Fetal genotype detection method based on haplotype

Abstract

The invention relates to the technical field of biology, and discloses a fetus genotype detection method based on haplotype. The method comprises the following steps: 1) performing cfDNA sequencing on a to-be-tested pregnant woman plasma sample with known haplotype parting of a mother and a father to obtain a sequencing read section; 2) comparing the sequencing read segment to a reference genome, and estimating depth distribution and mutation of the sample according to a comparison result; 3) for four types of allele sites, the concentration of the fetal sample is calculated according to the secondary allele frequency, and the four types of allele sites comprise mother homozygous fetal homozygous, mother homozygous fetal heterozygous, mother heterozygous fetal homozygous and mother heterozygous fetal heterozygous; and 4) for genotype combinations AAAB, ABAA and ABAB of the mother and the father, deducing fetal SNP and InDel genotypes. The method disclosed by the invention can be used for detecting the single-point variation in the whole genome range, so that the accuracy is improved, and the sequencing depth is effectively reduced.

Description

Fetal genotype detection method based on haplotype Technical Field

[0001] The present invention relates to the field of biotechnology, and more specifically, provides a method for detecting fetal genotype based on haplotype. Background Art

[0002] Since 2000, research and clinical application of non-invasive prenatal testing for monogenic diseases using PCR and related technologies on maternal plasma cfDNA have aimed to definitively diagnose certain monogenic diseases (Saito, Sekizawa et al. 2000), such as myotonic dystrophy (Amicucci, Gennarelli et al. 2000). This testing method primarily targets a single gene or variant site, detecting both paternally inherited mutations and fetal de novo mutations by determining the presence of foreign alleles in the mother's peripheral blood that differ from her own genotype.

[0003] At maternal heterozygous loci, the presence of a large amount of maternal DNA interference in maternal plasma, combined with the presence of fetal and maternal DNA fragments, makes it difficult to determine whether the fetus carries a maternally inherited mutation. Qualitative analysis cannot confirm the presence of alleles distinct from the mother's own, necessitating quantitative analysis of each allele type. Digital PCR, capable of detecting minor allelic imbalances, was the first method to infer maternally inherited mutations. Relative mutation dosage (RMD) analysis, which measures the ratio of the two alleles at maternal heterozygous loci and their relationship to the fetal fraction (cfDNA concentration) of fetal DNA, has been validated in various diseases (Tsui, Kadir et al. 2011, Barrett, McDonnell et al. 2012).

[0004] The methods mentioned above are all aimed at detecting specific variants or a small number of gene regions. With the development of next-generation sequencing technology, based on the presence of fetal cfDNA in maternal plasma, it has become possible to infer the fetal genome and single-gene variants and new mutations on a genome-wide scale by performing whole-exome sequencing (WES) or whole-genome sequencing (WGS) of cfDNA in maternal plasma.

[0005] Lo et al. (Lo, Chan et al. 2010) pioneered the use of high-depth sequencing of cfDNA and maternal haplotype analysis to develop a method based on haplotype dosage (RHDO) to reveal the entire fetal genome, laying the foundation for research in this field. This method exploits the imbalance in read coverage of the two alleles at heterozygous sites in maternal plasma cfDNA sequencing data to develop a relative haplotype dosage (RHDO) method to determine the sites in the fetal genotype where the mother is heterozygous and the father is homozygous.

[0006] In the years following the publication of Lo et al.'s findings, efforts to improve haplotype-based methods have focused on using various techniques to directly haplotype the mother, thereby avoiding reliance on haplotypes from the proband or other family members (Fan et al. 2012, Kitzman et al. 2012). The method of Fan et al. (2012) utilizes maternal plasma cfDNA data and single-cell sequencing data to infer the fetal genome. This method first extracts lymphocytes from maternal peripheral blood and performs single-cell sequencing. Direct deterministic phasing is then used to obtain highly accurate haplotypes at the maternal chromosome level. Whole-genome (or exome) sequencing is then performed on the maternal plasma cfDNA. The haplotype inherited from the fetus is determined by counting the number of reads at heterozygous loci in the maternal plasma. The number of reads at the father's specific heterozygous sites in the maternal plasma is then counted to determine the alleles inherited from the fetus at the paternal heterozygous sites. This information is then used to reconstruct the paternal haplotype, thereby inferring the haplotype inherited from the father. Combining these two haplotype results yields the fetal genome.

[0007] The aforementioned methods for inferring the fetal genome using maternal plasma cfDNA data and parental haplotype sequencing data share a common limitation: reliance on haplotype information: these methods either rely on haplotype information from the proband to reconstruct the maternal haplotype or rely on complex, less widely used experimental techniques to obtain the maternal haplotype, requiring considerable expertise and being time-consuming. For various reasons, these methods are unable to type the entire fetal genome and can only cover a subset of genomic loci, making it impossible to infer the fetal genotype for loci not covered by haplotypes. Using haplotype information requires addressing recombination events and typing errors. Some methods cannot identify parental diheterogeneous variants. These methods cannot detect de novo mutations, and they cannot reliably infer fetal genotypes for indel-type variants. Haplotype-based methods have numerous limitations. In the first trimester, inferring the fetal genotype at all loci remains unfeasible due to issues such as low FF and limitations of sequencing methods.

[0008] As the cost of high-throughput sequencing decreases, researchers are experimenting with more reliable sequencing methods for deep sequencing, such as PCR-free WGS sequencing of maternal plasma cfDNA. Chan et al. (Chan, Jiang et al. 2016) employed a PCR-free protocol to reduce sequencing errors and ultra-deep cfDNA WGS sequencing to develop a genome-wide relative allelic dosage (GRAD) analysis method. This approach pioneered two major frontiers in genome-wide NIPD for monogenic diseases: inference of maternal heterozygous sites and detection of de novo mutations. This method utilizes ultra-deep sequencing (~270×) of maternal plasma cfDNA and parental whole-genome sequencing (WGS) data to infer the fetal genome. This method first aligns and detects variants in maternal plasma cfDNA data and parental WGS data. Using a PCR-free protocol to reduce sequencing errors and ultra-deep cfDNA WGS, they developed a genome-wide relative allelic dosage (GRAD) analysis method to determine the number of reads at each locus in maternal plasma. This method then combines the maternal plasma read counts to generate fetal genomic data. The advantage is that it does not require parental haplotypes and can infer de novo mutations. However, limitations of this method include: it does not infer parental diheterozygous loci; it requires a high FF; and it covers far fewer maternal heterozygous loci than the theoretical number. Subsequently, Rabinowitz et al. (Rabinowitz, Polsky et al. 2019) used high-throughput sequencing technology to perform deep sequencing of maternal plasma cfDNA, eliminating the need for parental haplotype information. Using a Bayesian model combined with machine learning, they were able to detect indels and parental diheterozygous loci. However, the overall detection accuracy was lower than that of other algorithms.

[0009] Ultra-deep sequencing data from maternal plasma cfDNA is used to infer the fetal genome genotype. Although full coverage of fetal genomic sites can be achieved, extremely high sequencing depth (above 300×) is required because only a fixed threshold is used for each site to determine the fetal genotype. When the sequencing depth is low (100×) or the fetal concentration is low, the fetal genotype cannot be accurately inferred.

[0010] References:

[0011] Amicucci,P.,M.Gennarelli,G.Novelli and B.Dallapiccola(2000)."Prenatal Diagnosis of Myotonic Dystrophy Using Fetal DNA Obtained from Maternal Plasma."Clinical Chemistry 46(2):301-302.

[0012] Barrett,A.N.,T.C.McDonnell,K.C.Chan and L.S.Chitty(2012)."Digital PCR analysis of maternal plasma for noninvasive detection of sickle cell anemia."Clin Chem 58(6):1026-1032.

[0013] Chan,K.C.,P.Jiang,K.Sun,Y.K.Cheng,Y.K.Tong,S.H.Cheng,A.I.Wong,I.Hudecova,T.Y.Leung,R.W.Chiu and Y.M.Lo(2016)."Second generation noninvasive fetal genome analysis reveals de novo mutations,single-base parental inheritance,and preferred DNA ends."Proc Natl Acad Sci U S A 113(50):E8159-E8168.

[0014] Fan,H.C.,W.Gu,J.Wang,Y.J.Blumenfeld,Y.Y.El-Sayed and S.R.Quake(2012)."Non-invasive prenatal measurement of the fetal genome."Nature 487(7407):320-324.

[0015] Kitzman,J.O.,M.W.Snyder,M.Ventura,A.P.Lewis,R.Qiu,L.E.Simmons,H.S.Gammill,C.E.Rubens,D.A.Santillan,J.C.Murray,H.K.Tabor,M.J.Bamshad,E.E.Eichler and J.Shendure(2012)."Noninvasive whole-genome sequencing of a human fetus."Sci Transl Med 4(137):137ra176.

[0016] Lo,Y.M.,K.C.Chan,H.Sun,E.Z.Chen,P.Jiang,F.M.Lun,Y.W.Zheng,T.Y.Leung,T.K.Lau,C.R.Cantor and R.W.Chiu(2010)."Maternal plasma DNA sequencing reveals the genome-wide genetic and mutational profile of the fetus."Sci Transl Med 2(61):61ra91.

[0017] Rabinowitz,T.,A.Polsky,D.Golan,A.Danilevsky,G.Shapira,C.Raff,L.Basel-Salmon,R.T.Matar and N.Shomron(2019)."Bayesian-based noninvasive prenatal diagnosis of single-gene disorders."Genome Res 29(3):428-438.

[0018] Saito,H.,A.Sekizawa,T.Morimoto,M.Suzuki and T.Yanaihara(2000)."Prenatal DNA diagnosis of a single-gene disorder from maternal plasma."Lancet 356(9236):1170.

[0019] Tsui, NBY, RAKadir, KCAChan, C. Chi and YMDLo (2011). "Noninvasive prenatal diagnosis of hemophilia by microfluidics digital PCR analysis of maternal plasma DNA." Blood 117(13):3684-3691.

[0020] Summary of the Invention

[0021] In response to at least some of the problems in the prior art, the present invention aims to provide a method for fetal genotype detection, which can detect single-site variation across the entire genome, improve accuracy, and effectively reduce sequencing depth.

[0022] Therefore, the present invention provides a method for detecting fetal genotype based on haplotype, the method comprising:

[0023] 1) For maternal plasma samples with known haplotypes of the mother and father, cfDNA sequencing is performed to obtain sequencing reads;

[0024] 2) Align the sequencing reads to the reference genome and estimate the depth distribution and mutations of the sample based on the alignment results;

[0025] 3) for four types of allele loci, calculating the concentration of the fetal sample according to the minor allele frequency, the four types of allele loci include maternal homozygous fetal homozygous, maternal homozygous fetal heterozygous, maternal heterozygous fetal homozygous, and maternal heterozygous fetal heterozygous;

[0026] 4) For the maternal and paternal genotype combinations AAAB, ABAA, and ABAB, the fetal SNP and InDel genotypes were inferred as follows:

[0027] a) For the maternal and paternal genotype combination AAAB, the fetal SNP genotype was inferred using the Bayesian algorithm, and based on the haplotype type of the SNP inherited from the father, the CIVA method was used to infer the InDel allele inherited from the father by the fetus;

[0028] b) For the maternal and paternal genotype combination ABAA, the SPRT algorithm was first used to infer the sites covered by the haplotypes, and the CIVA method was used to infer the remaining sites;

[0029] c) For the maternal and paternal genotype combination ABAB, the CIVA method is first used to determine the type of locus inherited from the father. The SPRT algorithm is then used to infer the type of locus inherited from the mother. For loci that cannot be inferred by the SPRT algorithm, the CIVA method is used to infer the allele type inherited from the mother to the fetus. The paternal and maternal alleles are combined to obtain the fetal genotype at that locus.

[0030] d) For sites not covered by haplotype data in the above three genotype combinations, the Bayesian algorithm was used to determine their genotypes;

[0031] e) For cases where the fetal locus determined by the Bayesian algorithm is different from that of the father and mother, a new fetal mutation is determined.

[0032] In one embodiment, in 4),

[0033] The Bayesian algorithm includes: after obtaining the fetal concentration, establishing a Bayesian model based on the fetal concentration to infer the fetal genotype;

[0034] The SPRT algorithm includes: using the imbalance of the read coverage of the two alleles of the heterozygous site in the maternal plasma cfDNA sequencing data to determine the site in the fetal genotype that is heterozygous for the mother and homozygous for the father;

[0035] The CIVA method includes: for a fetal site to be predicted, respectively inferring two haplotypes where two alleles of the site are inherited from the father and the mother, extracting the alleles actually at the site from the two haplotypes, and inferring the fetal genotype.

[0036] In one embodiment, in 4), the Bayesian algorithm includes: for each site, when the depth of reads of different alleles in maternal plasma and the fetal concentration are known, the probability of obtaining the maternal and fetal genotype combination i is calculated as:

[0037]

[0038] Among them, P(A i ) is the prior probability of the occurrence of combination i among n combinations of maternal and fetal genotypes based on fetal concentration; P(B|A i ) is the cumulative probability of the maternal and fetal genotype combination i calculated based on the read depth and prior probability at the site; maternal and fetal genotype combination n = (10 maternal genotypes × 10 fetal genotypes); when the genotypes of the father and the mother are known, the combination of prior probabilities is modified based on the above Bayesian model to remove the maternal and fetal genotype combinations that do not conform to Mendel's law of inheritance, and the probability value of each combination is recalculated; finally, the combination with the highest probability is selected, that is, the fetal genotype.

[0039] In one embodiment, in the SPRT algorithm, all SNPs in the mother's genome are divided into two types: α and β, and the mother's two haplotypes are numbered as Hap I and Hap II. In α-type SNPs, the father's allele is the same as the mother's Hap I allele, while in β-type SNPs, the father's allele is the same as the mother's Hap II allele. For α-type SNPs, when the fetus inherits the mother's Hap I, the Hap I allele at this site will have an increase in the number of reads; when the fetus inherits the mother's Hap II, the number of reads of the two alleles at this site will approach equilibrium. For β-type SNPs, when the fetus inherits the mother's Hap I, the number of reads of the two alleles at this site will approach equilibrium; when the fetus inherits the mother's Hap II, the number of reads of the Hap II allele at this site will increase.

[0040] In one embodiment, in the SPRT algorithm, the following two thresholds are used to determine whether the two alleles at each site are balanced:

[0041]

[0042] and

[0043]

[0044] Here,

[0045]

[0046] and

[0047]

[0048] Here, q0 is the total number of reads of Hap I at the site when the fetus inherits Hap II from the mother, q1 is the total number of reads of Hap I at the site when the fetus inherits Hap I from the mother, and N is the total number of reads;

[0049] For α-type SNP, q0=0.5, q1=ff / 2+0.5; for β-type SNP, q1=0.5, q0=0.5-ff / 2.

[0050] In one embodiment, in 4 e), the fetus is further screened for new mutations in the following manner:

[0051] i) All variants with minor allele frequency (MAF) > 1% in East Asian populations in dbSNP147 and gnomAD (Genome Aggregation Database) databases and recorded in internal databases;

[0052] ii) The genotype quality (GQ) of the father, mother, or fetus at this locus is less than 30;

[0053] iii) The sequencing depth of the father, mother, or fetus at this site is <20;

[0054] iv) The genotype likelihood (PL) of the father, mother, or fetus at the locus is less than 30;

[0055] v) The minor allele frequency of the father, mother, or fetus at the locus is <0.3 or >0.7;

[0056] vi) The following quality control indicators were found in the VCF (variation detection result file, generated by GATK) of the combined parent and fetus samples of this locus: QualByDepth (QD) <2.0, RMSMappingQuality (MQ) <50.0, FisherStrand (FS) >60.0, StrandOddsRatio (SOR) >3.0, MappingQualityRankSumTest (MQRankSum) <-12.5, and ReadPosRankSumTest (ReadPosRankSum) <-8.0.

[0057] In one embodiment, in 1), the haplotype typing of the mother and the father is obtained by performing stLFR (single tube Long Fragment Read) sequencing on the maternal blood cells and the paternal blood cells, respectively.

[0058] In one embodiment, in 1), the depth of cfDNA sequencing is 70-200 times, for example 100 times.

[0059] The solution of the present invention expands the scope of fetal genome inference using whole-genome sequencing data of maternal plasma cell-free DNA to single-site variations across the whole genome, providing support for the expansion of fetal genetic disease screening; by combining the haplotype algorithm with the Bayesian model, the accuracy of fetal genotype inference using only maternal plasma is further improved; high accuracy is maintained when the fetal cfDNA concentration is 4% and the sequencing depth of maternal plasma cell-free DNA is 100×; and the scope of fetal genotype inference is expanded to InDel types. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] FIG1 shows a schematic diagram of a fetal genotype inference method. DETAILED DESCRIPTION

[0061] The solution of the present invention can accurately infer the fetal genotype by performing high-depth whole-genome sequencing on the free DNA in the peripheral blood of pregnant women and stLFR sequencing on the blood cells of the parents, thereby screening for abnormalities in the fetal genome. In the present invention, stLFR sequencing can be used to obtain parental haplotype information, and the inference of paternal and maternal haplotypes can be further improved by introducing hidden Markov models or based on methods such as fragment assembly, thereby improving the overall accuracy. In the present invention, other haplotype sequencing methods can be used to replace the stLFR sequencing method. The present invention is of great significance for the detection of pathogenic sites of various genetic diseases such as single-gene diseases, polygenic diseases, chromosomal diseases, etc. that the fetus may carry. At the same time, the solution of the present invention can be applied to fields such as pregnancy disease prediction and fetal phenotype inference.

[0062] The scheme of the present invention does not rely on the haplotype information of the proband; the region of fetal genome analysis is expanded to the whole genome, the detection accuracy is narrowed to single-base mutations, and the application scope of fetal genetic disease detection is expanded to include sites where the mother is homozygous and the father is heterozygous, the mother is heterozygous and the father is homozygous, and both parents are heterozygous; the types of mutations detected include SNPs, InDels, and new fetal mutations; it does not rely on the high or low fetal concentration, and the results can still achieve high accuracy when the fetal concentration is low (4%); it still achieves high accuracy when the sequencing depth is relatively low (~100×).

[0063] The present invention is described below with reference to an exemplary embodiment. In this embodiment, the haplotype-based fetal genotype detection method comprises:

[0064] 1) For each analyzed family, complete high-depth sequencing (e.g., 100x) of maternal plasma cfDNA and stLFR sequencing of maternal and fetal paternal blood cells;

[0065] 2) After quality control of all offline data (fq format), use BWA (mem mode) or other alignment software to align the data to the human reference genome; use Picard and other methods to remove duplicate reads from the alignment results and calculate the duplication rate; use the base quality score correction (BQSR) function in variation detection algorithms such as GATK to perform local correction of the alignment results; use the Depth of Coverage function in variation detection algorithms such as GATK to calculate the depth distribution of each sample; use the population variation detection mode in variation detection algorithms such as GATK to complete the detection of single-point (SNP) and small fragment insertion and deletion (Indel);

[0066] 3) For the single-tube long fragment reads (stLFR) sequencing data of paternal and maternal blood cells, use tools such as Hapcut2 or Longhap to complete the assembly of long fragments and sample haplotype analysis based on the barcode information;

[0067] 4) Calculation of fetal sample concentration: Calculate the minor allele frequency (MAF) distribution. It is expected that there will be four peaks in this distribution, corresponding to four types of sites: maternal homozygosity and fetal homozygosity, maternal homozygosity and fetal heterozygosity, maternal heterozygosity and fetal homozygosity, and maternal heterozygosity and fetal heterozygosity. It is known that the MAF distribution is composed of reads from both maternal and fetal sources. For each site, when the fetal concentration is f, the MAF corresponding to the peak of the maternal pure fetal heterozygosity site is f / 2. Therefore, for MAF∈[0,0.25], its distribution is simulated as a mixture model of two normal distributions. It can be inferred that the MAF corresponding to the peak of the fetal read distribution is f / 2, that is, the fetal concentration of the sample is obtained.

[0068] 5) Bayesian algorithm: After obtaining the fetal concentration, a Bayesian model is established based on this to infer the fetal genotype. For each locus, when the depth of reads of different alleles in maternal plasma and the fetal concentration are known, the probability of obtaining the maternal and fetal genotype combination i is calculated as:

[0069]

[0070] Among them, A i is the fetal genotype of the i-th combination among n combinations, B is the base support number of the allele observed in the data; P(A i ) is the prior probability of the occurrence of the i-th combination among n combinations of maternal and fetal genotypes based on fetal concentration; P(B|A i ) is the cumulative probability of maternal and fetal genotype combination i, calculated based on the base sequencing depth and prior probability at that site; maternal and fetal genotype combination n = (10 maternal genotypes × 10 fetal genotypes); when the paternal and maternal genotypes are known, the above Bayesian model is used to modify the prior probability combinations, removing maternal and fetal genotype combinations that do not conform to Mendelian inheritance, and recalculating the probability value of each combination; ultimately, the combination with the highest probability is selected to obtain the fetal genotype;

[0071] 6) Haplotype method a (SPRT (sequential probability ratio test) algorithm (Lo, Chan et al. 2010)): This method uses the imbalance in the number of read segments of the two alleles at the heterozygous sites in the maternal plasma cfDNA sequencing data to determine the sites in the fetal genotype that are heterozygous for the mother and homozygous for the father.

[0072] Specifically, all single-nucleotide polymorphism (SNP) sites in the mother's genome are divided into two types: α and β. The mother's two haplotypes are numbered Hap I and Hap II. In α-type SNPs, the father's allele is the same as the mother's Hap I allele, while in β-type SNPs, the father's allele is the same as the mother's Hap II allele. For α-type SNPs, when the fetus inherits the mother's Hap I, the Hap I allele at this site will have an increase in reads; when the fetus inherits the mother's Hap II, the reads of the two alleles at this site approach equilibrium. For β-type SNPs, when the fetus inherits the mother's Hap I, the reads of the two alleles at this site approach equilibrium; when the fetus inherits the mother's Hap II, the reads of the Hap II allele at this site increase.

[0073] Based on the above theory, two thresholds are introduced to determine whether the two alleles at each site are balanced. The calculation methods of these two thresholds are:

[0074]

[0075] and

[0076]

[0077] Here,

[0078]

[0079] and

[0080]

[0081] Here, q0 is the total number of reads of Hap I at the site when the fetus inherits Hap II from the mother, q1 is the total number of reads of Hap I at the site when the fetus inherits Hap I from the mother, and N is the total number of reads;

[0082] For α-type SNP, q0=0.5, q1=ff / 2+0.5; for β-type SNP, q1=0.5, q0=0.5-ff / 2.

[0083] By comparing the cumulative proportion of reads corresponding to the maternal Hap I and Hap II alleles at each locus with the two thresholds described above, the fetal genotype can be determined by inferring the paternal and maternal alleles. If the cumulative depth of the maternal Hap I allele as a percentage of the total cumulative sequencing depth of the two haplotypes is less than the lower boundary, the fetus has inherited the allele from the mother's Hap II, and if it is greater than the upper boundary, the fetus has inherited the allele from the mother's Hap I.

[0084] 7) Haplotype method b (closest informative variant algorithm, CIVA):

[0085] For the fetal locus that needs to be predicted, the two haplotypes of the two alleles inherited from the father and mother are inferred respectively, and the alleles actually present at the locus in the two haplotypes are extracted to infer the fetal genotype;

[0086] Specifically, based on linkage disequilibrium, the fetal genotype information inferred from the previous Bayesian or SPRT algorithm is used, along with the haplotype information inherited from the father or mother at the nearest significant variant within a certain interval upstream and downstream of the analyzed locus, to infer the haplotype inheritance of the locus. The selected region length can be 50kb, 100kb, or 200kb, for example.

[0087] At the same time, a more stringent method can be added to determine whether the value of the variant site is credible: when the two value variant sites upstream and downstream within a certain area near the analyzed site point to inconsistent haplotype types inherited from the father or mother of the fetus, it is deemed unreliable and CIVA cannot be used in this case.

[0088] Valuable variant sites include the following two types:

[0089] When inferring the paternal haplotype, the paternal heterozygous and maternal homozygous loci are selected: for example, when the father is AB, the mother is AA, and the fetus is AB, the haplotype of the father's B allele is determined to be inherited by the fetus; for example, when the father is AB, the mother is AA, and the fetus is AA, the haplotype of the father's A allele is determined to be inherited by the fetus;

[0090] Similarly, the maternal haplotype can be inferred using the maternal heterozygous and paternal homozygous loci: when the mother is AB, the father is BB, and the fetus is BB, then the haplotype of the mother's B allele is determined to be inherited to the fetus; when the mother is AB, the father is BB, and the fetus is AB, then the haplotype of the mother's A allele is determined to be inherited to the fetus;

[0091] 8) The combined method of haplotype and Bayesian algorithms: SNPs or indels are divided into four categories, and the maternal and paternal genotype combinations are AAAB, ABAA, ABAB, and AAAA. The fetal genotype is inferred from the SNPs and indels using the aforementioned Bayesian, haplotype-based SPRT, and nearest variant site algorithms (Figure 1).

[0092] For SNP-type variants:

[0093] a) AAAB type SNPs, using Bayesian model;

[0094] b) For ABAA type SNPs, the SPRT algorithm is first used to infer sites that can be covered by the haplotype; for the remaining sites that cannot be judged by SPRT (between the two thresholds mentioned in the SPRT method above), the CIVA method is used;

[0095] c) For ABAB SNPs, CIVA is first used to determine the locus type inherited from the father. The SPRT algorithm is then used to infer the locus type inherited from the mother. For loci that cannot be inferred by SPRT, CIVA is used to infer the allele type inherited from the mother to the fetus. The paternal and maternal alleles are combined to obtain the fetal genotype at that locus.

[0096] d) For sites in the three types of SNPs mentioned above that are not covered by haplotype data, a Bayesian algorithm is used;

[0097] For InDel type mutations:

[0098] a) For AAAB-type InDels, based on the haplotype type of the AAAC-type SNP inherited from the father inferred in the previous step, the CIVA method is used to infer the alleles inherited by the fetus from the father;

[0099] b) For ABAA-type InDels, the SPRT algorithm is first used to infer sites that can be covered by haplotypes. For the remaining sites for which SPRT cannot make a judgment (between the two thresholds mentioned in the SPRT method above), the CIVA method is used.

[0100] c) For ABAB-type InDels, CIVA was first used to determine the locus type inherited from the father. The SPRT algorithm was then used to infer the locus type inherited from the mother. For loci that could not be inferred by SPRT, CIVA was used to infer the allele type inherited from the mother to the fetus. The paternal and maternal alleles were combined to obtain the fetal genotype at that locus.

[0101] d) For sites in the three types of InDels mentioned above that are not covered by haplotype data, a Bayesian algorithm is used.

[0102] 9) Fetal novel SNP mutation: For fetal mutation sites predicted by Bayesian methods to be different from those of the father and mother, the following filtering process is used to obtain fetal novel mutations:

[0103] i) All variants with MAF > 1% in East Asian populations in dbSNP147 and gnomAD (Genome Aggregation Database) databases and recorded in internal databases;

[0104] ii) The genotype quality (GQ) of the father, mother, or fetus at this locus is less than 30;

[0105] iii) The sequencing depth of the father, mother, or fetus at this site is <20;

[0106] iv) The genotype likelihood (PL) of the father, mother, or fetus at the locus is less than 30;

[0107] v) The minor allele frequency of the father, mother, or fetus at the locus is <0.3 or >0.7;

[0108] vi) The following quality control indicators in the VCF results of the parent-fetus combination for this locus: QualByDepth (QD) < 2.0, RMSMappingQuality (MQ) < 50.0, FisherStrand (FS) > 60.0, StrandOddsRatio (SOR) > 3.0, MappingQualityRankSumTest (MQRankSum) < -12.5, ReadPosRankSumTest (ReadPosRankSum) < -8.0;

[0109] QD: QualByDepth, which is the quality value corrected by depth.

[0110] MQ: RMSMappingQuality, the root mean square of the alignment quality of all aligned reads, used to determine the average alignment quality of a site.

[0111] FS: FisherStrand, which uses Fisher's exact test to determine whether the current mutation has a strand-specific tendency.

[0112] MQRankSum: MappingQualityRankSumTest, the Mann-Whitney rank sum test result of the alignment quality between different bases at heterozygous sites, which evaluates the credibility of the site by the difference in alignment quality between Ref and Alt bases.

[0113] SOR: StrandOddsRatio, uses OR value to judge the degree of chain specificity of the mutation site. The larger the value, the higher the chain specificity.

[0114] ReadPosRankSum: ReadPosRankSumTest is still a rank sum test for heterozygous sites to see whether different bases tend to appear at specific positions on the read segment (for example, close to the start or end of the read segment).

[0115] The method of the present invention utilizes two kinds of haplotype algorithms to be combined with Bayesian model, and carries out fetal unit point inference based on maternal plasma depth information and parental haplotype information, realizes full coverage of genomic sites.In the present invention, Bayesian algorithm can also combine the fragment difference of maternal and fetal cfDNA, such as fragment length, the motif (Motif) type at the end of the fragment, the difference of the alignment position on the genome, increase the weight of fetal source fragment, promote the accuracy of Bayesian algorithm.For ABAB type sites, first judge the haplotype type that AAAB site father inherits to fetus on genome according to Bayesian algorithm, then adopt CIVA algorithm to determine the haplotype that ABAB site father inherits to fetus, adopt SPRT algorithm to predict the haplotype that ABAB site mother predicts to fetus, jointly infer the fetal genotype of ABAB type.

[0116] Example

[0117] The inventors have completed sample collection, sequencing and data analysis of 50 healthy families and 9 disease families in 2021, and based on this, completed fetal genome inference to test the feasibility and accuracy of this fetal genome inference method, and further clarify the scope of application.

[0118] The present invention completed the test on the maternal plasma cfDNA sequencing data (WGS paired-end 100bp sequencing) and the paternal and maternal stLFR data (WGS paired-end 100bp sequencing) of five families, and used the fetal umbilical cord blood WGS data (paired-end 100bp) as the true set to calculate the accuracy of fetal genotype inference (Tables 1, 2, and 3).

[0119] The prediction accuracy of AAAB type variation in the five family SNP type sites was 96.2%±5.8% (mean±standard deviation), ABAA was 96.2%±1.4%, and ABAB was 87.2%±4.7%.

[0120] The prediction accuracy of AAAB type variants in the five families' InDel type sites was (mean ± standard deviation) 94.6% ± 1.9%, ABAA was 80.2% ± 4.3%, and ABAB was 79.3% ± 3.3%.

[0121] When the fetal concentration was 4% and the cfDNA sequencing depth was 128× (JK-53), the prediction accuracy of SNPs was 95.6% (AAAB), 92% (ABAA), and 74.7% (ABAB).

[0122] When the fetal concentration was 4% and the cfDNA sequencing depth was 270× (JK-53), the sensitivity of predicting fetal de novo mutations was 91.3% and the positive predictive value (PPV) was 53.9%.

[0123] Table 1. Sequencing data statistics of tested families

[0124]

[0125] Table 2. Summary of the accuracy of fetal genome inference at SNP type sites using different schemes

[0126]

[0127] Table 3. Summary of the accuracy of fetal genome inference at InDel type sites using different schemes

[0128]

Claims

1. A method for detecting fetal genotype based on haplotype, the method comprising: 1) For maternal plasma samples with known haplotypes of the mother and father, cfDNA sequencing is performed to obtain sequencing reads; 2) Align the sequencing reads to the reference genome and estimate the depth distribution and mutations of the sample based on the alignment results; 3) for four types of allele loci, calculating the concentration of the fetal sample according to the minor allele frequency, the four types of allele loci include maternal homozygous fetal homozygous, maternal homozygous fetal heterozygous, maternal heterozygous fetal homozygous, and maternal heterozygous fetal heterozygous; 4) For the maternal and paternal genotype combinations AAAB, ABAA, and ABAB, the fetal SNP and InDel genotypes were inferred as follows: a) For the maternal and paternal genotype combination AAAB, the fetal SNP genotype was inferred using the Bayesian algorithm, and based on the haplotype type of the SNP inherited from the father, the CIVA method was used to infer the InDel allele inherited from the father by the fetus; b) For the maternal and paternal genotype combination ABAA, the SPRT algorithm was first used to infer the sites covered by the haplotypes, and the CIVA method was used to infer the remaining sites; c) For the maternal and paternal genotype combination ABAB, the CIVA method is first used to determine the type of locus inherited from the father. The SPRT algorithm is then used to infer the type of locus inherited from the mother. For loci that cannot be inferred by the SPRT algorithm, the CIVA method is used to infer the allele type inherited from the mother to the fetus. The paternal and maternal alleles are combined to obtain the fetal genotype at that locus. d) For sites not covered by haplotype data in the above three genotype combinations, the Bayesian algorithm was used to determine their genotypes; e) For cases where the fetal locus determined by the Bayesian algorithm is different from that of the father and mother, a new fetal mutation is determined.

2. The method according to claim 1, wherein in 4), The Bayesian algorithm includes: After obtaining the fetal concentration, a Bayesian model was established based on it to infer the fetal genotype; The SPRT algorithm includes: using the imbalance of the read coverage of the two alleles of the heterozygous site in the maternal plasma cfDNA sequencing data to determine the site in the fetal genotype that is heterozygous for the mother and homozygous for the father; The CIVA method includes: for a fetal site to be predicted, inferring two haplotypes of two alleles inherited from the father and the mother at the site, extracting the alleles actually at the site from the two haplotypes, and inferring the fetal genotype.

3. The method according to claim 1 or 2, in 4), the Bayesian algorithm comprises: For each locus, when the depth of reads of different alleles in maternal plasma and the fetal concentration are known, the probability of obtaining a maternal and fetal genotype combination i is calculated as: Among them, P(A i ) is the prior probability of the occurrence of combination i among n combinations of maternal and fetal genotypes based on fetal concentration; P(B|A i ) is the cumulative probability of the maternal and fetal genotype combination i calculated based on the read depth and prior probability at the site; maternal and fetal genotype combination n = (10 maternal genotypes × 10 fetal genotypes); when the genotypes of the father and the mother are known, the combination of prior probabilities is modified based on the above Bayesian model to remove the maternal and fetal genotype combinations that do not conform to Mendel's law of inheritance, and the probability value of each combination is recalculated; finally, the combination with the highest probability is selected, that is, the fetal genotype.

4. The method according to claim 1 or 2, wherein in the SPRT algorithm, all SNPs in the mother's genome are divided into two types: α-type and β-type, and the mother's two haplotypes are numbered as Hap I and Hap II; in α-type SNPs, the father's allele is the same as the mother's Hap I allele, while in β-type SNPs, the father's allele is the same as the mother's Hap II allele; for α-type SNPs, when the fetus inherits the mother's Hap I, the Hap I allele at the site will have an increase in the number of reads; when the fetus inherits the mother's Hap II, the number of reads of the two alleles at the site will approach equilibrium; for β-type SNPs, when the fetus inherits the mother's Hap I, the number of reads of the two alleles at the site will approach equilibrium; when the fetus inherits the mother's Hap II, the number of reads of the Hap II allele at the site will increase.

5. The method according to claim 4, wherein in the SPRT algorithm, the following two thresholds are used to determine whether the two alleles at each site are balanced: and Here, and Here, q0 is the total number of reads of Hap I at the site when the fetus inherits Hap II from the mother, q1 is the total number of reads of Hap I at the site when the fetus inherits Hap I from the mother, and N is the total number of reads; For α-type SNP, q0=0.5, q1=ff / 2+0.5; for β-type SNP, q1=0.5, q0=0.5-ff / 2.

6. The method according to claim 1 or 2, wherein in 4 e), the fetus is further screened for new mutations by: i) All variants with minor allele frequency (MAF) > 1% in East Asian populations in dbSNP147 and gnomAD (Genome Aggregation Database) databases and recorded in internal databases; ii) The genotype quality (GQ) of the father, mother, or fetus at this locus is less than 30; iii) The sequencing depth of the father, mother, or fetus at this site is <20; iv) The genotype likelihood (PL) of the father, mother, or fetus at the locus is less than 30; v) The minor allele frequency of the father, mother, or fetus at the locus is <0.3 or >0.7; vi) The following quality control indicators were found in the VCF (variation detection result file, generated by GATK) of the combined parent and fetus samples of this locus: QualByDepth (QD) <2.0, RMSMappingQuality (MQ) <50.0, FisherStrand (FS) >60.0, StrandOddsRatio (SOR) >3.0, MappingQualityRankSumTest (MQRankSum) <-12.5, and ReadPosRankSumTest (ReadPosRankSum) <-8.

0.

7. The method according to claim 1 or 2, wherein in 1), the haplotypes of the mother and the father are obtained by performing stLFR (single tube Long Fragment Read) sequencing on maternal blood cells and paternal blood cells, respectively.

8. The method according to claim 1 or 2, wherein in 1), the depth of cfDNA sequencing is 70-200 times.

9. The method according to claim 8, wherein the depth of cfDNA sequencing is 100 times.

Images