Method for constructing fetal concentration prediction model, method for predicting fetal concentration and electronic device

Abstract

The invention provides a fetal concentration prediction model construction method, a fetal concentration prediction method and an electronic device. The construction method comprises the following steps: respectively carrying out continuous region division on a sequencing file of a first sample and a sequencing file of a second sample according to physical positions of chromosomes to obtain a window set of the first sample and a window set of the second sample; comparing the reading number distribution of windows with consistent chromosome physical positions in the first sample window set and the second sample window set, retaining windows with statistical differences, and judging correlation to obtain a total window set; obtaining a training set sample vector corresponding to each training set sample according to the principal component corresponding to the reading number corresponding to each training set sample in the total window set; inputting vectors of all training set samples into the model for training by using XGBoost to obtain a fetal concentration prediction model; the method can solve the problem of limitation of fetal concentration detection in the prior art, and is suitable for the field of fetal concentration calculation.

Description

Technical Field

[0001] This invention relates to the field of fetal concentration calculation, and more specifically, to a method for constructing a fetal concentration prediction model, a method for predicting fetal concentration, and an electronic device. Background Technology

[0002] In 1997, Dennis Lo first revealed the presence of cell-free fetal DNA in the peripheral blood of pregnant women. This landmark discovery ushered in a new era of non-invasive prenatal testing (NIPT), making it possible to screen for fetal chromosomal aneuploidy and other genetic diseases simply by collecting a blood sample from the pregnant woman. Compared to traditional fetal trisomy screening methods, such as amniocentesis and chorionic villus sampling, NIPT has rapidly become a significant force in the field of prenatal diagnosis due to its lower false positive rate, non-invasive procedure, and higher detection accuracy. However, in the practical application of NIPT technology, a core parameter—the concentration of cell-free fetal DNA in maternal plasma (referred to as fetal concentration)—plays a decisive role in the reliability of the test results.

[0003] Currently, a range of solutions have been developed for fetal concentration measurement technology. One of the most well-known strategies is based on Y-chromosome analysis, utilizing the presence of Y-chromosome DNA fragments unique to male fetuses in maternal plasma as the basis for calculating fetal concentration. While this method demonstrates outstanding accuracy, it is limited to assessing concentrations in male fetuses and is ineffective for female fetuses. Furthermore, some algorithms attempt to circumvent the influence of sex chromosomes by utilizing DNA fragment length, methylation status, and information on specific single nucleotide polymorphism (SNP) sites. However, these strategies often require additional experimental steps and costs, such as using paired-end (PE) sequencing technology to obtain DNA fragment length information, or performing specialized high-depth sequencing and paternal sequencing for SNP site analysis. These factors increase the complexity and cost of NIPT, limiting its widespread adoption among the general population.

[0004] Against this backdrop, three new methods for calculating fetal concentration have emerged in recent years. These methods require only low-depth single-end (SE) whole-genome sequencing data collected during NIPT testing, attempting to simplify the process and reduce costs. One method, such as seqFF software, draws on the distribution characteristics of DNA fragments throughout the genome; another focuses on the distribution of nucleosome fragments, attempting to find clues for calculating fetal concentration; and the third method, hetFF, indirectly estimates fetal concentration by analyzing the distribution of heterozygous SNP sites in maternal plasma. Unfortunately, the first two methods either lack accuracy or fail to provide reliable results under low fetal concentration conditions. While hetFF is theoretically more reasonable, its accuracy is significantly affected when the actual sequencing depth is less than 0.5×.

[0005] In summary, existing technologies have significant limitations in calculating fetal concentrations: on the one hand, methods relying on additional experimental data, while accurate, are expensive; on the other hand, methods requiring only low-depth single-end sequencing data, while economical, cannot guarantee accuracy under certain conditions (such as extremely low fetal concentrations or sequencing depths). Therefore, developing a new technology that is both economical and covers fetuses of all sexes, and can accurately calculate fetal concentrations under various conditions, has become a crucial issue that urgently needs to be addressed in the NIPT field. Summary of the Invention

[0006] The main objective of this invention is to provide a method for constructing a fetal concentration prediction model, a method for predicting fetal concentration, and an electronic device to address the limitations of existing fetal concentration detection technologies.

[0007] To achieve the above objectives, according to a first aspect of the present invention, a method for constructing a fetal concentration prediction model is provided, the method comprising:

[0008] S1) Divide the sequencing files of the first sample and the second sample into continuous regions based on the physical location of chromosomes to obtain the first sample window set and the second sample window set; S2) Compare the read count distribution of windows with the same physical location of chromosomes in the first sample window set and the second sample window set, and retain windows with statistical differences to obtain the calibration window set; compare the read count distribution of each window in the calibration window set pairwise to determine the correlation. If a high correlation is found, retain only one window to obtain the total window set; high correlation is defined as a correlation coefficient ≥ 95%; S3) Construct a vector matrix of training set samples and total window set, and extract principal components. Then extract the principal components corresponding to the read count of each training set sample in the total window set to obtain the training set sample vector for each training set sample; S4) Use XGBoost to input the vectors of all training set samples into the model for training to obtain the fetal concentration prediction model; wherein, the first sample is a plasma sample from a non-pregnant woman; the second sample is a peripheral blood sample from a pregnant woman with a male singleton pregnancy; and the training set samples are peripheral blood samples from pregnant women.

[0009] Furthermore, the sequencing files of the first sample and the second sample are sequencing files obtained by low-depth sequencing of the DNA of the first sample and the DNA of the second sample.

[0010] Further, in S1), the continuous division of regions according to the physical location of chromosomes includes dividing autosomes into continuous regions in units of 2 to 5M, with the windows obtained from the division being adjacent and non-overlapping; preferably, autosomes do not include chromosomes 13, 18, and 21.

[0011] Furthermore, the method for calculating statistical differences includes the t-test; preferably, a statistically significant difference is defined as a p-value less than 0.05.

[0012] Furthermore, the number of the first sample is at least 100, the number of the second sample is at least 100, and the number of training set samples is at least 100; preferably, the principal components are the principal components ranked in the top 10 to 40; preferably, the correlation coefficient includes Pearson correlation coefficient, Spearman rank correlation coefficient, or dot-bicollinear correlation coefficient.

[0013] To achieve the above objective, according to a second aspect of the present invention, a method for calculating fetal concentration is provided, the method comprising: constructing a fetal concentration prediction model using the above-described method for constructing a fetal concentration prediction model; obtaining a vector of a sample to be tested based on principal components; and inputting the vector into the fetal concentration prediction model to obtain the fetal concentration of the sample to be tested.

[0014] To achieve the above objectives, according to a third aspect of the present invention, an electronic device for constructing a fetal concentration prediction model is provided, the electronic device comprising: a window set acquisition unit, a total window set acquisition unit, a training set sample vector calculation unit, and a model training unit;

[0015] The window set acquisition unit is used to divide the sequencing files of the first sample and the sequencing files of the second sample into continuous regions according to the physical location of chromosomes, respectively, to obtain the first sample window set and the second sample window set.

[0016] The total window set acquisition unit includes a window set correction unit and a judgment unit;

[0017] The correction window set unit is used to compare the distribution of read counts of windows with the same chromosome physical location in the first sample window set and the second sample window set, retain windows with statistical differences, and obtain the correction window set.

[0018] The judgment unit is used to compare the distribution of the number of read segments corresponding to each window in the correction window set pairwise to determine the correlation. If a high correlation is found, only one window is retained to obtain the total window set; high correlation is defined as a correlation coefficient ≥ 95%.

[0019] The training set sample vector calculation unit is used to construct a vector matrix of training set samples and total window set, extract principal components, and then extract the principal components corresponding to the number of read segments of each training set sample in the total window set to obtain the training set sample vector corresponding to each training set sample.

[0020] The model training unit is used to train the model by inputting the vectors of all training set samples into the model using XGBoost, thereby obtaining the fetal concentration prediction model.

[0021] The first sample consisted of plasma samples from non-pregnant women; the second sample consisted of peripheral blood samples from pregnant women with a single male fetus; and the training set consisted of peripheral blood samples from pregnant women.

[0022] Furthermore, the sequencing files of the first sample and the second sample are sequencing files obtained by low-depth sequencing of the DNA of the first sample and the DNA of the second sample.

[0023] Furthermore, the window set acquisition unit includes a chromosome window division unit; the chromosome window division unit includes continuous division of regions according to the physical location of chromosomes, including dividing autosomes into continuous regions in units of 2 to 5M, with the windows obtained being adjacent and non-overlapping; preferably, autosomes do not include chromosomes 13, 18, and 21; preferably, the statistical difference calculation method includes a t-test; preferably, a statistical difference is defined as a P-value less than 0.05; preferably, the number of the first sample is at least 100, the number of the second sample is at least 100, and the number of training set samples is at least 100; preferably, the principal components are the principal components ranked in the top 10 to 40; preferably, the correlation coefficient includes Pearson correlation coefficient, Spearman rank correlation coefficient, or dot-bivariate correlation coefficient.

[0024] To achieve the above objectives, according to a fourth aspect of the present invention, an electronic device for calculating fetal concentration is provided, the electronic device comprising: a fetal concentration prediction model building unit and a calculation unit; wherein, the fetal concentration prediction model building unit is used to build a fetal concentration prediction model using the above-described fetal concentration prediction model building method; the calculation unit is used to obtain a vector of the sample to be tested based on principal components, and input the vector into the fetal concentration prediction model to obtain the fetal concentration of the sample to be tested.

[0025] To achieve the above objectives, according to a fifth aspect of the present invention, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the above-described method for constructing a fetal concentration prediction model or the above-described method for calculating fetal concentration by executing the executable instructions.

[0026] To achieve the above objectives, according to a sixth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, the device on which the computer-readable storage medium is located executes the above-described method for constructing the fetal concentration prediction model or the above-described method for calculating fetal concentration.

[0027] By applying the technical solution of this invention, through refined window segmentation, statistical difference screening, and correlation coefficient optimization, redundant information is eliminated, focusing on genomic regions carrying pregnancy-specific information, thereby improving the density and quality of detection information. Furthermore, by utilizing the predictive performance of the XGBoost algorithm and combining it with principal components extracted from sample data, the constructed model can effectively distinguish and predict fetal concentrations in different samples, improving the model's computational efficiency and making fetal concentration calculations for large-scale samples possible. The fetal concentration prediction model of this application not only improves the accuracy and reliability of fetal concentration calculations but also simplifies the detection process, reduces overall costs, and is easily integrated with other NIPT analysis tools, promoting the dissemination and application of the technology. Attached Figure Description

[0028] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0029] Figure 1 A flowchart illustrating the construction method of the fetal concentration prediction model in this application specification is shown.

[0030] Figure 2 A schematic diagram of the electronic device used to construct a fetal concentration prediction model, as described in this application specification, is shown.

[0031] Figure 3 A hardware block diagram of the method for constructing a fetal concentration prediction model as described in this application specification is shown.

[0032] Figure 4 The correlation coefficient results of the fetal concentration prediction model in Example 4 of this application are shown in the figure. Detailed Implementation

[0033] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the embodiments.

[0034] Terminology Explanation:

[0035] Extreme Gradient Boosting (XGBoost) is an efficient, flexible, and powerful machine learning algorithm. It is an advanced version of the Gradient Boosting Decision Tree (GBDT) family. It builds multiple weaker decision tree models sequentially and combines their predictions to form a powerful final model. Each subsequent tree is dedicated to correcting the errors of the preceding tree's predictions.

[0036] Fetal concentration: In non-invasive prenatal testing (NIPT), a small amount of cell-free DNA fragments from the fetus are present in the pregnant woman's blood. The percentage of fetal cell-free DNA in the total cell-free DNA (usually ranging from 5% to 30%).

[0037] Principal Component Analysis (PCA) is an unsupervised linear dimensionality reduction technique that projects the original high-dimensional data onto a set of linearly independent new variables (i.e., principal components) ordered by variance through orthogonal transformation. This achieves dimensionality compression and information extraction while preserving as much of the global variance of the data as possible.

[0038] Correlation coefficient: A statistic used to quantify the strength and direction of a linear (or monotonic) association between two variables. The correlation coefficient measures the strength of the correlation between two variables. In this application, the correlation coefficient may be the Pearson correlation coefficient (r), the Spearman rank correlation coefficient (ρ), or the point-two correlation coefficient (rpb), or any correlation coefficient obtained by a method known to those skilled in the art.

[0039] As mentioned in the background section, existing methods for calculating fetal concentration have limitations, including being restricted by fetal sex, relying on specific detection conditions, high detection costs, or low detection accuracy. Therefore, in this application, the inventors attempt to develop a new method for calculating fetal concentration, and thus propose a series of protection schemes. Based on this, in this application, the inventors attempt to develop a new method for constructing a fetal concentration prediction model, and thus propose a series of protection schemes.

[0040] In a first typical embodiment of this application, a method for constructing a fetal concentration prediction model is provided. A flowchart illustrating the method for constructing the fetal concentration prediction model in this application is shown below. Figure 1As shown, the construction method includes: S1) dividing the sequencing files of the first sample and the second sample into continuous regions according to the physical location of chromosomes, respectively, to obtain the first sample window set and the second sample window set; S2) comparing the read number distribution of windows with the same physical location of chromosomes in the first sample window set and the second sample window set, retaining windows with statistical differences to obtain the correction window set; comparing the read number distribution of each window in the correction window set pairwise to determine the correlation, and if a high correlation is found, retaining only one window to obtain the total window set; high correlation refers to a high degree of correlation. The relation count is ≥95%; S3) Construct the vector matrix of the training set samples and the total window set, and extract the principal components. Then, extract the principal components corresponding to the number of reads in the total window set for each training set sample to obtain the training set sample vector corresponding to each training set sample; S4) Use XGBoost to input the vectors of all training set samples into the model for training to obtain the fetal concentration prediction model; where the first sample is the plasma sample of a non-pregnant woman; the second sample is the peripheral blood sample of a pregnant woman with a male singleton pregnancy (preferably a pregnant woman with a gestational age of 12 weeks + 0 days or more); the training set sample is the peripheral blood sample of a pregnant woman.

[0041] In constructing the fetal concentration prediction model of this application, the samples were first divided into windows. Sequencing data from the first sample (i.e., plasma samples from non-pregnant women) and the second sample (i.e., peripheral blood samples from pregnant women with singleton male fetuses) were processed. The sequencing files were mapped onto the human reference genome. Based on the physical location of the chromosomes, the sample data was continuously divided into multiple windows, ensuring that the size and position of each window were consistent in both sample groups. This yielded the first sample window set and the second sample window set, laying the foundation for subsequent differential analysis and model construction.

[0042] Then, by comparing the distribution of read counts for windows with the same chromosomal physical location in the first and second sample window sets, and by performing a t-test, windows that show statistically significant differences between the two groups of samples are selected, thus forming a preliminary set of correction windows. This step enables the windows selected when building the model to more realistically reflect the impact of pregnancy status.

[0043] Furthermore, the windows within the calibration window set are optimized. Specifically, the correlation between windows is evaluated by calculating the correlation coefficient between each pair of windows. When the correlation coefficient between two windows is greater than 0.95, it indicates that they carry highly similar information, and only one window needs to be retained. This removes redundant information, improves the efficiency and accuracy of the model, and ultimately forms the optimized total window set.

[0044] Based on the aforementioned total window set, a vector matrix containing the read distribution for each window was constructed for the training set samples (in specific cases, only peripheral blood samples from pregnant women were used as the training set samples). Principal component analysis (PCA) was used to orthogonally decompose the vector matrix, extracting principal components that summarize most of the variation information in the read distribution of the samples. The read distribution of each training set sample within the total window set corresponds to these principal components, thereby generating training set sample vectors that represent the key genetic characteristics of the samples.

[0045] Finally, the XGBoost algorithm was used to train the model, with the vectors of all training set samples as input features and fetal concentration calculated based on the Y chromosome as the label. XGBoost is an efficient and powerful machine learning algorithm that can learn the complex nonlinear relationship between sample vectors and fetal concentration through gradient boosting techniques, thereby constructing a highly accurate fetal concentration prediction model.

[0046] The fetal concentration prediction model constructed using the method described in this application enhances the accuracy and efficiency of fetal concentration calculation in non-invasive prenatal testing (NIPT). The method first eliminates irrelevant biological noise, improving the purity and specificity of the data. This is achieved by focusing on genomic windows that reflect changes in pregnancy status. Furthermore, this application transcends the limitations of traditional static window screening (i.e., seqFF and preface window screening are based on genomic sequence characteristics, excluding regions with low alignment, N content, and low complexity, with the remaining genomic sequences further divided into windows for subsequent detection), introducing dynamic biological response characteristics to capture the most sensitive pregnancy signals.

[0047] Specifically, by systematically comparing and analyzing the genomic data of non-pregnant and pregnant women, this application uses a t-test to identify whether there are significant differences in the screening windows between the two groups, thereby pinpointing the regions most affected by pregnancy. Next, to further optimize the window set, a redundancy removal step is performed. By evaluating the correlation coefficients between windows, if their correlation coefficients meet the "high correlation" criterion in this calculation method, only one window is retained, excluding highly similar or redundant windows. This ensures that each retained window carries highly specific biological information, increasing the density of detection information.

[0048] The total window set obtained through the above steps can accurately locate genetic features closely related to pregnancy, which are key to calculating fetal concentration. This application combines the powerful predictive capabilities of the Extreme Gradient Boosting (XGBoost) algorithm to avoid interference from invalid information, reduce computational complexity, and ensure data integrity. The fetal concentration prediction model constructed in this application can accurately calculate fetal concentration even with low-depth sequencing (sequencing depth of 0.1~0.4×) and low sample concentration (less than 4%). Compared with existing technologies, this improves the accuracy of fetal concentration calculation and provides technical support for the further development of the NIPT field.

[0049] In a preferred embodiment, the sequencing files of the first sample and the second sample are sequencing files obtained by low-depth sequencing of the DNA of the first sample and the DNA of the second sample. Preferably, the low-depth sequencing is whole-genome low-depth sequencing using SE single-end sequencing.

[0050] This application only requires SE single-end whole genome low-depth sequencing data to perform model training and prediction, reducing the sequencing cost of NIPT, and can still achieve accurate detection in the face of extreme samples, thus reducing the detection cost of NIPT and improving the accuracy of detection results.

[0051] In a preferred embodiment, in S1), the continuous division of regions according to the physical location of chromosomes includes dividing the autosomes into continuous regions in units of 2 to 5M, wherein the windows obtained from the division are adjacent and do not overlap; preferably, the autosomes do not include chromosomes 13, 18 and 21.

[0052] In a preferred embodiment, the method for calculating statistical differences includes a t-test; preferably, a statistical difference is defined as a p-value less than 0.05.

[0053] In a preferred embodiment, the number of first samples is at least 100, the number of second samples is at least 100, and the number of training set samples is at least 100. Preferably, the principal components are the top 10-40 principal components. Preferably, the correlation coefficient includes, but is not limited to, Pearson correlation coefficient, Spearman rank correlation coefficient, or point-bivariate correlation coefficient. Those skilled in the art can choose any correlation calculation method to calculate the correlation coefficient between the two windows of the pairwise comparisons mentioned above, and this application does not impose any restrictions. The calculated correlation coefficients include, but are not limited to, Pearson correlation coefficient, Spearman rank correlation coefficient, or point-bivariate correlation coefficient, and the threshold for judging high correlation is 0.95 for all of them.

[0054] In a second typical embodiment of this application, a method for calculating fetal concentration is provided. The method includes: constructing a fetal concentration prediction model using the above-described method for constructing a fetal concentration prediction model; obtaining a vector of the sample to be tested based on 10 principal components; and inputting the vector into the fetal concentration prediction model to obtain the fetal concentration of the sample to be tested.

[0055] This application ingeniously integrates fragment distribution characteristics with the Extreme Gradient Boosting (XGBoost) algorithm when constructing a predictive model for fetal concentration. This enables the model to detect samples at low data depth (0.4×). Even when the concentration of fetal cell-free DNA is at a low level (less than 4%) that is difficult to measure accurately using traditional methods, the correlation coefficient calculated using the prediction method of this application can still be stably maintained above 0.8. This result proves the high sensitivity and accuracy of the fetal concentration calculation method of this application.

[0056] Furthermore, this application only requires SE single-end whole-genome low-depth sequencing data, avoiding reliance on high-cost PE paired-end sequencing or specific chromosome detection data, as well as the intervention of additional experimental steps. This not only simplifies the detection process but also reduces the overall economic burden. Moreover, it is highly compatible with existing NIPT analysis workflows and can be easily integrated without complex system adjustments, which can further enhance the practicality and promotion potential of NIPT technology.

[0057] In a third typical embodiment of this application, an electronic device for constructing a fetal concentration prediction model is provided. A schematic diagram of the electronic device for constructing the fetal concentration prediction model is shown below. Figure 2 As shown, the electronic device includes: a window set acquisition unit 01, a total window set acquisition unit 02, a training set sample vector calculation unit 03, and a model training unit 04;

[0058] The window set acquisition unit is used to divide the sequencing files of the first sample and the sequencing files of the second sample into continuous regions according to the physical location of chromosomes, respectively, to obtain the first sample window set and the second sample window set.

[0059] The total window set acquisition unit includes a window set correction unit and a judgment unit;

[0060] The correction window set unit is used to compare the distribution of read counts of windows with the same chromosome physical location in the first sample window set and the second sample window set, retain windows with statistical differences, and obtain the correction window set.

[0061] The judgment unit is used to compare the distribution of the number of read segments corresponding to each window in the correction window set pairwise to determine the correlation. If a high correlation is found, only one window is retained to obtain the total window set; high correlation is defined as a correlation coefficient ≥ 95%.

[0062] The training set sample vector calculation unit is used to construct a vector matrix of training set samples and total window set, extract principal components, and then extract the principal components corresponding to the number of read segments of each training set sample in the total window set to obtain the training set sample vector corresponding to each training set sample.

[0063] The model training unit is used to train the model by inputting the vectors of all training set samples into the model using XGBoost to obtain the fetal concentration prediction model. The first sample is a plasma sample from a non-pregnant woman; the second sample is a peripheral blood sample from a pregnant woman with a male singleton pregnancy; and the training set samples are peripheral blood samples from pregnant women.

[0064] In a preferred embodiment, the sequencing files of the first sample and the second sample are sequencing files obtained by low-depth sequencing of the DNA of the first sample and the DNA of the second sample.

[0065] In a preferred embodiment, the window set acquisition unit includes a chromosome window partitioning unit;

[0066] The chromosome window division unit includes continuous division of regions based on the physical location of chromosomes, using units of 2-5M to continuously divide autosomes into regions, with the resulting windows being adjacent and non-overlapping. Preferably, autosomes do not include chromosomes 13, 18, and 21. Preferably, the statistical difference calculation method includes a t-test. Preferably, a statistical difference is defined as a p-value less than 0.05. Preferably, the number of the first sample is at least 100, the number of the second sample is at least 100, and the number of training set samples is at least 100. Preferably, the principal components are the top 10-40 principal components. Preferably, the correlation coefficient includes Pearson correlation coefficient, Spearman rank correlation coefficient, or dot-bivariate correlation coefficient.

[0067] In a fourth typical embodiment of this application, an electronic device for calculating fetal concentration is provided, the electronic device comprising: a fetal concentration prediction model building unit and a calculation unit;

[0068] The fetal concentration prediction model building unit is used to build a fetal concentration prediction model using the above-mentioned fetal concentration prediction model building method; the calculation unit is used to obtain the vector of the sample to be tested based on 10 principal components, and input the vector into the fetal concentration prediction model to obtain the fetal concentration of the sample to be tested.

[0069] In a fifth typical embodiment of this application, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the above-described method for constructing a fetal concentration prediction model or the above-described method for calculating fetal concentration by executing the executable instructions.

[0070] In a sixth typical embodiment of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the above-described method for constructing the fetal concentration prediction model or the above-described method for calculating fetal concentration.

[0071] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to this application.

[0072] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus hardware devices such as detection devices. Based on this understanding, the data processing part of the technical solution of this application can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of the embodiments of this application.

[0073] This application can be used in a wide range of general-purpose or special-purpose computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices.

[0074] The method provided in this application can be executed on a terminal, computer terminal, or similar computing device. Taking running on a terminal as an example, Figure 3 This is a hardware structure block diagram of the computational model for the fetal concentration prediction model construction method according to an embodiment of the present invention. For example... Figure 3 As shown, a terminal may include one or more ( Figure 3Only one is shown in the diagram. A processor A1 (processor A1 may include, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA), etc.) and a memory B1 for storing data are also shown. Optionally, the terminal may further include a transmission device C1 for communication functions and an input / output device D1. Those skilled in the art will understand that... Figure 3 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are more... Figure 3 The more or fewer components shown, or having the same Figure 3 The different configurations shown.

[0075] Memory B1 can be used to store computer programs, such as application software programs and modules, like the computer programs corresponding to the methods of segment concatenation, clustering, and consistency processing in this embodiment of the invention. Processor A1 executes various functional applications and data processing by running the computer programs stored in memory B1, thereby implementing the methods described above. Memory B1 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, memory B1 may further include memory remotely located relative to processor A1, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0076] Transmission device C1 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the terminal's communication provider. In one example, transmission device C1 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, transmission device C1 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0077] Obviously, those skilled in the art should understand that some modules or steps of this application described above can be implemented in general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps into a single integrated circuit module. Thus, this application is not limited to any particular hardware and software combination.

[0078] The beneficial effects of this application will be explained in more detail below with reference to specific embodiments.

[0079] Example 1

[0080] I. Sample Processing

[0081] 1. Data: First sample: Plasma samples from 300 non-pregnant women;

[0082] Second sample: peripheral blood samples from 300 pregnant women with singleton male fetuses and a gestational age of 12 weeks + 0 days or more.

[0083] DNA was extracted from plasma samples, libraries were constructed, and sequencing was performed on a sequencer to obtain FASTQ files with an SE50 value and a single sample data size of ≥3.5M.

[0084] The FASTQ file is aligned with the human reference genome to obtain the original BAM file. The original BAM file is then deduplicated of repetitive sequences and the uniquely aligned reads are extracted to obtain the final processed BAM file.

[0085] 2. Chromosome screening: Humans have 22 autosomes and 2 sex chromosomes. Since the number of sex chromosomes differs between males and females, including them in the calculation would be difficult to perform. Furthermore, there are many variations on the three autosomes, chromosomes 13, 18, and 21, which may interfere with the accuracy of fetal concentration calculation. Therefore, only the remaining 19 chromosomes are considered.

[0086] 3. Divide the remaining 19 chromosomes into windows, with each 2M region divided into a window, and adjacent windows having no overlapping areas.

[0087] The proportion of N bases in the reference genome sequence corresponding to each window is counted. If the proportion of N bases is greater than 10%, it is filtered out, and the remaining windows are used for the next calculation.

[0088] The final BAM file for each sample is used to count the number of read segments falling into each window, as well as the GC content of each window;

[0089] The `loess` function in R is used to perform GC content correction on the number of read segments in each window, resulting in the number of read segments in each window after GC content correction.

[0090] The total number of read segments with GC content across all windows for each sample is calculated, along with the proportion of read segments with GC content in each window to the total. The number of read segments with GC content in each window is then corrected for data volume based on a total read segment count of 40M, resulting in the corrected number of read segments for each window.

[0091] After the above-mentioned routine data quality control, the first sample window set and the second sample window set are obtained.

[0092] II. Filtering area.

[0093] The first step is to screen the two groups of samples and perform a t-test on the distribution of the number of read segments after data volume correction in each window to see if there is a statistically significant difference. If there is a statistically significant difference, the window is retained; otherwise, it is filtered out to obtain the corrected window set.

[0094] The second step involves screening the samples from group A (300 non-pregnant women) by counting the number of reads in each window after data volume correction. Pearson coefficients are calculated for each pair of windows. If the Pearson coefficient is greater than 0.95, only one window is retained. The Pearson coefficient is a statistical indicator that measures the degree of linear correlation between two variables.

[0095] Based on the above filtering, a total of 2347 windows were obtained, which is the total window set.

[0096] Example 2

[0097] I. Training Reference Set.

[0098] Single-end SE sequencing was performed on 300 male singleton fetuses at a sequencing depth of 0.4×.

[0099] The fetal concentration of 300 male fetuses was calculated based on the Y chromosome as the gold standard, and the next step of training is planned.

[0100] The sequencing data FASTQ file is aligned to the human reference genome to obtain the original BAM file. The original BAM file is then deduplicated of repetitive sequences and the uniquely aligned reads are extracted to obtain the final processed BAM file.

[0101] The number of read segments in 2347 windows for 300 samples was counted, and GC content correction and data volume correction were performed.

[0102] A matrix was constructed using the number of reads from 300 samples × 2347 windows after data volume correction. Orthogonal decomposition was performed to extract the top 10 principal components. The values ​​of the 10 principal components from the 300 samples were calculated. The concentration of the Y chromosome was used as the expected value, and the values ​​of the 10 principal components were used as the observed values. These values ​​were then input into the XGBoost model for training to obtain the fetal concentration prediction model.

[0103] II. Calculation of fetal concentration

[0104] The male fetal sample to be tested (with a known fetal concentration of 14.2%) was subjected to single-end SE sequencing at a sequencing depth of 0.4×.

[0105] The sequencing data of the sample is FASTQ file. The FASTQ file is aligned with the human reference genome to obtain the original BAM file. The original BAM file is then deduplicated of repetitive sequences and the uniquely aligned reads are extracted to obtain the final processed BAM file.

[0106] The number of read segments in the sample across 2347 windows was counted, and GC content correction and data volume correction were performed to obtain the number of read segments in each window after data volume correction.

[0107] Using the number of reads corrected for data volume in each window, the values ​​of the sample on the 10 principal components are calculated, forming a vector of 10 values.

[0108] Inputting the vector into the XGBoost model for fetal concentration prediction yielded a fetal concentration of 14.3% for this sample, which is almost identical to the gold standard.

[0109] Example 3

[0110] Single-end SE sequencing was performed on 300 male fetal samples at a sequencing depth of 0.4×.

[0111] Sequencing data from 300 samples were processed into FASTQ files. These FASTQ files were then aligned to the human reference genome to obtain the original BAM files. Repeated sequences were removed from the original BAM files, and uniquely aligned reads were extracted to obtain the final processed BAM files.

[0112] The concentration of Y chromosomes in 300 samples was calculated as the gold standard;

[0113] The number of read segments in 2347 windows for 300 samples was counted, and GC content correction and data volume correction were performed to obtain the number of read segments in each window after data volume correction.

[0114] Using the number of reads corrected for data volume in each window, the values ​​of the 10 principal components corresponding to the sample are calculated, forming a vector of 10 values.

[0115] The vector was input into the fetal concentration prediction model to obtain the fetal concentration of 300 samples, which were then compared with the known fetal concentrations to calculate the correlation coefficient.

[0116] The correlation coefficients of the same sample calculated by different software are shown in Table 1.

[0117] Table 1

[0118]

[0119] Example 4

[0120] Peripheral blood samples were collected from women with 10 male singleton pregnancies. Based on the Y concentration, the concentration in all 10 samples was less than 4%. Using the method described in this application embodiment for concentration detection, the correlation coefficient reached 0.854, as shown in the results. Figure 4 As shown, Figure 4 The horizontal axis “Y_ff” refers to the sex chromosome concentration (gold standard), and the vertical axis “cal_ff” refers to the fetal concentration calculated by the method of this application.

[0121] As can be seen from the above description, the embodiments of the present invention achieve the following technical effects: This application provides a method and apparatus for constructing a fetal concentration prediction model based on fragment distribution characteristics and the XGBoost algorithm. The constructed prediction model can calculate fetal concentration at extremely low sequencing depths (0.4×), and maintains high accuracy even when the fetal concentration is less than 4%, with a correlation coefficient exceeding 0.8. The fetal concentration calculation method of this application avoids the need for additional experimental data, reduces the cost of NIPT testing, simplifies the operation process, and improves testing efficiency, bringing progress to non-invasive prenatal diagnosis of genetic diseases, and has scientific significance and broad application prospects.

[0122] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a fetal concentration prediction model, characterized in that, The construction method includes: S1) Divide the sequencing files of the first sample and the sequencing files of the second sample into continuous regions according to the physical location of chromosomes to obtain the first sample window set and the second sample window set; S2) Compare the distribution of read counts of windows with consistent chromosome physical locations in the first sample window set and the second sample window set, retain windows with statistical differences, and obtain a correction window set; The correlation between the number of read segments corresponding to each window in the correction window set is determined by pairwise comparison. If a high correlation is found, only one window is retained to obtain the total window set. The term "high correlation" refers to a correlation coefficient ≥ 95%. S3) Construct a vector matrix of training set samples and the total window set, extract principal components, and then extract the principal components corresponding to the number of read segments for each training set sample in the total window set to obtain the training set sample vector corresponding to each training set sample. S4) Using XGBoost, input the vectors of all the training set samples into the model for training to obtain the fetal concentration prediction model; The first sample is a plasma sample from a non-pregnant woman; The second sample was a peripheral blood sample from a pregnant woman with a singleton male fetus; The training set samples are peripheral blood samples from pregnant women.

2. The construction method according to claim 1, characterized in that, The sequencing files of the first sample and the second sample are sequencing files obtained by low-depth sequencing of the DNA of the first sample and the DNA of the second sample.

3. The construction method according to claim 1, characterized in that, In S1), the continuous division of regions according to the physical location of chromosomes includes dividing autosomes into continuous regions in units of 2 to 5M, wherein the windows obtained by the division are adjacent and do not overlap. Preferably, the autosomes do not include chromosomes 13, 18, and 21.

4. The construction method according to claim 1, characterized in that, The methods for calculating the statistical differences include the t-test; Preferably, a statistically significant difference is defined as a p-value less than 0.

05.

5. The construction method according to claim 1, characterized in that, The number of the first sample is at least 100, the number of the second sample is at least 100, and the number of training set samples is at least 100; Preferably, the principal components are the principal components arranged in the first 10 to 40 positions; Preferably, the correlation coefficient includes Pearson correlation coefficient, Spearman rank correlation coefficient, or point-bicol correlation coefficient.

6. A method for calculating fetal concentration, characterized in that, The method includes: constructing the fetal concentration prediction model using the method for constructing the fetal concentration prediction model according to any one of claims 1 to 5. Based on the principal components, the vector of the sample to be tested is obtained, and the vector is input into the fetal concentration prediction model to obtain the fetal concentration of the sample to be tested.

7. An electronic device for constructing a fetal concentration prediction model, characterized in that, The electronic device includes: a window set acquisition unit, a total window set acquisition unit, a training set sample vector calculation unit, and a model training unit; The window set acquisition unit is used to divide the sequencing files of the first sample and the sequencing files of the second sample into continuous regions according to the physical location of chromosomes, respectively, to obtain the first sample window set and the second sample window set. The total window set acquisition unit includes a window set correction unit and a judgment unit; The correction window set unit is used to compare the distribution of read counts of windows with the same chromosome physical location in the first sample window set and the second sample window set, retain windows with statistical differences, and obtain the correction window set. The judgment unit is used to compare the distribution of the number of read segments corresponding to each window in the correction window set pairwise to determine the correlation. If a high correlation is found, only one window is retained to obtain the total window set. The term "high correlation" refers to a correlation coefficient ≥ 95%. The training set sample vector calculation unit is used to construct a vector matrix of training set samples and the total window set, extract principal components, and then extract the principal components corresponding to the number of read segments of each training set sample in the total window set to obtain the training set sample vector corresponding to each training set sample. The model training unit is used to train the model by inputting the vectors of all the training set samples into the model using XGBoost, so as to obtain the fetal concentration prediction model. The first sample is a plasma sample from a non-pregnant woman; The second sample was a peripheral blood sample from a pregnant woman with a singleton male fetus; The training set samples are peripheral blood samples from pregnant women.

8. The electronic device according to claim 7, characterized in that, The sequencing files of the first sample and the second sample are sequencing files obtained by low-depth sequencing of the DNA of the first sample and the DNA of the second sample.

9. The electronic device according to claim 7, characterized in that, The window set acquisition unit includes a chromosome window partitioning unit; The chromosome window division unit includes the continuous division of regions according to the physical location of chromosomes, including dividing autosomes into continuous regions with 2 to 5M as a unit, and the windows obtained by the division are adjacent and do not overlap. Preferably, the autosomes do not include chromosomes 13, 18, and 21; Preferably, the method for calculating the statistical difference includes a t-test; Preferably, a statistically significant difference is defined as a p-value less than 0.05; Preferably, the number of the first sample is at least 100, the number of the second sample is at least 100, and the number of training set samples is at least 100. Preferably, the principal components are the principal components arranged in the first 10 to 40 positions; Preferably, the correlation coefficient includes Pearson correlation coefficient, Spearman rank correlation coefficient, or point-bicol correlation coefficient.

10. An electronic device for calculating fetal concentration, characterized in that, The electronic device includes: a unit for constructing a fetal concentration prediction model and a calculation unit; The fetal concentration prediction model construction unit is used to construct the fetal concentration prediction model using the fetal concentration prediction model construction method according to any one of claims 1 to 5. The calculation unit is used to obtain the vector of the sample to be tested based on the principal components, and input the vector into the fetal concentration prediction model to obtain the fetal concentration of the sample to be tested.

11. An electronic device, characterized in that, The electronic device includes: A processor; and a memory for storing executable instructions of the processor; The processor is configured to execute the method for constructing the fetal concentration prediction model according to any one of claims 1 to 5, or the method for calculating fetal concentration according to claim 6, by executing the executable instructions.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the method for constructing the fetal concentration prediction model according to any one of claims 1 to 5, or the method for calculating fetal concentration according to claim 6.

Images