing data sample sequencing depth of 30X and tumor purity of 0.4.

[0058] Figure 9 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 30X and the tumor purity is 0.6.

[0059] Figure 10 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 30X and the tumor purity is 0.8.

[0060] Figure 11 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 50X and the tumor purity is 0.4.

[0061] Figure 12 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 50X and the tumor purity is 0.6.

[0062] Figure 13 The contour plot shows the comparison of the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV under the condition that the sequencing depth of the simulated sequencing data sample is 50X and the tumor purity is 0.8.

[0063] Figure 14 The bar chart shows the F1 scores of the present invention compared with existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV in four real sequencing data samples.

[0064] Figure 15 The above is a contour plot showing the comparison results of the present invention with existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV in real sequencing data sample NA19238.

[0065] Figure 16 The above is a contour plot comparing the detection results of this invention with existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV in real sequencing data sample NA19239.

[0066] Figure 17 This is a contour plot comparing the detection results of the present invention with existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV in real sequencing data sample NA19240;

[0067] Figure 18 This is a contour plot comparing the detection results of the present invention with existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV in real sequencing data sample HG002;

[0068] Figure 19 UpSet plot showing the number of true positive CNVs detected by the present invention and existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV in real sequencing data sample NA19238;

[0069] Figure 20 UpSet plot showing the number of true positive CNVs detected by the present invention and existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV in real sequencing data sample NA19239;

[0070] Figure 21UpSet plot showing the number of true positive CNVs detected by the present invention and existing detection tools CNVkit, GROM-RD, Matchclips2 and OTSUCNV in real sequencing data sample NA19240;

[0071] Figure 22 This is an UpSet graph showing the number of true positive CNVs detected by the present invention and existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV in real sequencing data sample HG002. Detailed Implementation

[0072] like Figure 1 As shown, this invention provides a copy number mutation detection method based on semi-supervised learning, which is mainly achieved through the following steps.

[0073] Step 1: Use the BWA-MEM function in the BWA software to align the sequencing data sample with the reference genome sequence to generate an initial BAM file. Then, use the SAMtools tool to sort the initial BAM file according to the position of the reference genome sequence, obtaining a sorted BAM file containing specific sequencing reads.

[0074] Step two: To ensure the accuracy and reliability of the analysis results, it is necessary to identify and process missing values ​​of N bases and alignment information of sequencing reads in the reference genome sequence. The reference genome sequence mainly includes four base types: A, G, C, and T. There is also a special N base used to represent undetermined bases; the N base can be any base. Since the positions corresponding to N bases lack precise base information, directly involving them in the analysis would introduce bias and affect the reliability of the results. Therefore, it is necessary to identify and delete regions containing N bases in the reference genome sequence. The reference genome sequence is divided into fixed-size, non-overlapping windows. Windows containing N bases are marked as abnormal windows, and the read counts corresponding to abnormal windows are set to a specific negative number (e.g., -9999) for marking. If a read count of a window is detected to be less than 0, the window is determined to contain N bases, considered an abnormal window, and deleted from the analysis; conversely, if a read count of a window is detected to be non-negative, the window is determined to not contain N bases, considered a normal window, and the start and end positions of the normal window are retained. For missing values ​​in alignment information resulting from failed alignment of sequencing data samples with the reference genome sequence or the lack of recording of key alignment information, zero-filling is used to replace the missing values ​​at the locations where they appear, ensuring consistency in subsequent statistical analysis.

[0075]

[0076] in, Indicates the first The read segment depth signal of a normal window. Indicates the length of a normal window. Indicates the first The first normal window The read depth signal of each base;

[0077] The alignment information of the sequencing reads is obtained. Based on the start and end positions of each normal window in the reference genome sequence, the mapping quality signal of each base within the normal window is calculated. The average mapping quality signal of all bases within the normal window is taken as the mapping quality signal of the normal window.

[0078]

[0079] in, Indicates the first The mapping quality signal of a normal window, Indicates the length of a normal window. Indicates the first The first normal window The mapping quality signal of each base.

[0080] Step four, GC content bias, refers to the uneven distribution of GC content in different regions of the sequencing data sample. In high-throughput sequencing, GC content can introduce systematic biases into the sequencing process: regions with high GC content are prone to enrichment during library construction and amplification, leading to increased sequencing depth; while regions with low GC content may experience insufficient sequencing coverage, resulting in decreased sequencing depth. To reduce the impact of GC content bias on the analysis results, it is necessary to correct the extracted read depth signal for GC content, i.e.,

[0081]

[0082] in, Indicates the first correction The read segment depth signal of a normal window. Indicates the first time before correction The read segment depth signal of a normal window. This represents the average read depth signal of all normal windows before correction. Indicates the difference between before and after correction. The average read depth signal of a normal window with highly consistent GC content.

[0083] Step 5: Segment all normal windows using a cyclic binary segmentation algorithm. The cyclic binary segmentation algorithm continuously segments the normal windows, eventually dividing them into segmented regions with consistent read depth signals. Each segmented region typically corresponds to a specific copy number state. For segmented regions containing gaps, the small windows are merged by cutting at the gaps according to a fixed length, thus restoring the continuity of the reference genome sequence coordinates. Secondly, for wide windows exceeding a preset length threshold (e.g., 5000 bp), a fixed-size sliding window is used, starting from the beginning of the wide window and sliding sequentially at a set step size, further subdividing the wide window into segmented regions of consistent length.

[0084] To further improve the accuracy of copy number variation breakpoint identification, a CIGAR string-based read splitting strategy is adopted. For alignment information of sequencing reads, alignments containing soft or hard truncation operations are prioritized. Truncation indicates that a portion of the sequencing read failed to match the reference genome sequence during alignment, and the truncated position may be a copy number variation breakpoint. By comparing the relative positions of the truncated portion in the CIGAR string with the matching region of the reference genome sequence, the possible breakpoint position is inferred. If the truncated portion is before the matching region, the breakpoint is located at the end of the truncated portion; if the truncated portion is after the matching region, the breakpoint is located at the beginning of the truncated portion.

[0085]

[0086] in, This indicates a contiguous region where the sequencing read matches the reference genome sequence. This indicates a soft truncation in the CIGAR string. This indicates a hard truncation in the CIGAR string. This indicates that there are sequencing reads. Each base matches the reference genome sequence. This indicates that the length of the sequencing read is... A single base did not align to the reference genome sequence but remains in the sequencing read. This indicates that the length of the sequencing read is... One base did not align to the reference genome sequence and was discarded in the sequencing read; and This indicates that the truncated portion is located after the matched region. and This indicates that the truncated portion is located before the matched region.

[0087] In step six, to ensure that the mapping quality signal for each segment region more objectively reflects the actual mapping quality and to eliminate systematic bias caused by differences in sequencing depth, the mapping quality signal needs to be normalized.

[0088]

[0089] in, Indicates the number of digits after normalization. Mapping quality signal of each segmented region Indicates the first digit before normalization. The average value of the mapped quality signal of each segmented region, Indicates the first digit before normalization. The average value of the read depth signal of each segmented region.

[0090] Step 7: A total variational model is used to smooth and denoise the read signals of all segmented regions. By minimizing the difference between the original read signal and the smoothed and denoised read signal, the variation in the difference between adjacent segmented regions is controlled, thereby highlighting copy number variation regions and suppressing local noise. During the smoothing and denoising process, a fidelity term is constructed by calculating the difference between the original read depth signal and the smoothed and denoised read depth signal. This term constrains the smoothing and denoising process, ensuring that the main characteristics of the original read depth signal are maintained, thus preventing the smoothing and denoising result from deviating excessively from the original read depth signal. The sum of the absolute values ​​of the differences between adjacent segmented regions in the read signal is calculated as a total variational regularization term, quantifying the variation in read depth signal between adjacent segmented regions. Larger differences usually correspond to potential copy number variation, while more frequent, minor fluctuations are often noise. Based on the difference in read depth signals between adjacent segments and the noise level, different optimization weights are assigned to different segments. Segments with detected true breakpoints or significant abrupt changes in read depth signals are assigned higher weights to enhance the preservation of key changes; while segments with more noise fluctuations are assigned lower weights to smooth out meaningless noise fluctuations. The formula for smoothing and noise reduction is as follows:

[0091]

[0092] in This represents the read depth signal of the segmented region after smoothing and noise reduction. This represents the read depth signal of the segmented region before smoothing and noise reduction, i.e., the original read depth signal. This represents the regularization parameter, which is used during the smoothing process. It is set to 0.25 to balance the fidelity and smoothness of the read segment depth signal. Indicates the number of segmented regions. It is the fidelity term of the total variation model, preventing the read depth signal after smoothing and denoising from deviating excessively from the original read depth signal. Indicates the first The segmented region and the first The degree of deviation of the read depth signal in each segmented region. This represents the total variation regularization term, which penalizes sharp fluctuations in the read depth signal, smooths the read depth signal, and preserves true inflection points or abrupt changes.

[0093] Step 8: Improve the accuracy of copy number variant (CNV) identification by leveraging the results of existing detection tools. Introduce a seed file to perform preliminary pseudo-labeling on segmented regions. The seed file is constructed from the intersection of the outputs of several representative existing CNV detection tools (CNVkit, GROM-RD, Matchclips2, and OTSUCNV). Segmented regions labeled with pseudo-labels are considered high-confidence pseudo-labeled regions if they meet the following two screening criteria: the pseudo-labeled segmented region belongs to the intersection of the detection results of existing CNV detection tools, ensuring strong cross-method consistency; the read depth signal corresponding to the pseudo-labeled segmented region must exceed [a certain threshold]. The principle is that the read depth signal of the segmented region labeled with the pseudo-label must exceed three standard deviations of the average read depth signal of all segmented regions. That is,

[0094]

[0095] in Indicates the first The read depth signal of each segmented region This represents the average value of the read depth signal across all segmented regions. This represents the standard deviation of the reading depth signal for all segmented regions.

[0096] Step nine involves using high-confidence pseudo-label regions as clustering cores and the read depth signals of all segmented regions as clustering features. Clustering analysis is then performed on all segmented regions based on the clustering cores and an improved density clustering algorithm. After obtaining the high-confidence pseudo-label regions, an improved density clustering algorithm incorporating z-score constraints is used to perform clustering analysis on all segmented regions to identify potential copy number variation regions. Traditional density clustering algorithms rely on two key parameters: neighborhood radius and minimum neighbor count. By introducing z-score as a third constraint into the density clustering algorithm, the z-score quantifies the deviation of the read depth signal of a segmented region from the average read depth signal of the global region. Under the conditions of satisfying the neighborhood radius and minimum neighbor count, a deviation threshold for the z-score is set (e.g., ...). Using clustering labels, segment regions where the read depth signal deviates significantly from the global read depth signal average are marked as abnormal segment regions, while segment regions where the read depth signal deviates less from the global read depth signal average are marked as normal segment regions.

[0097]

[0098] For the first z-score of each segmented region Indicates the first The read depth signal of each segmented region This represents the average value of the read depth signal across all segmented regions. This represents the standard deviation of the reading depth signal for all segmented regions.

[0099] Step 10: Sort the anomalous segment regions selected from the cluster analysis results according to their starting positions. Iterate through the sorted anomalous segment regions sequentially. For two adjacent anomalous segment regions, if the difference between the ending position of the previous anomalous segment region and the starting position of the next anomalous segment region is no more than 1, merge the two anomalous segment regions. Take the average of the read depth signal, GC content, and mapping quality signal of the merged anomalous segment region as the feature information of the new anomalous segment region, replacing the original features. If the interval between two adjacent anomalous segment regions is greater than 1, record the previous anomalous segment region as an independent variant region and start a new merging process. For each merged anomalous segment region, if the read depth signal of the anomalous segment region is higher than the overall average of the read depth signals of all merged segment regions, its variant type is determined to be duplicate; otherwise, its variant type is determined to be missing. Sort the starting position, ending position, read depth signal, and determined variant type of each merged anomalous segment region, and save the detection results to a TXT file for output.

[0100] To verify the effectiveness and fairness of the technical solution of this invention, the detection tool was evaluated from multiple perspectives using simulated sequencing data samples and real sequencing data samples, and a unified evaluation index was used for comparative analysis. For existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV, default parameters were used for detection to ensure the objectivity and consistency of the comparative experiments.

[0101] In the simulated sequencing data experiments, 50 sets of samples were generated for each experimental configuration to reduce the randomness of the experimental results. In these experiments, the copy number variant detection method (SSLCNV) based on semi-supervised learning of this invention was compared with existing detection tools CNVkit, GROM-RD, Matchclips2, and OTSUCNV in terms of precision, recall, and F1 score. In the experimental evaluation, precision was defined as the proportion of true positive CNVs to the total number of detected CNVs, and recall was defined as the proportion of true positive CNVs to the total number of real CNV samples. The F1 score is an indicator that comprehensively considers both precision and recall, and can be regarded as the harmonic mean of the two, used to comprehensively evaluate the detection performance of the tool.

[0102]

[0103] In this context, TP represents a true positive CNV, defined as a CNV whose detection result matches the true CNV type and whose length range overlaps with the true CNV length range by at least 80%; FP represents a false positive CNV, defined as a CNV whose detection result does not match the true CNV type, or whose length range overlaps with the true CNV length range by less than 80%; and FN represents a false negative CNV, defined as a true CNV that was not identified by the detection tool. TP+FP represents the total number of CNVs detected by the detection tool, and TP+FN represents the total number of true CNVs.

[0104] The experimental results of the simulated sequencing data are as follows Figures 2-13 As shown, SSLCNV achieved the highest F1 score in all configurations of the simulated sequencing data experiments. Figure 2 , Figure 5 , Figure 8 and Figure 11 As shown, when the tumor purity is 0.4, existing detection tools face the problem of low recall due to the weak CNV signal itself. OTSUCNV has a precision close to 1, but its recall is significantly insufficient, resulting in an F1 score of only 0.5-0.7. In contrast, the SSLCNV proposed in this invention can significantly improve recall while maintaining high precision, ultimately achieving an F1 score of 0.8, which is superior to the four existing detection tools overall. Figure 3 , Figure 6 , Figure 9 and Figure 12 As shown, at a tumor purity of 0.6, the precision and recall of the SSLCNV method of this invention are further improved, with the F1 score increasing to 0.89, significantly higher than the four existing detection tools. This result indicates that SSLCNV can maintain a better detection balance even as the CNV signal gradually increases. Figure 4 , Figure 7, Figure 10 and Figure 13 As shown, when the tumor purity is 0.8, the precision and recall of the SSLCNV of this invention both exceed 0.80, maintaining optimal performance. Although GROM-RD has a higher recall under this condition, its precision is lower than that of SSLCNV, resulting in an overall F1 score that is still lower than that of SSLCNV. CNVkit and OTSUCV still have relatively high precision, but their recall is low, so their overall detection performance is not as good as that of the SSLCNV of this invention.

[0105] To further verify the adaptability and robustness of the proposed SSLCNV at different sequencing depths, this invention systematically compared the performance of SSLCNV with four existing detection tools—CNVkit, GROM-RD, Matchclips2, and OTSUCNV—at four sequencing depths of 10x, 20x, 30x, and 50x, based on simulated sequencing data samples. For each sequencing depth, statistical analysis was performed on 150 sets of simulated sequencing data samples, and the average F1 score of these 150 sets was calculated to evaluate the improvement in F1 score of SSLCNV compared to the four existing detection tools. The improvement in F1 score of SSLCNV proposed in this invention compared to the four existing detection tools is shown in Table 1.

[0106] Table 1. Improvement in F1 score of SLLCNV compared to the four existing detection tools

[0107]

[0108] As shown in Table 1, SSLCNV achieved the highest F1 score at all sequencing depths. At a sequencing depth of 10x, SSLCNV achieved an F1 score of 0.739, an improvement of approximately 10% over GROM-RD, 14.09% over CNVkit, and over 100 times better than Matchclips2, fully demonstrating SSLCNV's robustness in environments with high noise and sparse CNV signals. At medium sequencing depths of 20x and 30x, SSLCNV maintained its leading position. At 20x, the F1 score was 0.807, an improvement of approximately 0.86% over Matchclips2, and improvements of 9.44%, 15.37%, and 23.96% over GROM-RD, OTSUCNV, and CNVkit, respectively. At a sequencing depth of 30x, SSLCNV achieved an F1 score of 0.856, an improvement of 6.29% over the second-best Matchclips2, and over 36.75% over CNVkit. At a high sequencing depth of 50x, the SSLCNV achieved an F1 score of 0.889, a 14.18% improvement over Matchclips2, and 17.75%, 18.73%, and 50.45% improvement over GROM-RD, OTSUCNV, and CNVkit, respectively. Experiments demonstrate that the SSLCNV of this invention achieves high precision and recall at low, medium, and high sequencing depths, consistently outperforming the F1 scores of the four existing detection tools. This fully validates the stability and reliability of this invention at different sequencing depths and complex tumor purity levels, exhibiting significant advantages, especially in scenarios with weak CNV signals and high detection difficulty.

[0109] Simulated sequencing data experiments are an effective method for evaluating algorithm performance, enabling systematic comparison of the performance of different detection tools under controlled conditions. However, simulated sequencing data samples still cannot fully reflect the complexity of genome structure and the diversity of variations in real sequencing data samples. To further verify the detection effect of the proposed SSLCNV method in real-world scenarios, this invention also applies SSLCNV along with four existing detection tools to real sequencing data samples. Specifically, this includes three real sequencing data samples NA19238, NA19239, and NA19240 from the 1000 Genomes Project, and real sequencing data sample HG002 from the NCBI database.

[0110] Real sequencing data experimental results as follows Figures 14-18 As shown in Table 2, the F1 scores of SLLCNV compared to the four existing detection tools in real sequencing data samples are as follows:

[0111] Table 2. F1 scores of SLLCNV compared to four existing detection tools in real sequencing data samples.

[0112]

[0113] Depend on Figure 14 As shown in Table 2, SSLCNV achieved the highest F1 score in all four real-world sequencing data samples, maintaining the best overall detection performance and demonstrating its stability and adaptability across different datasets and data sample backgrounds. Specifically, in the real-world sequencing data sample HG002, SSLCNV achieved an F1 score of 0.196, significantly higher than GROM-RD, demonstrating strong detection capabilities for complex genomic regions. In the two real-world sequencing data samples NA19238 and NA19239, SSLCNV achieved F1 scores of 0.133 and 0.149 respectively, both higher than the F1 scores of the four existing detection tools. In the NA19240 sample, despite the greater overall detection difficulty, SSLCNV maintained an F1 score of 0.114, while the F1 scores of the four existing detection tools were all below 0.05. Compared to GROM-RD, SSLCNV not only improved F1 scores by approximately 9% and 4% in NA19238 and NA19239 samples, respectively, but also showed more significant improvements in HG002 and NA19240 samples. Compared to CNVkit and OTSUCNV, SSLCNV demonstrated a more pronounced performance advantage in all real-world sequencing data samples, with F1 scores improved several times over. Figures 15-17 As shown, Matchclips2 exhibits extremely low precision and recall in real sequencing data samples with low sequencing depths (NA19238, NA19239, and NA19240), while SSLCNV still maintains high detection accuracy, demonstrating good robustness for detecting low CNV signals. In... Figure 18 As shown, CNVkit and OTSUCNV exhibit extremely high accuracy, but due to the small number of true positive CNVs detected, the recall rate is too low, and the F1 score is lower than that of SSLCNV.

[0114] Figures 19-22 UpSet plots were used to illustrate the number of overlaps among true positive CNVs identified by each detection tool in four real-world sequencing data samples. The results show that SSLCNV not only detected the most true positive CNVs in all real-world sequencing data samples, but also had the largest number of true positive CNVs overlapping with the four existing detection tools, demonstrating more comprehensive detection coverage and stronger consistency. Figures 19-21 As shown, in sample NA19238, the number of true positive CNVs between SSLCNV and CNVkit reached 300; in samples NA19239 and NA19240, the number of true positive CNVs between SSLCNV and CNVkit were 353 and 461, respectively; Figure 22As shown, in the HG002 sample, the number of true positive CNVs intersecting with GROM-RD reached 569. Notably, SSLCNV also detected true positive CNVs that the four existing detection tools failed to identify, further demonstrating its higher detection accuracy and ability to discover potential new CNVs. Figures 19-22 As shown, in sample NA19238, SSLCNV independently detected 312 true positive CNVs; in samples NA19239 and NA19240, 637 and 1699 true positive CNVs were detected, respectively; and in sample HG002, a remarkable 2667 true positive CNVs were detected. Therefore, the SSLCNV of this invention exhibits superior recall and precision in real sequencing data samples, consistently outperforming the four existing detection tools in F1 scores. This fully demonstrates that this invention not only performs significantly in simulated sequencing data samples but also demonstrates strong detection capabilities and reliability in real sequencing data samples, making it more suitable for practical scientific research and clinical testing needs.

[0115] The above results demonstrate that the copy number variant (SSLCNV) detection method based on semi-supervised learning proposed in this invention effectively addresses the problems of insufficient detection accuracy, limited boundary recognition precision, and excessive reliance on labeled data in complex sequencing environments. To ensure fairness and objectivity, the experiment was systematically tested on 600 simulated sequencing data samples and 4 real sequencing data samples. The 600 simulated sequencing data samples were derived from combinations of four different sequencing depths (10x, 20x, 30x, and 50x) and three tumor purities (0.4, 0.6, and 0.8). Specifically, by combining the four sequencing depths with the three tumor purities, 12 experimental conditions were created, generating 50 replicates under each condition, totaling 4 × 3 × 50 = 600 simulated sequencing data samples to comprehensively evaluate the performance of SSLCNV under different sequencing depths and tumor purities. In addition, four real sequencing data samples, corresponding to real sequencing data samples NA19238, NA19239, NA19240, and HG002, were selected to verify the applicability and stability of SSLCNV in real genomic variant backgrounds. Test results show that, whether in simulated or real sequencing data samples, the SSLCNV method proposed in this invention achieves the best balance in core indicators such as detection precision, recall, and F1 score. Especially under complex conditions such as low sequencing depth or low tumor purity, it maintains high detection accuracy and stability, significantly outperforming the four existing detection tools.

[0116] In summary, the experimental results fully demonstrate that this invention is a reliable tool for detecting copy number variations based on second-generation sequencing data samples. It is particularly suitable for practical application scenarios that are greatly affected by noise, sequencing depth, and the complexity of tumor purity, and has high detection accuracy and wide application value.

Claims

1. A method for detecting copy number variations based on semi-supervised learning, characterized in that, Includes the following steps, Step 1: Input the reference genome sequence, the sequencing data sample to be tested, and the seed file. Align and sort the sequencing data sample with the reference genome sequence to obtain the sorted BAM file. Step 2: Under the reference genome coordinate system, the reference genome sequence is divided into fixed-size, non-overlapping windows. Sequencing reads are obtained from the sorted BAM file. Based on the alignment information of the sequencing reads, the read count of all windows in the reference genome sequence is counted. Abnormal windows containing N bases in the reference genome sequence are identified and removed, while normal windows that do not contain N bases are retained. For missing values ​​in the alignment information of sequencing reads, zero-padding is used to replace the missing values. Step 3: Obtain the read depth signal and mapping quality signal of each normal window of the reference genome sequence from the alignment information of the sequencing reads, and save the characteristics of all normal windows of the reference genome sequence; wherein, the read depth signal is the RD signal, which is used to represent the sequencing depth of bases in the sequencing data sample; the mapping quality signal is the MQ signal, which is used to represent the quality of the alignment between the sequencing data sample and the reference genome sequence. Step 4: Correct the GC content deviation for the read segment depth signal of all normal windows; Step 5: Based on the corrected read segment depth signal, a cyclic binary segmentation algorithm is used to divide all normal windows into segmented regions with uniform read segment depth signals. Combined with the segment splitting strategy, potential copy number variation breakpoint locations are identified. Step 6: Normalize the mapping quality signals of all segmented regions; Step 7: Perform smoothing and noise reduction processing on the reading depth signals of all segmented regions; Step 8: Based on the start and end positions of copy number mutations in the seed file, label the corresponding segment regions with pseudo-labels. The seed file is constructed from the intersection of detection results from multiple existing copy number mutation detection tools. A segment region labeled with a pseudo-label will be selected if it meets the following two criteria: the pseudo-labeled segment region belongs to the intersection of detection results from existing copy number mutation detection tools, and the read depth signal corresponding to the pseudo-labeled segment region exceeds [a certain threshold]. The principle is that if the read depth signal of a segmented region labeled with a false label exceeds three standard deviations of the average read depth signal of all segmented regions, it is considered a high-confidence false-labeled region. in Indicates the first The read depth signal of each segmented region This represents the average value of the read depth signal across all segmented regions. This represents the standard deviation of the read depth signal for all segmented regions; Step nine involves using the pseudo-labeled segmented regions as the clustering core of the semi-supervised learning clustering process, and the read depth signals of all segmented regions as clustering features. An improved density clustering algorithm is used to perform cluster analysis on all segmented regions. The resulting cluster analysis results are labeled as normal or abnormal segmented regions using clustering labels. The improved density clustering algorithm, based on the traditional density clustering algorithm, introduces z-score as a third constraint, in addition to the original neighborhood radius and minimum neighbor count parameters. This z-score quantifies the deviation of the read depth signal of a given segment from the average read depth signal of all segmented regions. Under the condition of satisfying the neighborhood radius and minimum neighbor count, segmented regions with z-scores higher than the deviation threshold are labeled as abnormal segmented regions, while those with z-scores lower are labeled as normal segmented regions. For the first z-score of each segmented region Indicates the first The read depth signal of each segmented region This represents the average value of the read depth signal across all segmented regions. This represents the standard deviation of the read depth signal for all segmented regions; Step 10: Integrate the abnormal segmented regions and determine the mutation type, then output the copy number mutation detection results. 2.The method of claim 1, wherein, In step two, positions containing N bases in the reference genome sequence are identified. The reference genome sequence is divided into fixed-size, non-overlapping windows. Windows containing N bases are detected and defined as abnormal windows. The read count of abnormal windows is marked with a specific negative value for distinction. The read count refers to the number of times the sequencing read in the BAM file appears within the reference genome sequence window, reflecting the sequencing coverage of the window. If the read count of a window is negative, the window is determined to be an abnormal window and deleted; otherwise, it is determined to be a normal window, and the start and end positions of the normal window are recorded. For missing values ​​in the alignment information of sequencing reads, zero-padding is used to replace the missing values.

3. The copy number mutation detection method based on semi-supervised learning according to claim 2, characterized in that, In step three, the alignment information of the sequencing reads is obtained, the read depth signal of each base in each normal window is counted according to the start and end positions of each normal window in the reference genome sequence, and the average value of the read depth signals of all bases in the normal window is taken as the read depth signal of the normal window, that is, wherein, represents the read depth signal of the th normal window, represents the length of the normal window, represents the read depth signal of the th base in the th normal window, th base in the th normal window.​​ The alignment information of the sequencing reads is obtained. Based on the start and end positions of each normal window in the reference genome sequence, the mapping quality signal of each base within the normal window is calculated. The average mapping quality signal of all bases within the normal window is used as the mapping quality signal of the normal window. in, Indicates the first The mapping quality signal of a normal window, Indicates the length of a normal window. Indicates the first The first normal window The mapping quality signal of each base.

4. The copy number mutation detection method based on semi-supervised learning according to claim 3, characterized in that, In step four, GC content deviation correction is performed on the read segment depth signals of all normal windows, that is, in, Indicates the first correction The read segment depth signal of a normal window. Indicates the first time before correction The read segment depth signal of a normal window. This represents the average read depth signal of all normal windows before correction. Indicates the difference between before and after correction. The average read depth signal of a normal window with highly consistent GC content.

5. The method of claim 4, wherein, In step five, a cyclic binary segmentation algorithm is used to segment all normal windows, dividing them into segmented regions with uniform read depth signals. During the segmentation process, for discontinuous windows containing gaps, the discontinuous windows are cut and merged according to the fixed length of the segmented regions to restore the continuity of the reference genome sequence coordinates. For wide windows whose window length exceeds a preset length threshold, a fixed-size sliding window is used, starting from the beginning of the wide window and sliding sequentially according to a set step size to further subdivide the wide window into segmented regions of uniform length. The split-read strategy based on CIGAR string is adopted to identify the breakpoint position of copy number variation. The breakpoint is inferred by comparing the relative position relationship between the truncated part in the CIGAR string and the matched region of the reference genome sequence. If the truncated part is located before the matched region, the breakpoint position is located at the end position of the truncated part. If the truncated part is located after the matched region, the breakpoint position is located at the start position of the truncated part. That is, wherein, represents a continuous region matched by the sequencing read and the reference genome sequence, represents a soft truncation in the CIGAR string, represents a hard truncation in the CIGAR string, represents that bases in the sequencing read are matched with the reference genome sequence, represents that bases in the sequencing read are not aligned to the reference genome sequence, but are still retained in the sequencing read, represents that bases in the sequencing read are not aligned to the reference genome sequence, and are discarded in the sequencing read; and represents that the truncated part is located after the matched region, and represents that the truncated part is located before the matched region.

6. The copy number mutation detection method based on semi-supervised learning according to claim 5, characterized in that, In step six, the mapping quality signals of all segmented regions are normalized to eliminate systematic biases introduced by differences in sequencing depth, making the mapping quality signals more accurately reflect the mapping quality. in, Indicates the number of digits after normalization. Mapping quality signal of each segmented region Indicates the first digit before normalization. The average value of the mapped quality signal of each segmented region, Indicates the first digit before normalization. The average value of the read depth signal of each segmented region.

7. The copy number mutation detection method based on semi-supervised learning according to claim 6, characterized in that, In step seven, a total variational model is used to smooth and denoise the read depth signals of all segmented regions. A fidelity term is constructed by calculating the difference between the original read depth signal and the smoothed and denoised read depth signal to constrain the smoothing and denoising process, ensuring that the main characteristics of the original read depth signal are preserved during the process. A total variational regularization term is used by calculating the sum of the absolute values ​​of the differences between adjacent segmented regions in the read signal to quantify the changes in read depth signals between adjacent segmented regions. Different optimization weights are assigned to different segmented regions based on the magnitude of the differences in read depth signals between adjacent segmented regions. in This represents the read depth signal of the segmented region after smoothing and noise reduction. This represents the read depth signal of the segmented region before smoothing and noise reduction, i.e., the original read depth signal. This represents the regularization parameter, used to balance the fidelity and smoothness of the read segment depth signal. Indicates the number of segmented regions. It is the fidelity term of the total variation model, preventing the read depth signal after smoothing and denoising from deviating excessively from the original read depth signal. Indicates the first The segmented region and the first The degree of deviation of the read depth signal in each segmented region. This represents the total variation regularization term, which penalizes sharp fluctuations in the read depth signal, smooths the read depth signal, and preserves true inflection points or abrupt changes.

8. The copy number mutation detection method based on semi-supervised learning according to claim 7, characterized in that, In step ten, the abnormal segment regions selected from the cluster analysis results are sorted according to their starting positions. If the interval between two adjacent abnormal segment regions is very small, the two abnormal segment regions are merged, and the average values ​​of the read depth signal, GC content, and mapping quality signal are taken as the new features of the merged segment region. If the interval is large, the abnormal segment region is recorded as an independent variant region. For the merged segment region, if the read depth signal of the abnormal segment region is higher than the average read depth signal of all merged segment regions, the variant type of the abnormal segment region is defined as duplicate; otherwise, the variant type is defined as missing. The start and end positions, read depth signals, and mutation types of the merged abnormal segment regions are integrated and saved as the detection results in a TXT file for output.

Images