WO2018201805A1 - 用于计算癌症样本纯度和染色体倍性的方法和装置 - Google Patents

用于计算癌症样本纯度和染色体倍性的方法和装置 Download PDF

Info

Publication number
WO2018201805A1
WO2018201805A1 PCT/CN2018/078908 CN2018078908W WO2018201805A1 WO 2018201805 A1 WO2018201805 A1 WO 2018201805A1 CN 2018078908 W CN2018078908 W CN 2018078908W WO 2018201805 A1 WO2018201805 A1 WO 2018201805A1
Authority
WO
WIPO (PCT)
Prior art keywords
tre
value
peak
genome
calculated
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2018/078908
Other languages
English (en)
French (fr)
Inventor
黄宇
罗志辉
苏瑶
范新平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Institute of Materia Medica of CAS
Original Assignee
Shanghai Institute of Materia Medica of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Institute of Materia Medica of CAS filed Critical Shanghai Institute of Materia Medica of CAS
Publication of WO2018201805A1 publication Critical patent/WO2018201805A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16BBIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B15/00ICT specially adapted for analysing two-dimensional [2D] or three-dimensional [3D] molecular structures, e.g. structural or functional relations or structure alignment
    • CCHEMISTRY; METALLURGY
    • C12BIOCHEMISTRY; BEER; SPIRITS; WINE; VINEGAR; MICROBIOLOGY; ENZYMOLOGY; MUTATION OR GENETIC ENGINEERING
    • C12QMEASURING OR TESTING PROCESSES INVOLVING ENZYMES, NUCLEIC ACIDS OR MICROORGANISMS; COMPOSITIONS OR TEST PAPERS THEREFOR; PROCESSES OF PREPARING SUCH COMPOSITIONS; CONDITION-RESPONSIVE CONTROL IN MICROBIOLOGICAL OR ENZYMOLOGICAL PROCESSES
    • C12Q1/00Measuring or testing processes involving enzymes, nucleic acids or microorganisms; Compositions therefor; Processes of preparing such compositions
    • C12Q1/68Measuring or testing processes involving enzymes, nucleic acids or microorganisms; Compositions therefor; Processes of preparing such compositions involving nucleic acids
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16BBIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B30/00ICT specially adapted for sequence analysis involving nucleotides or amino acids

Definitions

  • the present invention is in the field of cancer research, and in particular relates to a method and apparatus for calculating cancer cell purity and intracellular chromosome ploidy in a cancer sample.
  • Cancer research is an important research area in life medicine and has a major impact on human health. Cancer is a kind of malignant proliferation of cells. Because of its complicated pathology, humans cannot overcome such diseases. Second generation sequencing provides the possibility to quickly detect patient genetic information. However, sequencing requires the extraction of samples from patient tissue, but usually cancer tissue does not simply contain cancer cells, it also has a very rich microenvironment. Cancer cell microenvironment refers to the environment of non-cancerous cells surrounding or accompanying cancer cells. When the cancer cell samples are extracted, these microenvironments are extracted together with the cancer cells and sequenced along with the cancer cells [1]. The proportion of cancer cells in a cancer sample is defined as the purity of the cancer sample.
  • the cancer genome usually contains a large number of somatic cell copy number variations, which are mainly caused by amplification or deletion of genomic fragments. Identifying changes in the copy number of genomic fragments of a particular tumor genome is an important topic in cancer genome research. Accurate identification of genomic fragment copy number has certain challenges, because the cancer fragment copy number is mainly determined by a mixture of two factors, one is the purity of cancer samples, that is, the proportion of cancer cells in cancer samples, and the other is chromosome ploidy [2, 3 ]. Traditional methods for identifying cancer sample purity and ploidy are using experimental techniques such as quantitative image analysis [4] or single cell sequencing [5]. But in large projects, such an approach can cost a lot of manpower, money, and time. With the development of sequencing technology, the rapid growth of sequencing data, and the accumulation of sequencing data analysis techniques, various cancer sample purity algorithms have been proposed and corresponding software has been developed.
  • PyLOH [11] used frequency information of hybrid SNV (single nucleotide variation) sites on the genome, and copy number of genomic fragments. PyLOH solves the problem of “difficulties in identification” to a certain extent, and can give a more reasonable solution. However, its accuracy is poor, especially in the case of subclone in the genome. Patchwork uses both types of information, but in the intermediate steps of calculating genotypes, manual identification is required, the results of manual judgments lack accuracy, and such semi-automated software brings a lot of inconvenience to the application.
  • Sequencing depth The ratio of the total number of bases (bp) obtained by sequencing to the size of the genome (Genome), which is one of the indicators for evaluating the amount of sequencing.
  • Window A genomic fragment divided by a length that represents the size of the window.
  • the window size can be freely set by the user, and is usually set to several hundred bases.
  • a large genomic fragment S can contain a large number of windows.
  • Tumor Read Enrichment Cancer long fragment reads the enrichment e s, refers to the ratio of the number of read number of segments within a cancerous samples read with a corresponding segment S corresponding normal sample, the following formula is defined:
  • N t represents the total number of reads of the cancer sample whole genome sequencing
  • N n represents the corresponding normal sample whole genome sequencing obtained read The total number.
  • Heterozygous Germline Single Nucleotide Variants Single base variation in heterozygous germline cells. Since human chromosomes are diploid, somatic cells are derived from embryonic cells, while HGSNV sites in germ cells have only two base types. A and B, one of which comes from the father and the other from the mother.
  • Major Allele Fraction The major allele score.
  • the HGSNV used in the present invention has only two alleles, one allele is identical to the reference genome and the other is different from the reference genome.
  • the two allelic scores are calculated by covering the number of reads of an allele divided by the ratio of the total number of reads covering the site, and MAF is the larger of the two allelic scores.
  • the calculation formula is as shown in (1.1), where n r is the number of reads containing the same allele as the reference genome, n a is the number of reads containing another allele, and n t is the total read covering the HGSNV site.
  • the quantity, C is the MAF value of the HGSNV.
  • MAF is a concept relative to HGSNV.
  • “MAF of a fragment” refers to the MAF mean of all HGSNVs in a fragment
  • MAF of peak refers to the MAF mean of HGSNV contained in all fragments in a peak.
  • Major allele copy number the copy number of the major allele, which refers to the value of the copy number of the major allele in the fragment with copy number i, which is greater than or equal to The integer.
  • Peak refers to the TRE clusters that are clustered together in the TRE distribution of all windows in the genome.
  • Figure A shows the TRE distribution of all windows on the genome, and the vertical axis shows the total number of windows corresponding to a TRE site.
  • the figure shows the TRE distribution before the genomic GC content correction.
  • Figure B shows the GC content correction.
  • the TRE distribution, in Figure B, can be seen that the window is obviously clustered.
  • This method defines the TRE cluster identified by the autoregressive model as peak, which is essentially the aggregation of the window within the genome segment with the same copy.
  • Cancer sample A cancer tissue taken from an individual with cancer, which contains a portion of cancer cells and a portion of normal cells.
  • P refers to the spacing between two adjacent peaks. Since peak is a cluster, the TRE of peak here is represented by the TRE mean of peak, so it is actually the difference of the TRE mean of two adjacent peaks. Since peak is an aggregation of windows within the same copy number genome segment, it is also expressed here as the difference of adjacent copy number segments TRE.
  • the object of the present invention is to overcome the deficiencies of the prior art and to provide a fully automatic, high efficiency, high accuracy method and apparatus for calculating cancer sample purity and chromosome ploidy.
  • the invention has broad application prospects in the calculation of cancer sample purity and chromosome ploidy.
  • the technical solution adopted by the present invention is to construct a mixed Gaussian model of the MAF distribution of TRE and HGSNV of different copy number fragments by using whole genome sequencing data of cancer samples and matched normal samples, and calculate the purity of the cancer sample and Chromosome ploidy.
  • the invention mainly uses the TRE information of the whole genome sequencing data and the MAF information of the HGSNV.
  • TRE basically reflects the copy number variation of cancer samples
  • MAF information of HGSNV basically reflects the genotype of cancer samples.
  • the difference in TRE is mainly due to the difference in copy number of genomic fragments.
  • the number of reads obtained by sequencing in high copy number genomic fragments must be greater than the number of reads obtained by sequencing of low copy number genomic fragments.
  • the difference in copy number of fragments is calculated by the difference in the number of reads in the fragments. Common methods in genome copy number detection. However, in most studies, the difference in the number of reads in the cancer sample segment was divided by the ratio of the number of reads in the normal sample to calculate the difference in the number of reads.
  • the present invention uses the TRE shown in the formula (1) to evaluate the difference in the number of reads of different segments of the slice.
  • the calculated ratio of the traditional method is not only affected by the purity and chromosome ploidy of the cancer sample, but also by the depth of sequencing of the cancer sample and the normal sample, and the TRE is not affected by the depth of the sample sequencing.
  • the genotype of each copy number fragment cannot be determined by relying solely on the difference in the number of reads, and more importantly, the compensation effect of sample purity and sample ploidy cannot be distinguished.
  • HGSNV combined with copy number difference fragments can provide genotype information and help to solve the compensation effect of purity and ploidy.
  • the genotypes that may correspond to different copy number fragments are listed one by one, and then the results of the permutation and combination are calculated to select the most reliable results.
  • the common feature of these methods is that the method has a long calculation time and poor accuracy, and the sample with high copy number or large genomic variation has a poor effect.
  • the invention calculates the purity and chromosome ploidy of the cancer sample according to the mixed Gaussian model of MAF and TRE of HGSNV, can significantly reduce the calculation time and improve the accuracy of the calculation result.
  • the proportion of normal cells in the cancer sample is 1- ⁇ .
  • the normal cell has a chromosome ploidy of 2 in the cancer sample, and the cancer cell has a chromosome ploidy of ⁇ .
  • the chromosome ploidy ⁇ of the cancer sample is as shown in the formula (2).
  • the TRE is calculated as shown in equation (1).
  • the derivation formula of the expectation E(e s ) of TRE is as shown in the following formula (4), N n and N t have the same meanings as in formula (1).
  • the method defines a parameter of some help understanding.
  • Length of fragment S L S length of human reference genome L gw , depth of sequencing of cancer samples
  • Sampling depth of normal samples Then the fragment S is sequenced in a cancer sample to a depth of Fragment S is sequenced in a normal sample to a depth of ⁇ S refers to a parameter related to the characteristics of the segment S (such as the GC content and the like which cause the depth of preference of the sequencing), so it is the same in cancer and normal samples.
  • e s is represented by ⁇ , ⁇ , C s as shown in the formula (5).
  • C s represents the copy number of the fragment S in the cancer cell
  • the corresponding TRE mean S i and S i+1 when the copy number of the fragment S is i and i+1 are as shown in the formula (6), respectively.
  • the cancer sample purity ⁇ and chromosome ploidy ⁇ can be calculated by determining P and Q.
  • the peak distribution of the peak shown in Figure B is because the TRE value of the genomic fragment with the same copy number (the mean value of the TRE of all windows in the fragment) is not completely equal, and the copy number fragment TRE is mutually There is an error between them. This error obeys a Gaussian distribution, so the clustered distribution in Figure B is considered to be a Gaussian distribution.
  • the Q site represents the TRE value corresponding to the segment of copy number 2.
  • the number of windows corresponding to the distance P and the distance 2P before the Q site is 0.
  • the distribution map of TRE for X f , that is, the first peak that appears, it may correspond to a copy number of 2 (the copy corresponding to the peak of copy number 1 and 0 is 0), and the corresponding copy number may be 1 (The fragment with a copy number of 0 corresponds to the window of the peak is 0.) It is also possible that the corresponding copy number is 0.
  • the copy number of the first peak in Figure 2 that is, the X f corresponding segment has several different possibilities, and each of them may make Q correspond to a different peak.
  • the present invention calculates the copy number of the most likely X f corresponding segment by mixing the Gaussian model, thereby determining the value of Q, and finally obtaining the purity and chromosome ploidy of the cancer sample.
  • the present invention determines the value of X f by the following formula (13.1).
  • C(X f +P) represents the number of windows corresponding to the position where TRE is X f + P, and n represents the maximum number of peaks within M t .
  • f(X f ) takes the maximum value, X f is the TRE mean of the first peak.
  • Equation (13.2) uses Equation (13.2) to find up to a maximum of several peaks before X f .
  • X f represents a mean peak of TRE
  • P denotes the distance between the copy number of segments corresponding to an adjacent peak
  • floor represents rounded down
  • N 0, indicating no prior peak X f
  • X f corresponds
  • a corresponding Q value can be calculated by the following formula (13.3).
  • Q is the TRE value corresponding to the peak of the fragment with a copy number of 2.
  • the TRE value of the fragment with a copy number of 0 is X f -n ⁇ P, where n represents the number of peaks before X f , the value ranges from 0 to N, and P represents the adjacent copy number.
  • the spacing between the peaks corresponding to the segments, the meaning of X f is the same as in the formula (13.1), and the formula (13.3) is as follows, where Q n represents the value of Q when there are n peaks before X f .
  • the MAF actually obeys the binomial distribution with f as the probability and d as the number of experiments.
  • the present invention corrects f by the following formula (16) to obtain the expected value f b of the MAF.
  • k represents the number of alleles (A or B) at a certain HGSNV site, and the total amount of alleles measured is d (equal to the sequencing depth).
  • Equation (13.2) indicates that there are N possibilities for the copy number of each genomic fragment.
  • Equation (14) shows that a certain copy number fragment can have multiple major allele copy numbers, so for each genome peak, multiple f can be calculated, and multiple f b can be calculated, and the average observed MAF value is taken from the distance peak. The most recent f b is expected as the MAF of the peak. The whole genome has multiple peaks, and the MAF expectation of each peak is different, corresponding to calculating multiple MAF expected values ⁇ f b ⁇ . Considering that there is a certain error in the MAF of HGSNV in a certain peak, but also obeys the Gaussian distribution, the expected value of MAF of all HGSNV in peak can be directly calculated from the actual data.
  • the copy number and genotype of the peak can be judged by comparing the MAF observation of peak with the value of ⁇ f b ⁇ of peak.
  • the TRE value corresponding to the peak of the copy number 2 that is, the position of Q can be calculated.
  • the present invention proposes a hybrid Gaussian model to fit the observation data of TRE and HGSNV.
  • Equation 17 The Gaussian distribution model of TRE is shown in Equation 17:
  • L(e s ; ⁇ , ⁇ ) represents the likelihood function of the genomic fragment TRE.
  • N represents the number of all windows on the genome.
  • I represents the largest copy number of all fragments in the genome.
  • ⁇ i represents the standard deviation of the TRE of all segments of copy number i.
  • e s is the TRE observation of the sth window, and
  • S i represents the TRE mean of the ith peak.
  • p i represents the weight of the copy number of the sth window, i, and all i, p i in this formula take the value 1.
  • the formula shows that the size of the likelihood function is related to the value of S i . When S i and e S are closer, the value of the likelihood function is larger, and the closer the P value is to the true value. A reasonable P value can be calculated using the maximum likelihood estimation of L(e s ; ⁇ , ⁇ ).
  • the present invention further determines the P value and determines the Q value by combining the Gaussian distribution model of HGSNV as shown in the formula (18).
  • L(f s ; ⁇ , ⁇ ) represents the likelihood function of HGSNV.
  • M represents all HGSNV in the genome.
  • S represents the Sth HGSNV.
  • I represents the largest copy number of all fragments in the genome.
  • F i,j is the expected value of MAF of HGSNV in the fragment whose copy number is i and the copy number of the major allele is j, that is, f b calculated by the formula (16).
  • f s represents the mean value of the observed values of the MAF within the segment
  • ⁇ i,j represents the standard deviation of the MAF observations of the HGSNV within the segment.
  • Equation (18) shows that the size of the likelihood function is related to the value of F i,j .
  • F i,j is closer to f s , the larger the likelihood function, the more accurate f s is , and the formula (14) is also shown. The more accurate f is, the corresponding C cp and C mcp of each fragment can be obtained. Then the value of Q is determined.
  • the method adds equation (17) and formula (18) to obtain a mixed Gaussian model.
  • the present invention further uses a Bayesian Information Criterion (BIC) method to give a mixed Gaussian model a penalty function for controlling over-fitting of the model.
  • BIC Bayesian Information Criterion
  • BIC(S s , f s ; ⁇ , ⁇ ) represents the likelihood function of the mixed model
  • I is the number of Gaussian distributions in equation (17)
  • J is the number of Gaussian distributions in equation (18)
  • N is the number of windows in the genome
  • M is the number of HGSNVs in the genome.
  • one aspect of the invention provides a method for calculating cancer cell purity and ploidy ploidy in a cancer sample, the method comprising the steps of:
  • step A From the comparison result file obtained in step A, the read position and length information, the HGSNV site and the read quantity information covering the site are extracted, and the MAF of all HGSNVs is calculated, wherein the calculation formula is as shown in (1.1):
  • n r is the number of reads containing the same allele as the reference genome
  • n a is the number of reads containing another allele
  • n t is the total number of reads covering the HGSNV site
  • C The MAF value of the HGSNV
  • the number of reads contained in each window is counted in units of window, and the number of reads in all windows is corrected by using the genomic GC content;
  • the TRE of each window is calculated using equation (1), and then the genome is fragmented by BIC-seq software using TRE to obtain a genomic fragment divided by copy number:
  • the mean, variance, and number of windows in the window of all windows in the segment are counted, and the number of windows per segment of the genome is smoothed according to the mean and variance (smooth Processing, the distribution of TRE is more uniform, and then the window distribution of all segments after smoothing is summarized to obtain the distribution result of window change with TRE on the genome; and the mean value of MAF of all HGSNVs in the segment is calculated in units of fragments. variance;
  • Interval C(X t ) represents the number of windows corresponding to the position where TRE is X t ; C(X t+1000 ⁇ P ) represents the number of windows corresponding to the position where TRE is X t+1000 ⁇ P ;
  • Y(P) represents the function value of the auto-regressive model under the variable P;
  • step F calculate the TRE mean of the first actual observed peak in the TRE distribution, and then calculate the maximum number of theoretical peaks N before the first actual peak, and finally there are n before the first actual peak.
  • the value of Q is calculated, denoted by Q n , where step G may include:
  • i represents the ith peak
  • C(X f + P ⁇ i) represents the position where TRE is X f + P ⁇ i, the corresponding number of windows
  • n represents the maximum number of peaks within M t
  • M t represents the maximum value of TRE
  • X f represents the mean of the first peak
  • P represents the spacing between peaks corresponding to adjacent copy number segments
  • floor represents an integer below
  • n represents the number of peaks before X f , the value ranges from 0 to N, P represents the spacing between peaks of adjacent copy number segments, and X f represents the first actual observation.
  • the TRE mean of peak, Q n represents the Q value when there are theoretically n peaks before X f ;
  • represents the purity of the sample, and ⁇ represents the ploidy of the chromosome, so that the corresponding ( ⁇ , ⁇ ) can be obtained for all (P, Q N );
  • n represents the number of peaks before X f
  • the value range is an integer between 0 and N
  • P represents the spacing between peaks corresponding to adjacent copy number segments
  • X f represents the first actual observation.
  • the TRE mean of peak T i represents the TRE mean of the ith peak
  • the fragment For a fragment that falls near T i , the fragment is considered to have a copy number i; for a fragment that does not fall near T i , it is classified as a subcloned fragment, and all subcloned fragments are eliminated in subsequent analysis; Calculate the copy number of the cancer sample purity ⁇ and peak, calculate the expected f b of the MAF of peak, and the MAF expectation of different peaks. For all peaks on the genome, finally obtain the desired set of MAF ⁇ f b ⁇ ; TRE mean and variance (or standard deviation) for each peak;
  • Step J can include the following steps:
  • L(e s ; ⁇ , ⁇ ) represents the likelihood function of the genomic fragment TRE
  • N represents the number of all windows on the genome
  • I represents the maximum copy number of all fragments in the genome
  • ⁇ i represents a copy
  • the standard deviation of the TRE of all segments of number i is obtained by step I
  • e s is the TRE observation value of the sth window
  • S i represents the TRE mean value of the i th peak, that is, T i in step I
  • p i represents the first
  • the copy number of s windows is the weight of i, and the value of all i, p i is 1;
  • L(f s ; ⁇ , ⁇ ) represents the likelihood function of HGSNV
  • M represents the number of all HGSNVs in the genome
  • S represents the Sth HGSNV
  • I represents the maximum copy number of all fragments in the genome
  • i, j represents the expected value of the MAF of the HGSNV in the fragment whose copy number is i, the copy number of the major allele is j, obtained from step I
  • f s represents the mean value of the observed value of the MAF of all HGSNVs in the fragment, obtained from step E
  • ⁇ i,j represents the standard deviation of the MAF observations of all HGSNVs in the segment, obtained from step E
  • p i,j represents the weight of the Gaussian distribution when the copy number of the primary allele is j, for all i And j, p i, j have a value of 1
  • p i represents the weight of the copy of the segment where the S H HSNSNV is i, and the value of all
  • BIC(e s , f s ; ⁇ , ⁇ ) represents the likelihood function of the mixed model
  • I represents the maximum copy number of all the fragments in the genome
  • J is the value of j in the formula (18).
  • Number, N is the number of windows in the genome, and M is the number of HGSNVs in the genome.
  • Q n can be obtained by step G, or the desired set of MAFs of all peaks ⁇ f b ⁇ can be obtained by step I, and a pair (P, ⁇ f b ⁇ ) It is possible to construct a model shown in equation (19), essentially for each pair (P, Q n ), to construct a model as shown in equation (19);
  • step H the corresponding cancer sample purity and chromosome ploidy can be found under (P, Q) obtained in step K.
  • the invention provides an apparatus for calculating cancer cell purity and chromosome ploidy in a cancer sample, comprising a processor for running a program, the program running the following steps:
  • step A From the comparison result file obtained in step A, the read position and length information, the HGSNV site and the read quantity information covering the site are extracted, and the MAF of all HGSNVs is calculated, wherein the calculation formula is as shown in (1.1):
  • n r is the number of reads containing the same allele as the reference genome
  • n a is the number of reads containing another allele
  • n t is the total number of reads covering the HGSNV site
  • C The MAF value of the HGSNV
  • the number of reads contained in each window is counted in units of window, and the number of reads in all windows is corrected by using the genomic GC content;
  • the TRE of each window is calculated using equation (1), and then the genome is fragmented by BIC-seq software using TRE to obtain a genomic fragment divided by copy number:
  • the mean, variance, and number of windows in the window of all windows in the segment are counted, and the number of windows per segment of the genome is smoothed according to the mean and variance (smooth Processing, the distribution of TRE is more uniform, and then the window distribution of all segments after smoothing is summarized to obtain the distribution result of window change with TRE on the genome; and the mean value of MAF of all HGSNVs in the segment is calculated in units of fragments. variance;
  • Interval C(X t ) represents the number of windows corresponding to the position where TRE is X t ; C(X t+1000 ⁇ P ) represents the number of windows corresponding to the position where TRE is X t+1000 ⁇ P ;
  • Y(P) represents the function value of the auto-regressive model under the variable P;
  • step F calculate the TRE mean of the first actual observed peak in the TRE distribution, and then calculate the maximum number of theoretical peaks N before the first actual peak, and finally there are n before the first actual peak.
  • the value of Q is calculated, denoted by Q n , where step G may include:
  • i represents the ith peak
  • C(X f + P ⁇ i) represents the position where TRE is X f + P ⁇ i, the corresponding number of windows
  • n represents the maximum number of peaks within M t
  • M t represents the maximum value of TRE
  • X f represents the mean of the first peak
  • P represents the spacing between peaks corresponding to adjacent copy number segments
  • floor represents an integer below
  • n represents the number of peaks before X f
  • the value range is an integer between 0 and N
  • P represents the spacing between peaks corresponding to adjacent copy number segments
  • X f represents the TRE mean of the first actually observed peak
  • Q n represents the theoretical existence before X f Q value at n peaks;
  • represents the purity of the sample, and ⁇ represents the ploidy of the chromosome, so that the corresponding ( ⁇ , ⁇ ) can be obtained for all (P, Q N );
  • n represents the number of peaks before X f
  • the value range is an integer between 0 and N
  • P represents the spacing between peaks corresponding to adjacent copy number segments
  • X f represents the first actual observation.
  • the TRE mean of peak T i represents the TRE mean of the ith peak.
  • the fragment For a fragment that falls near T i , the fragment is considered to have a copy number i; for a fragment that does not fall near T i , it is classified as a subcloned fragment, and all subcloned fragments are eliminated in subsequent analysis; Calculated the copy number of the cancer sample purity ⁇ and peak, the expected f b of the MAF of peak can be calculated, the MAF of different peaks is expected to be different, and for all peaks on the genome, the desired set of MAF ⁇ f b ⁇ ; Calculate the TRE mean and variance (or standard deviation) of each peak at the same time;
  • Step J can include the following steps:
  • L(e s ; ⁇ , ⁇ ) represents the likelihood function of the genomic fragment TRE
  • N represents the number of all windows on the genome
  • I represents the maximum copy number of all fragments in the genome
  • ⁇ i represents a copy
  • the standard deviation of the TRE of all segments of number i is obtained by step I
  • e s is the TRE observation value of the sth window
  • S i represents the TRE mean value of the i th peak, that is, T i in step I
  • p i represents the first
  • the copy number of s windows is the weight of i, and the value of all i, p i is 1;
  • L(f s ; ⁇ , ⁇ ) represents the likelihood function of HGSNV
  • M represents the number of all HGSNVs in the genome
  • S represents the Sth HGSNV
  • I represents the maximum copy number of all fragments in the genome
  • i, j represents the expected value of the MAF of the HGSNV in the fragment whose copy number is i, the copy number of the major allele is j, obtained from step I
  • f s represents the mean value of the observed value of the MAF of all HGSNVs in the fragment, obtained from step E
  • ⁇ i,j represents the standard deviation of the MAF observations of all HGSNVs in the segment, obtained from step E
  • p i,j represents the weight of the Gaussian distribution when the copy number of the primary allele is j, for all i And j, p i, j have a value of 1
  • p i represents the weight of the copy of the segment where the S H HSNSNV is i, and the value of all
  • BIC(e s , f s ; ⁇ , ⁇ ) represents the likelihood function of the mixed model
  • I represents the maximum copy number of all the fragments in the genome
  • J is the value of j in the formula (18).
  • Number, N is the number of windows in the genome, and M is the number of HGSNVs in the genome.
  • Q n can be obtained by step G, or the desired set of MAFs of all peaks ⁇ f b ⁇ can be obtained by step I, and a pair (P, ⁇ f b ⁇ ) It is possible to construct a model shown in equation (19), essentially for each pair (P, Q n ), to construct a model as shown in equation (19);
  • step H the corresponding cancer sample purity and chromosome ploidy can be found under (P, Q) obtained in step K.
  • the reference genome hs37d5 (ftp://ftp) used in the phase 3 project of the 1000 genome project is adopted.
  • the comparison software uses Burrows-Wheeler Aligner (BWA), and the comparison method uses bwa mem, and finally obtains the bam format file of the comparison result of cancer and normal samples.
  • step B samtools software is used to extract the position and length information of the read, the HGSNV site and the read covering the site. Quantity information.
  • the read information is extracted using the samtools view command, the sequence whose sequence alignment quality (MAPQ) is lower than 31 is filtered out (parameter -q 31, q indicates that the sequence with poor sequencing quality is filtered out), and the read that fails to match correctly is filtered out.
  • the parameter -f 0x2 -F 0x18, f indicates that the sequence that meets certain requirements is extracted, and F indicates that the sequence that meets certain requirements is filtered.
  • the method of the present invention collects in advance the 1000 genome (genome) program (http://www.internationalgenome.org/), the heterozygous allele locus based on a large number of samples, and filters out the B-allele frequency ( B-allele frequence) is less than 0.05 and is then made into a bed file.
  • the use of the "-1" parameter greatly accelerates the extraction speed of the HGSNV site and improves the operating efficiency of the device on the basis of ensuring sufficient HGSNV sites.
  • the step C may include four steps:
  • the whole genome is divided according to a window of a certain base length, and the number of reads of the window is covered for each window statistics, and the position of the read is represented by the midpoint of each read in the statistics;
  • the present invention creates a GC content index file for the reference genome.
  • G guanine
  • C cytosine
  • the present invention stores the index file in a binary format, which greatly speeds up the extraction of the GC content of a specific region.
  • the present invention uses the respective window GC contents extracted in steps C1 and C2 to fit the read through the following elastic network model.
  • the amount varies with GC content.
  • the invention uses the GC content of the window as the variable x, uses x, x 2 , x 3 , x 4 , x 5 , x 6 as the input variables of the elastic network model, and uses the number of reads as the output variable to construct an elastic network model such as a formula ( 20) shown.
  • y represents the number of reads observed in the window
  • X represents the input variable matrix
  • represents the variable coefficient matrix
  • j represents the variable coefficient subscript
  • P represents the total number of coefficients
  • ⁇ 1 and ⁇ 2 represent the penalty coefficients.
  • the model in step C3 is used to predict the theoretical read number ⁇ gc of each window, and the average GC content of the genome.
  • the number of reads observed in the window
  • y the number of corrected reads in the window
  • the present invention calculates the TRE value of each window using the formula (1).
  • the whole genome was then segmented using the BIC-seq software using the value of TRE.
  • the idea of BIC-seq is to use the Bayesian Information Criterion (BIC) algorithm to count the BIC values of adjacent windows. The smaller the value, the more similar the two windows are, and then merge the windows with BIC values less than 0. Finally, BIC-seq will follow the fragment.
  • BIC-seq The difference in copy number divides the whole genome into different segments. Each segment has a different TRE mean value than the adjacent segment, that is, there is a difference in copy number.
  • the number of windows included in the fragment is calculated by using the genomic fragment after the BIC-seq processing in the step D, The mean and variance of the TRE.
  • the TRE of the segment is then subjected to smooth processing.
  • the processing method is as shown in formula (22). For each genomic fragment, the mean value of TRE is taken as the mean ⁇ of the normal distribution, and the variance of TRE is taken as the variance ⁇ of the normal distribution, and the distribution of the window number of TRE in the range of [ ⁇ -2 ⁇ , ⁇ +2 ⁇ ] is calculated.
  • v is the TRE coordinate
  • the value range is [ ⁇ -2 ⁇ , ⁇ +2 ⁇ ]
  • the resolution is 0.000
  • C win is the number of windows allocated to the v-site
  • C T is the total number of windows in the segment.
  • step F traversing all Ps in the range of (0, 1) with a resolution of 0.001, using an autoregressive model.
  • Y(P) appears as a multimodal distribution, similar to that shown in Figure 3, where the horizontal axis is P and the vertical axis represents Y(P), and the present invention uses the second peak Y (P)
  • the maximum value corresponds to P as the calculation result of P, and M t is the maximum value of TRE, where M t is set to 3.
  • the step G includes three steps, and in the step G1, the TRE interval of [0, 1] is used as the variable X f .
  • the TRE interval of [0, 1] is used as the variable X f .
  • Step I according to the copy number of the cancer sample purity ⁇ and peak calculated according to the step H, the expectation of the MAF of the peak can be calculated.
  • Step I can include three steps of I1, I2, and I3.
  • Equation (15) m is the mean of the number of reads in all windows in the fragment, and v is the variance of the number of reads in all windows in the fragment.
  • the obtained p is the probability of success of the random variable used for the negative binomial distribution, r For the number of times a random variable fails, the random variable is the number of reads in a certain HGSNV.
  • HGSNV has only two genotypes, subject to the binomial distribution law, and uses equation (16) to calculate the correction value f b of f (ie, the expectation of f).
  • f b the correction value of the MAF observed as the peak of the peak.
  • k represents the number of alleles (A or B) at a certain HGSNV site
  • d is the number of reads covering the HGSNV
  • r is the number of failed random variables
  • p is used for The probability of success of a random variable with a negative binomial distribution
  • m is 0.02
  • the traversal interval of the P value is [P-0.02, P+0.02].
  • the hierarchical mixed Gaussian model provided by the invention, the rapid and accurate calculation of the purity of the cancer sample is realized, the time and economic cost of the purity estimation are saved, and the accuracy of the calculation result is improved.
  • Figure 1 shows the distribution of the number of windows in the whole genome on the TRE.
  • Figure A shows the TRE distribution that has not undergone GC correction
  • Figure B shows the TRE distribution after GC content correction.
  • Fig. 2 shows a model of TRE distribution in a cancer cell.
  • the figure shows that after the smooth treatment, the peak in the figure satisfies the distribution with a period of P, and a small number of small peaks which do not satisfy the periodic distribution are considered to be subcloned fragments.
  • Q represents a peak with a copy number of 2, and there is no segment with a copy number of 1, so the number of windows of the peak at a position of about 0.6 is zero.
  • Figure 3 shows that the horizontal axis is the vertical axis of the P-class autoregressive model.
  • Figure 4 shows a flow chart of the method and apparatus of the present invention.
  • FIG. 1 A flow chart for calculating cancer sample purity and chromosome ploidy using the apparatus of the present invention is shown in FIG.
  • the experimental material used was the normal tissue TCGA-AD-A5EJ-10A and the cancer tissue TCGA-AD of the sample (TCGA-AD-A5EJ) downloaded from the TCGA (https://cancergenome.nih.gov/) database.
  • TCGA-AD-A5EJ cancer tissue TCGA-AD of the sample
  • TCGA-AD-A5EJ cancer tissue TCGA-AD of the sample
  • TCGA-AD-A5EJ whole genome sequencing data.
  • the computing platform is ubuntu 16.04, and the specific implementation of the method is C++, Python, and R programs.
  • EXAMPLES The purity and ploidy of cancer samples were calculated using a hierarchical mixed Gaussian model based on genome-wide sequencing data for cancer tissues and normal tissues of sample TCGA-CM-4746.
  • Each column of data represents the positional information of a genomic fragment and the mean value of the TRE, the variance and the number of windows in the fragment.
  • Chromosome number Start termination TRE mean TRE variance Number of Windows Chr1 13001 45265500 0.944508 0.0014742 86128 Chr1 45265501 85978000 0.945454 0.00133201 80970 Chr1 85978001 86011500 1.27362 0.0981321 68 Chr1 86011501 116069000 0.94915 0.00153058 58775
  • step 6 the mean and variance of the TRE of each segment are obtained, and the number of windows included in the segment.
  • the TRE mean and variance of each segment are used as the mean and variance of the normal distribution, and the windows in the segment are smoothed according to the normal distribution. Summarize the TRE of all clips and the corresponding number of windows.
  • step 10 show that when P is 0.382, the mixed model takes the maximum value, and the Q at this time is 0.948. According to this, the purity of the cancer sample can be calculated to be 0.80, and the chromosome ploidy of the cancer cell is 2.14.

Landscapes

  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Organic Chemistry (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Proteomics, Peptides & Aminoacids (AREA)
  • Biotechnology (AREA)
  • Theoretical Computer Science (AREA)
  • Zoology (AREA)
  • Evolutionary Biology (AREA)
  • Analytical Chemistry (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Medical Informatics (AREA)
  • Wood Science & Technology (AREA)
  • Molecular Biology (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Microbiology (AREA)
  • Immunology (AREA)
  • Biochemistry (AREA)
  • General Engineering & Computer Science (AREA)
  • Genetics & Genomics (AREA)
  • Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种计算癌症样本纯度和染色体倍性的方法和装置。通过提供的层次混合高斯模型,实现了对癌症样本纯度和染色体倍性的快速和准确计算,节约了纯度估算的时间和经济成本,同时提高了计算结果的准确性,在癌症样本纯度和染色体倍性计算上具有广阔的运用前景。

Description

用于计算癌症样本纯度和染色体倍性的方法和装置 技术领域
本发明属于癌症研究领域,具体涉及用于计算癌症样本中癌症细胞纯度和细胞内染色体倍性的方法和装置。
背景技术
癌症的研究是生命医学中的重要研究领域,并对人类健康生活有重大影响。癌症是一类细胞恶性增殖的疾病,因其病理十分复杂,人类尚无法攻克这类疾病。二代测序(next generation sequencing)为快速检测病人遗传信息提供了可能。然而测序需要从病人组织中提取样本,但通常癌症组织并不是只单纯地包含癌症细胞,它还有非常丰富的微环境。癌症细胞微环境指包围或伴随癌症细胞的正常细胞(non-cancerous cells)环境。癌细胞样本提取时,这些微环境会和癌细胞一起被提取,并会伴随癌细胞一起被测序[1]。癌症细胞在癌症样本中的比例被定义为癌症样本的纯度。癌症基因组通常包含着大量体细胞序列拷贝数变异,这些变异主要由基因组片段扩增或删除造成。识别特定肿瘤基因组的基因组片段拷贝数变化,是癌症基因组研究的一个重要课题。要准确鉴定基因组片段拷贝数具有一定挑战,因为癌症片段拷贝数主要由两个因素混合决定,一是癌症样本纯度,即癌症细胞在癌症样本中所占比例,二是染色体倍性[2,3]。传统中鉴定癌症样本纯度和染色体倍性的方法是使用实验技术,如定量图像分析[4]或单细胞测序[5]。但是在大型项目中,这样的方法会耗费大量人力、资金和时间。随着测序技术地发展,测序数据地快速增长,以及测序数据分析技术的积累,各种各样的癌症样本纯度算法被提出,并开发出了相对应的软件。
基于基因组片段拷贝数变异和基于等位基因频率(突变位点的B-等位基因(B-allele))的计算方法被相继提出。基于等位基因频率的方法有PurityEst[6]和PurBayes[7],主要是依赖于随着肿瘤样本纯度和肿瘤基因组倍性的不同,等位基因的频率会有所不同。基于拷贝数变异的方法有CNAnorm[8]、THetA[9]、和ABSOLUTE[10]等。然而这两种方法都有不同程度的问题,使用等位基因频率的方法由于数据量的问题会有较大的误差,而运用拷贝数变异的方法虽然较稳定,但却无法区分样本纯度和染色体倍性的补偿效应,即存在识别问题。以上基于片段拷贝数的软件都没有解决这一问题,CNAnorm倾向于选择染色体倍性离二倍体最近的解决方案,ABSOLUTE结合了其他的经验数据,THetA则直接将所有可能结果都列出来了。
更优的方案应该是结合等位基因频率信息和片段拷贝数信息共同计算肿瘤样本纯度。PyLOH[11],patchwork[12]使用了基因组上杂合SNV(单核苷酸变异)位点的频率信息,和基因组片段的拷贝数。PyLOH一定程度上解决了“识别困难”的问题,可以更合理给出 唯一解决方案。但是其准确性较差,特别是遇到基因组中存在亚克隆(subclone)的情况下。Patchwork同时使用了两种信息,但是在计算基因型的中间步骤中,需要人工识别,人工判断的结果缺乏准确性,并且这种半自动化软件给应用带来很多不便。
如何充分利用现有的二代测序数据准确计算癌症样本纯度和癌症细胞基因组倍性问题仍然是一项具有挑战性的工作。
参考文献
1、Junttila M R,de Sauvage F J.Influence of tumour micro-environment heterogeneity on therapeutic response[J].Nature,2013,501(7467):346-354.
2、Carter S L,Cibulskis K,Helman E,et al.Absolute quantification of somatic DNA alterations in human cancer[J].Nature Biotechnology,2012,30(5):413-21.
3、Oesper L,Mahmoody A,Raphael B J.Inferring intra-tumor heterogeneity from high-throughput DNA sequencing data[J].Genome Biology,2013,14(7):R80.
4、Yuan Y,Failmezger H,Rueda O M,et al.Quantitative Image Analysis of Cellular Heterogeneity in Breast Tumors Complements Genomic Profiling[J].Science Translational Medicine,2012,4(157):157ra143.
5、Navin N,Kendall J,Troge J,et al.Tumour evolution inferred by single-cell sequencing[J].Nature,2011,472(7341):90-4.
6、Su X,Zhang L,Zhang J,et al.PurityEst:estimating purity of human tumor samples using next-generation sequencing data.[J].Bioinformatics,2012,28(17):2265-2266.
7、Larson N B.PurBayes:estimating tumor cellularity and subclonality in next-generation sequencing data[J].Bioinformatics,2013,29(15):1888-9.
8、Gusnanto A,Wood H M,Pawitan Y,et al.Correcting for cancer genome size and tumour cell content enables better estimation of copy number alterations from next-generation sequence data[J].Bioinformatics,2012,28(1):40-47.
9、Oesper L,Mahmoody A,Raphael B J.Inferring intra-tumor heterogeneity from high-throughput DNA sequencing data[J].Genome Biology,2013,14(7):R80.
10、Carter S L,Cibulskis K,Helman E,et al.Absolute quantification of somatic DNA alterations in human cancer[J].Nature Biotechnology,2012,30(5):413-21.
11、Li Y,Xie X.Deconvolving tumor purity and ploidy by integrating copy number alterations and loss of heterozygosity[J].Bioinformatics,2015,30(4):2121.
12、Mayrhofer M,Dilorenzo S,Isaksson A.Patchwork:allele-specific copy number analysis of whole-genome sequenced tumor tissue[J].Genome Biology,2013,14(3):R24.
发明内容
术语定义
为了更好地理解本发明,下面提供相关的解释和说明:
Whole Genome Sequencing(WGS):使用二代测序技术的全基因组测序。
read:高通量测序平台产生的测序序列。
测序深度:测序得到的碱基(bp)总量与基因组(Genome)大小的比值,它是评价测序量的指标之一。
window(窗口):按照一定长度划分的基因组片段,该长度代表window大小。本方法中window大小可由使用者自由设置,通常设置为几百碱基。一个大基因组片段S可以包含大量window。
Tumor Read Enrichment(TRE):癌症片段读长富集程度e s,指癌症样本中某一片段S内read数量与相应正常样本中对应片段read数量的比值,定义公式如下:
Figure PCTCN2018078908-appb-000001
公式(1)中,
Figure PCTCN2018078908-appb-000002
Figure PCTCN2018078908-appb-000003
分别表示在癌症样本中覆盖片段s的read数量和相匹配的正常样本中覆盖片段s的read数量,N t表示癌症样本全基因组测序获得read总数量,N n表示相应正常样本全基因组测序获得read总数量。
Heterozygous Germline Single Nucleotide Variants(HGSNV):杂合生殖系细胞单碱基变异,由于人类染色体属于二倍体,体细胞均由胚胎细胞发育而来,而生殖细胞中HGSNV位点只有两种碱基类型A和B,其中一种来源于父本,另一种来源于母本。
Major Allele Fraction(MAF):主要等位基因(allele)分数,本发明中使用的HGSNV只有两种等位基因,一种等位基因与参考基因组相同,另一种与参考基因组不同。这两种等位基因型分数的计算方法为覆盖某一等位基因的read数量除以覆盖该位点总read数量的比值,MAF就是两种等位基因分数中的较大值。计算公式如(1.1)所示,n r为包含与参考基因组相同等位基因的read数量,n a为包含另一种等位基因的read的数量,n t表示覆盖该HGSNV位点的总read数量,C为该HGSNV的MAF值。MAF是相对于HGSNV的概念,本发明中“片段的MAF”指片段内所有HGSNV的MAF均值,“peak的MAF”指peak中所有片段包含的HGSNV的MAF均值。
Figure PCTCN2018078908-appb-000004
major allele copy number:主要等位基因拷贝数,指在拷贝数为i的片段中,主要等位基因拷贝数的取值,它的取值范围为大于等于
Figure PCTCN2018078908-appb-000005
的整数。
peak:指基因组所有window的TRE分布中,聚集在一起的TRE簇。如附图1所示,图A表示基因组上所有window的TRE分布,纵轴表示对应某TRE位点的window 总数量,该图为基因组GC含量校正之前的TRE分布,图B表示GC含量校正之后的TRE分布,图B中可以看到window明显以簇聚集,本方法将通过类自回归模型鉴定出来的TRE簇定义为peak,本质上是具有相同拷贝的基因组片段内window的聚集。
癌症样本:指从某患癌症的个体身上取下的癌症组织,它包含了一部分癌症细胞和一部分正常细胞。
P:指两个相邻peak之间的间距,由于peak是一个簇,这里peak的TRE由peak的TRE均值来表示,所以实际上是两个相邻peak的TRE均值的差。由于peak是具有相同拷贝数基因组片段内window的聚集,所以这里也表述为相邻拷贝数片段TRE的差值。
发明目的
本发明的目的在于克服现有技术的缺陷,提供一种全自动、高效率、高准确性的计算癌症样本纯度和染色体倍性的方法和装置。本发明在癌症样本纯度和染色体倍性计算上具有广阔的运用前景。
技术方案
为实现上述发明目的,本发明采取的技术方案为:通过癌症样本和匹配的正常样本的全基因组测序数据,对不同拷贝数片段的TRE和HGSNV的MAF分布构建混合高斯模型,计算癌症样本纯度和染色体倍性。
本发明主要运用了全基因组测序数据的TRE信息和HGSNV的MAF信息。TRE基本反映了癌症样本拷贝数变异情况,HGSNV的MAF信息基本反映了癌症样本的基因型。
TRE的差别主要来源于基因组片段的拷贝数差异,高拷贝数基因组片段内测序获得的read数量一定大于低拷贝数基因组片段测序获得的read数量,通过片段内read数量差异计算片段拷贝数差异是基因组拷贝数检测中的常用方法。但是大多数研究中,直接使用癌症样本片段内read数量除以正常样本该片段内read数量的比值(ratio)来进行read数量差异评估。本发明使用公式(1)所示的TRE来评估片不同片段的read数量差异。传统方法计算所得ratio不仅受癌症样本纯度与染色体倍性的影响,还受到癌症样本和正常样本测序深度的影响,而TRE不会受到样本测序深度的影响。
单独依赖read数量差异,无法确定各拷贝数片段的基因型,更重要的是无法区分样本纯度与样本倍性的补偿效应。而结合拷贝数差异片段内的HGSNV可以提供基因型信息,并帮助解决纯度与倍性的补偿效应,然而在前人的研究中,并没有高效利用HGSNV的方法,大部分方法采用枚举的方式将不同拷贝数片段可能对应的基因型一一列出,然后对排列组合的结果进行计算,挑选最可信的结果。这些方法共同的特点是,方法计算时间长,准确性差,对拷贝数很高或基因组变化较大的样本效果很差。本发明依据HGSNV的MAF与TRE的混合高斯模型计算癌症样本纯度和染色体倍性,能显著减少计算时间,并提高计算结果准确率。
假设某癌症样本的纯度为γ,那么癌症样本中的正常细胞比例为1-γ。癌症样本中正常细胞的染色体倍性为2,癌症细胞的染色体倍性为κ。那么癌症样本的染色体倍性ω如下公式(2)所示。
ω=(1-γ)×2+γ×κ   (2)
假设在癌症细胞中某一片段S的拷贝数为C S。那么癌症样本的片段S的拷贝数C t应该为如下公式(3)所示。
C t=(1-γ)×2+γ×C s   (3)
对于基因组片段S,TRE的计算方式如公式(1)所示。而TRE的期望值(expectation)E(e s)的推导公式如下公式(4)所示,式中的
Figure PCTCN2018078908-appb-000006
N n和N t与公式(1)中含义相同。
Figure PCTCN2018078908-appb-000007
为了更进一步引出e s,本方法定义了一些帮助理解的参数。片段S的长度L S,人类参考基因组的长度L gw,癌症样本的测序深度
Figure PCTCN2018078908-appb-000008
正常样本的测序深度
Figure PCTCN2018078908-appb-000009
那么片段S在癌症样本中测序深度为
Figure PCTCN2018078908-appb-000010
片段S在正常样本中测序深度为
Figure PCTCN2018078908-appb-000011
λ S是指与片段S特性(如GC含量等引起测序深度偏好的特性)有关的参数,所以在癌症和正常样本中是一样的。进一步通过γ,κ,C s来表示e s,如公式(5)所示。
Figure PCTCN2018078908-appb-000012
公式(5)中C s表示在癌症细胞中片段S的拷贝数,那么当片段S的拷贝数为i时和i+1时对应的TRE均值S i和S i+1分别如公式(6)和公式(7)所示:
Figure PCTCN2018078908-appb-000013
Figure PCTCN2018078908-appb-000014
通过公式(6)和公式(7),对于相邻的拷贝数对应的片段,它们的TRE的差值P如公式(8)所示,可见P值的大小与片段具体拷贝数没有关系,它只决定于癌症样本纯度和染色体倍性。通过附图2可以直观的看到在TRE分布图中,peak之间的距离是恒定的。
Figure PCTCN2018078908-appb-000015
此外,对于i=2即拷贝数为2的片段,它们的TRE值Q如公式9所示。附图2中Q对应的peak的TRE值略大于1。
Figure PCTCN2018078908-appb-000016
通过上述公式(8)和(9),可以解得癌症样本的纯度(γ)和染色体倍性(κ)分别为:
Figure PCTCN2018078908-appb-000017
Figure PCTCN2018078908-appb-000018
通过以上分析可以得知,通过确定P和Q可以计算出癌症样本纯度γ和染色体倍性κ。
如附图2所示,计算出全基因组所有片段的TRE分布后,可以计算peak间的间距得到P。在前人的研究方法中,patchwork[12]使用相邻拷贝数片段的对应的read数量的比值的间距来辅助计算癌症样本纯度,但是该研究无法自动识别read数量比值之间的间距,需要人工识别图像确定read数量比值间距来进行下一步计算,效率和准确性都比较低。本发明开创性的使用类自回归模型鉴定TRE之间的间距,如公式(12)、(13)所示。公式(12)和(13)中,X t表示0到M t之间的TRE值;t表示扩大了1000倍的TRE值;M t表示TRE的最大值;P表示两个TRE位点的间隔;C(X t)表示在TRE为X t的位点,对应的window数量;C(X t+1000×P)表示在TRE为X t+1000×P的位点,对应的window数量;Y(P)表示在P下,类自回归模型的函数值;显而易见当P=0时,Y(P)的取值最大,但这时候的P并不是实际peak之间的间距。
Figure PCTCN2018078908-appb-000019
Figure PCTCN2018078908-appb-000020
以0.001为分辨率,遍历0到1之间的所有P值,然后求Y(P)。Y(P)的值分布如附图3所示。根据公式(13)的特点,我们可以知道,当P等于0时,Y(P)的值会是最大的,但此时的P并不是peak之间的间距。我们选择图中第二高峰中Y(P)的最大值对应的x轴坐标值P作为peak之间的间距P的计算结果。
如图1中B图所示的peak之所以簇状分布,是因为具有相同拷贝数的基因组片段的TRE值(指片段内所有window的TRE的均值)并不完全相等,同拷贝数片段TRE相互之间存在误差。该误差服从高斯分布,所以图B中的簇状分布被认为是高斯分布。
如附图2所示,P确定以后,peak会被识别出来,但是有部分基因组片段没有落在识别出的peak上,这些片段被称作亚克隆片段(subclone segmentation)。在考虑亚克隆片段的情况下,会对后面公式(17)和(18)所示的高斯模型取值有影响,进而会影响最终混合高斯模型的取值。由于在后续的分析中,本发明只需考虑落在peak位点的片段,由此排除了亚克隆片段的干扰。
在如图2所示的TRE分布图中,Q位点表示拷贝数为2的片段对应的TRE值。首先 我们可以推测,如果癌症细胞基因组中,存在部分片段的拷贝数为1,部分片段的拷贝数为0,那么在Q位点之前应该存在两个peak分别对应拷贝数1和0。如果不存在拷贝数为1的片段,只存在拷贝数为0的片段,那么在Q位点之前距离2P的位点存在一个peak,而在Q位点之前距离P的位点peak的window数为0,这也就是图2所示的情形。另一种情况假若拷贝数为1和0的片段都没有,那么在Q位点之前距离P和距离2P的位点对应的window数都为0。那么对于TRE的分布图,对于X f,即第一个出现的peak,它可能对应的拷贝数为2(拷贝数1和0的peak对应的window就是0),也可能对应的拷贝数是1(拷贝数为0的片段对应peak的window为0),也有可能对应的拷贝数是0。
通过上述分析,我们可以知道,图2中的第一个peak即X f对应的片段的拷贝数有几种不同的可能,而每一种可能都会使Q对应不同的peak。本发明通过混合高斯模型计算出了最可能的X f对应的片段的拷贝数,从而确定了Q的取值,最终得到了癌症样本纯度和染色体倍性。首先我们需要鉴定出X f对应片段的拷贝数的所有可能值。本发明通过如下公式(13.1)来确定X f的值。(13.1)中,C(X f+P)表示在TRE为X f+P的位点,对应的window数量,n表示M t以内peak的最大数量。当f(X f)取最大值时X f为第一个peak的TRE均值。
Figure PCTCN2018078908-appb-000021
然后使用公式(13.2)求X f之前最多可能有几个peak。其中X f表示第一个peak的TRE均值,P表示相邻拷贝数片段对应的peak之间的间距,floor表示向下取整数,当N=0时,表示X f之前没有peak,X f对应的片段拷贝数为0;当N=1时,表示X f之前最多可能有1个peak,也可能没有peak;当N=2时,表示X f之前最多可能有2个peak,也可能只有一个peak或者没有peak;
Figure PCTCN2018078908-appb-000022
对于X f之前可能有(1,2,3...N)个peak的情形,每一种情形下都可以通过如下公式(13.3)计算出一个对应的Q值。根据Q的定义,我们知道Q为拷贝数为2的片段的peak对应的TRE值。首先可以推断拷贝数为0的片段对应的TRE值为X f-n×P,其中n表示X f之前peak的个数,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f含义与公式(13.1)中相同,公式(13.3)如下所示,其中Q n表示在X f之前存在n个peak时,Q的取值。
Q n=X f-n×P+2×P=X f+(2-n)×P,n∈[0,N]  (13.3)
根据以上分析,可以得到对于X f之前可能有(0,1,2,3...N)个peak的情形时,Q n取值可能为(Q 0,Q 1,Q 2,Q 3...Q N)。而前面的类自回归模型已经计算出了P,那么对于每一个 可能的Q,我们可以通过公式(10)和(11)计算出相应的γ和κ。本发明通过混合高斯模型计算出X f之前最可能的peak数量n,从而确定了X f对应的片段的拷贝数,进而确定了Q的取值,最终得到了癌症样本纯度和染色体倍性。具体方法如下说明。
对于Q n的每一个可能的取值,结合P,我们可以计算得到相应的γ,然后计算各个拷贝数片段内HGSNV的MAF的理论值。其理论计算方式如公式(14)所示。C mcp表示主要等位基因的拷贝数(major allele copy number),C cp表示peak的整体拷贝数。f表示该peak内MAF的理论值。
Figure PCTCN2018078908-appb-000023
但是在实际情况下当测序深度比较低时,f与真实值(期望值)会有较大的误差。这里需要进一步校正f,校正方法如公式(15)、(16)所示。式(15)中m是某peak内所有window中read数量的均值,v是peak内所有window中read数量的方差,所求得的p是用于负二项分布的随机变量成功的概率,r为随机变量失败的次数,这里的随机变量d为测序获取到的测序深度(read coverage)。
Figure PCTCN2018078908-appb-000024
在某一个测序深度即d下,MAF实际上是服从以f为概率、d为实验次数的二项分布。本发明以如下公式(16)对f进行校正,得到MAF的期望值f b。公式中k表示在某个HGSNV位点,某一种等位基因(A或B)的数量,测得的等位基因总量为d(与测序深度相等)。
Figure PCTCN2018078908-appb-000025
公式(13.2)表明每个基因组片段的拷贝数有N种可能。公式(14)表明,某一拷贝数片段可以有多种主要等位基因拷贝数,于是对每个基因组peak,可以算出多个f,也可以算出多个f b,取距离peak实际观测MAF均值最近的f b作为该peak的MAF期望。而全基因组有多个peak,每个peak的MAF期望不同,对应算出多个MAF期望值{f b}。考虑到某一peak内HGSNV的MAF会存在一定误差但也近似服从高斯分布,peak内所有HGSNV的MAF期望值可以由实际数据直接计算获得。假设某peak的基因型一定,那么通过比较peak的MAF观测值与peak的{f b}的值就可以判断该peak的拷贝数和基因型。也就可以计算出拷贝数为2的peak对应的TRE值即Q的位置。同时也为了进一步校正P,本发明提出了一种混合高斯模型对TRE和HGSNV的观测数据进行拟合。
通过前面的分析可以得知因为公式(12)中ε t的存在,X t并不能十分准确的代表各个peak的TRE均值。TRE的高斯分布模型如公式17所示:
Figure PCTCN2018078908-appb-000026
其中,L(e s;γ,κ)表示基因组片段TRE的似然函数。N表示基因组上的所有window的数量。I表示基因组中所有片段的最大的拷贝数。σ i表示拷贝数为i的所有片段的TRE的标准差。e s为第s个window的TRE观测值,S i表示第i个peak的TRE均值。p i表示第s个window的拷贝数为i的权重,本公式中对所有的i,p i均取值为1。该公式表明似然函数的大小与S i的取值相关,当S i与e S越接近,似然函数的取值越大,同时也表示P值越接近真实值。使用L(e s;γ,κ)的极大似然估计可以计算出较合理的P值。
然而在部分情况下,P在较小的区间内波动时(如[-0.005,0.005]),对应的似然函数值可能相同。本发明通过结合HGSNV的高斯分布模型如公式(18)所示,来更近一步确定P值,同时确定Q值。
Figure PCTCN2018078908-appb-000027
其中,L(f s;γ,κ)表示HGSNV的似然函数。M表示基因组中所有HGSNV。S表示第S个HGSNV。I表示基因组中所有片段的最大的拷贝数。F i,j为拷贝数为i、主要等位基因的拷贝数为j的片段内HGSNV的MAF期望值,即公式(16)计算出的f b。f s表示该片段内MAF的观测值均值,σ i,j表示该片段内HGSNV的MAF观测值的标准差。p i,j表示在主要等位基因的拷贝数为j时,高斯分布的权重,对所有的i和j,p i,j取值均为1。p i表示第S个HGSNV所在片段的拷贝数为i的权重,对所有的i,p i取值均为1。公式(18)表明,似然函数的大小与F i,j值相关,当F i,j与f s越接近,似然函数越大,表明f s越准确,同时表明公式(14)中的f越准确,从而可以得到各片段对应的C cp与C mcp。于是确定了Q的取值。为了得到最准确的P与Q,本方法将公式(17)和公式(18)相加,得到混合高斯模型。
但对该混合模型统计计算容易发生模型过拟合现象。本发明进一步使用了贝叶斯信息准则(Bayesian Information Criterion,BIC)方法,给混合高斯模型一个罚分函数,用于控制模型的过拟合,最终混合高斯模型如公式(19)所示:
BIC(e s,f s;γ,κ)=-2×log L(f s;γ,κ)-2×log L(e s;γ,κ)+I×log(N)+J×log(M)  (19)
其中,BIC(S s,f s;γ,κ)表示混合模型的似然函数,I是公式(17)中高斯分布的个数,J是公式(18)中高斯分布的个数。N是基因组中窗口的数量,M是基因组中HGSNV的个数。
通过遍历[P-0.02,P+0.02]的区间,对所有的(P,Q n)求极大似然估计,可以得到最合适的P值与Q值然后根据公式(10)和(11)可计算癌症样本的纯度和染色体倍性。
因此,本发明一方面提供了一种用于计算癌症样本中癌症细胞纯度和染色体倍性的方法,所述方法包括以下步骤:
步骤A:
获取配对的(来自同一癌症病人的)癌症组织样本和正常组织样本的全基因组测序 (WGS)数据,并将测序数据比对到参考基因组;
步骤B:
从步骤A得到的比对结果文件中,提取read位置和长度信息,HGSNV位点和覆盖该位点的read数量信息,计算所有HGSNV的MAF,其中,计算公式如(1.1)所示:
Figure PCTCN2018078908-appb-000028
公式(1.1)中,n r为包含与参考基因组相同等位基因的read数量,n a为包含另一种等位基因的read的数量,n t表示覆盖该HGSNV位点的总read数量,C为该HGSNV的MAF值;
步骤C:
根据步骤B得到的read位置和长度信息,以window为单位统计各window内包含的read数量,使用基因组GC含量校正所有window内read数量;
步骤D:
使用步骤C校正后的read数量,使用公式(1)计算每一个window的TRE,然后运用TRE,通过BIC-seq软件对基因组进行片段化,获得以拷贝数划分的基因组片段:
Figure PCTCN2018078908-appb-000029
公式(1)中,
Figure PCTCN2018078908-appb-000030
Figure PCTCN2018078908-appb-000031
分别表示在癌症样本中覆盖片段s(这里指window)的read数量和在正常样本中覆盖片段s的read数量,N t表示癌症样本总read数量,N n表示相应正常样本总read数量,e s为TRE值;
步骤E:
以步骤D中BIC-seq处理后的基因组片段为单位,统计片段内所有window的TRE的均值、方差和该片段内window数量,根据均值和方差对基因组每个片段的window数量进行平滑化(smooth)处理,使TRE的分布更均匀,然后将平滑化处理后所有片段的window分布汇总,得到基因组上window随TRE变化的分布结果;同时以片段为单位,计算片段中所有HGSNV的MAF的均值和方差;
步骤F:
使用如公式(12)、(13)所示的类自回归模型,计算相邻拷贝数片段内TRE的差值即P,具体方法为遍历一定范围的P,计算Y(P),在Y(P)的分布中,选择第二高峰内Y(P)的最大值对应的P作为P的计算结果:
Figure PCTCN2018078908-appb-000032
Figure PCTCN2018078908-appb-000033
公式(12)和(13)中,X t表示0到M t之间的TRE值;t表示扩大了1000倍的TRE值;M t表示TRE的最大值;变量P表示两个TRE位点的间隔;C(X t)表示在TRE为X t的位点,对应的window数量;C(X t+1000×P)表示在TRE为X t+1000×P的位点,对应的window数量;Y(P)表示在变量P下,类自回归模型的函数值;
步骤G:
根据步骤F得到的P,计算TRE分布中第一个实际观测peak的TRE均值,然后计算在第一个实际peak之前最多可能存在理论peak的数量N,最后当第一个实际peak之前存在n个理论peak时,计算Q的值,以Q n表示,其中步骤G可以包括:
G1:
根据步骤F计算的P,使用公式(13.1),选取使公式(13.1)取最大值的X f作为第一个实际观测peak的TRE均值:
Figure PCTCN2018078908-appb-000034
公式(13.1)中,i表示第i个peak,C(X f+P×i)表示在TRE为X f+P×i的位点,对应的window数量,n表示M t以内peak的最大数量,M t表示TRE的最大值;
G2:
使用公式(13.2),根据步骤F计算的P和步骤G1计算的X f,计算在X f之前最多可能存在的peak数量N:
Figure PCTCN2018078908-appb-000035
公式(13.2)中,X f表示第一个peak的均值,P表示相邻拷贝数片段对应的peak之间的间距,floor表示向下取整数;
G3:
利用步骤G2计算的N值,当n取0到N之间的整数时,使用公式(13.3)计算Q n的值:
Q n=X f-n×P+2×P=X f+(2-n)×P,n∈[0,N]  (13.3)
公式(13.3)中,n表示X f之前peak的数量,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f表示第一个实际观测peak的TRE均值,Q n表示在X f之前理论上存在n个peak时的Q值;
步骤H:
使用步骤F计算的P与步骤G计算的所有可能的Q n,使用公式(10)、(11)计算癌 症样本纯度γ和染色体倍性κ:
Figure PCTCN2018078908-appb-000036
Figure PCTCN2018078908-appb-000037
公式(10)、(11)中,γ表示样本纯度,κ表示染色体倍性,那么对所有的(P,Q N)都可以得到对应的(γ,κ);
步骤I:
当n取[0,N]之间的某个整数值时,使用公式(13.4)计算第i个peak的TRE均值:
T i=X f-n×P+i×P=X f+(i-n)×P,n∈[0,N]  (13.4)
公式(13.4)中,n表示X f之前peak的数量,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f表示第一个实际观测peak的TRE均值,T i表示第i个peak的TRE均值,
对于落在T i附近的片段,认为该片段具有拷贝数i;对于没有落在T i附近的片段,将其归类为亚克隆片段,在后续分析中剔除所有亚克隆片段;然后根据步骤H计算的癌症样本纯度γ和peak对应的拷贝数,可计算peak的MAF的期望f b,不同peak的MAF期望不同,对基因组上的所有peak,最终得到MAF期望的集合{f b};同时计算各个peak的TRE均值和方差(或标准差);
步骤J:
根据步骤F计算的P和步骤I计算的{f b}构建如公式(19)所示的用“贝叶斯信息准则”校正后的混合高斯分布模型,然后对模型极大似然估计;其中,步骤J可以包括如下几步:
J1:
以步骤F计算的P构建如公式(17)所示的高斯分布模型:
Figure PCTCN2018078908-appb-000038
公式(17)中,L(e s;γ,κ)表示基因组片段TRE的似然函数,N表示基因组上的所有window的数量,I表示基因组中所有片段的最大的拷贝数,σ i表示拷贝数为i的所有片段的TRE的标准差由步骤I得到,e s为第s个window的TRE观测值,S i表示第i个peak的TRE均值即步骤I中的T i,p i表示第s个window的拷贝数为i的权重,对所有的i,p i均取值为1;
J2:
以步骤I计算的f b构建如公式(18)所示的高斯分布模型:
Figure PCTCN2018078908-appb-000039
公式(18)中,L(f s;γ,κ)表示HGSNV的似然函数,M表示基因组中所有HGSNV数量,S表示第S个HGSNV,I表示基因组中所有片段的最大的拷贝数;F i,j表示拷贝数为i,主要等位基因的拷贝数为j的片段内HGSNV的MAF期望值,由步骤I得到;f s表示该片段内所有HGSNV的MAF的观测值均值,由步骤E得到,σ i,j表示该片段内所有HGSNV的MAF观测值的标准差,由步骤E得到;p i,j表示在主要等位基因的拷贝数为j时,高斯分布的权重,对所有的i和j,p i,j取值均为1,p i表示第S个HGSNV所在片段的拷贝数为i的权重,对所有的i,p i取值均为1;
J3:
将(17)与(18)相加得到混合高斯模型,然后对混合模型进行BIC(Bayesian Information Criterion)校正得到最终混合模型如公式(19):
BIC(e s,f s;γ,κ)=-2×log L(f s;γ,κ)-2×log L(e s;γ,κ)+I×log(N)+J×log(M)  (19)
公式(19)中,BIC(e s,f s;γ,κ)表示混合模型的似然函数,I表示基因组中所有片段的最大的拷贝数,J是公式(18)中j的取值个数,N是基因组中window的数量,M是基因组中HGSNV的个数,
对[0,N]范围内的每一个整数值n,可以通过步骤G得到Q n,也可以通过步骤I得到所有peak的MAF期望的集合{f b},而一对(P,{f b})可以构建一个公式(19)所示的模型,实质上是对每一对(P,Q n),可以构建一个公式(19)所示的模型;
步骤K:
以0.001为分辨率,对[P-m,P+m]区间的所有P值,重复步骤G~J,可以得到一系列不同的(P,Q n)与对应的似然函数值,取最大的似然函数值对应的(P,Q n)作为最合适的P和Q值,m是0到0.5之间的一个值;
步骤L:
查询步骤H的结果,可以找到在步骤K得到的(P,Q)下,对应的癌症样本纯度和染色体倍性。
另一方面,本发明提供了一种用于计算癌症样本中癌症细胞纯度和染色体倍性的装置,其包括处理器,所述处理器用于运行程序,所述程序运行时执行以下步骤:
步骤A:
获取配对的(来自同一癌症病人的)癌症组织样本和正常组织样本的全基因组测序(WGS)数据,并将测序数据比对到参考基因组;
步骤B:
从步骤A得到的比对结果文件中,提取read位置和长度信息,HGSNV位点和覆盖该位点的read数量信息,计算所有HGSNV的MAF,其中,计算公式如(1.1)所示:
Figure PCTCN2018078908-appb-000040
公式(1.1)中,n r为包含与参考基因组相同等位基因的read数量,n a为包含另一种等位基因的read的数量,n t表示覆盖该HGSNV位点的总read数量,C为该HGSNV的MAF值;
步骤C:
根据步骤B得到的read位置和长度信息,以window为单位统计各window内包含的read数量,使用基因组GC含量校正所有window内read数量;
步骤D:
使用步骤C校正后的read数量,使用公式(1)计算每一个window的TRE,然后运用TRE,通过BIC-seq软件对基因组进行片段化,获得以拷贝数划分的基因组片段:
Figure PCTCN2018078908-appb-000041
公式(1)中,
Figure PCTCN2018078908-appb-000042
Figure PCTCN2018078908-appb-000043
分别表示在癌症样本中覆盖片段s(这里指window)的read数量和在正常样本中覆盖片段s的read数量,N t表示癌症样本总read数量,N n表示相应正常样本总read数量,e s为TRE值;
步骤E:
以步骤D中BIC-seq处理后的基因组片段为单位,统计片段内所有window的TRE的均值、方差和该片段内window数量,根据均值和方差对基因组每个片段的window数量进行平滑化(smooth)处理,使TRE的分布更均匀,然后将平滑化处理后所有片段的window分布汇总,得到基因组上window随TRE变化的分布结果;同时以片段为单位,计算片段中所有HGSNV的MAF的均值和方差;
步骤F:
使用如公式(12)、(13)所示的类自回归模型,计算相邻拷贝数片段内TRE的差值即P,具体方法为遍历一定范围的P,计算Y(P),在Y(P)的分布中,选择第二高峰内Y(P)的最大值对应的P作为P的计算结果:
Figure PCTCN2018078908-appb-000044
Figure PCTCN2018078908-appb-000045
公式(12)和(13)中,X t表示0到M t之间的TRE值;t表示扩大了1000倍的TRE值;M t表示TRE的最大值;变量P表示两个TRE位点的间隔;C(X t)表示在TRE为X t的位点,对应的window数量;C(X t+1000×P)表示在TRE为X t+1000×P的位点,对应的window数量;Y(P)表示在变量P下,类自回归模型的函数值;
步骤G:
根据步骤F得到的P,计算TRE分布中第一个实际观测peak的TRE均值,然后计算在第一个实际peak之前最多可能存在理论peak的数量N,最后当第一个实际peak之前存在n个理论peak时,计算Q的值,以Q n表示,其中步骤G可以包括:
G1:
根据步骤F计算的P,使用公式(13.1),选取使公式(13.1)取最大值的X f作为第一个实际观测peak的TRE均值:
Figure PCTCN2018078908-appb-000046
公式(13.1)中,i表示第i个peak,C(X f+P×i)表示在TRE为X f+P×i的位点,对应的window数量,n表示M t以内peak的最大数量,M t表示TRE的最大值;
G2:
使用公式(13.2),根据步骤F计算的P和步骤G1计算的X f,计算在X f之前最多可能存在的peak数量N:
Figure PCTCN2018078908-appb-000047
公式(13.2)中,X f表示第一个peak的均值,P表示相邻拷贝数片段对应的peak之间的间距,floor表示向下取整数;
G3:
利用步骤G2计算的N值,当n取0到N之间的整数时,使用公式(13.3)计算Q n的值:
Q n=X f-n×P+2×P=X f+(2-n)×P,n∈[0,N] (13.3) 公式(13.3)中,n表示X f之前peak的数量,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f表示第一个实际观测peak的TRE均值,Q n表示在X f之前理论上存在n个peak时的Q值;
步骤H:
使用步骤F计算的P与步骤G计算的所有可能的Q n,使用公式(10)、(11)计算癌症样本纯度γ和染色体倍性κ:
Figure PCTCN2018078908-appb-000048
Figure PCTCN2018078908-appb-000049
公式(10)、(11)中,γ表示样本纯度,κ表示染色体倍性,那么对所有的(P,Q N)都可以得到对应的(γ,κ);
步骤I:
当n取[0,N]之间的某个整数值时,使用公式(13.4)计算第i个peak的TRE均值:
T i=X f-n×P+i×P=X f+(i-n)×P,n∈[0,N]  (13.4)
公式(13.4)中,n表示X f之前peak的数量,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f表示第一个实际观测peak的TRE均值,T i表示第i个peak的TRE均值。
对于落在T i附近的片段,认为该片段具有拷贝数i;对于没有落在T i附近的片段,将其归类为亚克隆片段,在后续分析中剔除所有亚克隆片段;然后根据步骤H计算的癌症样本纯度γ和peak对应的拷贝数,可计算peak的MAF的期望f b,不同peak的MAF期望不同,对基因组上的所有peak,最终得到MAF期望的集合{f b};。同时计算各个peak的TRE均值和方差(或标准差);
步骤J:
根据步骤F计算的P和步骤I计算的{f b}构建如公式(19)所示的用“贝叶斯信息准则”校正后的混合高斯分布模型,然后对模型极大似然估计;其中,步骤J可以包括如下几步:
J1:
以步骤F计算的P构建如公式(17)所示的高斯分布模型:
Figure PCTCN2018078908-appb-000050
公式(17)中,L(e s;γ,κ)表示基因组片段TRE的似然函数,N表示基因组上的所有window的数量,I表示基因组中所有片段的最大的拷贝数,σ i表示拷贝数为i的所有片段的TRE的标准差由步骤I得到,e s为第s个window的TRE观测值,S i表示第i个peak的TRE均值即步骤I中的T i,p i表示第s个window的拷贝数为i的权重,对所有的i,p i均取值为1;
J2:
以步骤I计算的f b构建如公式(18)所示的高斯分布模型:
Figure PCTCN2018078908-appb-000051
公式(18)中,L(f s;γ,κ)表示HGSNV的似然函数,M表示基因组中所有HGSNV数量,S表示第S个HGSNV,I表示基因组中所有片段的最大的拷贝数;F i,j表示拷贝数为i,主要等位基因的拷贝数为j的片段内HGSNV的MAF期望值,由步骤I得到;f s表示该片段内所有HGSNV的MAF的观测值均值,由步骤E得到,σ i,j表示该片段内所有HGSNV的MAF观测值的标准差,由步骤E得到;p i,j表示在主要等位基因的拷贝数为j时,高斯分布的权重,对所有的i和j,p i,j取值均为1,p i表示第S个HGSNV所在片段 的拷贝数为i的权重,对所有的i,p i取值均为1;
J3:
将(17)与(18)相加得到混合高斯模型,然后对混合模型进行BIC(Bayesian Information Criterion)校正得到最终混合模型如公式(19):
BIC(e s,f s;γ,κ)=-2×log L(f s;γ,κ)-2×log L(e s;γ,κ)+I×log(N)+J×log(M) (19)
公式(19)中,BIC(e s,f s;γ,κ)表示混合模型的似然函数,I表示基因组中所有片段的最大的拷贝数,J是公式(18)中j的取值个数,N是基因组中window的数量,M是基因组中HGSNV的个数。
对[0,N]范围内的每一个整数值n,可以通过步骤G得到Q n,也可以通过步骤I得到所有peak的MAF期望的集合{f b},而一对(P,{f b})可以构建一个公式(19)所示的模型,实质上是对每一对(P,Q n),可以构建一个公式(19)所示的模型;
步骤K:
以0.001为分辨率,对[P-m,P+m]区间的所有P值,重复步骤G~J,可以得到一系列不同的(P,Q n)与对应的似然函数值,取最大的似然函数值对应的(P,Q n)作为最合适的P和Q值,m是0到0.5之间的一个值;
步骤L:
查询步骤H的结果,可以找到在步骤K得到的(P,Q)下,对应的癌症样本纯度和染色体倍性。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤A中,采用1000基因组计划第三期(phase 3)项目使用的参考基因组hs37d5(ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/reference/phase2_reference_assembly_sequence/hs37d5.fa.gz)作为本发明的参考基因组,它包含了GRCh37中的所有染色体和零散序列(decoy sequences)。比对软件使用Burrows-Wheeler Aligner(BWA),比对方法使用其中的bwa mem,最终获得癌症和正常样本的比对结果bam格式文件。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤B中,采用了samtools软件提取read的位置和长度信息,HGSNV的位点和覆盖该位点的read数量信息。使用samtools view命令提取read信息时,过滤掉序列比对质量(MAPQ)低于31的序列(参数-q 31,q表示过滤掉测序质量差的序列),同时过滤掉未能正确匹配的read(参数-f 0x2 -F 0x18,f表示提取符合一定要求的序列,F表示过滤符合一定要求的序列)。使用samtools mpileup命令提取HGSNV信息时,过滤掉序列比对质量(MAPQ)低于20的序列(参数-q 20),并过滤掉碱基质量小于20的序列(参数-Q 20,Q表示过滤掉碱基质量差的序列)。选取等位基因频率(allele frequence)时,本发明使用samtools mpileup的-1参数。使用该参数需要提前准备一个包含SNP位点信息的bed格式文件。本 发明方法提前收集了1000基因组(genome)计划(http://www.internationalgenome.org/)中,根据大量样本统计出来的杂合等位基因位点,并且过滤掉B-等位基因频率(B-allele frequence)小于0.05的位点,然后做成bed文件。使用“-1”参数在确保能提供充足的HGSNV位点基础上,大大加快了HGSNV位点的提取速度,提高了装置运行效率。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤C中,步骤C可以包括4步:
C1、将全基因组按照一定碱基长度的window为单位进行划分,对每个window统计覆盖该window的read数量,统计时以每条read的中点代表该read的位置;
C2、对参考基因组创建索引文件,提高GC含量的统计速度;
C3、以每个window的GC含量为自变量,以每个window的read数量为因变量,拟合read数量随GC含量变化的函数;
C4、使用拟合出的模型对全基因组read数量进行调整。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤C2中,本发明为参考基因组创建GC含量索引文件。对每一条染色体分别统计1、5、25、125个碱基间隔的区域内,鸟嘌呤(G)和胞嘧啶(C)的累积数量。那么在统计某一个window中的GC含量时,可以用a*125+b*25+c*5+d*1(其中a,b,c,d表示系数变量)的快速算法提取。例如想要统计某380bp区域内的GC含量时,可以分解为3*125+1*5形式,那么只需要读取特定索引文件的某个区域5碱基中的GC含量和某个区域125碱基中的GC含量即可。同时本发明将索引文件存储为二进制的格式,极大的加快了对特定区域的GC含量的提取。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤C3中,本发明使用步骤C1和步骤C2提取的各window GC含量,通过如下弹性网络模型拟合read数量随GC含量变化。本发明使用window的GC含量为变量x,使用x,x 2,x 3,x 4,x 5,x 6作为弹性网络模型的输入变量,以read数量为输出变量,构建弹性网络模型如公式(20)所示。式中,y表示window内观测到的read数量,X表示输入变量矩阵,β表示变量系数矩阵,j表示变量系数下标,P表示系数总数,λ 1和λ 2表示罚分系数。
Figure PCTCN2018078908-appb-000052
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤C4中,使用步骤C3中的模型预测每一个window理论上的read数量μ gc,基因组的平均GC含量定义为μ,window内观测到的read数量定义为y,window内校正后的read数量为Y。那么校正公式如下(21)所示:
Figure PCTCN2018078908-appb-000053
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤D中,本发明使用公式(1)计算每个window的TRE取值。然后运用TRE的值,使用BIC-seq软件对全基因组进行片段化(segmentation)。BIC-seq的思路是使用Bayesian Information Criterion(BIC)算法,统计相邻窗口的BIC值,值越小说明两个窗口越相似,然后将BIC值小于0的window合并,最终BIC-seq会按照片段拷贝数的差异,将全基因组分割为不同片段。每一个片段与相邻片段有不同的TRE均值,即拷贝数存在差异。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤E中,使用步骤D中BIC-seq处理后的基因组片段为单位,计算片段所包含的window数量,TRE的平均值以及方差。然后对片段的TRE进行smooth处理。处理方式如公式(22)所示。针对每一个基因组片段,以TRE的均值作为正态分布的均值μ,以TRE的方差作为正态分布的方差σ,计算出TRE在[μ-2σ,μ+2σ]范围内window数量的分布,定义v为TRE坐标,取值范围为[μ-2σ,μ+2σ],分辨率为0.000,C win为该片段分配到v位点的window数量,C T表示该片段内window的总数。将所有片段的window根据TRE值smooth后,可使片段内的window数量呈现正态分布,对所有片段各TRE位点对应的window数求和汇总,可以得到基因组范围的window随TRE变化的分布。
Figure PCTCN2018078908-appb-000054
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤F中,以0.001为分辨率,遍历(0,1]范围内的所有P,使用类自回归模型,计算Y(P)的值。Y(P)表现为多峰分布,类似图3所示,图中横轴为P,纵轴表示Y(P),本发明使用第二高峰内Y(P)的最大值对应的P作为P的计算结果,M t是TRE的最大取值,这里将M t设置为3。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤G中,步骤G包括3个步骤,步骤G1中,[0,1]的TRE区间作为变量X f的取值范围,过滤掉C(X f)小于1000的TRE位点,计算使公式(13.1)取最大值时的X f作为第一个peak的均值点。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤I中,根据步骤H计算的癌症样本纯度γ和peak对应的拷贝数,可计算peak的MAF的期望f b。其中步骤I可以包含I1、I2、I3三个步骤。
I1,使用公式(14)计算peak内HGSNV的MAF理论值,公式(14)中,C mcp表示主要等位基因的拷贝数(major allele copy number),C cp表示peak的整体拷贝数,由步骤I得到,f表示该peak内MAF的理论值,可见当C cp较大时,f有多种不同的可能值。
Figure PCTCN2018078908-appb-000055
I2,利用负二项分布估计覆盖每个HGSNV位点的read总数的概率,使用公式(15)计算负二项分布的概率p和失败次数r。公式(15)中,m是片段内所有window中read数量的均值,v是片段内所有window中read数量的方差,所求得的p是用于负二项分布的随机变量成功的概率,r为随机变量失败的次数,随机变量为覆盖某个HGSNV中的read数量。
Figure PCTCN2018078908-appb-000056
I3,利用二项分布求得的覆盖某个HGSNV的read数的概率。结合在一定read数量下,HGSNV只有两种基因型,服从二项分布规律,利用公式(16)计算f的校正值f b(即f的期望)。同一个peak中,不同的C mcp可以计算得到不同的f b,选择与该peak的MAF观测均值最接近的f b作为该peak的f b。公式(16)中,k表示在某个HGSNV位点,某一种等位基因(A或B)的数量,d为覆盖该HGSNV的read数量,r为随机变量失败的次数,p是用于负二项分布的随机变量成功的概率;
Figure PCTCN2018078908-appb-000057
对每一个Q n,可推断获得基因组所有peak对应的拷贝数和癌症样本纯度,从而对每一个peak可求f b,进而可以得到所有peak的MAF的期望值得集合{f b}。
作为一种优选方案,上述计算癌症样本纯度和染色体倍性的方法和装置中,所述步骤K中,m取0.02,P值的遍历区间为[P-0.02,P+0.02]。
通过本发明提供的层次混合高斯模型,实现了对癌症样本纯度的快速和准确计算,节约了纯度估算的时间和经济成本,同时提高了计算结果的准确性。
附图说明
图1表示全基因组中window数量在TRE上的分布。其中,图A表示的是未进过GC校正的TRE分布,图B表示经过GC含量校正后的TRE分布图。
图2表示一种癌症细胞中的TRE分布的模型,图为smooth处理以后,图中的peak满足以P为周期的分布,少量不满足周期性分布的小peak被认为是亚克隆片段。Q表示拷贝数为2的peak,不存在拷贝数为1的片段,所以在大约0.6的位置的peak的window数量为0。
图3表示横轴为P纵轴为类自回归模型计算值得分布。
图4表示本发明方法和装置的流程图。
具体实施方式
为更好地说明本发明的目的、技术方案和优点,下面将结合附图和具体实施例对本发 明作进一步说明。但是提供实施例仅用于说明目的,而本发明的范围不限于实施例。
使用本发明所述的装置计算癌症样本纯度和染色体倍性的流程图如图4所示。
实施例中,所使用的实验材料为TCGA(https://cancergenome.nih.gov/)数据库下载的样本(TCGA-AD-A5EJ)的正常组织TCGA-AD-A5EJ-10A和癌症组织TCGA-AD-A5EJ-01A的全基因组测序数据。计算平台为ubuntu 16.04,方法的具体实现为C++,python,R程序。
实施例:根据样本TCGA-CM-4746的癌症组织和正常组织的全基因组测序数据,使用层次性混合高斯模型计算癌症样本的纯度和染色体倍性。
一、收集样本数据,在TCGA中下载TCGA-CM-4746-01A的肿瘤样本和正常样本的全基因组测序数据。癌症样本bam文件大小为12.6G,正常样本bam文件大小为10.1G。将bam文件用PICARD软件处理为fastq文件。将fastq使用bwa mem比对到参考基因组hs37d5得到新的癌症样本和正常样本bam文件,文件大小分别为12.4G和9.9G。
二、下载1000genome项目提供的1到22号染色体的vcf文件(ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/release/20130502/),使用GATK的SelectVariants方法,提取参考基因组hs37d5.fa中的等位基因频率大于5%的BIALLELIC位点作为潜在的HGSNV位点,最终得到5633774个biallele位点。
三、提取正常样本和癌症样本的read信息,同时提取癌症样本的HGSNV信息。使用samtools提取癌症样本的序列覆盖度和HGSNV,得到HGSNV 67732个。提取HGSNV时,采用上述步骤一获得的biallele位点作为备选位点表单。使用samtools view方法,直接从备选表单中提取HGSNV,加快提取速度。
四、以500bp为window对参考基因组建立GC含量的索引文件,对上述步骤一下载得到的参考基因组hs37d5.fa文件,建立1,5,25,125区段中的GC含量索引文件。存储为二进制格式。
五、以500bp为一个窗口,统计全基因组范围内每个窗口中的read数量。同时使用步骤四中产生的索引文件计算每个窗口中的GC含量。通过弹性网络模型对read数量进行GC含量校正。
六、对每一个窗口,使用矫正后的read数量计算TRE。并依据TRE,通过BIC-seq对基因组进行片段化。片段化的结果如表一所示,每一列数据表示了一个基因组片段的位置信息和TRE的均值,方差和片段内window的数量。
表1 BIC-seq对基因组片段化后的结果
染色体号 起始 终止 TRE均值 TRE方差 Windows数量
chr1 13001 45265500 0.944508 0.0014742 86128
chr1 45265501 85978000 0.945454 0.00133201 80970
chr1 85978001 86011500 1.27362 0.0981321 68
chr1 86011501 116069000 0.94915 0.00153058 58775
chr1 116069001 120339500 1.01323 0.00437891 8488
chr1 120339501 143744000 1.07492 0.016337 2442
chr1 143744001 144707500 1.36461 0.0514154 469
chr1 144707501 145290000 1.46903 0.0252744 887
chr1 145290001 145833000 1.73666 0.0262944 936
chr1 145982001 148248000 1.33895 0.0131223 2306
chr1 148248001 149200000 1.68214 0.0381542 725
chr1 149200001 249240500 1.3383 0.00114954 196613
chr2 10001 41403000 0.948769 0.00134201 81865
chr2 41403001 51317000 0.563251 0.00183306 19717
chr2 51317001 91701500 0.950325 0.00145809 75550
chr2 91701501 91824500 1.23428 0.0605257 188
chr2 91824501 233387500 0.952652 0.000744377 271202
chr2 233387501 243186500 0.936487 0.00308593 19032
chr3 60001 197956500 0.952134 0.000613478 386567
chr4 10001 6890000 0.959657 0.00417921 13520
chr4 6890001 191044500 0.952703 0.000633657 358360
chr5 11501 12616000 0.55999 0.00180406 24895
chr5 12616001 180901000 0.948663 0.000676533 323787
chr6 124001 24827000 0.941509 0.00171221 49041
chr6 24827001 25982500 0.612532 0.00544556 2311
chr6 25982501 171051000 0.953312 0.000725761 280819
chr7 10001 5630500 0.954414 0.00461892 10976
chr7 5630501 38306500 0.955796 0.00149665 64732
chr7 38306501 38394000 1.24215 0.0368986 176
chr7 38394001 55087500 0.950529 0.00212248 33130
chr7 55087501 73283500 0.947545 0.00272214 27156
chr7 73283501 142340000 0.955659 0.00107445 134626
chr7 142340001 142491000 1.14391 0.0259809 303
chr7 142491001 159128500 0.965812 0.00251871 31914
chr8 11501 46857500 1.32635 0.00175707 84373
chr8 46857501 47744500 1.01383 0.015617 1735
chr8 47744501 146304000 1.33053 0.00112019 194805
chr9 10501 89624000 0.956959 0.00120825 119638
chr9 89624001 90043500 1.25975 0.0167381 836
chr9 90043501 92343000 1.91959 0.010442 4544
chr9 92343001 93352000 3.24803 0.0257886 1498
chr9 93352001 93696000 3.54267 0.0399231 683
chr9 93696001 94232000 2.92273 0.0283735 1068
chr9 94232001 95076500 3.23526 0.0249389 1668
chr9 95076501 95080000 1.47383 0.2343 8
chr9 95080001 95099000 0.580555 0.0461102 39
chr9 95099001 124413000 0.94853 0.00161715 58162
chr9 124413001 124419500 1.57438 0.201191 14
chr9 124419501 141128000 0.944109 0.00248041 32615
chr10 66001 135525000 0.946217 0.000787218 255445
chr11 113001 134946500 0.950948 0.000777276 259377
chr12 60501 93500 1.88481 0.206704 52
chr12 93501 100378000 0.949564 0.000874433 193005
chr12 100378001 133841500 0.943986 0.00155671 65852
chr13 19020501 115110000 0.984116 0.000893338 189923
chr14 19000001 22533500 0.981344 0.00661204 5638
chr14 22533501 23038000 1.12146 0.0151437 1009
chr14 23038001 74253500 0.948794 0.00119784 101769
chr14 74253501 74254500 3.63901 1.12718 3
chr14 74254501 107289500 0.943291 0.00152386 65750
chr15 20000001 29346000 0.940128 0.00397412 13933
chr15 29346001 102521500 0.946193 0.00104706 141616
chr16 60001 32646500 0.950261 0.00184498 60296
chr16 32646501 33796500 0.889339 0.0178243 1172
chr16 33796501 90282000 0.945044 0.00135985 89305
chr17 1 6109000 0.91987 0.00396313 11836
chr17 6109001 6125500 1.97935 0.163898 34
chr17 6125501 16653500 0.922942 0.00286745 20945
chr17 16653501 16738500 0.739932 0.0500593 142
chr17 16738501 22262500 0.959384 0.00557573 9958
chr17 22262501 27097500 1.71012 0.0106513 3647
chr17 27097501 34432500 0.93298 0.00339514 14570
chr17 34432501 34508500 1.28216 0.050695 121
chr17 34509001 36215500 0.91458 0.00729353 2890
chr17 36215501 36415000 1.0867 0.0380984 263
chr17 36415001 39421500 0.949891 0.00582802 5984
chr17 39421501 39432000 0.249188 0.0386297 20
chr17 39432001 51996000 0.937382 0.00264461 24210
chr17 51996001 52037000 1.3725 0.0543202 83
chr17 52037001 55322000 0.938207 0.00465989 6539
chr17 55322001 55632500 0.607987 0.0102 622
chr17 55632501 68609500 0.942895 0.00246878 25575
chr17 68609501 68745000 1.26374 0.0235519 272
chr17 68745001 81195000 0.943562 0.00287259 24285
chr18 10001 9988500 1.32762 0.00351405 19817
chr18 9988501 78017500 0.944828 0.00106534 128567
chr19 89001 23964500 0.940128 0.00211755 46618
chr19 23964501 24028000 0.547803 0.0243662 128
chr19 24028001 24621000 0.915363 0.0118983 1155
chr19 24628501 59119000 1.27099 0.00216296 61940
chr20 60001 29423000 0.944634 0.00166596 51929
chr20 29423001 62965500 1.6413 0.0024125 66170
chr21 9411001 40067000 0.938475 0.00164672 52525
chr21 40067001 40442000 0.605613 0.00897434 747
chr21 40442001 48120000 0.958554 0.00369137 14996
chr22 16050001 51235000 0.944787 0.00174455 66591
七、通过步骤六得到了每个片段的TRE的均值和方差,以及该片段内包含的窗口数量。使用正态分布的方法,以每个片段的TRE均值和方差作为正态分布的均值和方差,将片段中的窗口按照正态分布进行平滑化。汇总所有片段smooth后的TRE以及对应窗口数的信息。
八、对smooth后的TRE的窗口数进行自回归分析,得到P的取值为0.386。
九、P等于0.386时,第一个实际观测peak的TRE均值为0.562,第一个实际观测peak前最多可以存在1个理论peak,即N=1。可能的Q为:Q 0=1.334,Q 1=0.948,这两种Q的混合高斯模型的似然函数值分别为1.77E+07,1.78E+07。
十、计算P在取值范围[P-0.02,P+0.02]内,BIC校正后的混合高斯模型的极大似然值,计算结果如表2所示。
表2在P的取值范围内混合高斯模型的结果
P值 Q值 似然函数值 P值 Q值 似然函数值
0.366 0.932 1.12E+06 0.386 1.334 1.77E+07
0.366 1.298 1.12E+06 0.387 0.948 1.78E+07
0.367 0.933 849906 0.387 1.335 1.77E+07
0.367 1.3 847125 0.388 0.948 1.84E+07
0.368 0.934 795735 0.388 1.336 1.84E+07
0.368 1.302 792858 0.389 0.948 1.88E+07
0.369 0.935 832922 0.389 1.337 1.87E+07
0.369 1.304 830037 0.39 0.948 1.87E+07
0.37 0.936 1.05E+06 0.39 1.338 1.87E+07
0.37 1.306 1.04E+06 0.391 0.948 1.88E+07
0.371 0.937 1.18E+06 0.391 1.339 1.87E+07
0.371 1.308 1.17E+06 0.392 0.949 1.84E+07
0.372 0.938 1.72E+06 0.392 1.341 1.83E+07
0.372 1.31 1.71E+06 0.393 0.95 1.84E+07
0.373 0.939 3.74E+06 0.393 1.343 1.83E+07
0.373 1.312 3.73E+06 0.394 0.951 1.80E+07
0.374 0.94 5.34E+06 0.394 1.345 1.79E+07
0.374 1.314 5.31E+06 0.395 0.952 1.75E+07
0.375 0.941 7.56E+06 0.395 1.347 1.74E+07
0.375 1.316 7.52E+06 0.396 0.953 1.63E+07
0.376 0.942 8.71E+06 0.396 1.349 1.62E+07
0.376 1.318 8.67E+06 0.397 0.954 1.53E+07
0.377 0.943 1.08E+07 0.397 1.351 1.52E+07
0.377 1.32 1.07E+07 0.398 0.955 1.34E+07
0.378 0.944 1.58E+07 0.398 1.353 1.33E+07
0.378 1.322 1.57E+07 0.399 0.956 1.17E+07
0.379 0.945 1.69E+07 0.399 1.355 1.17E+07
0.379 1.324 1.68E+07 0.4 0.957 8.63E+06
0.38 0.946 1.78E+07 0.4 1.357 8.57E+06
0.38 1.326 1.77E+07 0.401 0.958 6.79E+06
0.381 0.947 1.88E+07 0.401 1.359 6.74E+06
0.381 1.328 1.87E+07 0.402 0.959 1.77E+06
0.382 0.948 1.93E+07 0.402 1.361 1.76E+06
0.382 1.33 1.92E+07 0.403 0.96 831069
0.383 0.948 1.92E+07 0.403 1.363 826993
0.383 1.331 1.91E+07 0.404 0.961 412408
0.384 0.948 1.92E+07 0.404 1.365 411139
0.384 1.332 1.91E+07 0.405 0.962 351299
0.385 0.948 1.88E+07 0.405 1.367 350311
0.385 1.333 1.87E+07 0.406 0.963 352988
0.386 0.948 1.78E+07 0.406 1.369 352002
十一、步骤十中的结果显示,P为0.382时,混合模型取极大值,此时的Q为0.948,据此可计算获得癌症样本纯度为0.80,癌症细胞染色体倍性为2.14。

Claims (13)

  1. 一种用于计算癌症样本中癌症细胞纯度和染色体倍性的方法,所述方法包括以下步骤:
    步骤A:
    获取配对的癌症组织样本和正常组织样本的全基因组测序数据,并将测序数据比对到参考基因组;
    步骤B:
    从步骤A得到的比对结果文件中,提取read位置和长度信息,HGSNV位点和覆盖该位点的read数量信息,计算所有HGSNV的MAF,其中,计算公式如(1.1)所示:
    Figure PCTCN2018078908-appb-100001
    公式(1.1)中,n r为包含与参考基因组相同等位基因的read数量,n a为包含另一种等位基因的read的数量,n t表示覆盖该HGSNV位点的总read数量,C为该HGSNV的MAF值;
    步骤C:
    根据步骤B得到的read位置和长度信息,以window为单位统计各window内包含的read数量,使用基因组GC含量校正所有window内read数量;
    步骤D:
    使用步骤C校正后的read数量,使用公式(1)计算每一个window的TRE,然后运用TRE,通过BIC-seq软件对基因组进行片段化,获得以拷贝数划分的基因组片段:
    Figure PCTCN2018078908-appb-100002
    公式(1)中,
    Figure PCTCN2018078908-appb-100003
    Figure PCTCN2018078908-appb-100004
    分别表示在癌症样本中覆盖片段s的read数量和在正常样本中覆盖片段s的read数量,N t表示癌症样本总read数量,N n表示相应正常样本总read数量,e s为TRE值;
    步骤E:
    以步骤D中BIC-seq处理后的基因组片段为单位,统计片段内所有window的TRE的均值、方差和该片段内window数量,根据均值和方差对基因组每个片段的window数量进行平滑化处理,使TRE的分布更均匀,然后将平滑化处理后所有片段的window分布汇总,得到基因组上window随TRE变化的分布结果;同时以片段为单位,计算片段中所有HGSNV的MAF的均值和方差;
    步骤F:
    使用如公式(12)、(13)所示的类自回归模型,计算相邻拷贝数片段内TRE的差值即P,其中,遍历一定范围的P,计算Y(P),在Y(P)的分布中,选择第二高峰内Y(P)的最大值对应的P作为P的计算结果:
    Figure PCTCN2018078908-appb-100005
    Figure PCTCN2018078908-appb-100006
    公式(12)和(13)中,X t表示0到M t之间的TRE值;t表示扩大了1000倍的TRE值;M t表示TRE的最大值;变量P表示两个TRE位点的间隔;C(X t)表示在TRE为X t的位点,对应的window数量;C(X t+1000×P)表示在TRE为X t+1000×P的位点,对应的window数量;Y(P)表示在变量P下,类自回归模型的函数值;
    步骤G:
    根据步骤F得到的P,计算TRE分布中第一个实际观测peak的TRE均值,然后计算在第一个实际peak之前最多可能存在理论peak的数量N,最后当第一个实际peak之前存在n个理论peak时,计算Q的值,以Q n表示,其中步骤G包括:
    G1:
    根据步骤F计算的P,使用公式(13.1),选取使公式(13.1)取最大值的X f作为第一个实际观测peak的TRE均值:
    Figure PCTCN2018078908-appb-100007
    公式(13.1)中,i表示第i个peak,C(X f+P×i)表示在TRE为X f+P×i的位点,对应的window数量,n表示M t以内peak的最大数量,M t表示TRE的最大值;
    G2:
    使用公式(13.2),根据步骤F计算的P和步骤G1计算的X f,计算在X f之前最多可能存在的peak数量N:
    Figure PCTCN2018078908-appb-100008
    公式(13.2)中,X f表示第一个peak的均值,P表示相邻拷贝数片段对应的peak之间的间距,floor表示向下取整数;
    G3:
    利用步骤G2计算的N值,当n取0到N之间的整数时,使用公式(13.3)计算Q n的值:
    Q n=X f-n×P+2×P=X f+(2-n)×P,n∈[0,N]  (13.3)
    公式(13.3)中,n表示X f之前peak的数量,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f表示第一个实际观测peak的TRE均值,Q n表示在X f之前理论上存在n个peak时的Q值;
    步骤H:
    使用步骤F计算的P与步骤G计算的Q n,使用公式(10)、(11)计算癌症样本纯度γ和染色体倍性k:
    Figure PCTCN2018078908-appb-100009
    Figure PCTCN2018078908-appb-100010
    公式(10)、(11)中,γ表示样本纯度,k表示染色体倍性,由此对(P,Q N)得到对应的(γ,k);
    步骤I:
    当n取[0,N]之间的某个整数值时,使用公式(13.4)计算第i个peak的TRE均值:
    T i=X f-n×P+i×P=X f+(i-n)×P,n∈[0,N]  (13.4)
    公式(13.4)中,n表示X f之前peak的数量,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f表示第一个实际观测peak的TRE均值,T i表示第i个peak的TRE均值,
    对于落在T i附近的片段,认为该片段具有拷贝数i;对于没有落在T i附近的片段,将其归类为亚克隆片段,在后续分析中剔除所有亚克隆片段;然后根据步骤H计算的癌症样本纯度γ和peak对应的拷贝数,计算peak的MAF的期望f b,不同peak的MAF期望不同,对基因组上的所有peak,最终得到MAF期望的集合{f b};同时计算各个peak的TRE均值和方差或标准差;
    步骤J:
    根据步骤F计算的P和步骤I计算的{f b}构建如公式(19)所示的用“贝叶斯信息准则”校正后的混合高斯分布模型,然后对模型极大似然估计;其中,步骤J包括如下几步:
    J1:
    以步骤F计算的P构建如公式(17)所示的高斯分布模型:
    Figure PCTCN2018078908-appb-100011
    公式(17)中,L(e s;γ,k)表示基因组片段TRE的似然函数,N表示基因组上的所有window的数量,I表示基因组中所有片段的最大的拷贝数,σ i表示拷贝数为i的所有片段的TRE的标准差由步骤I得到,e s为第s个window的TRE观测值,S i表示第i个peak的TRE均值即步骤I中的T i,p i表示第s个window的拷贝数为i的权重,对所有的i,p i均取 值为1;
    J2:
    以步骤I计算的f b构建如公式(18)所示的高斯分布模型:
    Figure PCTCN2018078908-appb-100012
    公式(18)中,L(f s;γ,k)表示HGSNV的似然函数,M表示基因组中所有HGSNV数量,S表示第S个HGSNV,I表示基因组中所有片段的最大的拷贝数;F i,j表示拷贝数为i,主要等位基因的拷贝数为j的片段内HGSNV的MAF期望值,由步骤I得到;f s表示该片段内所有HGSNV的MAF的观测值均值,由步骤E得到;σ i,j表示该片段内所有HGSNV的MAF观测值的标准差,由步骤E得到;p i,j表示在主要等位基因的拷贝数为j时,高斯分布的权重,对所有的i和j,p i,j取值均为1,p i表示第S个HGSNV所在片段的拷贝数为i的权重,对所有的i,p i取值均为1;
    J3:
    将(17)与(18)相加得到混合高斯模型,然后对混合模型进行BIC(Bayesian Information Criterion)校正得到最终混合模型如公式(19):
    BIC(e s,f s;γ,k)=-2×log L(f s;γ,k)-2×log L(e s;γ,k)+I×log(N)+J×log(M) (19)
    公式(19)中,BIC(e s,f s;γ,k)表示混合模型的似然函数,I表示基因组中所有片段的最大的拷贝数,J是公式(18)中j的取值个数,N是基因组中window的数量,M是基因组中HGSNV的个数,
    对[0,N]范围内的每一个整数值n,通过步骤G得到Q n,或者通过步骤I得到所有peak的MAF期望的集合{f b},由一对(P,{f b})构建一个公式(19)所示的模型;
    步骤K:
    以0.001为分辨率,对[P-m,P+m]区间的所有P值,重复步骤G~J,得到一系列不同的(P,Q n)与对应的似然函数值,取最大的似然函数值对应的(P,Q n)作为最合适的P和Q值,m是0到0.5之间的一个值;
    步骤L:
    查询步骤H的结果,找到在步骤K得到的(P,Q)下,对应的癌症样本纯度和染色体倍性。
  2. 一种用于计算癌症样本中癌症细胞纯度和染色体倍性的装置,其包括处理器,所述处理器用于运行程序,所述程序运行时执行以下步骤:
    步骤A:
    获取配对的癌症组织样本和正常组织样本的全基因组测序数据,并将测序数据比对到参考基因组;
    步骤B:
    从步骤A得到的比对结果文件中,提取read位置和长度信息,HGSNV位点和覆盖该位点的read数量信息,计算所有HGSNV的MAF,其中,计算公式如(1.1)所示:
    Figure PCTCN2018078908-appb-100013
    公式(1.1)中,n r为包含与参考基因组相同等位基因的read数量,n a为包含另一种等位基因的read的数量,n t表示覆盖该HGSNV位点的总read数量,C为该HGSNV的MAF值;
    步骤C:
    根据步骤B得到的read位置和长度信息,以window为单位统计各window内包含的read数量,使用基因组GC含量校正所有window内read数量;
    步骤D:
    使用步骤C校正后的read数量,使用公式(1)计算每一个window的TRE,然后运用TRE,通过BIC-seq软件对基因组进行片段化,获得以拷贝数划分的基因组片段:
    Figure PCTCN2018078908-appb-100014
    公式(1)中,
    Figure PCTCN2018078908-appb-100015
    Figure PCTCN2018078908-appb-100016
    分别表示在癌症样本中覆盖片段s的read数量和在正常样本中覆盖片段s的read数量,N t表示癌症样本总read数量,N n表示相应正常样本总read数量,e s为TRE值;
    步骤E:
    以步骤D中BIC-seq处理后的基因组片段为单位,统计片段内所有window的TRE的均值、方差和该片段内window数量,根据均值和方差对基因组每个片段的window数量进行平滑化处理,使TRE的分布更均匀,然后将平滑化处理后所有片段的window分布汇总,得到基因组上window随TRE变化的分布结果;同时以片段为单位,计算片段中所有HGSNV的MAF的均值和方差;
    步骤F:
    使用如公式(12)、(13)所示的类自回归模型,计算相邻拷贝数片段内TRE的差值即P,其中,遍历一定范围的P,计算Y(P),在Y(P)的分布中,选择第二高峰内Y(P)的最大值对应的P作为P的计算结果:
    Figure PCTCN2018078908-appb-100017
    Figure PCTCN2018078908-appb-100018
    公式(12)和(13)中,X t表示0到M t之间的TRE值;t表示扩大了1000倍的TRE值;M t表示TRE的最大值;变量P表示两个TRE位点的间隔;C(X t)表示在TRE为X t的位点,对应的window数量;C(X t+1000×P)表示在TRE为X t+1000×P的位点,对应的window数量;Y(P)表示在变量P下,类自回归模型的函数值;
    步骤G:
    根据步骤F得到的P,计算TRE分布中第一个实际观测peak的TRE均值,然后计算在第一个实际peak之前最多可能存在理论peak的数量N,最后当第一个实际peak之前存在n个理论peak时,计算Q的值,以Q n表示,其中步骤G包括:
    G1:
    根据步骤F计算的P,使用公式(13.1),选取使公式(13.1)取最大值的X f作为第一个实际观测peak的TRE均值:
    Figure PCTCN2018078908-appb-100019
    公式(13.1)中,i表示第i个peak,C(X f+P×i)表示在TRE为X f+P×i的位点,对应的window数量,n表示M t以内peak的最大数量,M t表示TRE的最大值;
    G2:
    使用公式(13.2),根据步骤F计算的P和步骤G1计算的X f,计算在X f之前最多可能存在的peak数量N:
    Figure PCTCN2018078908-appb-100020
    公式(13.2)中,X f表示第一个peak的均值,P表示相邻拷贝数片段对应的peak之间的间距,floor表示向下取整数;
    G3:
    利用步骤G2计算的N值,当n取0到N之间的整数时,使用公式(13.3)计算Q n的值:
    Q n=X f-n×P+2×P=X f+(2-n)×P,n∈[0,N]  (13.3)公式(13.3)中,n表示X f之前peak的数量,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f表示第一个实际观测peak的TRE均值,Q n表示在X f之前理论上存在n个peak时的Q值;
    步骤H:
    使用步骤F计算的P与步骤G计算的Q n,使用公式(10)、(11)计算癌症样本纯度γ和染色体倍性k:
    Figure PCTCN2018078908-appb-100021
    Figure PCTCN2018078908-appb-100022
    公式(10)、(11)中,γ表示样本纯度,k表示染色体倍性,由此对(P,Q N)得到对应的(γ,k);
    步骤I:
    当n取[0,N]之间的某个整数值时,使用公式(13.4)计算第i个peak的TRE均值:
    T i=X f-n×P+i×P=X f+(i-n)×P,n∈[0,N]  (13.4)
    公式(13.4)中,n表示X f之前peak的数量,取值范围是0到N之间的整数,P表示相邻拷贝数片段对应的peak之间的间距,X f表示第一个实际观测peak的TRE均值,T i表示第i个peak的TRE均值,
    对于落在T i附近的片段,认为该片段具有拷贝数i;对于没有落在T i附近的片段,将其归类为亚克隆片段,在后续分析中剔除所有亚克隆片段;然后根据步骤H计算的癌症样本纯度γ和peak对应的拷贝数,计算peak的MAF的期望f b,不同peak的MAF期望不同,对基因组上的所有peak,最终得到MAF期望的集合{f b};同时计算各个peak的TRE均值和方差或标准差;
    步骤J:
    根据步骤F计算的P和步骤I计算的{f b}构建如公式(19)所示的用“贝叶斯信息准则”校正后的混合高斯分布模型,然后对模型极大似然估计;其中,步骤J包括如下几步:
    J1:
    以步骤F计算的P构建如公式(17)所示的高斯分布模型:
    Figure PCTCN2018078908-appb-100023
    公式(17)中,L(e s;γ,k)表示基因组片段TRE的似然函数,N表示基因组上的所有window的数量,I表示基因组中所有片段的最大的拷贝数,σ i表示拷贝数为i的所有片段的TRE的标准差由步骤I得到,e s为第s个window的TRE观测值,S i表示第i个peak的TRE均值即步骤I中的T i,p i表示第s个window的拷贝数为i的权重,对所有的i,p i均取值为1;
    J2:
    以步骤I计算的f b构建如公式(18)所示的高斯分布模型:
    Figure PCTCN2018078908-appb-100024
    公式(18)中,L(f s;γ,k)表示HGSNV的似然函数,M表示基因组中所有HGSNV数量,S表示第S个HGSNV,I表示基因组中所有片段的最大的拷贝数;F i,j表示拷贝数 为i,主要等位基因的拷贝数为j的片段内HGSNV的MAF期望值,由步骤I得到;f s表示该片段内所有HGSNV的MAF的观测值均值,由步骤E得到,σ i,j表示该片段内所有HGSNV的MAF观测值的标准差,由步骤E得到;p i,j表示在主要等位基因的拷贝数为j时,高斯分布的权重,对所有的i和j,p i,j取值均为1,p i表示第S个HGSNV所在片段的拷贝数为i的权重,对所有的i,p i取值均为1;
    J3:
    将(17)与(18)相加得到混合高斯模型,然后对混合模型进行BIC校正得到最终混合模型如公式(19):
    BIC(e s,f s;γ,k)=-2×log L(f s;γ,k)-2×log L(e s;γ,k)+I×log(N)+J×log(M) (19)
    公式(19)中,BIC(e s,f s;γ,k)表示混合模型的似然函数,I表示基因组中所有片段的最大的拷贝数,J是公式(18)中j的取值个数,N是基因组中window的数量,M是基因组中HGSNV的个数,
    对[0,N]范围内的每一个整数值n,通过步骤G得到Q n,或者通过步骤I得到所有peak的MAF期望的集合{f b},由一对(P,{f b})构建一个公式(19)所示的模型;
    步骤K:
    以0.001为分辨率,对[P-m,P+m]区间的所有P值,重复步骤G~J,得到一系列不同的(P,Q n)与对应的似然函数值,取最大的似然函数值对应的(P,Q n)作为最合适的P和Q值,m是0到0.5之间的一个值;
    步骤L:
    查询步骤H的结果,找到在步骤K得到的(P,Q)下,对应的癌症样本纯度和染色体倍性。
  3. 根据权利要求1所述的方法或者权利要求2所述的装置,其中,所述步骤A中,采用1000基因组计划第三期(phase 3)项目使用的参考基因组hs37d5(ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/reference/phase2_reference_assembly_sequ ence/hs37d5.fa.gz)作为所述参考基因组;和/或,比对软件使用Burrows-Wheeler Aligner(BWA),比对方法使用其中的bwa mem,最终获得癌症和正常样本的比对结果bam格式文件。
  4. 根据权利要求1所述的方法或者权利要求2所述的装置,其中,所述步骤B中,采用samtools软件提取read的位置和长度信息,HGSNV的位点和覆盖该位点的read数量信息,其中,使用samtools view命令提取read信息时,使用参数-q 31过滤掉序列比对质量(MAPQ)低于31的序列,其中q表示过滤掉测序质量差的序列,同时使用参数-f 0x2-F 0x18过滤掉未能正确匹配的read,其中f表示提取符合一定要求的序列,F表示过滤符合一定要求的序列,使用samtools mpileup命令提取HGSNV信息时,使用参数-q  20过滤掉序列比对质量低于20的序列,并使用参数-Q 20过滤掉碱基质量小于20的序列,其中Q表示过滤掉碱基质量差的序列;选取等位基因频率时,使用samtools mpileup的-l参数;使用该参数需要提前准备一个包含SNP位点信息的bed格式文件。
  5. 根据权利要求1所述的方法或者权利要求2所述的装置,其中,
    所述步骤C包括4步:
    C1、将全基因组按照一定碱基长度的window为单位进行划分,对每个window统计覆盖该window的read数量,统计时以每条read的中点代表该read的位置;
    C2、对参考基因组创建索引文件,提高GC含量的统计速度;
    C3、以每个window的GC含量为自变量,以每个window的read数量为因变量,拟合read数量随GC含量变化的函数;
    C4、使用拟合出的模型对全基因组read数量进行调整。
  6. 根据权利要求5所述的方法或者装置,其中,所述步骤C2中,为参考基因组创建GC含量索引文件,对每一条染色体分别统计1、5、25、125个碱基间隔的区域内,鸟嘌呤(G)和胞嘧啶(C)的累积数量,其中,在统计某一个window中的GC含量时,用a*125+b*25+c*5+d*1的快速算法提取,其中a,b,c,d表示系数变量。
  7. 根据权利要求5所述的方法或者装置,其中,所述步骤C3中,使用步骤C1和步骤C2提取的各window的GC含量,通过如下弹性网络模型拟合read数量随GC含量变化,其中,使用window的GC含量为变量x,使用x,x 2,x 3,x 4,x 5,x 6作为弹性网络模型的输入变量,以read数量为输出变量,构建弹性网络模型如公式(20)所示:
    Figure PCTCN2018078908-appb-100025
    公式(20)中,y表示window内观测到的read数量,X表示输入变量矩阵,β表示变量系数矩阵,j表示变量系数下标,P表示系数总数,λ 1和λ 2表示罚分系数。
  8. 根据权利要求5所述的方法或者装置,其中,所述步骤C4中,使用步骤C3中的模型预测每一个window理论上的read数量μ gc,基因组的平均GC含量定义为μ,window内观测到的read数量定义为y,window内校正后的read数量为Y,那么校正公式如下(21)所示:
    Figure PCTCN2018078908-appb-100026
  9. 根据权利要求1所述的方法或者权利要求2所述的装置,其中,所述步骤E中,使用步骤D中BIC-seq处理后的基因组片段为单位,计算片段所包含的window数量,TRE的平均值以及方差,然后对片段的TRE进行平滑化处理,处理方式如公式(22)所示,针对每一个基因组片段,以TRE的均值作为正态分布的均值μ,以TRE的方差作为正态分布的方差σ,计算出TRE在[μ-2σ,μ+2σ]范围内window数量的分布,定义v为TRE 坐标,取值范围为[μ-2σ,μ+2σ],分辨率为0.001,C win为该片段分配到v位点的window数量,C T表示该片段内window的总数,将所有片段的window根据TRE值平滑化后,可使片段内的window数量呈现正态分布,对所有片段各TRE位点对应的window数求和汇总,得到基因组范围的window随TRE变化的分布:
    Figure PCTCN2018078908-appb-100027
  10. 根据权利要求1所述的方法或者权利要求2所述的装置,其中,所述步骤F中,以0.001为分辨率,遍历[0,1]范围内的所有P,使用类自回归模型,计算Y(P)的值,Y(P)表现为多峰分布,使用第二高峰内Y(P)的最大值对应的P作为P的计算结果,M t是TRE的最大取值,这里将M t设置为3。
  11. 根据权利要求1所述的方法或者权利要求2所述的装置,其中,所述步骤G中,步骤G包括3个步骤,步骤G1中,遍历[0,1]的TRE区间作为X f,过滤掉C(X f)小于1000的TRE位点,计算使公式(13.1)取最大值时的X f作为第一个实际观测peak的均值。
  12. 根据权利要求1所述的方法或者权利要求2所述的装置,其中,所述步骤I中,然后根据步骤H计算的癌症样本纯度γ和peak对应的拷贝数,计算peak的MAF的期望f b,其中步骤I包括:
    I1,使用公式(14)计算peak内HGSNV的MAF理论值:
    Figure PCTCN2018078908-appb-100028
    公式(14)中,C mcp表示主要等位基因的拷贝数,C cp表示peak的整体拷贝数,由步骤I得到,f表示该peak内MAF的理论值,可见当C cp较大时,f有多种不同的可能值;
    I2,利用负二项分布估计覆盖每个HGSNV位点的read总数的概率,使用公式(15)计算负二项分布的概率p和失败次数r:
    Figure PCTCN2018078908-appb-100029
    公式(15)中,m是peak内所有window中read数量的均值,v是peak内所有window中read数量的方差,所求得的p是用于负二项分布的随机变量成功的概率,r为随机变量失败的次数,随机变量为覆盖某个HGSNV中的read数量;
    I3,利用二项分布求得的覆盖某个HGSNV的read数的概率,结合在一定read数量下,HGSNV只有两种基因型,服从二项分布规律,利用公式(16)计算f的校正值f b,同一个peak中,不同的C mcp计算得到不同的f b,选择与该peak的MAF观测均值最接近的f b作为该peak的f b
    Figure PCTCN2018078908-appb-100030
    公式(16)中,k表示在某个HGSNV位点,某一种等位基因A或B的数量,d为覆盖该HGSNV的read数量,r为随机变量失败的次数,p是用于负二项分布的随机变量成功的概率;
    对每一个Q n,可推断获得基因组所有peak对应的拷贝数和癌症样本纯度,从而对每一个peak求f b,进而得到所有peak的MAF的期望值得集合{f b}。
  13. 根据权利要求1所述的方法或者权利要求2所述的装置,其中,所述步骤K中,m取0.02作为P值得遍历区间为[P-0.02,P+0.02]。
PCT/CN2018/078908 2017-05-05 2018-03-14 用于计算癌症样本纯度和染色体倍性的方法和装置 Ceased WO2018201805A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710312237.7 2017-05-05
CN201710312237.7A CN108804876B (zh) 2017-05-05 2017-05-05 用于计算癌症样本纯度和染色体倍性的方法和装置

Publications (1)

Publication Number Publication Date
WO2018201805A1 true WO2018201805A1 (zh) 2018-11-08

Family

ID=64016930

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/078908 Ceased WO2018201805A1 (zh) 2017-05-05 2018-03-14 用于计算癌症样本纯度和染色体倍性的方法和装置

Country Status (2)

Country Link
CN (1) CN108804876B (zh)
WO (1) WO2018201805A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110808084A (zh) * 2019-09-19 2020-02-18 西安电子科技大学 一种基于单样本二代测序数据的拷贝数变异检测方法
CN112767999A (zh) * 2021-01-05 2021-05-07 中国科学院上海药物研究所 一种全基因组测序数据的分析方法及装置
CN114708905A (zh) * 2022-05-11 2022-07-05 臻和(北京)生物科技有限公司 基于ngs的染色体非整倍体检测方法、装置、介质和设备

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111627498B (zh) * 2020-05-21 2022-10-04 北京吉因加医学检验实验室有限公司 一种测序数据gc偏向性校正的方法及其装置
CN112216344B (zh) * 2020-09-05 2024-12-31 西安翻译学院 肿瘤纯度和平均倍体信息的预测方法、系统、存储介质
CN113808009A (zh) * 2021-09-24 2021-12-17 熵智科技(深圳)有限公司 一种峰值初相位估计方法、装置、计算机设备及存储介质
CN115948521B (zh) * 2022-12-29 2024-06-25 东北林业大学 一种检测非整倍体缺失染色体信息的方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104560697A (zh) * 2015-01-26 2015-04-29 上海美吉生物医药科技有限公司 一种基因组拷贝数不稳定性的检测装置
US20150337298A1 (en) * 2014-05-23 2015-11-26 Fluidigm Corporation Haploidome determination by digitized transposons
CN106460070A (zh) * 2014-04-21 2017-02-22 纳特拉公司 检测染色体片段中的突变和倍性

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106520940A (zh) * 2016-11-04 2017-03-22 深圳华大基因研究院 一种染色体非整倍体和拷贝数变异检测方法及其应用

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106460070A (zh) * 2014-04-21 2017-02-22 纳特拉公司 检测染色体片段中的突变和倍性
US20150337298A1 (en) * 2014-05-23 2015-11-26 Fluidigm Corporation Haploidome determination by digitized transposons
CN104560697A (zh) * 2015-01-26 2015-04-29 上海美吉生物医药科技有限公司 一种基因组拷贝数不稳定性的检测装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
LI, YI ET AL.: "Deconvolving Tumor Purity and Ploidy by Integrating Copy Number Alterations and Loss of Heterozygosity", BIOINFORMATICS, vol. 30, no. 15, 1 August 2014 (2014-08-01), pages 2121 - 2129, XP055546291 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110808084A (zh) * 2019-09-19 2020-02-18 西安电子科技大学 一种基于单样本二代测序数据的拷贝数变异检测方法
CN110808084B (zh) * 2019-09-19 2023-02-28 西安电子科技大学 一种基于单样本二代测序数据的拷贝数变异检测方法
CN112767999A (zh) * 2021-01-05 2021-05-07 中国科学院上海药物研究所 一种全基因组测序数据的分析方法及装置
CN114708905A (zh) * 2022-05-11 2022-07-05 臻和(北京)生物科技有限公司 基于ngs的染色体非整倍体检测方法、装置、介质和设备

Also Published As

Publication number Publication date
CN108804876B (zh) 2022-03-15
CN108804876A (zh) 2018-11-13

Similar Documents

Publication Publication Date Title
WO2018201805A1 (zh) 用于计算癌症样本纯度和染色体倍性的方法和装置
US12584176B2 (en) Integrated machine-learning framework to estimate homologous recombination deficiency
Hasan et al. Performance evaluation of indel calling tools using real short-read data
KR102447079B1 (ko) 유전적 변이의 비침습 평가를 위한 방법 및 프로세스
JP2021040655A (ja) 無細胞核酸の多重解像度分析のための方法
CN107423578B (zh) 检测体细胞突变的装置
WO2018009723A1 (en) Methods for fragmentome profiling of cell-free nucleic acids
CN112218957A (zh) 用于确定在无细胞核酸中的肿瘤分数的系统及方法
IL258999A (en) Methods for detecting copy-number variations in next-generation sequencing
CN108475300B (zh) 利用癌症患者的基因组碱基序列突变信息和生存信息的定制型药物选择方法及系统
CN108603229A (zh) 用于高保真测序的方法和系统
CN110016497B (zh) 一种检测肿瘤单细胞基因组拷贝数变异的方法
CN105986011B (zh) 一种杂合性缺失的检测方法
CN106951731A (zh) 一种大片段插入或缺失的预测方法及系统
CN114974432A (zh) 一种生物标志物的筛选方法及其相关应用
CN109461473B (zh) 胎儿游离dna浓度获取方法和装置
WO2016141517A1 (zh) 检测突变簇的方法和装置
CN109712671B (zh) 基于ctDNA的基因检测装置、存储介质及计算机系统
US20090287420A1 (en) Method and system to characterize transcriptionally active regions and quantify sequence abundance for large scale sequencing data
CN107563152A (zh) 基于生物云平台的甲基化数据分析应用系统
Ritz et al. Detection of recurrent rearrangement breakpoints from copy number data
WO2025167582A1 (zh) 判断样本污染的方法、装置、电子设备和存储设备
CN106326689A (zh) 确定群体中受到选择作用的位点的方法和装置
WO2024020036A1 (en) Dynamically selecting sequencing subregions for cancer classification
Krsmanovic Discovery and optimization of cell-type-specific DNA methylation markers for in silico deconvolution

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18793979

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18793979

Country of ref document: EP

Kind code of ref document: A1