A cnv detection method based on an isolation forest algorithm, medium and equipment
By combining the isolated forest algorithm with RD and PEM methods, the accuracy problem of CNV detection in single-cell DNA sequencing data was solved, the CNV event identification capability was improved, and the optimization of ITH research and cancer treatment was supported.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XI AN JIAOTONG UNIV
- Filing Date
- 2024-04-08
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies cannot accurately identify copy number variations (CNVs) in single-cell DNA sequencing data. In particular, under low sequencing depth conditions, the RD signal is easily interfered with by noise, which limits the application of ITH research.
A CNV detection method based on the isolated forest algorithm is adopted, which combines the RD method and the PEM method. Through variable window strategy and machine learning technology, multi-feature calculation and analysis of RD signal values and PEM signal values are performed to identify CNV events in single-cell DNA sequencing data.
It improves the accuracy and reliability of CNV event detection, provides a powerful tool for ITH research, and optimizes cancer treatment strategies.
Smart Images
Figure CN118230825B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of gene sequencing technology, and in particular to a CNV detection method, medium, and device based on the isolated forest algorithm. Background Technology
[0002] With the rapid development of modern biology and medicine, researchers are gradually gaining a deeper understanding of the pathogenesis and treatment of cancer at the molecular level. Among these advancements, intratumor heterogeneity (ITH), a common phenomenon in cancer, has become a research hotspot in recent years. ITH refers to the presence of multiple subclones with different mutation sets within a single tumor. The existence of these subclones not only increases the difficulty of cancer treatment but is also a key factor leading to treatment failure and drug resistance.
[0003] In ITH research, accurately identifying the mutation status of each clone has become a crucial scientific challenge. While traditional batch DNA sequencing can provide a wealth of genomic information, it often masks the genomic signals of individual cells or a few individuals due to its homogenization of DNA content across multiple cells, making it difficult to accurately reflect the actual situation in ITH. Therefore, single-cell DNA sequencing technology has emerged, capable of querying DNA at the single-cell level, thus revealing the mutation status of each cell more accurately.
[0004] In single-cell DNA sequencing data, mutations are an important molecular level indicator, including single nucleotide variants (SNVs), insertion-deletion variants (INDELs), and structural variants (SVs). Copy number variants (CNVs), as an important form of SV, have been extensively studied for their widespread distribution in the human genome and their association with various cancers. However, inferences based on CNVs in single-cell DNA sequencing data still face significant challenges compared to inferences based on SNVs.
[0005] Currently, CNV detection methods mainly include read splitting methods, paired-end alignment methods, de novo assembly methods, and sequencing depth-based methods. Among these, read depth (RD)-based methods are currently the mainstream approach for CNV detection. However, in single-cell DNA sequencing data, due to the relatively low sequencing depth, the RD signal is often affected by noise, limiting its application in ITH research. Summary of the Invention
[0006] The technical problem to be solved by the embodiments of the present invention is to provide a CNV detection method, medium and device based on the isolated forest algorithm, so as to solve the problem that the prior art cannot accurately identify CNVs in single-cell DNA sequencing data.
[0007] This invention discloses a CNV detection method based on the isolated forest algorithm, comprising:
[0008] Obtain genome sequencing data that introduces CNV events in the reference genome from Bam files of single-cell DNA sequencing;
[0009] The reference genome is divided into windows using a variable window strategy, and the windows are adjusted by a consistent number of reads to obtain a segmentation site information file.
[0010] The RD signal value within each window is calculated based on the segmentation site information file and the genome sequencing data;
[0011] The PEM signal is extracted from the Bam file, and the PEM signal is compared with the segmentation site information file corresponding to the window of the RD signal value to obtain the PEM signal value;
[0012] The isolated forest algorithm is used to perform multi-feature calculation and analysis on the RD signal value and the PEM signal value, and CNV event identification is performed on each window based on the analysis results.
[0013] Optionally, the method for generating the Bam file of the single-cell DNA sequencing is also included:
[0014] Establish a human reference genome;
[0015] The reference genome was sequenced using sequencing tools to obtain the sequencing results of the reference genome, and the sequencing results of the reference genome were set as the sequencing results of a normal single strand of DNA.
[0016] A CNV event is introduced into the reference genome to obtain a variant genome. The variant genome is sequenced using a sequencing tool to obtain sequencing results containing the CNV event. The sequencing results containing the CNV event are set as sequencing results of a single strand of variant DNA.
[0017] The sequencing results of normal DNA single strands and the sequencing results of variant DNA single strands are integrated to obtain the FastQ raw file. The FastQ raw file is then aligned back to the reference genome to obtain the Bam file of the single-cell DNA sequencing.
[0018] Optionally, the method for calculating the RD signal value within each window based on the segmentation site information file and the genome sequencing data includes:
[0019] The segmentation site information file and the genome sequencing data are used as input data for the Bedtools tool. The Bedtools tool is used to statistically analyze the sequencing depth information within each window, and the RD signal value within each window is calculated based on the statistically analyzed sequencing depth information. The RD signal value includes the start point, end point, and sequencing depth after GC correction within the window.
[0020] Optionally, the method of dividing the reference genome into windows using a variable window strategy and adjusting the windows by a consistent number of reads to obtain a segmentation site information file includes:
[0021] The number of reads within the window is preset, and the number of reads within each window is set to be consistent.
[0022] A variable window strategy is used to divide the reference genome into windows, and the number of aligned reads in each window is obtained by comparing the genome sequencing data within the window.
[0023] The size of the corresponding window is adjusted according to the comparison between the number of reads being compared and the preset number of reads, until the number of reads being compared within the window reaches the preset number of reads.
[0024] According to the window that has reached the preset number of reads, the dividing sites are recorded at the start and end of the window respectively, until the recorded dividing sites cover the entire reference genome;
[0025] The recorded segmentation sites are integrated to obtain a segmentation site information file containing the coordinate information of each window region.
[0026] Optionally, the method of extracting the PEM signal from the Bam file and comparing the PEM signal with the segmentation site information file corresponding to the window of the RD signal value to obtain the PEM signal value includes:
[0027] Extract the insert size of the two matching reads from the integrated single-cell DNA double-strand sequencing results from the Bam file;
[0028] Based on the segment covered by the inserted fragment on the single-cell DNA, each segment covered by the inserted fragment is considered a new PEM signal;
[0029] Based on the site information of the PEM signal on the single-cell DNA, if more than half of the PEM signal is contained within the window corresponding to a certain RD signal value, the PEM signal is aligned to the segmentation site information file of the window corresponding to the RD signal value to obtain the PEM signal value.
[0030] Optionally, the method for aligning the PEM signal to the segmentation site information file corresponding to the window of a certain RD signal value and obtaining the PEM signal value when more than half of the PEM signal is contained within it includes:
[0031] The site information of the PEM signal on the single-cell DNA is extracted from the segmentation site information file, and the starting site of the PEM signal is set as x. i The termination site is y i Simultaneously, the starting point of the j-th window in the segmentation site information file is set to m. j The termination site is n j ;
[0032] The PEM signal is matched according to the PEM signal and the window's set parameters. The matching method is as follows:
[0033] If m j <x i <y i <n j Then, in the j-th window, the quantity is incremented by one, and the PEM signal value is incremented by y. i -x i ;
[0034] If the PEM signal does not satisfy m j <x i <y i <n j When, find max(x) i ) <m j and reset x i =max(x i );
[0035] According to the reset of the PEM signal start point, when When the time is right, the quantity is incremented by one in the (j-1)th window, and the PEM signal value is incremented by y. i -x i ;
[0036] when When the time is right, the quantity is incremented by one in the j-th window, and the PEM signal value is incremented by y. i -xi ;
[0037] After the PEM signal matching is completed, the sum of the PEM signal values is calculated and divided by all the counted quantities to obtain the average PEM signal value corresponding to the window.
[0038] Optionally, the method for performing multi-feature calculation and analysis on the RD signal value and the PEM signal value based on the isolated forest algorithm, and for performing CNV event recognition on each window based on the analysis results, includes:
[0039] The RD signal value and PEM signal value corresponding to each window are used as input objects through different combinations. The isolated forest algorithm is used to calculate and analyze the combined input objects to obtain the abnormality score of the genome sequencing data in the corresponding window.
[0040] The calculated anomaly score is compared with a preset threshold:
[0041] If the abnormal score is greater than a preset threshold, then a CNV event is determined to exist within the window;
[0042] If the abnormal score is less than a preset threshold, then the window is determined to be within the normal range.
[0043] Optionally, it also includes a method for performing cell clustering after CNV event identification for each window based on the isolated forest algorithm:
[0044] Based on the determination result of CNV events within the window using the isolated forest algorithm, a recognition result is assigned to each window;
[0045] Based on each identification result, multiple cells are clustered and a cluster tree is drawn. The lineage of a single cell is then inferred from the cluster tree.
[0046] Based on the above-described CNV detection method based on the isolated forest algorithm, this invention also discloses a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described CNV detection method based on the isolated forest algorithm.
[0047] Based on the above-described CNV detection method based on the isolated forest algorithm, this invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described CNV detection method based on the isolated forest algorithm.
[0048] Compared with existing technologies, the CNV detection method based on the isolated forest algorithm provided in this invention has the following advantages:
[0049] By combining the advantages of the RD and PEM methods and introducing machine learning techniques such as the isolated forest algorithm, multi-feature calculation and analysis of RD and PEM signal values can effectively identify CNV events in single-cell DNA sequencing data. This not only overcomes the limitations of traditional methods in single-cell DNA sequencing data, but also improves the accuracy and reliability of CNV event detection. It provides a powerful tool for revealing various ITHs and offers new ideas and methods for ITH research and optimization of cancer treatment strategies. Attached Figure Description
[0050] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. In the accompanying drawings:
[0051] Figure 1 A flowchart illustrating the CNV detection method based on the isolated forest algorithm provided in this embodiment of the invention;
[0052] Figure 2 This is a flowchart illustrating the CNV detection method based on the isolated forest algorithm provided in an embodiment of the present invention.
[0053] Figure 3 This is a schematic diagram of the segmentation site information file on which the PEM signal is compared to the corresponding window of the RD signal value, as provided in an embodiment of the present invention.
[0054] Figure 4 A schematic diagram of ROC curves for the amplified and deleted groups obtained by testing at different thresholds with a simulated 0.2x sequencing depth in an embodiment of the present invention.
[0055] Figure 5 A schematic diagram of ROC curves obtained by testing amplified and deleted groups at different thresholds at simulated 1x sequencing depth provided in an embodiment of the present invention.
[0056] Figure 6 A schematic diagram of ROC curves for the amplified and deleted groups obtained by testing at different thresholds at simulated 5x sequencing depth provided in this embodiment of the invention.
[0057] Figure 7 This is a schematic diagram illustrating the difference in AUC values between two combinations of RD and PEM signal values at three sequencing depths provided in this embodiment of the invention and a single RD signal.
[0058] Figure 8 This is a schematic diagram of a clustering tree of multiple cells at a simulated sequencing depth of 0.2x, provided in an embodiment of the present invention.
[0059] Figure 9 This is a schematic diagram of a clustering tree of multiple cells at a simulated 1x sequencing depth provided in an embodiment of the present invention;
[0060] Figure 10 This is a schematic diagram of a clustering tree of multiple cells at a simulated 5x sequencing depth provided in an embodiment of the present invention;
[0061] Figure 11 This is a schematic diagram of the hardware structure of a computer device for a CNV detection method based on the isolated forest algorithm provided in an embodiment of the present invention. Detailed Implementation
[0062] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0063] This invention provides a CNV detection method based on the isolated forest algorithm, such as... Figure 1 and Figure 2 As shown, it includes:
[0064] Obtain genome sequencing data that introduces CNV events in the reference genome from Bam files of single-cell DNA sequencing;
[0065] A variable window strategy was used to divide the reference genome into windows, and the window was adjusted by a consistent number of reads to obtain the segmentation site information file.
[0066] The RD signal value within each window is calculated based on the segmentation site information file and genome sequencing data;
[0067] Extract the PEM signal from the Bam file and align the PEM signal to the segmentation site information file corresponding to the RD signal value to obtain the PEM signal value;
[0068] The isolated forest algorithm is used to perform multi-feature calculation and analysis on RD signal values and PEM signal values, and CNV event identification is performed on each window based on the analysis results.
[0069] By implementing the above method, genome sequencing data incorporating CNV events were obtained, and the reference genome was divided into windows to ensure a consistent number of reads within each window, thus eliminating the potential influence of other factors on the number of reads. Genome sequencing data and segmentation site information files were used as input to calculate RD and PEM signal values, combining the advantages of both methods. Furthermore, machine learning techniques using the isolated forest algorithm were introduced to perform multi-feature calculation and analysis on the RD and PEM signal values, effectively identifying CNV events in single-cell DNA sequencing data. This not only overcomes the limitations of traditional methods in single-cell DNA sequencing data but also improves the accuracy and reliability of CNV event detection.
[0070] Furthermore, looking back Figure 2 It also includes methods for generating Bam files for single-cell DNA sequencing:
[0071] Establish a human reference genome;
[0072] The reference genome was sequenced using sequencing tools, and the sequencing results of the reference genome were set as the sequencing results of a normal single strand of DNA.
[0073] CNV events were introduced into the reference genome to obtain a variant genome. Sequencing tools were used to sequence the variant genome to obtain sequencing results with CNV events. The sequencing results with CNV events were set as sequencing results of a single strand of variant DNA.
[0074] The sequencing results of normal DNA single strands and the sequencing results of variant DNA single strands are integrated to obtain the FastQ raw file. The FastQ raw file is aligned back to the reference genome, and after cleaning, deduplication and other operations, the Bam file of single-cell DNA sequencing is obtained.
[0075] The above method preferably uses the hg19 version of the human genome as the reference genome and serves as a benchmark. The ART Illumina software is used to simulate genome sequencing. An improved CNV-sim function is used to introduce CNV events. Unlike traditional methods, this embodiment ensures that CNV events are introduced only on one strand of the DNA double helix. Specifically, one strand of the DNA double helix can be randomly selected, and CNV events, including increases or decreases in copy number, can be introduced therein. For the other strand of the DNA double helix without CNV events, the reference genome is used directly for sequencing. This ensures that the simulation data includes the influence of CNV events while preserving the integrity of the reference genome. Finally, appropriate bioinformatics tools are used to integrate the sequencing data of both strands to form the final sequencing result of the DNA double helix. This successfully simulates a human chromosome with ploidy of 2, making the simulation results more realistic. This improved CNV-sim introduction function not only improves the realism of the simulation data but also provides a more reliable data foundation for subsequent genomics research and disease diagnosis. In implementing the above methods, the parameters of the ART illumina software and the improved CNV-sim function can be adjusted and optimized according to specific research needs and data characteristics to further improve the accuracy and reliability of simulation data.
[0076] Furthermore, the method of dividing the reference genome into windows using a variable window strategy and adjusting the window size by a consistent number of reads to obtain the segmentation site information file includes:
[0077] The preset number of reads in a window is set, and the number of reads in each window is kept consistent.
[0078] A variable window strategy was used to divide the reference genome into windows, and the number of aligned reads in each window was obtained by comparing the genome sequencing data within the window.
[0079] Adjust the size of the corresponding window based on the comparison between the number of reads being compared and the preset number of reads, until the number of reads being compared in the window reaches the preset number of reads.
[0080] Based on the window that has reached the preset number of reads, record the dividing sites at the beginning and end of the window respectively, until the recorded dividing sites cover the entire reference genome;
[0081] The recorded division points are integrated to obtain a division point information file containing the coordinate information of each window region.
[0082] The method for adjusting the size of the corresponding window based on the comparison between the number of reads and the preset number of reads is as follows:
[0083] If the number of aligned reads in the window is less than the preset number of reads, gradually increase the size of the window and re-align the genome sequencing data in the window until the number of aligned reads reaches the preset number of reads;
[0084] If the number of aligned reads in the window exceeds the preset number of reads, the window size is gradually reduced, and the genome sequencing data in the window is re-aligned until the number of aligned reads reaches the preset number of reads.
[0085] By implementing the above method, it is preferable to ensure that the number of reads that can be aligned within each variable window is 50,000. This setting is based on a comprehensive consideration of the quality of genome sequencing data, aiming to reduce the impact of data bias and noise on subsequent analysis by standardizing the number of reads within the window. By performing sequencing alignment operations on the reference genome, ensuring that the number of reads aligned back in each window is 50,000, window partitioning is performed, and the partitioning sites are recorded. This divides the reference genome into multiple windows with the same number of reads, providing a standardized dataset for subsequent analysis. Window partitioning of the reference genome using the above method can effectively eliminate the influence of the number of reads on genome analysis, improving the accuracy and reliability of the analysis. It is applicable to various genomics research scenarios, especially those requiring precise control of data quality. The final site partitioning results obtained through the method in this embodiment can be referenced from the research on the hg19 version chromosome by Baslan et al., which provides a partitioning site information file for study.
[0086] Furthermore, methods for calculating the RD signal value within each window based on the segmentation site information file and genome sequencing data include:
[0087] The segmentation site information file and genome sequencing data are used as input data for the Bedtools tool. The Bedtools tool is used to statistically analyze the sequencing depth information within each window, and the RD signal value within each window is calculated based on the statistically analyzed sequencing depth information. The RD signal value includes the start point, end point, and sequencing depth after GC correction within the window.
[0088] Furthermore, the method for extracting the PEM signal from the Bam file and aligning the PEM signal to the segmentation site information file corresponding to the RD signal value to obtain the PEM signal value includes:
[0089] Extract the insert size of the two matching reads from the integrated single-cell DNA double-stranded sequencing results from the Bam file;
[0090] Based on the segment covered by the inserted fragment on the single-cell DNA, each segment covered by the inserted fragment is regarded as a new PEM signal;
[0091] Based on the location information of the PEM signal on the single-cell DNA, if more than half of the PEM signal is contained within the window corresponding to a certain RD signal value, the PEM signal is aligned to the segmentation site information file of the window corresponding to the RD signal value and the PEM signal value is obtained.
[0092] Through the implementation of the above methods, such as Figure 3As shown, each segment covered by an inserted fragment is considered a new PEM signal, allowing the PEM signal to reflect the sequencing depth and quality of different regions in the reference genome, providing important information for subsequent analysis. The PEM signal is aligned to the segmentation site information file corresponding to the RD signal value window. In this way, two corresponding features can be obtained: the RD signal and the PEM signal. These two features represent different aspects of sequencing information, providing a rich data foundation for subsequent analysis. Therefore, by calculating the PEM signal value within each window based on the RD method and combined with the PEM method, the accuracy and precision of genome sequencing data analysis can be improved.
[0093] Furthermore, if more than half of the PEM signal is contained within a window corresponding to a certain RD signal value, the method for aligning the PEM signal to the segmentation site information file corresponding to the RD signal value window and obtaining the PEM signal value includes:
[0094] The PEM signal site information on single-cell DNA was extracted from the segmentation site information file, and the starting site of the PEM signal was set as x. i The termination site is y i Simultaneously, the starting point of the j-th window in the segmentation site information file is set to m. j The termination site is n j ;
[0095] The PEM signal is matched according to the PEM signal and the window's set parameters. The matching method is as follows:
[0096] If m j <x i <y i <n j Then, in the j-th window, the quantity is incremented by one, and the PEM signal value is incremented by y. i -x i ;
[0097] If the PEM signal does not satisfy m j <x i <y i <n j When, find max(x) i ) <m j and reset x i =max(x i );
[0098] According to the reset of the PEM signal start point, when When the time is right, the quantity is incremented by one in the (j-1)th window, and the PEM signal value is incremented by y. i -x i ;
[0099] when When the time is right, the quantity is incremented by one in the j-th window, and the PEM signal value is incremented by y. i -x i ;
[0100] After PEM signal matching is completed, the sum of PEM signal values is calculated and divided by all the counts to obtain the average PEM signal value corresponding to the window.
[0101] By implementing the above method and based on the PEM signal matching method, the accuracy and reliability of PEM signal alignment are ensured, avoiding information loss or errors caused by improper PEM signal segmentation. After completing PEM signal alignment and feature extraction, the window, RD signal value, and PEM signal value are combined in different ways as the input matrix for the Isolation Forest algorithm. This matrix contains the start and end points of each window, the RD signal value, and the PEM signal value, providing more data information for subsequent algorithm analysis. Preferably, this method embodiment proposes two combination methods for different data, which can ensure good recognition results under data of various sequencing depths. The first combination method is to directly use the RD signal value and the PEM signal value as the same input to the algorithm, in which case a window (sample) has two features; the second combination method is to input the RD signal value and the PEM signal value twice, and then integrate their recognition results. When any signal identifies a CNV event, its result is accepted.
[0102] Furthermore, a method for performing multi-feature calculation and analysis on RD and PEM signal values based on the isolated forest algorithm, and then identifying CNV events for each window based on the analysis results, includes:
[0103] The RD signal value and PEM signal value corresponding to each window are used as input objects in different combinations. The isolated forest algorithm is used to calculate and analyze the combined input objects to obtain the abnormality score of the genome sequencing data in the corresponding window.
[0104] The calculated anomaly score is compared with a preset threshold:
[0105] If the abnormal score is greater than the preset threshold, then a CNV event is determined to exist within the window;
[0106] If the abnormal score is less than the preset threshold, then the range within the judgment window is considered normal.
[0107] By implementing the above method, the Isolation Forest algorithm is used as the analysis tool. It accepts RD and PEM signal values as two features, and each variable window is treated as an identification object. Through processing by the Isolation Forest algorithm, CNV events in the reference genome can be identified. For the first combination of RD and PEM signal values: using RD and PEM signal values as two features in one window, the Isolation Forest algorithm performs only one analysis. For the second combination of RD and PEM signal values: RD and PEM signal values are input twice, and the Isolation Forest algorithm performs two analyses, distinguishing the thresholds of the two signals (finding the best effect within a threshold difference of 0.1), and then integrating the CNV event results. For each variable window, the Isolation Forest algorithm calculates its anomaly score, which reflects the degree of anomaly in the genome sequencing data within the window and is an important basis for determining the existence of CNV events.
[0108] As mentioned above, after performing CNV event determination for each window based on the isolated forest algorithm, the following steps are also included:
[0109] Creating window labels: Based on the segmentation site information file, the cnv-sim tool is used to output the CNV event site information on single-cell DNA. Labels are then created for each window based on the output CNV event site information. Specifically, for each window, if half of the window is covered by a CNV event, the window is marked as positive (CNV present); otherwise, it is marked as negative (normal range).
[0110] Comparison of judgment results with labels: The judgment results based on the isolated forest algorithm are compared with the window labels, and indicators such as true positive rate and false positive rate are calculated to measure the algorithm's detection effect on CNV.
[0111] Based on the judgment results and window labels of the Isolation Forest algorithm, windows are divided into the following four categories: True Positive (TP) indicates a window correctly identified as a CNV by the algorithm and consistent with the window label; True Negative (TN) indicates a window correctly identified as a non-CNV by the algorithm and consistent with the window label; False Positive (FP) indicates a window incorrectly identified as a CNV by the algorithm but inconsistent with the window label; False Negative (FN) indicates a window incorrectly identified as a non-CNV by the algorithm but inconsistent with the window label.
[0112] By implementing the above method, the method of this embodiment can be tested by simulating Bam files of three depths (sequencing depths of 0.2x, 1x, and 5x) using simulation software, statistically analyzing the results at different thresholds, and plotting ROC graphs (see reference). Figures 4-6 The amplified group represents CNV amplification events, using only the RD signal as input, while the deleted group represents CNV deletion events. This demonstrates the impact of incorporating the PEM signal on the algorithm's classification performance under the first combination of RD and PEM signal values. It can be observed that at low sequencing depths, the first combination—treating the PEM and RD signals as two features of a single sample—actually reduces the algorithm's recognition performance. Therefore, for low-depth data, the second combination of RD and PEM signal values is recommended for discrimination, i.e., separating the PEM and RD signals and performing two separate algorithmic discriminations. Figure 7 As shown, based on the AUC value as a measure of effectiveness, the differences in performance between the first and second combinations of RD and PEM signal values and a single RD signal are illustrated. It can be observed that at medium to high depths, the first combination of RD and PEM signal values yields the best results. However, for low depths (0.2x), the PEM signal threshold needs to be increased by 0.09 to achieve better results. Therefore, when using the CNV detection method based on the isolated forest algorithm of this invention for CNV detection, the second combination of RD and PEM signal values is recommended for low-depth data, with a higher threshold for the PEM signal; for medium to high depth data, the first combination of RD and PEM signal values is recommended directly.
[0113] Furthermore, it also includes a method for cell clustering after CNV event identification for each window based on the isolated forest algorithm:
[0114] Based on the determination results of CNV events within a window using the isolated forest algorithm, each window is assigned a recognition result;
[0115] Based on each identification result, multiple cells are clustered and a cluster tree is drawn. The lineage of a single cell is then inferred from the cluster tree.
[0116] By implementing the above method, the isolated forest assigns an identification result to each identified object, thus allowing clustering of a large number of cells based on this result to infer the relationships between cells. Through simulation, subclones with three different CNV intervals were obtained based on different CNV events, and 30 cells were generated for each subclone. The CNV amplification data obtained by simulating the method of this embodiment using Bam files of three depths (sequencing depths of 0.2x, 1x, and 5x) are as follows: Figures 8-10 The figure shows the clustering results of 90 cells at three sequencing depths. In the figure, a, b, and c represent three subclones, demonstrating that the three cell types are clearly assigned to three branches of the cell lineage, proving the feasibility of the method proposed in this patent for inferring cell lineage.
[0117] Based on the CNV missing data obtained from tests on datasets of three depths, and using the same clustering tree drawing method as the CNV amplification method, clustering results are drawn based on the dual-feature detection results of the combination of RD signal values and PEM signal values. Specifically, when it is not necessary to detect CNVs in specific segments, but only to perform lineage inference on single cells, this embodiment of the invention also provides a new method: directly using the anomaly scores given by the isolated forest to assign values to the segmentation site information corresponding to the RD signal values, and then performing clustering based on the scores of each segmentation site. This method provides a judgment on the probability of CNV existence for each segmentation site, avoiding errors that may be introduced manually when selecting a threshold. Figure c shows the results of selecting a reasonable threshold (0.6) and directly using anomaly scores for lineage inference. It can be seen that although the distance values of the three subclones are larger in the clustering results based on CNV detection results, directly using anomaly scores for clustering can also achieve correct classification results.
[0118] Based on the above-mentioned CNV detection method based on the isolated forest algorithm, this invention also discloses an identification device, which includes:
[0119] The genome sequencing module is used to obtain genome sequencing data from the Bam file of single-cell DNA sequencing;
[0120] The genome windowing module is used to divide the reference genome into windows using a variable window strategy and to obtain the segmentation site information file by adjusting the window through a consistent number of reads.
[0121] The RD signal value calculation module is used to calculate the RD signal value within each window based on the segmentation site information file and genome sequencing data.
[0122] The PEM signal calculation module is used to extract the PEM signal from the Bam file and compare the PEM signal with the segmentation site information file corresponding to the RD signal value to obtain the PEM signal value.
[0123] Based on the above-described CNV detection method based on the isolated forest algorithm, this invention also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described CNV detection method based on the isolated forest algorithm.
[0124] Based on the above-described CNV detection method based on the isolated forest algorithm, this invention also discloses a computer device, such as... Figure 11 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the CNV detection method based on the isolated forest algorithm described above.
[0125] This invention is described based on flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to specific embodiments. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the flowcharts and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0126] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0127] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
Claims
1. A CNV detection method based on the isolated forest algorithm, characterized in that, The CNV detection method includes: Obtain genome sequencing data that introduces CNV events in the reference genome from Bam files of single-cell DNA sequencing; The reference genome is divided into windows using a variable window strategy, and the windows are adjusted by a consistent number of reads to obtain a segmentation site information file. The RD signal value within each window is calculated based on the segmentation site information file and the genome sequencing data; The PEM signal is extracted from the Bam file, and the PEM signal is compared with the segmentation site information file corresponding to the window of the RD signal value to obtain the PEM signal value; The isolated forest algorithm is used to perform multi-feature calculation and analysis on the RD signal value and the PEM signal value, and CNV event identification is performed on each window based on the analysis results; The method for extracting the PEM signal from the Bam file and comparing the PEM signal with the segmentation site information file corresponding to the window of the RD signal value to obtain the PEM signal value includes: Extract the insert size of the two matching reads from the integrated single-cell DNA double-strand sequencing results from the Bam file; Based on the segment covered by the inserted fragment on the single-cell DNA, each segment covered by the inserted fragment is considered a new PEM signal; Based on the site information of the PEM signal on the single-cell DNA, if more than half of the PEM signal is contained within the window corresponding to a certain RD signal value, the PEM signal is compared to the segmentation site information file of the window corresponding to the RD signal value to obtain the PEM signal value. The method for aligning the PEM signal to the segmentation site information file corresponding to the window of a certain RD signal value and obtaining the PEM signal value when more than half of the PEM signal is contained within it includes: The site information of the PEM signal on the single-cell DNA is extracted from the segmentation site information file, and the start site of the PEM signal is set as... The termination site is Simultaneously, the starting point of the j-th window in the segmentation site information file is set as... The termination site is ; The PEM signal is matched according to the PEM signal and the window's set parameters. The matching method is as follows: like Then, in the j-th window, the quantity is incremented by one, and the PEM signal value is incremented. ; If the PEM signal does not meet the requirements At that time, searching and reset = ; According to the reset of the PEM signal start point, when When the time is right, the quantity is incremented by one in the (j-1)th window, and the PEM signal value is incremented. ; when When the time is right, the quantity is incremented by one in the j-th window, and the PEM signal value is incremented. ; After the PEM signal matching is completed, the sum of the PEM signal values is calculated and divided by all the counted quantities to obtain the average PEM signal value corresponding to the window.
2. The CNV detection method based on the isolated forest algorithm according to claim 1, characterized in that, It also includes a method for generating the Bam file for the single-cell DNA sequencing: Establish a human reference genome; The reference genome was sequenced using sequencing tools to obtain the sequencing results of the reference genome, and the sequencing results of the reference genome were set as the sequencing results of a normal single strand of DNA. A CNV event is introduced into the reference genome to obtain a variant genome. The variant genome is then sequenced using a sequencing tool to obtain sequencing results containing the CNV event. The sequencing results containing the CNV event are then set as sequencing results of a single strand of variant DNA. The sequencing results of normal DNA single strands and the sequencing results of variant DNA single strands are integrated to obtain the FastQ raw file. The FastQ raw file is then aligned back to the reference genome to obtain the Bam file of the single-cell DNA sequencing.
3. The CNV detection method based on the isolated forest algorithm according to claim 1, characterized in that, The method of dividing the reference genome into windows using a variable window strategy and adjusting the windows by a consistent number of reads to obtain a segmentation site information file includes: The number of reads within the window is preset, and the number of reads within each window is set to be consistent. A variable window strategy is used to divide the reference genome into windows, and the number of aligned reads in each window is obtained by comparing the genome sequencing data within the window. The size of the corresponding window is adjusted according to the comparison between the number of reads being compared and the preset number of reads, until the number of reads being compared within the window reaches the preset number of reads. According to the window that has reached the preset number of reads, the dividing sites are recorded at the start and end of the window respectively, until the recorded dividing sites cover the entire reference genome; The recorded segmentation sites are integrated to obtain a segmentation site information file containing the coordinate information of each window region.
4. The CNV detection method based on the isolated forest algorithm according to claim 3, characterized in that, The method for calculating the RD signal value within each window based on the segmentation site information file and the genome sequencing data includes: The segmentation site information file and the genome sequencing data are used as input data for the Bedtools tool. The Bedtools tool is used to statistically analyze the sequencing depth information within each window, and the RD signal value within each window is calculated based on the statistically analyzed sequencing depth information. The RD signal value includes the start point, end point, and sequencing depth after GC correction within the window.
5. The CNV detection method based on the isolated forest algorithm according to claim 1, characterized in that, The method for performing multi-feature calculation and analysis on the RD signal value and the PEM signal value based on the isolated forest algorithm, and for performing CNV event identification on each window based on the analysis results, includes: The RD signal value and PEM signal value corresponding to each window are used as input objects through different combinations. The isolated forest algorithm is used to calculate and analyze the combined input objects to obtain the abnormality score of the genome sequencing data in the corresponding window. The calculated anomaly score is compared with a preset threshold: If the abnormal score is greater than a preset threshold, then a CNV event is determined to exist within the window; If the abnormal score is less than a preset threshold, then the window is determined to be within the normal range.
6. The CNV detection method based on the isolated forest algorithm according to claim 5, characterized in that, It also includes a method for performing cell clustering after CNV event identification for each window based on the isolated forest algorithm: Based on the determination result of CNV events within the window using the isolated forest algorithm, a recognition result is assigned to each window; Based on each identification result, multiple cells are clustered and a cluster tree is drawn. The lineage of a single cell is then inferred from the cluster tree.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the CNV detection method based on the isolated forest algorithm as described in any one of claims 1-6.
8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the CNV detection method based on the isolated forest algorithm as described in any one of claims 1-6.