Method and device for determining fetal concentration
Through reference genome alignment and nucleosome center position processing, the problem of differences in fetal concentration assessment on the third-generation sequencing platform was solved, and the applicability and accuracy of various sequencing platforms and genders were achieved.
Patent Information
- Application Number
- CN202411323873.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-30
- Publication Date
- 2024-11-08
AI Technical Summary
The data characteristics of the third-generation sequencing technology are different from the second-generation sequencing technology. The existing fetal concentration evaluation method for second-generation sequencing is not suitable for third-generation sequencing platforms, especially in the evaluation of the gender of female fetus.
By obtaining the sequencing sequence of cfDNA samples and the reference genome, nucleosome center scores are calculated, the nucleosome center locations are screened, the nucleosome center locations are aligned and added, and the dimension reduction and normalization are performed to determine the fetal concentration, which is suitable for various sequencing Platform and gender assessment.
It improves the accuracy of fetal concentration evaluation and is suitable for single-molecular sequencing platforms, making up for the shortcomings of the third-generation sequencing platform for fetal concentration evaluation of female fetal concentration.
Smart Images

Figure CN118918956A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of bioinformatics, and more specifically to a method and apparatus for determining fetal concentration. Background Technology
[0002] In non-invasive prenatal testing (NIPT), fetal concentration refers to the proportion of fetal deoxyribonucleic acid (DNA) in the pregnant woman's peripheral blood cell-free DNA (cfDNA). Fetal concentration is an important parameter in NIPT.
[0003] For second-generation sequencing (NGS) technology, there are already methods for assessing fetal concentration, such as using the distribution characteristics of cfDNA length, methylation characteristics, SNP (single nucleotide polymorphism) characteristics, and the XY chromosome content of male fetuses to determine fetal concentration. Third-generation sequencing (NGS) technology is a single-molecule sequencing technology, and its sequencing principle differs from that of NGS. The data characteristics generated by this sequencing differ significantly from those of NGS. Therefore, the fetal concentration assessment methods mentioned above for NGS are not applicable to NGS platforms. Summary of the Invention
[0004] The sequencing principles of third-generation sequencing (NGS) differ from those of second-generation sequencing (NGS), resulting in data characteristics that also differ from those obtained from NGS. These differences lead to variations in their applications. For example, NGS platforms cannot measure insert sequence length, thus making it impossible to determine fetal concentration based on cfDNA length distribution. Furthermore, NGS platforms cannot measure methylation information, rendering methods that extract methylation features unsuitable for determining fetal concentration. Additionally, the differences in the sequencing principles of NGS platforms result in different error characteristics compared to NGS, making the SNP method, applicable to NGS, unsuitable for determining fetal concentration on NGS platforms. Determining fetal concentration using XY chromosome content is only applicable to male fetuses and is also unsuitable for female fetuses. Therefore, based on these considerations, there is a need to develop a fetal concentration assessment method that is compatible with various sequencing platforms and applicable to fetuses of different sexes.
[0005] In view of the above, this application provides the following technical solution:
[0006] A method for determining fetal concentration, comprising:
[0007] Obtain the sequencing sequence of the cfDNA sample to be tested;
[0008] The sequencing sequence is compared with a reference genome to obtain the comparison results, including:
[0009] The count of each base site in the reference genome that is the starting position of the sequencing sequence alignment is determined as the second alignment number, and based on the second alignment number, the nucleosome center fraction corresponding to each base site in the reference genome is calculated;
[0010] The nucleosome center position is determined based on the nucleosome center score and the center score screening threshold.
[0011] Based on the nucleosome center position, the second logarithms of corresponding positions in all nucleosome regions are aligned and summed to obtain the summed second logarithm.
[0012] The second logarithm after summation is subjected to dimensionality reduction processing to obtain the dimensionality-reduced normalized second logarithm, and the dimensionality-reduced normalized second logarithm is determined as the alignment result;
[0013] Based on the comparison results, the fetal concentration was determined.
[0014] A device for determining fetal concentration, comprising:
[0015] The first acquisition unit is used to acquire the sequencing sequence of the cfDNA sample to be tested;
[0016] The first alignment unit is used to align the sequencing sequence with a reference genome to obtain alignment results, including:
[0017] The count of each base site in the reference genome that is the starting position of the sequencing sequence alignment is determined as the second alignment number, and based on the second alignment number, the nucleosome center fraction corresponding to each base site in the reference genome is calculated;
[0018] The nucleosome center position is determined based on the nucleosome center score and the center score screening threshold.
[0019] Based on the nucleosome center position, the second logarithms of corresponding positions in all nucleosome regions are aligned and summed to obtain the summed second logarithm.
[0020] The second logarithm after summation is subjected to dimensionality reduction processing to obtain the dimensionality-reduced normalized second logarithm, and the dimensionality-reduced normalized second logarithm is determined as the alignment result;
[0021] The first determining unit is used to determine the fetal concentration based on the comparison results.
[0022] This application utilizes a method to determine fetal concentration based on the difference in the distribution of the two ends of maternal and fetal cfDNA on nucleosomes. This method can be applied to various sequencing platforms, including single-molecule sequencing platforms, thus improving the accuracy of fetal concentration determination. Furthermore, the fetal concentration determination method provided in this application does not rely on sex-dependent characteristics for identification and confirmation; therefore, it is applicable to the determination of fetal concentration in both male and female fetuses, overcoming the limitation of third-generation sequencing platforms in effectively assessing fetal concentration in female fetuses. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0024] Figure 1 A schematic flowchart illustrating a method for determining fetal concentration provided in an embodiment of this application;
[0025] Figure 2 A schematic diagram illustrating the relationship between fetal concentration calculated using the Enet model and a reference fetal concentration, provided for an embodiment of this application;
[0026] Figure 3 A schematic diagram illustrating the relationship between fetal concentration calculated by a WRSC model and a reference fetal concentration, provided for an embodiment of this application;
[0027] Figure 4 A schematic diagram illustrating the relationship between fetal concentration calculated from the original gmFF_V7 parameters and a reference fetal concentration, provided for an embodiment of this application;
[0028] Figure 5 A schematic diagram illustrating the relationship between the fetal concentration calculated by linear correction of gmFF_V7 and the reference fetal concentration, provided for an embodiment of this application;
[0029] Figure 6 A schematic diagram illustrating the distribution of a set of data obtained by aligning and summing them according to the nucleosome regions, as provided in an embodiment of this application;
[0030] Figure 7 A schematic diagram illustrating the prediction performance of a model after randomly splitting the test set and training set, as provided in an embodiment of this application;
[0031] Figure 8 A flowchart illustrating another method for determining fetal concentration provided in an embodiment of this application;
[0032] Figure 9A schematic diagram of a device for determining fetal concentration provided in an embodiment of this application;
[0033] Figure 10 This is a schematic diagram of another device for determining fetal concentration provided in an embodiment of this application. DETAILED DESCRIPTION
[0034] The technical solutions of this application will now be clearly and completely described with reference to the embodiments and accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0035] The terms "first" and "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order, and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. For example, the expressions "first comparison number" and "second comparison number" in this application can be interchanged without departing from the scope of the embodiments of this application.
[0036] The terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may include steps or units not listed.
[0037] The term "sequencing" can also be called "nucleic acid sequencing" or "gene sequencing," meaning the three terms are interchangeable. It refers to the determination of the base types and sequence in a nucleic acid sequence. It includes sequencing by synthesis (sequencing by synthesis, SBS) and / or sequencing by ligation (sequencing by ligation, SBL), including DNA sequencing and / or RNA sequencing, including long fragment sequencing and / or short fragment sequencing. The terms "long fragment" and "short fragment" are relative. For example, nucleic acid molecules longer than 1Kb, 2Kb, 5Kb, or 10Kb can be called long fragments, while those shorter than 1Kb or 800bp can be called short fragments.
[0038] The term "single-molecule sequencing technology" refers to sequencing technology based on a single copy of a base signal acquisition unit. This means it allows for direct reading of nucleic acid molecule sequences without the need for specific amplification techniques to amplify the signal. The term "single-molecule sequencing platform" refers to sequencing platforms based on single-molecule sequencing technology, including various sequencers based on this technology, such as Genocare sequencing platform from Genocare Biotech.
[0039] This application provides a method for determining fetal concentration. This method can be applied to various sequencing platforms to assess fetal concentration. In particular, this method can be used to establish a fetal concentration assessment method based on a single-molecule sequencing platform, and is especially suitable for assessing fetal concentration in female fetuses.
[0040] See also Figure 1 This is a flowchart illustrating a method for determining fetal concentration according to an embodiment of this application. The method may include the following steps:
[0041] S101. Obtain the sequencing sequence of the cfDNA sample to be tested.
[0042] Here, cfDNA refers to highly fragmented DNA that exists outside cells and is circulating in the human bloodstream. In this embodiment of the application, the cfDNA sample to be tested refers to a cfDNA sample for which the fetal concentration needs to be determined. For example, the cfDNA sample can be peripheral blood of a pregnant woman containing free deoxyribonucleic acid (or free DNA).
[0043] In this embodiment, the cfDNA sample to be tested refers to a sample containing cfDNA whose fetal concentration needs to be determined. The sequencing sequence of the cfDNA sample to be tested refers to the sequence obtained by sequencing the cfDNA in the sample, i.e., the sequencing sequence of fragmented DNA in the sample. It should be understood that the sequencing sequence of the cfDNA sample to be tested has no requirements on the sequencing instrument and sequencing platform; that is, the sequencing sequence of the cfDNA sample to be tested can be obtained from sequencing data obtained from any sequencing platform. In some embodiments, the sequencing sequence of the cfDNA sample to be tested can be obtained by a single-molecule sequencing platform (such as Genocare sequencing platform or nanopore sequencing platform from Genocare Biotech) or by a second sequencing platform (such as the Illumina platform or BGI platform for next-generation sequencing). For example, the sequencing sequence of the cfDNA sample to be tested is data obtained by DNA sequencing of the cfDNA sample to be tested based on a single-molecule sequencing platform. Thus, the sequencing sequence of the cfDNA sample to be tested derived from a single-molecule sequencing platform can be obtained through the method provided in this embodiment, thereby enabling the analysis of the fetal concentration of the cfDNA sample to be tested.
[0044] S102. Align the sequencing sequence with the reference genome to obtain the alignment results.
[0045] In this step, the reference genome refers to the human reference genome, for example, such as the human reference genome hg19, but not limited to this.
[0046] S103. Based on the comparison results, determine the fetal concentration.
[0047] In this embodiment of the application, in order to accurately obtain the final fetal concentration, the sequencing sequence is compared with the reference genome. The comparison result can be the comparison probability of the maternal and infant cfDNA in different positions on the reference genome, the comparison result corresponding to the difference in the distribution of the two ends of the maternal and infant cfDNA on the nucleosome, the comparison result corresponding to the comparison probability of the maternal and infant cfDNA in different positions on the reference genome and the difference in the distribution of the two ends of the maternal and infant cfDNA on the nucleosome, or the comparison result newly formed after processing the above two comparison results.
[0048] The method for determining fetal concentration provided in the embodiments of this application will be described in detail below.
[0049] The fetal concentration was determined by the difference in alignment probability between maternal and infant cfDNA at different locations on the reference genome.
[0050] In cell-free DNA (cfDNA) obtained from maternal peripheral blood, part comes from the mother and part comes from the fetus. When the DNA from the mother and the fetus is compared with the reference genome, there will be certain differences in the coverage depth of different regions of the reference genome. Therefore, the fetal concentration can be determined based on the comparison results corresponding to this difference.
[0051] In one embodiment of this application, the fetal concentration is determined by the difference in alignment probability between maternal and infant cfDNA at different locations on a reference genome. Specifically, the method for determining fetal concentration includes:
[0052] S111. Obtain the sequencing sequence of the cfDNA sample to be tested.
[0053] In this step, the cfDNA sample to be tested is the cfDNA sample for which the fetal concentration needs to be determined. The method provided in this application does not specify requirements regarding the source of the cfDNA sample to be tested or its characteristic parameters.
[0054] S112. Align the sequencing sequence with the reference genome to obtain the alignment results.
[0055] This step includes:
[0056] S1121. The reference genome is segmented to obtain multiple segment intervals.
[0057] S1122. Determine the first alignment number of each sequencing sequence falling into each of the said segment intervals, such that the first alignment number is determined as the alignment result.
[0058] It should be noted that the "first" in the first alignment number is only used to distinguish it from subsequent alignment numbers; there is no sequential relationship between "first" and "second". This method divides the reference genome into segments of fixed length and then counts the number of sequencing sequences falling into each segment interval. That is, it calculates how many sequencing sequences fall into the corresponding segment interval. The number of sequences falling into each segment interval is the first alignment number, which is also the alignment result for this scenario. Correspondingly, in some scenarios, the number of sequencing sequences falling into each segment interval can also be called the "coverage number." For ease of description, it is referred to as the alignment number in subsequent embodiments of this application, and to distinguish it from other types of alignment results, it is called the "first alignment number".
[0059] It should be understood that the "first alignment number" mentioned in the embodiments of this application can be the absolute number of sequencing sequences falling into the corresponding segment interval, or it can be the median of the sequencing sequences falling into the corresponding segment interval. In addition, during the statistical process, the number of sequencing sequences falling into each segment interval is counted, allowing a certain error tolerance ratio, that is, allowing sequencing sequences with a preset number of base errors within a certain length range to be included in the alignment.
[0060] In order to obtain accurate alignment results, the reference genome can be segmented into segments of a fixed length, such as determining the segment length of the reference genome based on the characteristics of the sequencing sequence.
[0061] The reference genome contains certain special chromosomes that may introduce bias or other unique characteristics during the determination of fetal concentrations through reference genome alignment. In one implementation, the bias introduced by these special chromosomes (hereinafter referred to as specific chromosomes) is reduced by removing segmented regions from the reference genome, thereby minimizing their impact on the test results. In some embodiments, specific chromosomes include at least one of the following: X chromosome, Y chromosome, mitochondrial chromosome, chromosome 13, chromosome 18, and chromosome 21. Specifically, removing segmented regions corresponding to the X and Y chromosomes avoids the impact of differences in these chromosomes between male and female fetuses on the results of subsequent concentration quantification models; chromosomes 13, 18, and 21 are removed because these chromosomes have a higher probability of duplication or deletion. One of the screening objectives of NIPT is to screen for duplications and deletions on chromosomes 13, 18, and 21. When chromosomes 13, 18, and 21 are used as reference chromosomes, biases may occur when normalizing the data, such as affecting the normalized logarithm, which in turn affects the calculation results and the accuracy of NIPT screening results.
[0062] The embodiments of this application can adjust the range of specific chromosomes according to the expected needs of fetal concentration detection, such as removing the X and Y chromosomes, or removing chromosomes 13, 18, and 21, or removing mitochondrial chromosomes. Of course, multiple situations mentioned above can be removed simultaneously, such as removing all X, Y, mitochondrial chromosomes, chromosomes 13, 18, and 21 or their corresponding segments, so that the segmented intervals obtained after segmenting the reference genome do not contain the segmented intervals corresponding to the aforementioned chromosomes.
[0063] This application allows for the removal of specific chromosomes into segmented intervals through various methods. The segmented intervals can be filtered based on specific chromosomes, or these specific chromosomes can be removed before segmentation. Specifically, the appropriate processing method can be selected based on actual application requirements; this application does not impose any restrictions on this.
[0064] In one implementation, the reference genome is segmented to obtain multiple segmented intervals, including: segmenting the reference genome based on a preset segmentation length to obtain multiple initial segmented intervals; and removing segmented intervals corresponding to specific chromosomes from the initial segmented intervals to obtain multiple segmented intervals. That is, after segmenting the reference genome into multiple segmented intervals, the segmented intervals corresponding to specific chromosomes in the reference genome are removed and not included in the calculation. It should be understood that the "segmented intervals corresponding to specific chromosomes" refers to the multiple segmented intervals formed after the specific chromosome is segmented according to the preset segmentation length.
[0065] In another implementation, the reference genome is segmented to obtain multiple segment intervals, including: removing specific chromosomes from the reference genome to obtain a removed reference genome; and segmenting the removed reference genome based on a preset segmentation length to obtain multiple segment intervals. That is, specific chromosomes are removed before segmenting the reference genome.
[0066] The preset segmentation length can be adjusted according to the amount of data processed and the expected relative accuracy. For example, the reference genome can be segmented into segments of 50k (5000) bases each. To make the obtained first alignment number more accurate, the first alignment number can be corrected to obtain a corrected first alignment number, which can then be used as the alignment result.
[0067] In some embodiments, correction includes GC correction. GC correction is an operation in the bioinformatics analysis process. GC correction is performed because gene sequencers and their corresponding sequencing procedures can introduce GC bias. For example, some sequencers are more likely to detect high-GC sequences, while others are more likely to detect low-GC sequences. In this embodiment, the determination of fetal concentration is based on the alignment probabilities of different regions on the reference genome. The GC bias of the sequencer will affect the judgment of the actual alignment probabilities of different regions; therefore, GC correction is necessary to improve accuracy.
[0068] At this point, the first logarithm is corrected to obtain the corrected first logarithm, which includes: normalizing the first logarithm of each segment interval based on the first logarithm of each segment interval and the average first logarithm of all segment intervals to obtain the normalized first logarithm of each segment interval; and performing GC correction on the normalized first logarithm of each segment interval to obtain the GC-corrected first logarithm. The average first logarithm of all segment intervals refers to the ratio of the sum of the first logarithms of all segment intervals to the total number of segment intervals.
[0069] After obtaining the first logarithm of all segmented intervals, to facilitate calculation and processing, the first logarithm of all segmented intervals will be normalized. The corresponding formula is:
[0070] Normalized first logarithm of each segment interval = First logarithm of each segment interval / Average first logarithm of all segment intervals.
[0071] After obtaining the first logarithm of each segment interval, GC correction will be performed to obtain the first logarithm after GC correction.
[0072] In one implementation, GC correction is performed on the normalized first logarithm for each segment interval to obtain the GC-corrected first logarithm, including:
[0073] (1) Generate the first relationship curve based on the GC content and the normalized first logarithm corresponding to each segment interval.
[0074] In this implementation, the GC content can be represented by various parameters, such as the GC content of each segment interval or the median GC content, or other parameters that can reflect the GC content of the segment interval.
[0075] In this embodiment of the application, GC filtering can be performed on the segmented intervals that form the first relationship curve, such as filtering the segmented intervals according to the GC content, to filter out some abnormal or statistically insignificant segments, thereby optimizing the first relationship curve and reducing the consumption of data processing resources during data processing.
[0076] In one implementation, GC filtering is achieved as follows: before generating a first relationship curve based on GC content and the normalized first logarithm corresponding to each segment interval, the following processing is performed: the segment intervals are filtered based on GC content to remove segment intervals whose GC content does not meet the preset requirements, so that the first relationship curve is generated based on the GC content of each segment interval after GC filtering and the normalized first logarithm of each segment interval.
[0077] In another implementation, GC filtering is achieved as follows: during the process of normalizing the first logarithm based on the first relationship curve, the segmented intervals are filtered based on the GC content, and the normalized first logarithm after GC filtering is GC corrected based on the first relationship curve to obtain the GC-corrected first logarithm.
[0078] In the above embodiments, as an example, a first relationship curve is generated based on the GC content and the normalized first logarithm corresponding to each segment interval. This is achieved as follows: the x-axis represents the GC content of each segment interval (i.e., the proportion of GC bases corresponding to that segment interval on the reference genome), and the y-axis represents the normalized first logarithm corresponding to each segment interval. Data on each segment interval and its GC content within that interval are integrated, i.e., a scatter plot is generated based on the GC content of each segment interval and its corresponding normalized first logarithm. The scatter plot data is smoothed to obtain a smoothed curve. Smoothing the scatter plot data can be performed using algorithms such as the LOWESS algorithm to obtain the smoothed curve, i.e., the first relationship curve, such as y... s =f(x) s ).
[0079] Specifically, as an example, the first relationship curve can be generated as follows: divide the horizontal axis into different segments, then calculate the median GC content of the scatter points within the same segment, use this median GC content to represent the value corresponding to that segment, and then plot all these GC content medians on a graph and smooth them to obtain the first relationship curve. In another example, smoothing can be performed directly based on all the data, that is, all data points are considered and smoothed together to obtain a smooth curve, thus obtaining a higher Spearman correlation coefficient for the test set. It should be noted that this smoothing method is also applicable to the smoothing and correction of the contrast probabilities described later.
[0080] (2) Based on the first curve, perform GC correction on the normalized first logarithm to obtain the GC-corrected first logarithm.
[0081] In some implementations, GC correction is performed based on a subtraction or division relationship between the normalized first logarithm and the GC content of the segmented intervals corresponding to the normalized first logarithm, and the GC-corrected first logarithm is determined.
[0082] In one example, the GC correction method is "subtraction," and the corresponding correction formula is: y GC = yf(x); In another example, the GC correction method is "division", and the corresponding correction formula is: y GC = y / f(x).
[0083] In the two correction formulas above, x is the GC content of the segmented interval, and y is the normalized first logarithm corresponding to the segmented interval. GC This is the normalized first logarithm after GC correction.
[0084] In some embodiments, the correction further includes alignment probability correction to improve the accuracy of the obtained first alignment count. That is, in some embodiments, the correction includes both GC correction and alignment probability correction.
[0085] In one embodiment of this application, correcting the first alignment number to obtain a corrected first alignment number further includes: truncating the reference genome based on a specific sliding window length, aligning the truncated sequence with the reference genome, and counting the first alignment number in each segment of the reference genome, determining the first alignment number as the sliding window alignment number; determining the normalized sliding window alignment number for each segment based on the sliding window alignment number; and performing alignment probability correction based on the normalized sliding window alignment number for each segment and the normalized first alignment number to determine the alignment probability corrected first alignment number.
[0086] This application employs an "alignment probability correction" method to correct the first alignment count, based on the following considerations: Different regions of the reference sequence corresponding to the reference genome, if truncated to a specific length and treated as a single sequencing sequence, will have varying alignment counts when aligned back to the reference genome. This is because for some regions, similar sequences exist in multiple locations on the reference genome, making these regions more likely to align. For example, the human genome reference sequence hg19 can be truncated using a sliding window with a step size of 2 bases and a window size of 37 bases. These sequences are then aligned with the human reference genome hg19, and the alignment counts for different segments of the human reference genome are counted. This alignment count can be called the "sliding window alignment count." Specifically, the specific sliding window length can be determined based on the actual application scenario, such as the characteristics of the reference genome used or the characteristics of the sample to be tested. After determining a specific sliding window length, the reference genome is truncated. The truncated sequence is then aligned with the reference genome, and the first alignment number for each segment of the reference genome is counted. This first alignment number is defined as the sliding window alignment number. To facilitate calculation and processing, a normalized sliding window alignment number is determined for each segment. Alignment probability correction is then applied to the normalized sliding window alignment number. The method for alignment probability correction can refer to the GC correction method described earlier.
[0087] In one implementation, the alignment probability is corrected based on the normalized sliding window alignment count and the normalized first alignment count for each segment interval, and the first alignment count after alignment probability correction is determined, including:
[0088] (1) Generate the second relationship curve based on the normalized sliding window logarithm of each segment interval and the normalized first logarithm corresponding to each segment interval.
[0089] The normalized sliding window logarithm for each segment interval can be represented by various parameters, such as the normalized sliding window logarithm for each segment interval or the median of the sliding window logarithm, or other parameters that can reflect the normalized sliding window logarithm for each segment interval.
[0090] In this embodiment of the application, the normalized sliding window logarithm that forms the second relationship curve can be filtered to remove some abnormal or statistically insignificant segments, thereby optimizing the second relationship curve and improving the processing accuracy when performing related data processing based on the second relationship curve.
[0091] In one implementation, the normalized sliding window logarithm used to form the second relationship curve is filtered by the following method: before generating the second relationship curve based on the normalized sliding window logarithm for each segment interval and the normalized first logarithm corresponding to each segment interval, the following processing is performed: the segment intervals are filtered based on the normalized sliding window logarithm to obtain segment intervals where the normalized sliding window logarithm is not less than a first target threshold, so that the second relationship curve is generated in segment intervals where the normalized sliding window logarithm is not less than the first target threshold.
[0092] In one implementation, the normalized sliding window logarithms forming the second relationship curve are filtered by means of the following: during the process of performing alignment probability correction on the normalized first logarithms based on the second relationship curve to obtain the alignment probability-corrected first logarithms, the segmented intervals are filtered based on the normalized sliding window logarithms, retaining the second relationship curve intervals in the second relationship curve where the normalized sliding window logarithms are not less than a first target threshold, so as to obtain the alignment probability-filtered second relationship curve; the alignment probability is corrected on the sliding window normalized logarithms corresponding to each segmented interval based on the alignment probability-filtered second relationship curve to obtain the alignment probability-corrected first logarithms.
[0093] In the above embodiments, based on repeated verification, the first target threshold can be selected as 0.8, that is, removing the segment intervals where the normalized sliding window logarithm is less than 0.8.
[0094] In the above embodiments, as an example, a second relationship curve is generated based on the normalized sliding window logarithm of each segment interval and the normalized first logarithm corresponding to each segment interval. This is achieved by: using the x-axis to represent the normalized sliding window logarithm of each segment interval and the y-axis to represent the normalized first logarithm corresponding to each segment interval; integrating the data of each segment interval and its corresponding GC content, i.e., generating a scatter plot based on the GC content of each segment interval and its corresponding normalized first logarithm; and smoothing the scatter data in the scatter plot to obtain a smooth curve. For example, the smoothing of the scatter data in the scatter plot can be achieved using the LOWESS algorithm.
[0095] (2) Based on the second relationship curve, the normalized first logarithm is corrected for alignment probability to obtain the first logarithm after alignment probability correction.
[0096] In some implementations, the alignment probability is corrected by subtraction or division of the sliding window alignments of the segmented intervals corresponding to the normalized first alignment number and the first alignment number, and the first alignment number after alignment probability correction is determined.
[0097] In one example, the alignment probability correction method is "subtraction," and the corresponding correction formula is: y比对概率 = yf(x); In another example, the alignment probability correction method is "division", and the corresponding correction formula is: y 比对概率 = y / f(x).
[0098] In the two correction formulas above, x is the normalized sliding window logarithm for each segmented interval, and y corresponds to the normalized first logarithm for each segmented interval. 比对概率 This is the normalized first logarithm after comparison probability correction.
[0099] S113. Based on the comparison results, determine the fetal concentration.
[0100] In this application embodiment, machine learning methods combining neural networks can be used to determine fetal concentration. In one embodiment of this application, determining fetal concentration based on comparison results includes:
[0101] Obtain first training sample data, where each sample in the first training sample data is labeled with a first feature value and a first target value, where the first feature value is the number of sample comparisons and the first target value is the actual fetal concentration of the sample.
[0102] Based on the results of a specific model, machine learning modeling is performed on the first training sample data to obtain the first fetal concentration quantification model;
[0103] The comparison results are input into the first fetal concentration quantification model to obtain the first fetal concentration, and the first fetal concentration is determined as the fetal concentration.
[0104] In some embodiments, machine learning modeling is performed on the first training sample data based on a specific model structure to obtain a first fetal concentration quantification model, including:
[0105] The first training sample data is divided into a training set and a test set; machine learning modeling is performed based on the training set to obtain the initial model;
[0106] The test set is processed based on the initial model to obtain the estimated fetal concentration for each test sample in the test set.
[0107] The comparison results are obtained by comparing the estimated fetal concentration with the actual fetal concentration of each test sample in the test set.
[0108] Based on the comparison results, the model parameters of the initial model were adjusted to obtain the first fetal concentration quantitative model.
[0109] The first training sample data can be data from a database of pregnant women's peripheral blood cfDNA with known fetal concentrations obtained from various sequencing platforms, including single-molecule sequencing platforms. For example, using a certain number of samples with known fetal concentrations, the reference genome chromosome reference sequence is segmented into intervals (e.g., by 5k bases, 50k bases, 100k bases, 300k bases, or 800k bases, etc.), and the alignment count of each sample in different segment intervals is counted. The Lowes method is used to normalize, GC, and correct the alignment probability of these alignment counts. The final normalized and corrected alignment counts in different segment intervals are used as feature values, and the known actual fetal concentration corresponding to each sample is used as the target value. A specific model is used for learning and modeling. The specific model can be a classification model, a convolutional neural network model, etc. This application embodiment does not limit the specific model structure, as long as it can learn from the training sample data and predict the fetal concentration. The model learning and training process is an iterative process. Therefore, the training samples can be divided into a training set and a test set. The accuracy of the model obtained by the current training is tested through the test set. If the accuracy is lower than the set accuracy threshold, the model parameters in the current model structure are adjusted through the test set until the error between the fetal concentration output by the model and the actual fetal concentration meets the corresponding error range.
[0110] When the number of samples used for machine learning is small, but the number of features is large (e.g., approximately 3,000 to 600,000 features, depending on the size of the chromosome reference sequence intervals), overfitting is unavoidable during both dimensionality reduction and learning processes. Consequently, the fetal concentration quantification model will show a certain deviation between its prediction of fetal concentration on the test set and the actual fetal concentration. For example, in a fetal concentration quantification model built using 615 samples with known fetal concentrations, the model corresponds to the test set R... 2 The Pearson correlation coefficient between the actual and predicted values is only around 0.3, and approximately 0.5. However, when the logarithm is normalized and corrected (GC correction, or GC and alignment probability correction) using the method described in the embodiments of this application, the correlation coefficient between the predicted and actual values of the test set is improved. Comparing modeling directly using normalized segmented interval logarithms, and modeling using normalized segmented interval logarithms corrected for GC and alignment probability, the modeling effect after normalization, GC correction, and alignment probability correction is better, i.e., the correlation coefficient between the predicted and actual values of the test set is higher.
[0111] In another implementation, determining the fetal concentration based on the comparison results includes:
[0112] The comparison results are input into the first preset model to obtain the initial fetal concentration;
[0113] The initial fetal concentration is corrected according to the second preset model to obtain the fetal concentration.
[0114] The first preset model is a linear relationship pattern established based on the alignment results in the second sample data and the initial fetal concentration corresponding to the alignment results. The second sample data is the alignment results obtained by comparing the sequencing sequence in the cfDNA sample with the reference genome and the fetal concentration corresponding to the alignment results. The second preset model is a model obtained by processing the first preset model with constants determined by the phenomenon fitting.
[0115] For example, the first preset model can be in, For comparison, FF1 represents the initial fetal concentration. The second preset model can be FF2 = FF1·c + d, where c and d are constants determined by linear fitting, and FF2 represents the fetal concentration.
[0116] For example, a model that uses the difference in alignment probability between samples in different regions of a reference genome to build a model and estimate fetal concentration is called SeqFF. SeqFF uses two machine learning models: Weighted Ranking Selection Criterion (WRSC) and Elastic Network (Enet). The final output fetal concentration result is the average of the predictions from the two models. When testing the SeqFF model with data from sequencing platforms, especially single-molecule sequencing platforms, the WRSC model outperforms Enet. Specifically, the linear correlation coefficient between the fetal concentration estimated by WRSC and the actual fetal concentration is higher than that of Enet or SeqFF. Therefore, the model parameters involved in the WRSC algorithm are preferred. For details, please refer to... Figure 2 and Figure 3 , Figure 2 This is a schematic diagram illustrating the relationship between fetal concentration calculated using the Enet model and a reference fetal concentration, provided in an embodiment of this application. Figure 3 This diagram illustrates the relationship between fetal concentration calculated by the WRSC model and a reference fetal concentration, as provided in an embodiment of this application. As can be seen from the diagram, the positive correlation between the fetal concentration calculated by the WRSC model and the reference fetal concentration is superior to the positive correlation between the fetal concentration calculated by the Enet model and the reference fetal concentration.
[0117] For example, in the WRSC model, the chromosome reference sequence is divided into intervals of a specific number of bases, such as 50k bases. Through training with the WRSC algorithm, a linear relationship between coverage density and fetal concentration in different reference sequence intervals was found, as shown in the following formula. In the following formula, FF represents the fetal concentration. To represent the coverage density in different regions of the reference sequence, To and A constant vector of equal dimensions, where B is the parameter matrix used for linear transformation, and c and d are constants.
[0118]
[0119] Because SeqFF is a model trained on second-generation sequencing data, its sequencing principle differs significantly from single-molecule sequencing. Therefore, directly using the SeqFF model parameters to calculate fetal concentrations from other sequencing platforms, including single-molecule sequencing, will result in a large discrepancy between the calculated and actual fetal concentrations. Using the 615 NIPT clinical samples with known fetal concentrations from the above examples for analysis and testing, we directly processed the raw alignment file data (e.g., SAM files) from other sequencing platforms, including single-molecule sequencing, using the SeqFF algorithm and parameters. The Pearson correlation coefficient between the obtained intermediate parameters and the actual fetal concentration was 0.714.
[0120] The calculated fetal concentration from the SeqFF model shows a good linear relationship with the actual fetal concentration; therefore, the formula above can be used. This section includes model parameters that calculate intermediate parameters of fetal concentration, and then uses linear regression to fit the values of c and d, thereby obtaining the final model parameters suitable for other sequencing platforms, including single-molecule sequencing platforms.
[0121] The analysis and testing were still conducted using 615 NIPT clinical samples with known fetal concentrations from the above examples. The analysis was modified to use only the first alignment count of the sequencing sequence falling within each segment interval; regions with abnormal alignment probabilities on the X, Y, M, 13, 18, 21 chromosomes and mitochondrial chromosomes in the reference genome were excluded; and the correction method (GC correction, or GC and alignment probability correction) was optimized. The data were re-analyzed using the optimized parameters, and the Pearson correlation coefficient between the intermediate parameters and fetal concentration was 0.755. 100 random splits of the test set (30%) and training set (70%) were used to build a linear model and determine the aforementioned constant parameters c and d. The average test set prediction R obtained from 100 modeling iterations was... 2 The value is 0.561.
[0122] It should be noted that when actually building a linear model, the splitting of the test set and training set is to evaluate the reliability of the model built on our currently randomly split training set in limited data. Therefore, after randomly splitting the test set (30%) and training set (70%) 100 times, the average R² value of the test set prediction obtained from 100 modeling runs was 0.561.
[0123] Table 1 below shows nine examples of determining fetal concentration based on the differences in alignment probabilities between maternal and infant cfDNA at different locations on the reference genome.
[0124] In Table 1, Example 1 shows the conditions used in the original seqFF WRSC method. In the table, "Bin" in "Bin screening" refers to the segmented intervals on the genomic reference sequence; "Mappability correction" refers to the "correction of alignment probability"; "Mappability" refers to the normalized alignment number of the sliding window method mentioned above; and "groupby" refers to a GC correction method in the seqFF method. Groupby is equivalent to performing Lowess smoothing by segmenting the GC content into intervals, with each segment represented by a GC content value retained to three decimal places, and the normalized statistic for that GC content represented by the median of the normalized statistics corresponding to all segmented intervals within that GC content range. This processing method is marked as "median" in the table. Correspondingly, the table also includes a method marked as "all". The normalization method is as follows: "All" means that the GC content of all segments and their corresponding normalized alignments are considered together for Lowess smoothing, without dividing the GC content into intervals; that is, data from all segments are used during smoothing. "Median" refers to the median, "Frac" refers to the parameter used for Lowess smoothing, "all mapped" refers to all aligned sequences, "unique" refers to the sequence that is uniquely aligned (i.e., the sequence that only aligned to one location on the chromosome), "XYM" or "XYM chromosome" refers to the X chromosome, Y chromosome, and mitochondrial chromosome, and "flag filtering" refers to the filtering performed on different segments in the original seqFF paper, for which a filtering file was provided. "seqFF bin flag filtering" is filtering according to the labels marked in the original literature. The scheme we ultimately selected did not follow the labels in this original literature for filtering.
[0125] Table 1
[0126]
[0127]
[0128] The Peel correlation coefficient assesses the linear correlation between the numerical value directly calculated by the algorithm and a reference fetal concentration. See also Figure 4 Although the two are highly linearly correlated, their absolute values differ significantly. Therefore, a linear model is needed to correct the algorithm's calculations, ensuring accurate estimation of fetal concentration. The aforementioned [model / method / mechanism] can be used to achieve this. The formula is used for fitting correction, in which... This whole section contains the numerical results calculated by gmFF_V7. c and d are two parameters fitted in the linear fit, which will be used as constants in the subsequent model.
[0129] Table 1 evaluates the merits of different treatment parameters and conditions by using the Pearson correlation coefficient between the predicted fetal concentration and the reference fetal concentration (fetal concentration provided by other platforms).
[0130] As shown in Table 1, compared to the algorithm in Example 1 (Pearson correlation coefficient of 0.71425), the Pearson correlation coefficients of all other algorithms are greater than 0.74, and the Pearson correlation coefficients of the algorithms in Examples 6, 8, and 9 are greater than 0.75. Compared to other algorithms, the algorithm in Example 8 achieves the highest Pearson correlation coefficient of 0.75471.
[0131] The graph showing the relationship between the calculated fetal concentration and the reference fetal concentration in Example 8 after correction is shown below. Figure 5 As shown. By Figure 5 It is evident that there is a strong positive correlation between the calculated fetal concentration and the reference fetal concentration.
[0132] The fetal concentration can be calculated by utilizing the difference in the distribution of the two ends of the cfDNA on the nucleosome between the mother and the baby.
[0133] Nucleosomes are the basic units that make up human chromatin. When DNA is cleaved by enzymes, cutting inside nucleosomes is more difficult than cutting outside the junctions between nucleosomes. Therefore, during digestion, DNA is more likely to be cleaved at the junctions between nucleosomes. Furthermore, due to differences in the structural characteristics of nucleosomes in maternal and fetal DNA during apoptosis, the binding strength of nucleosomes also differs. Consequently, the probability of maternal and fetal DNA being cleaved inside nucleosomes or at the junctions between nucleosomes during digestion also differs. Specifically, the probability of fetal DNA being cleaved inside nucleosomes is higher than that of maternal DNA. Therefore, by assessing the distribution of the endpoints of cfDNA in a sample relative to the nucleosomes, information about fetal concentration can be obtained.
[0134] In another embodiment of this application, a method is provided to determine fetal concentration by utilizing the difference in the distribution of the two ends of maternal and infant cfDNA on nucleosomes. Specifically, the method for determining fetal concentration includes:
[0135] S121. Obtain the sequencing sequence of the cfDNA sample to be tested.
[0136] In this step, the cfDNA sample to be tested is the cfDNA sample for which the fetal concentration needs to be determined. The method provided in this application does not specify requirements regarding the source of the cfDNA sample to be tested or its characteristic parameters.
[0137] S122. Align the sequencing sequence with the reference genome to obtain the alignment results.
[0138] This step includes:
[0139] S1221. The count of each base site in the reference genome that is the starting position for sequencing sequence alignment is determined as the second alignment number, and based on the second alignment number, the nucleosome center fraction corresponding to each base site in the reference genome is calculated;
[0140] S1222. Determine the nucleosome center location based on the nucleosome center score and the center score screening threshold;
[0141] S1223. Based on the nucleosome center position, the second logarithms of corresponding positions in all nucleosome regions are aligned and summed to obtain the summed second logarithm;
[0142] S1224. Perform dimensionality reduction on the second pair of comparisons after summation to obtain the dimensionality-reduced normalized second pair of comparisons, and determine the dimensionality-reduced normalized second pair of comparisons as the comparison result.
[0143] It should be noted that the second alignment number in this embodiment is for the purpose of distinguishing it from the first alignment number in the above embodiments, and there is no order between the two. The second alignment number is determined by counting the number of base sites on the reference genome that are the start positions of the sequencing sequence alignment. Correspondingly, this second alignment number can also be called a "count," that is, counting the alignment start positions of different sequences and then obtaining the number of times that different sites on the reference genome are exactly the sequence alignment start positions.
[0144] Since the probability of the nucleosome's internal region being a cut in cfDNA (i.e., the start of the sequencing sequence) is lower than that of the nucleosome's linker region being a cut in cfDNA, it is necessary to identify and confirm the region of the nucleosome.
[0145] In step S1221, counting the base sites at the alignment start positions of sequencing sequences in the reference genome refers to analyzing the alignment start positions of all sequencing sequences and then counting the occurrences of each site on the reference genome as the alignment start position of these sequences. In this embodiment, this count is determined as the second alignment number. It should be understood that the second alignment number can be the absolute number of times each site on the reference genome is used as the alignment start position of these sequences, or it can be the median number of times each site on the reference genome is used as the alignment start position of these sequences. Furthermore, during the statistical process, counting each site on the reference genome as the alignment start position allows for a certain tolerance ratio, that is, allowing sequencing sequences with a predetermined number of base errors within a certain length range to be included in the alignment.
[0146] After determining this count as the second logarithm, a nucleosome center score is calculated for each base site in the reference genome based on the second logarithm. The "nucleosome center score" can be understood as a probability score of each site being a nucleosome center location. The "nucleosome center location" refers to the location of this site at the very center of the nucleosome region.
[0147] In some embodiments, calculating the nucleosome center fraction corresponding to each base site in the reference genome includes:
[0148] (1) Calculate the first average count of the first specific number of base ranges to the left and right of each base site in the reference genome;
[0149] (2) Calculate the second average count of the second specific number of bases corresponding to each base site on the left and right of the reference genome;
[0150] (3) Determine the nucleosome center score for each site based on the first average count and the second average count.
[0151] The first and second specific quantities are determined based on the number of bases in the nucleosome region and the number of bases in the nucleosome linker region. Let x represent a base site, the first specific quantity = (number of bases in the nucleosome region - 1) / 2, and the second specific quantity = (number of bases in the nucleosome region - 1) / 2 + (number of bases in the nucleosome linker region / 2).
[0152] For example, based on literature reports, the nucleosome region is 147 bases long, with approximately 20 bases on each end of the nucleosome. In this case, the first specific quantity is (147-1) / 2 = 73, and the second specific quantity is (147-1) / 2 + 20 = 93. That is, assuming x is the center position of the nucleosome, the range of 73 bases to the left and right of this position constitutes the nucleosome region, and the range of the 74th to 93rd bases to the left and right of this position respectively constitutes the nucleosome connection region.
[0153] The above formula calculates the ratio of the sum of the average counts of the 74-93 base ranges to the left and right of the position under investigation to the average counts of the 73 base ranges to the left and right of the position under investigation. Clearly, for a specific nucleosome region, the closer the position under investigation is to the center of the nucleosome, the larger this value will be.
[0154] Furthermore, based on the first and second average counts, the nucleosome center score corresponding to each site is determined. The formula for calculating the nucleosome center score can be expressed as:
[0155]
[0156] In the formula, x is a base site, [x-93,x-74] represents the interval from 93 nucleotides away from x on one side to 74 nucleotides away from x on the same side, [x+93,x+74] represents the interval from 93 nucleotides away from x on the other side to 74 nucleotides away from x on the same side, and [x-73,x+73] represents the interval from 73 nucleotides away from x on the other side to 73 nucleotides away from x on the other side.
[0157] In some embodiments, considering that the sequencing error rate of single-molecule sequencers tends to be higher than that of next-generation sequencing technology, when performing probabilistic analysis and comparison of each base in the nucleosome region as the nucleosome center location, the accuracy and efficiency of determining the nucleosome center location can be improved by truncating the regions with relatively high error rates at both ends of the sequence.
[0158] In some embodiments, the nucleosome region is shortened by n bases at each end before calculating the "nucleosome center fraction" and the subsequent "nucleosome ratio". n is a natural number less than or equal to 5, for example, n is 1, 2, 3, 4 or 5.
[0159] At this point, the formula for calculating the nucleosome center fraction can be expressed as:
[0160]
[0161] In the formula, x is a base site, [x-93,x-74-n] represents the interval from 93 nucleotides away from x on one side to 74-n nucleotides away from x on the same side, [x+93,x+74-n] represents the interval from 93 nucleotides away from x on the other side to 74-n nucleotides away from x on the same side, and [x-73-n,x+73-n] represents the interval from 73-n nucleotides away from x on the other side to 73-n nucleotides away from x on the other side.
[0162] For example, when n=5, the formula for calculating the nucleosome center fraction can be expressed as:
[0163]
[0164] In the formula, x is a base site, [x-93,x-69] represents the interval from 93 nucleotides away from x on one side to 69 nucleotides away from x on the same side, [x+93,x+69] represents the interval from 93 nucleotides away from x on the other side to 69 nucleotides away from x on the same side, and [x-68,x+68] represents the interval from 68 nucleotides away from x on the other side to 68 nucleotides away from x on the other side.
[0165] In step S1223, the nucleosome center score is determined based on the nucleosome center score and the center score screening threshold, including:
[0166] (1) Based on the maximum value of the nucleosome center fraction, determine the position of the maximum value on the reference genome and determine the position as the first position;
[0167] (2) Clear the nucleosome center score of the bases on both sides of the first position to zero, and determine the maximum value again based on the remaining nucleosome center score after clearing. Then, determine the position of the maximum value on the reference genome until the nucleosome center score is less than the second target threshold. The selected position is then determined as the candidate nucleosome center position.
[0168] (3) Determine the nucleosome center position by using the base center score screening threshold and the nucleosome center score corresponding to the candidate nucleosome center position.
[0169] The first position is based on the location of the maximum nucleosome center score on the reference genome. Specifically, the maximum nucleosome center score calculated above is found, its location on the reference genome is recorded, and the nucleosome center scores of specific bases flanking this position (e.g., 147 bases on each side) are reset to zero. This is because the found position is considered the nucleosome center location, and there cannot be another nucleosome center location within at least 147 bases on either side of it. Then, the position corresponding to the current maximum nucleosome center score is found again, and the step of resetting the next year's score in the nucleosomes on both sides is repeated. This process is repeated until the maximum nucleosome center score is lower than the second target threshold. The position selected based on this process is the nucleosome center location. Theoretically, the average technical score of the nucleosome connection region should be higher than that of the nucleosome region. Based on the above formula, the nucleosome center score should be greater than 2. To facilitate processing and improve accuracy, in one embodiment, the second target threshold can be set to 2.2. That is, when screening the nucleosome center region, the loop continues until the value of the largest nucleosome center score is lower than 2.2, at which point the iterative processing stops.
[0170] After obtaining all candidate nucleosome center locations, some of these locations have abnormally high "nucleosome center scores," which may be affected by various positional factors such as alignment probability bias. As a result, the corresponding second logarithm may be unreliable. Therefore, using a portion of these candidate nucleosome center locations as the final selected nucleosome center locations allows for the screening of nucleosome center scores.
[0171] Table 2 below provides several score screening criteria and the corresponding Pearson correlation coefficients of fetal concentration and nucleosome ratio under these criteria.
[0172] Table 2
[0173]
[0174] In Table 2, "score" refers to the nucleosome center score, "bp" refers to the number of bases, "linker" refers to the nucleosome linker region, and "nucleosome ratio" refers to the nucleosome ratio. The Pearson correlation coefficient between the nucleosome ratio and the reference fetal concentration was used to initially screen the nucleosome center location and the calculation method for the nucleosome ratio. [10,) represents a range greater than or equal to 10, and the right parentheses are open intervals.
[0175] As shown in Table 2, compared to not processing the nucleosome region, shortening the left and right ends of the nucleosome region by 5 bases before calculating the "nucleosome center fraction" and the subsequent "nucleosome ratio" resulted in a higher Pearson correlation coefficient between fetal concentration and nucleosome ratio.
[0176] In step S1223, the second logarithms of corresponding positions in all nucleosome regions are aligned and summed based on the nucleosome center position to obtain the summed second logarithms.
[0177] In some embodiments, when aligning and stacking according to the nucleosome region, a range of 200 bases to the left and right of the "nucleosome center position" is taken for alignment and summation, resulting in a vector of 401 elements, with the 201st element representing the nucleosome center position. This operation has no impact on the calculation of the "nucleosome ratio," but since machine learning algorithms are subsequently introduced to build a model, information from a larger range around the nucleosome region is extracted to improve the potential information content and predictive ability of the model.
[0178] After determining the nucleosome center location, for the sample data to be analyzed, the sequence start position count corresponding to each position on the reference genome can be calculated first using the method described above, i.e., the second alignment number is obtained. Then, based on the processing procedure of the nucleosome center location, the second alignment numbers of 147 bases to the left and right of the nucleosome center location are extracted. This operation is performed for all nucleosome center locations. After this operation, all extracted data are summed, i.e., the second alignment numbers of corresponding positions within all nucleosome regions are aligned and summed based on the nucleosome center location. The corresponding positions within the nucleosome region include a specific number of base ranges to the left and right of the nucleosome center location. After repeated verification, processing of 200 base ranges to the left and right yields the best results. That is, each time a vector of length 295 is removed, the 148th element of the vector is the second alignment number of the nucleosome center location, and the other elements of the vector are the second alignment numbers corresponding to the positions to the left and right of the nucleosome center location. These vectors are summed, and the summation result can be found in [reference needed]. Figure 6 As shown, it illustrates the distribution obtained by aligning and summing one set of data according to nucleosome regions. Figure 6 As can be seen, the statistical data of the aligned nucleosome region obtained according to our analysis and processing method shows that the probability of the sequence start site falling into the central region of the nucleosome is much lower than the probability of falling into the nucleosome junction region, which is consistent with the biochemical principle and indirectly confirms the reliability of our above analysis method.
[0179] The mean count of the 147 bases in the nucleosome region corresponding to the summed vector (that is, the range of 73 bases to the left and right of the 148th element of the vector) is calculated. Then, the ratio of this mean to the mean count of the left and right nucleosome connection regions is calculated (e.g., the range of 74-93 bases to the left and right of the nucleosome center is the nucleosome connection region). This ratio can be called the "nucleosome ratio", which is the comparison result determined based on nucleosome-related information in the embodiments of this application. The higher the corresponding "nucleosome ratio", the higher the fetal concentration should theoretically be. A linear model is established through the relationship between the nucleosome ratio and the reference fetal concentration to obtain the relationship between the comparison result and the fetal concentration, thus achieving the purpose of determining the fetal concentration based on the comparison result.
[0180] Specifically, a linear model between the second logarithm and fetal concentration can also be established, which can be achieved through machine learning processing. The determination of fetal concentration based on the logarithm results includes:
[0181] (1) Obtain the second training sample data. Each sample in the second training sample data is labeled with a feature value and a target value. The feature value is the normalized second logarithm after dimensionality reduction, and the target value is the actual fetal concentration.
[0182] (2) Create a second fetal concentration quantification model based on the second training sample; input the comparison results into the second fetal concentration quantification model to obtain the fetal concentration.
[0183] The process of generating the second fetal concentration quantitative model is similar to that of generating the first fetal concentration quantitative model, except that the feature values learned by the two models are different. Therefore, the creation process of the first fetal concentration quantitative model can be referred to, and will not be described in detail here.
[0184] For example, a linear model was established using the ratio of the average count in the nucleosome region to the average count in the nucleosome-connected region as a parameter, against known fetal concentrations. Due to the low accuracy of the initiation site in single-molecule sequencing data, the Pearson correlation coefficient was only around 0.3 without truncating the high-error-rate bases at both ends of the nucleosome region (see Table 2). Therefore, in one embodiment, the above... Figure 6 All 401 counts were used as raw features, and a machine learning model (principal component analysis dimensionality reduction + elastic network modeling) was built to estimate fetal concentration. For example, for a certain number of existing reference fetal concentration datasets (e.g., 615 sets), a summation vector of counts aligned to the nucleosome center positions was calculated. Each set corresponds to a vector containing 401 elements, which are the feature values of the original input. After 100 random splits of the test set (30%) and training set (70%), the Pearson correlation coefficient between the estimated fetal concentration on the test set and the actual fetal concentration was 0.58, and R0 was [not specified]. 2 It is 0.33. See also Figure 7 The model's prediction performance was demonstrated after one random split of the test set and training set. Figure 7 In the figure, the horizontal axis represents the fetal concentration calculated by the nucleosome method, and the vertical axis represents the reference fetal concentration.
[0185] In some embodiments, processing the raw feature numbers can improve the accuracy of fetal concentration calculations based on sequencing data obtained from various sequencing platforms, particularly single-molecule sequencing platforms.
[0186] In one embodiment, the original features are normalized. Specifically, 200 counts (second logarithms) are taken on each side of the center position, and these counts are divided by the count at the center position to obtain 401 normalized features. This normalization process can reduce the impact of noise on the model, thereby improving the model's performance.
[0187] In another embodiment, new features are created based on the original features. For example, based on the 401 original features mentioned above, two new features are created: the first new feature is the ratio of the average count of the 20 most distant positions to the average count of the 147 center positions; the second new feature is the ratio of the average count of the 20 rightmost positions to the average count of the 147 center positions. This results in 403 features.
[0188] In another embodiment, principal component dimensionality reduction can be performed on the original features (obtaining a principal component transformation model on the training set and performing the same transformation on the test set) to simplify the complex relationships between the variables, making the model more concise and reducing overfitting.
[0189] The three embodiments described above can be used individually or in combination to process the original features.
[0190] For example, after creating 403 new features based on the original features, principal component analysis is performed on the 403 features to reduce their dimensionality (the principal component transformation model is obtained on the training set and the same transformation is performed on the test set). By balancing the number of principal components and the information retained, the 403 features are finally reduced to 33 while retaining 90% of the effective information.
[0191] Machine learning models are built, and to avoid overfitting, L1 and L2 regularization are applied (L1 and L2 regularization are regularization methods that add additional terms to the loss function to prevent overfitting), and the optimal combination of L1 and L2 regularization parameters is selected iteratively. During model training, to more accurately evaluate the model's predictive performance, the test set (30%) and training set (70%) are randomly split 100 times for model training. The final Pearson correlation coefficient between the fetal concentration estimated on the test set and the actual fetal concentration is 0.58, R0. 2 It is 0.33. See also Figure 7 The model's prediction performance was demonstrated after one random split of the test set and training set.
[0192] Another embodiment of this application provides a different method for determining fetal concentration, see [link to relevant documentation]. Figure 8 The method may include the following steps:
[0193] S201. Obtain the sequencing sequence of the cfDNA sample to be tested.
[0194] The sequencing results are as described above, and will not be repeated here to save space.
[0195] S202. The reference genome is segmented to obtain multiple segment intervals, and the first alignment number of each sequencing sequence falling into each segment interval is determined.
[0196] In this step, in order to obtain accurate alignment results, the reference genome can be segmented according to a fixed length, such as determining the segment length of the reference genome based on the characteristics of the sequencing sequence.
[0197] The reference genome contains certain special chromosomes that may introduce bias or other unique characteristics during the determination of fetal concentrations through reference genome alignment. In one implementation, the bias introduced by these special chromosomes (hereinafter referred to as specific chromosomes) is reduced by removing segmented regions from the reference genome, thereby minimizing their impact on the test results. In some embodiments, specific chromosomes include at least one of the following: X chromosome, Y chromosome, mitochondrial chromosome, chromosome 13, chromosome 18, and chromosome 21. Specifically, removing segmented regions corresponding to the X and Y chromosomes avoids the impact of differences in these chromosomes between male and female fetuses on the results of subsequent concentration quantification models; chromosomes 13, 18, and 21 are removed because these chromosomes have a higher probability of duplication or deletion. One of the screening objectives of NIPT is to screen for duplications and deletions on chromosomes 13, 18, and 21. When chromosomes 13, 18, and 21 are used as reference chromosomes, biases may occur when normalizing the data, such as affecting the normalized logarithm, which in turn affects the calculation results and the accuracy of NIPT screening results.
[0198] The embodiments of this application can adjust the range of specific chromosomes according to the expected needs of fetal concentration detection, such as removing the X and Y chromosomes, or removing chromosomes 13, 18, and 21, or removing mitochondrial chromosomes. Of course, multiple situations mentioned above can be removed simultaneously, such as removing all X, Y, mitochondrial chromosomes, chromosomes 13, 18, and 21 or their corresponding segments, so that the segmented intervals obtained after segmenting the reference genome do not contain the segmented intervals corresponding to the aforementioned chromosomes.
[0199] This application allows for the removal of specific chromosomes into segmented intervals through various methods. The segmented intervals can be filtered based on specific chromosomes, or these specific chromosomes can be removed before segmentation. Specifically, the appropriate processing method can be selected based on actual application requirements; this application does not impose any restrictions on this.
[0200] In one implementation, the reference genome is segmented to obtain multiple segmented intervals, including: segmenting the reference genome based on a preset segmentation length to obtain multiple initial segmented intervals; and removing segmented intervals corresponding to specific chromosomes from the initial segmented intervals to obtain multiple segmented intervals. That is, after segmenting the reference genome into multiple segmented intervals, the segmented intervals corresponding to specific chromosomes in the reference genome are removed and not included in the calculation. It should be understood that the "segmented intervals corresponding to specific chromosomes" refers to the multiple segmented intervals formed after the specific chromosome is segmented according to the preset segmentation length.
[0201] In another implementation, the reference genome is segmented to obtain multiple segment intervals, including: removing specific chromosomes from the reference genome to obtain a removed reference genome; and segmenting the removed reference genome based on a preset segmentation length to obtain multiple segment intervals. That is, specific chromosomes are removed before segmenting the reference genome.
[0202] The preset segmentation length can be adjusted according to the amount of data processed and the expected relative accuracy. For example, the reference genome can be segmented into segments of 50k (5000) bases each. To make the obtained first alignment number more accurate, the first alignment number can be corrected to obtain a corrected first alignment number, which can then be used as the alignment result.
[0203] In some embodiments, correction includes GC correction. GC correction is an operation in the bioinformatics analysis process. GC correction is performed because gene sequencers and their corresponding sequencing procedures can introduce GC bias. For example, some sequencers are more likely to detect high-GC sequences, while others are more likely to detect low-GC sequences. In this embodiment, the determination of fetal concentration is based on the alignment probabilities of different regions on the reference genome. The GC bias of the sequencer will affect the judgment of the actual alignment probabilities of different regions; therefore, GC correction is necessary to improve accuracy.
[0204] At this point, the first logarithm is corrected to obtain the corrected first logarithm, which includes: normalizing the first logarithm of each segment interval based on the first logarithm of each segment interval and the average first logarithm of all segment intervals to obtain the normalized first logarithm of each segment interval; and performing GC correction on the normalized first logarithm of each segment interval to obtain the GC-corrected first logarithm. The average first logarithm of all segment intervals refers to the ratio of the sum of the first logarithms of all segment intervals to the total number of segment intervals.
[0205] After obtaining the first logarithm of all segmented intervals, to facilitate calculation and processing, the first logarithm of all segmented intervals will be normalized. The corresponding formula is:
[0206] Normalized first logarithm of each segment interval = First logarithm of each segment interval / Average first logarithm of all segment intervals.
[0207] After obtaining the first logarithm of each segment interval, GC correction will be performed to obtain the first logarithm after GC correction.
[0208] In one implementation, GC correction is performed on the normalized first logarithm for each segment interval to obtain the GC-corrected first logarithm, including:
[0209] (1) Generate the first relationship curve based on the GC content and the normalized first logarithm corresponding to each segment interval.
[0210] In this implementation, the GC content can be represented by various parameters, such as the GC content of each segment interval or the median GC content, or other parameters that can reflect the GC content of the segment interval.
[0211] In this embodiment of the application, GC filtering can be performed on the segmented intervals that form the first relationship curve, such as filtering the segmented intervals according to the GC content, to filter out some abnormal or statistically insignificant segments, thereby optimizing the first relationship curve and reducing the consumption of data processing resources during data processing.
[0212] In one implementation, GC filtering is achieved as follows: before generating a first relationship curve based on GC content and the normalized first logarithm corresponding to each segment interval, the following processing is performed: the segment intervals are filtered based on GC content to remove segment intervals whose GC content does not meet the preset requirements, so that the first relationship curve is generated based on the GC content of each segment interval after GC filtering and the normalized first logarithm of each segment interval.
[0213] In another implementation, GC filtering is achieved as follows: during the process of normalizing the first logarithm based on the first relationship curve, the segmented intervals are filtered based on the GC content, and the normalized first logarithm after GC filtering is GC corrected based on the first relationship curve to obtain the GC-corrected first logarithm.
[0214] In the above embodiments, as an example, a first relationship curve is generated based on the GC content and the normalized first logarithm corresponding to each segment interval. This is achieved as follows: the x-axis represents the GC content of each segment interval (i.e., the proportion of GC bases corresponding to that segment interval on the reference genome), and the y-axis represents the normalized first logarithm corresponding to each segment interval. Data on each segment interval and its GC content within that interval are integrated, i.e., a scatter plot is generated based on the GC content of each segment interval and its corresponding normalized first logarithm. The scatter plot data is smoothed to obtain a smoothed curve. Smoothing the scatter plot data can be performed using algorithms such as the LOWESS algorithm to obtain the smoothed curve, i.e., the first relationship curve, such as y... s =f(x) s ).
[0215] Specifically, as an example, the first relationship curve can be generated as follows: divide the horizontal axis into different segments, then calculate the median GC content of the scatter points within the same segment, use this median GC content to represent the value corresponding to that segment, and then plot all these GC content medians on a graph and smooth them to obtain the first relationship curve. In another example, smoothing can be performed directly based on all the data, that is, all data points are considered and smoothed together to obtain a smooth curve, thus obtaining a higher Spearman correlation coefficient for the test set. It should be noted that this smoothing method is also applicable to the smoothing and correction of the contrast probabilities described later.
[0216] (2) Based on the first curve, perform GC correction on the normalized first logarithm to obtain the GC-corrected first logarithm.
[0217] In some implementations, GC correction is performed based on a subtraction or division relationship between the normalized first logarithm and the GC content of the segmented intervals corresponding to the normalized first logarithm, and the GC-corrected first logarithm is determined.
[0218] In one example, the GC correction method is "subtraction," and the corresponding correction formula is: y GC = yf(x); In another example, the GC correction method is "division", and the corresponding correction formula is: y GC = y / f(x).
[0219] In the two correction formulas above, x is the GC content of the segmented interval, and y is the normalized first logarithm corresponding to the segmented interval. GC This is the normalized first logarithm after GC correction.
[0220] In some embodiments, the correction further includes alignment probability correction to improve the accuracy of the obtained first alignment count. That is, in some embodiments, the correction includes both GC correction and alignment probability correction.
[0221] In one embodiment of this application, correcting the first alignment number to obtain a corrected first alignment number further includes: truncating the reference genome based on a specific sliding window length, aligning the truncated sequence with the reference genome, and counting the first alignment number in each segment of the reference genome, determining the first alignment number as the sliding window alignment number; determining the normalized sliding window alignment number for each segment based on the sliding window alignment number; and performing alignment probability correction based on the normalized sliding window alignment number for each segment and the normalized first alignment number to determine the alignment probability corrected first alignment number.
[0222] This application employs an "alignment probability correction" method to correct the first alignment count, based on the following considerations: Different regions of the reference sequence corresponding to the reference genome, if truncated to a specific length and treated as a single sequencing sequence, will have varying alignment counts when aligned back to the reference genome. This is because for some regions, similar sequences exist in multiple locations on the reference genome, making these regions more likely to align. For example, the human genome reference sequence hg19 can be truncated using a sliding window with a step size of 2 bases and a window size of 37 bases. These sequences are then aligned with the human reference genome hg19, and the alignment counts for different segments of the human reference genome are counted. This alignment count can be called the "sliding window alignment count." Specifically, the specific sliding window length can be determined based on the actual application scenario, such as the characteristics of the reference genome used or the characteristics of the sample to be tested. After determining a specific sliding window length, the reference genome is truncated. The truncated sequence is then aligned with the reference genome, and the first alignment number for each segment of the reference genome is counted. This first alignment number is defined as the sliding window alignment number. To facilitate calculation and processing, a normalized sliding window alignment number is determined for each segment. Alignment probability correction is then applied to the normalized sliding window alignment number. The method for alignment probability correction can refer to the GC correction method described earlier.
[0223] In one implementation, the alignment probability is corrected based on the normalized sliding window alignment count and the normalized first alignment count for each segment interval, and the first alignment count after alignment probability correction is determined, including:
[0224] (1) Generate the second relationship curve based on the normalized sliding window logarithm of each segment interval and the normalized first logarithm corresponding to each segment interval.
[0225] The normalized sliding window logarithm for each segment interval can be represented by various parameters, such as the normalized sliding window logarithm for each segment interval or the median of the sliding window logarithm, or other parameters that can reflect the normalized sliding window logarithm for each segment interval.
[0226] In this embodiment of the application, the normalized sliding window logarithm that forms the second relationship curve can be filtered to remove some abnormal or statistically insignificant segments, thereby optimizing the second relationship curve and improving the processing accuracy when performing related data processing based on the second relationship curve.
[0227] In one implementation, the normalized sliding window logarithm used to form the second relationship curve is filtered by the following method: before generating the second relationship curve based on the normalized sliding window logarithm for each segment interval and the normalized first logarithm corresponding to each segment interval, the following processing is performed: the segment intervals are filtered based on the normalized sliding window logarithm to obtain segment intervals where the normalized sliding window logarithm is not less than a first target threshold, so that the second relationship curve is generated in segment intervals where the normalized sliding window logarithm is not less than the first target threshold.
[0228] In one implementation, the normalized sliding window logarithms forming the second relationship curve are filtered by means of the following: during the process of performing alignment probability correction on the normalized first logarithms based on the second relationship curve to obtain the alignment probability-corrected first logarithms, the segmented intervals are filtered based on the normalized sliding window logarithms, retaining the second relationship curve intervals in the second relationship curve where the normalized sliding window logarithms are not less than a first target threshold, so as to obtain the alignment probability-filtered second relationship curve; the alignment probability is corrected on the sliding window normalized logarithms corresponding to each segmented interval based on the alignment probability-filtered second relationship curve to obtain the alignment probability-corrected first logarithms.
[0229] In the above embodiments, based on repeated verification, the first target threshold can be selected as 0.8, that is, removing the segment intervals where the normalized sliding window logarithm is less than 0.8.
[0230] In the above embodiments, as an example, a second relationship curve is generated based on the normalized sliding window logarithm of each segment interval and the normalized first logarithm corresponding to each segment interval. This is achieved by: using the x-axis to represent the normalized sliding window logarithm of each segment interval and the y-axis to represent the normalized first logarithm corresponding to each segment interval; integrating the data of each segment interval and its corresponding GC content, i.e., generating a scatter plot based on the GC content of each segment interval and its corresponding normalized first logarithm; and smoothing the scatter data in the scatter plot to obtain a smooth curve. For example, the smoothing of the scatter data in the scatter plot can be achieved using the LOWESS algorithm.
[0231] (2) Based on the second relationship curve, the normalized first logarithm is corrected for alignment probability to obtain the first logarithm after alignment probability correction.
[0232] In some implementations, the alignment probability is corrected by subtraction or division of the sliding window alignments of the segmented intervals corresponding to the normalized first alignment number and the first alignment number, and the first alignment number after alignment probability correction is determined.
[0233] In one example, the alignment probability correction method is "subtraction," and the corresponding correction formula is: y 比对概率 = yf(x); In another example, the alignment probability correction method is "division", and the corresponding correction formula is: y 比对概率 = y / f(x).
[0234] In the two correction formulas above, x is the normalized sliding window logarithm for each segmented interval, and y corresponds to the normalized first logarithm for each segmented interval. 比对概率 This is the normalized first logarithm after comparison probability correction.
[0235] S203. Determine the first fetal concentration based on the first logarithm.
[0236] In this application embodiment, machine learning methods combining neural networks can be used to determine fetal concentration. In one embodiment of this application, determining fetal concentration based on comparison results includes:
[0237] Obtain first training sample data, where each sample in the first training sample data is labeled with a first feature value and a first target value, where the first feature value is the number of sample comparisons and the first target value is the actual fetal concentration of the sample.
[0238] Based on the results of a specific model, machine learning modeling is performed on the first training sample data to obtain the first fetal concentration quantification model;
[0239] The comparison results are input into the first fetal concentration quantification model to obtain the first fetal concentration, and the first fetal concentration is determined as the fetal concentration.
[0240] In some embodiments, machine learning modeling is performed on the first training sample data based on a specific model structure to obtain a first fetal concentration quantification model, including:
[0241] The first training sample data is divided into a training set and a test set; machine learning modeling is performed based on the training set to obtain the initial model;
[0242] The test set is processed based on the initial model to obtain the estimated fetal concentration for each test sample in the test set.
[0243] The comparison results are obtained by comparing the estimated fetal concentration with the actual fetal concentration of each test sample in the test set.
[0244] Based on the comparison results, the model parameters of the initial model were adjusted to obtain the first fetal concentration quantitative model.
[0245] The first training sample data can be data from a database of pregnant women's peripheral blood cfDNA with known fetal concentrations obtained from various sequencing platforms, including single-molecule sequencing platforms. For example, using a certain number of samples with known fetal concentrations, the reference genome chromosome reference sequence is segmented into intervals (e.g., by 5k bases, 50k bases, 100k bases, 300k bases, or 800k bases, etc.), and the alignment count of each sample in different segment intervals is counted. The Lowes method is used to normalize, GC, and correct the alignment probability of these alignment counts. The final normalized and corrected alignment counts in different segment intervals are used as feature values, and the known actual fetal concentration corresponding to each sample is used as the target value. A specific model is used for learning and modeling. The specific model can be a classification model, a convolutional neural network model, etc. This application embodiment does not limit the specific model structure, as long as it can learn from the training sample data and predict the fetal concentration. The model learning and training process is an iterative process. Therefore, the training samples can be divided into a training set and a test set. The accuracy of the model obtained by the current training is tested through the test set. If the accuracy is lower than the set accuracy threshold, the model parameters in the current model structure are adjusted through the test set until the error between the fetal concentration output by the model and the actual fetal concentration meets the corresponding error range.
[0246] When the number of samples used for machine learning is small, but the number of features is large (e.g., approximately 3,000 to 600,000 features, depending on the size of the chromosome reference sequence intervals), overfitting is unavoidable during both dimensionality reduction and learning processes. Consequently, the fetal concentration quantification model will show a certain deviation between its prediction of fetal concentration on the test set and the actual fetal concentration. For example, in a fetal concentration quantification model built using 615 samples with known fetal concentrations, the model corresponds to the test set R... 2The Pearson correlation coefficient between the actual and predicted values is only around 0.3, and approximately 0.5. However, when the logarithm is normalized and corrected (GC correction, or GC and alignment probability correction) using the method described in the embodiments of this application, the correlation coefficient between the predicted and actual values of the test set is improved. Comparing modeling directly using normalized segmented interval logarithms, and modeling using normalized segmented interval logarithms corrected for GC and alignment probability, the modeling effect after normalization, GC correction, and alignment probability correction is better, i.e., the correlation coefficient between the predicted and actual values of the test set is higher.
[0247] In another implementation, determining the fetal concentration based on the comparison results includes:
[0248] The comparison results are input into the first preset model to obtain the initial fetal concentration;
[0249] The initial fetal concentration is corrected according to the second preset model to obtain the fetal concentration.
[0250] The first preset model is a linear relationship pattern established based on the alignment results in the second sample data and the initial fetal concentration corresponding to the alignment results. The second sample data is the alignment results obtained by comparing the sequencing sequence in the cfDNA sample with the reference genome and the fetal concentration corresponding to the alignment results. The second preset model is a model obtained by processing the first preset model with constants determined by the phenomenon fitting.
[0251] For example, the first preset model can be in, For comparison, FF1 represents the initial fetal concentration. The second preset model can be FF2 = FF1·c + d, where c and d are constants determined by linear fitting, and FF2 represents the fetal concentration.
[0252] For example, a model that uses the difference in alignment probability between samples in different regions of a reference genome to build a model and estimate fetal concentration is called SeqFF. SeqFF uses two machine learning models: Weighted Ranking Selection Criterion (WRSC) and Elastic Network (Enet). The final output fetal concentration result is the average of the predictions from the two models. When testing the SeqFF model with data from sequencing platforms, especially single-molecule sequencing platforms, the WRSC model outperforms Enet. Specifically, the linear correlation coefficient between the fetal concentration estimated by WRSC and the actual fetal concentration is higher than that of Enet or SeqFF. Therefore, the model parameters involved in the WRSC algorithm are preferred. For details, please refer to... Figure 2 and Figure 3 , Figure 2This is a schematic diagram illustrating the relationship between fetal concentration calculated using the Enet model and a reference fetal concentration, provided in an embodiment of this application. Figure 3 This diagram illustrates the relationship between fetal concentration calculated by the WRSC model and a reference fetal concentration, as provided in an embodiment of this application. As can be seen from the diagram, the positive correlation between the fetal concentration calculated by the WRSC model and the reference fetal concentration is superior to the positive correlation between the fetal concentration calculated by the Enet model and the reference fetal concentration.
[0253] For example, in the WRSC model, the chromosome reference sequence is divided into intervals of a specific number of bases, such as 50k bases. Through training with the WRSC algorithm, a linear relationship between coverage density and fetal concentration in different reference sequence intervals was found, as shown in the following formula. In the following formula, FF represents the fetal concentration. To represent the coverage density in different regions of the reference sequence, To and A constant vector of equal dimensions, where B is the parameter matrix used for linear transformation, and c and d are constants.
[0254]
[0255] Because SeqFF is a model trained on second-generation sequencing data, its sequencing principle differs significantly from single-molecule sequencing. Therefore, directly using the SeqFF model parameters to calculate fetal concentrations from other sequencing platforms, including single-molecule sequencing, will result in a large discrepancy between the calculated and actual fetal concentrations. Using the 615 NIPT clinical samples with known fetal concentrations from the above examples for analysis and testing, we directly processed the raw alignment file data (e.g., SAM files) from other sequencing platforms, including single-molecule sequencing, using the SeqFF algorithm and parameters. The Pearson correlation coefficient between the obtained intermediate parameters and the actual fetal concentration was 0.714.
[0256] The calculated fetal concentration from the SeqFF model shows a good linear relationship with the actual fetal concentration; therefore, the formula above can be used. This section includes model parameters that calculate intermediate parameters of fetal concentration, and then uses linear regression to fit the values of c and d, thereby obtaining the final model parameters suitable for various sequencing platforms, including single-molecule sequencing platforms.
[0257] The analysis and testing were still conducted using 615 NIPT clinical samples with known fetal concentrations from the above examples. The analysis was modified to use only the first alignment count of the sequencing sequence falling within each segment interval; regions with abnormal alignment probabilities on the X, Y, M, 13, 18, 21 chromosomes and mitochondrial chromosomes in the reference genome were excluded; and the correction method (GC correction, or GC and alignment probability correction) was optimized. The data were re-analyzed using the optimized parameters, and the Pearson correlation coefficient between the intermediate parameters and fetal concentration was 0.755. 100 random splits of the test set (30%) and training set (70%) were used to build a linear model and determine the aforementioned constant parameters c and d. The average test set prediction R obtained from 100 modeling iterations was... 2 The value is 0.561.
[0258] It should be noted that when actually building a linear model, the splitting of the test set and training set is to evaluate the reliability of the model built on our currently randomly split training set in limited data. Therefore, after randomly splitting the test set (30%) and training set (70%) 100 times, the average R² value of the test set prediction obtained from 100 modeling runs was 0.561.
[0259] Table 1 above shows nine examples of determining fetal concentration based on the differences in alignment probabilities between maternal and infant cfDNA at different locations on the reference genome.
[0260] As shown in Table 1, compared to the algorithm in Example 1 (Pearson correlation coefficient of 0.71425), the Pearson correlation coefficients of all other algorithms are greater than 0.74, and the Pearson correlation coefficients of the algorithms in Examples 6, 8, and 9 are greater than 0.75. Compared to other algorithms, the algorithm in Example 8 achieves the highest Pearson correlation coefficient of 0.75471.
[0261] The graph showing the relationship between the calculated fetal concentration and the reference fetal concentration in Example 8 after correction is shown below. Figure 5 shown.
[0262] S204. Based on each base site of the reference genome, the second alignment number is the starting position of the sequencing sequence alignment, which determines the nucleosome center position.
[0263] This step includes:
[0264] The count of each base site in the reference genome that is the starting position for sequencing sequence alignment is determined as the second alignment number, and the nucleosome center fraction corresponding to each base site in the reference genome is calculated based on the second alignment number.
[0265] The nucleosome center location is determined based on the nucleosome center score and the center score screening threshold.
[0266] In some embodiments, calculating the nucleosome center fraction corresponding to each base site in the reference genome includes:
[0267] (1) Calculate the first average count of the first specific number of base ranges to the left and right of each base site in the reference genome;
[0268] (2) Calculate the second average count of the second specific number of bases corresponding to each base site on the left and right of the reference genome;
[0269] (3) Determine the nucleosome center score for each site based on the first average count and the second average count.
[0270] The first and second specific quantities are determined based on the number of bases in the nucleosome region and the number of bases in the nucleosome linker region. Let x represent a base site, the first specific quantity = (number of bases in the nucleosome region - 1) / 2, and the second specific quantity = (number of bases in the nucleosome region - 1) / 2 + (number of bases in the nucleosome linker region / 2).
[0271] For example, based on literature reports, the nucleosome region is 147 bases long, with approximately 20 bases on each end of the nucleosome. In this case, the first specific quantity is (147-1) / 2 = 73, and the second specific quantity is (147-1) / 2 + 20 = 93. That is, assuming x is the center position of the nucleosome, the range of 73 bases to the left and right of this position constitutes the nucleosome region, and the range of the 74th to 93rd bases to the left and right of this position respectively constitutes the nucleosome connection region.
[0272] The above formula calculates the ratio of the sum of the average counts of the 74-93 base ranges to the left and right of the position under investigation to the average counts of the 73 base ranges to the left and right of the position under investigation. Clearly, for a specific nucleosome region, the closer the position under investigation is to the center of the nucleosome, the larger this value will be.
[0273] Furthermore, based on the first and second average counts, the nucleosome center score corresponding to each site is determined. The formula for calculating the nucleosome center score can be expressed as:
[0274]
[0275] In the formula, x is a base site, [x-93,x-74] represents the interval from 93 nucleotides away from x on one side to 74 nucleotides away from x on the same side, [x+93,x+74] represents the interval from 93 nucleotides away from x on the other side to 74 nucleotides away from x on the same side, and [x-73,x+73] represents the interval from 73 nucleotides away from x on the other side to 73 nucleotides away from x on the other side.
[0276] In some embodiments, considering that the sequencing error rate of single-molecule sequencers tends to be higher than that of next-generation sequencing technology, when performing probabilistic analysis and comparison of each base in the nucleosome region as the nucleosome center location, the accuracy and efficiency of determining the nucleosome center location can be improved by truncating the regions with relatively high error rates at both ends of the sequence.
[0277] In some embodiments, the nucleosome region is shortened by n bases at each end before calculating the "nucleosome center fraction" and the subsequent "nucleosome ratio". n is a natural number less than or equal to 5, for example, n is 1, 2, 3, 4 or 5.
[0278] At this point, the formula for calculating the nucleosome center fraction can be expressed as:
[0279]
[0280] In the formula, x is a base site, [x-93,x-74-n] represents the interval from 93 nucleotides away from x on one side to 74-n nucleotides away from x on the same side, [x+93,x+74-n] represents the interval from 93 nucleotides away from x on the other side to 74-n nucleotides away from x on the same side, and [x-73-n,x+73-n] represents the interval from 73-n nucleotides away from x on the other side to 73-n nucleotides away from x on the other side.
[0281] For example, when n=5, the formula for calculating the nucleosome center fraction can be expressed as:
[0282]
[0283] In the formula, x is a base site, [x-93,x-69] represents the interval from 93 nucleotides away from x on one side to 69 nucleotides away from x on the same side, [x+93,x+69] represents the interval from 93 nucleotides away from x on the other side to 69 nucleotides away from x on the same side, and [x-68,x+68] represents the interval from 68 nucleotides away from x on the other side to 68 nucleotides away from x on the other side.
[0284] Based on the nucleosome center score and the center score screening threshold, the nucleosome center score is determined, including:
[0285] (1) Based on the maximum value of the nucleosome center fraction, determine the position of the maximum value on the reference genome and determine the position as the first position;
[0286] (2) Clear the nucleosome center score of the bases on both sides of the first position to zero, and determine the maximum value again based on the remaining nucleosome center score after clearing. Then, determine the position of the maximum value on the reference genome until the nucleosome center score is less than the second target threshold. The selected position is then determined as the candidate nucleosome center position.
[0287] (3) Determine the nucleosome center position by using the base center score screening threshold and the nucleosome center score corresponding to the candidate nucleosome center position.
[0288] The first position is based on the location of the maximum nucleosome center score on the reference genome. Specifically, the maximum nucleosome center score calculated above is found, its location on the reference genome is recorded, and the nucleosome center scores of specific bases flanking this position (e.g., 147 bases on each side) are reset to zero. This is because the found position is considered the nucleosome center location, and there cannot be another nucleosome center location within at least 147 bases on either side of it. Then, the position corresponding to the current maximum nucleosome center score is found again, and the step of resetting the next year's score in the nucleosomes on both sides is repeated. This process is repeated until the maximum nucleosome center score is lower than the second target threshold. The position selected based on this process is the nucleosome center location. Theoretically, the average technical score of the nucleosome connection region should be higher than that of the nucleosome region. Based on the above formula, the nucleosome center score should be greater than 2. To facilitate processing and improve accuracy, in one embodiment, the second target threshold can be set to 2.2. That is, when screening the nucleosome center region, the loop continues until the value of the largest nucleosome center score is lower than 2.2, at which point the iterative processing stops.
[0289] After obtaining all candidate nucleosome center locations, some of these locations have abnormally high "nucleosome center scores," which may be affected by various positional factors such as alignment probability bias. As a result, the corresponding second logarithm may be unreliable. Therefore, using a portion of these candidate nucleosome center locations as the final selected nucleosome center locations allows for the screening of nucleosome center scores.
[0290] Table 2 above provides several score screening criteria and the corresponding Pearson correlation coefficients of fetal concentration and nucleosome ratio under these criteria.
[0291] S205. Based on the nucleosome center position, the second logarithms of corresponding positions in all nucleosome regions are aligned and summed to obtain the summed second logarithms.
[0292] In some embodiments, when aligning and stacking according to the nucleosome region, a range of 200 bases to the left and right of the "nucleosome center position" is taken for alignment and summation, resulting in a vector of 401 elements, with the 201st element representing the nucleosome center position. This operation has no impact on the calculation of the "nucleosome ratio," but since machine learning algorithms are subsequently introduced to build a model, information from a larger range around the nucleosome region is extracted to improve the potential information content and predictive ability of the model.
[0293] After determining the nucleosome center location, for the sample data to be analyzed, the sequence start position count corresponding to each position on the reference genome can be calculated first using the method described above, i.e., the second alignment number can be obtained. Then, based on the processing procedure of the nucleosome center location, the second alignment number of 147 bases to the left and right of the nucleosome center location is extracted. This operation is performed for all nucleosome center locations.
[0294] S206. Determine the comparison result based on the first fetal concentration and the second comparison number after summation.
[0295] In this embodiment, the processing method for the feature values provided by the second processing mode is the same as when using the second processing mode alone. The difference is that the fetal concentration predicted in the first processing mode is added as one of the features, and principal component analysis dimensionality reduction is performed together. During model training, to ensure the stability of the model and the reliability of the model evaluation results, the test set (30%) and training set (70%) are randomly split 100 times. The elastic network model is also subjected to L1 and L2 regularization, and the optimal combination of L1 and L2 regularization parameters is selected iteratively. The specific effects of different models are detailed in Tables 3 and 4. In the table below, r2_test_enst represents the R value of the test set when modeling with the enst model. 2 r2_train_enst represents the R-value of the training set when modeling with the enst model. 2 `corr_test_enst` represents the Pearson correlation coefficient of the test set when modeling with the `enst` model; `corr_train_enst` represents the Pearson correlation coefficient of the training set when modeling with the `enst` model. Other values follow the same logic.
[0296] It should be noted that "randomly splitting the test set (30%) and training set (70%) 100 times" and calculating the R-value obtained from these 100 splits is... 2 The mean and STD of the Pearson correlation coefficient are used to evaluate the reliability of the model and its performance on a currently randomly split training set within a limited dataset. When building the model, only one split of the training and test sets was performed, and the model was built using R... 2The correlation coefficient was 0.61, and the Pearson correlation coefficient was 0.79. As shown in the table below, after another 100 regressions with random splitting of the training and test sets, the R-values were... 2 The mean correlation coefficients with Pearson were 0.61 and 0.78, respectively, and the STD values were 0.04 and 0.03, respectively. Therefore, it is considered that the model obtained by performing modeling with only one split (R²) is effective. 2 The correlation coefficient is 0.61, and the Pearson correlation coefficient is 0.79 (which is reliable).
[0297] Table 3
[0298] mean r2_test_enst 0.6033 r2_train_enst 0.6478 corrtestenst 0.7835 corr_train_enst 0.8063 r2_test_lasso 0.6078 r2_train_lasso 0.6565 corr_test_lasso 0.7839 corr_train_lasso 0.8111 r2_test_liner 0.6059 r2_train_liner 0.6614 corr_test_liner 0.7824 corr_train_liner 0.8132 r2_test_linerstep 0.6028 r2trainlinerstep 0.6286 corr_test_linerstep 0.7810 corr_train_linerstep 0.7927 r2_test_ridge 0.6066 r2_train_ridge 0.6604 corr_test_ridge 0.7824 corr_train_ridge 0.8132 r2_test_keras 0.5565 r2_train_keras 0.6408 corr_test_keras 0.7724 corr_train_keras 0.8209 r2_test_lgb 0.5340 r2_train_lgb 0.9127 corr_test_lgb 0.7488 corrtrainlgb 09724
[0299] Table 4
[0300]
[0301]
[0302] In some embodiments, as a possible implementation of this application, the alignment result is determined based on the first fetal concentration and the summed second alignment number, including: performing dimensionality reduction processing on the summed second alignment number corresponding to each base site in the nucleosome region to obtain a dimensionality-reduced normalized second alignment number; and determining the alignment result based on the first fetal concentration and the dimensionality-reduced normalized second alignment number. Alternatively, an initial alignment result is determined based on the first fetal concentration and the summed second alignment number; the initial alignment result is then dimensionality-reduced, and the dimensionality-reduced initial alignment result is determined as the alignment result.
[0303] For example, the 401 original features from the second processing mode (i.e., the 401 counts obtained by aligning and summing according to the nucleosome center positions) are combined with the fetal concentration values directly obtained by the WRSC method in the first determination mode to form 402 parameter data points, which are used as the original input feature values. Principal component analysis is used for dimensionality reduction, and a machine learning algorithm is used for modeling. The machine learning model can be various models such as elastic networks, Lasso, linear regression, stepwise regression, ridge regression, Keras, and lightGBM. Compared with other models, ridge regression performs better. The final R-value for fetal concentration estimation on the test set is... 2 It is 0.61.
[0304] S207. Based on the comparison results, determine the fetal concentration.
[0305] In another implementation, determining the fetal concentration based on the comparison results includes:
[0306] Obtain second training sample data. Each sample in the second training sample data is labeled with a feature value and a target value. The feature value is the normalized second alignment number after dimensionality reduction, and the target value is the actual fetal concentration. Create a second fetal concentration quantitative model based on the second training sample data. Input the alignment results into the second fetal concentration quantitative model to obtain the fetal concentration.
[0307] This embodiment is based on the method described in the previous embodiments that determines fetal concentration by utilizing the difference in alignment probability between maternal and infant cfDNA at different positions on the reference genome sequence (referred to as the first determination mode for ease of hereinafter); and the method that determines fetal concentration by utilizing the difference in the distribution of the two ends of maternal and infant cfDNA at the nucleosome positions (referred to as the second determination mode). Since these two methods use different principles to determine fetal concentration, combining the parameters of both methods to determine fetal concentration provides more effective information in the feature values and should result in a stronger estimation capability for fetal concentration. For example, merging the 401 original features from the second processing mode (i.e., the 401 counts obtained by aligning and summing according to the nucleosome center position) with the fetal concentration values directly obtained by the WRSC method in the first determination mode into 402 parameter data points, which are used as the original input feature values, using principal component analysis for dimensionality reduction, and modeling with machine learning algorithms, is beneficial for obtaining more accurate fetal concentration data.
[0308] Based on the foregoing embodiments, another embodiment of this application also provides a device for determining fetal concentration, see [link to relevant documentation]. Figure 9 The device may include:
[0309] The first acquisition unit 10 is used to acquire the sequencing sequence of the cfDNA sample to be tested;
[0310] The first alignment unit 11 is used to align the sequencing sequence with the reference genome to obtain the alignment result;
[0311] The first determining unit 12 is used to determine the fetal concentration based on the comparison results.
[0312] As one possible implementation of this application, the first comparison unit includes:
[0313] The first segmentation subunit is used to segment the reference genome into multiple segment intervals;
[0314] The first determining subunit is used to determine the first alignment number of each sequencing sequence falling into each segment interval, and the first alignment number is determined as the alignment result.
[0315] As one possible implementation of this application, the first segmented subunit is specifically used for:
[0316] The reference genome is segmented based on a preset segmentation length to obtain multiple initial segmentation intervals;
[0317] By removing the segment intervals corresponding to specific chromosomes from the initial segment intervals, multiple segment intervals are obtained;
[0318] or,
[0319] By removing specific chromosomes from the reference genome, a reference genome with the chromosomes removed is obtained;
[0320] The removed reference genome is segmented based on a preset segmentation length to obtain multiple segmentation intervals.
[0321] Among them, the specific chromosomes include at least one of the following: X chromosome, Y chromosome, mitochondrial chromosome, chromosome 13, chromosome 18, and chromosome 21.
[0322] As one possible implementation of this application, the apparatus further includes:
[0323] The first correction unit is used to correct the first alignment number to obtain the corrected first alignment number, so that the corrected first alignment number is determined as the alignment result.
[0324] As one possible implementation of this application, the first correction unit includes:
[0325] The first processing subunit is used to normalize the first logarithm of each segment interval based on the first logarithm of each segment interval and the average first logarithm of all segment intervals, so as to obtain the normalized first logarithm of each segment interval.
[0326] The GC correction subunit is used to perform GC correction on the normalized first logarithm of each segment interval to obtain the GC-corrected first logarithm.
[0327] As one possible implementation of this application, the GC correction subunit is specifically used for:
[0328] Based on the GC content and the normalized first logarithm corresponding to each segment interval, a first relationship curve is generated;
[0329] Based on the first relationship curve, GC correction is performed on the normalized first logarithm to obtain the GC-corrected first logarithm.
[0330] Prior to the step of generating the first relationship curve based on the GC content and the normalized first logarithm corresponding to each segment interval, the process also includes:
[0331] The segmented intervals are filtered based on GC content, so that the GC content of each segmented interval after GC filtering and the normalized first logarithm of each segmented interval are used to generate the first relationship curve.
[0332] or,
[0333] Based on the first relation curve, GC correction is performed on the normalized first logarithm to obtain the GC-corrected first logarithm, including:
[0334] The segmented intervals are filtered based on GC content, and the normalized first logarithm after GC filtering is GC corrected based on the first relationship curve to obtain the GC corrected first logarithm.
[0335] Specifically, GC correction involves determining the first logarithm after GC correction based on the subtraction or division relationship between the normalized first logarithm and the GC content of the segmented intervals corresponding to the normalized first logarithm.
[0336] In another embodiment, the first correction unit further includes:
[0337] The truncation subunit is used to truncate the reference genome based on a specific sliding window length, compare the truncated sequence with the reference genome, and count the first alignment number in each segment interval of the reference genome. The first alignment number is determined as the sliding window alignment number.
[0338] The second determining subunit is used to determine the normalized sliding window logarithm for each segment interval based on the sliding window logarithm.
[0339] The third determining subunit is used to perform alignment probability correction based on the normalized sliding window alignment number and the normalized first alignment number of each segment interval, and to determine the first alignment number after alignment probability correction.
[0340] As one possible implementation of this application, the third determining subunit is specifically used for:
[0341] A second relationship curve is generated based on the normalized sliding window logarithm of each segment interval and the normalized first logarithm corresponding to each segment interval.
[0342] Based on the second relationship curve, the normalized first logarithm is corrected for alignment probability to obtain the first logarithm after alignment probability correction.
[0343] In one implementation, before the step of generating the second relationship curve based on the normalized sliding window logarithm of each segment interval and the normalized first logarithm corresponding to each segment interval, the method further includes:
[0344] The segmented intervals are filtered based on the normalized sliding window comparison number to obtain segmented intervals where the normalized sliding window comparison number is not less than the first target threshold, so that the second relationship curve is generated in the segmented intervals where the normalized sliding window comparison number is not less than the first target threshold.
[0345] or,
[0346] Based on the second relationship curve, the normalized first logarithm is corrected for alignment probability to obtain the first logarithm after alignment probability correction, including:
[0347] Based on the normalized sliding window logarithm, the segmented intervals are filtered, and the intervals of the second relationship curve in which the normalized sliding window logarithm is not less than the first target threshold are retained, so as to obtain the second relationship curve after the comparison probability is filtered.
[0348] Based on the second relationship curve after alignment probability filtering, the alignment probability of the sliding window normalized alignment number is corrected to obtain the first alignment number after alignment probability correction.
[0349] As one possible implementation of this application, the first determining unit includes:
[0350] The first sample acquisition subunit is used to obtain the first training sample data. Each sample in the first training sample data is labeled with a first feature value and a first target value. The first feature value is the log number and the first target value is the actual fetal concentration.
[0351] The first modeling subunit is used to perform machine learning modeling on the first training sample data based on a specific model structure to obtain the first fetal concentration quantitative model.
[0352] The model processing subunit is used to input the comparison results into the first fetal concentration quantification model to obtain the first fetal concentration and determine the first fetal concentration as the fetal concentration.
[0353] As one possible implementation of this application, the first modeling subunit is specifically used for:
[0354] The first training sample data is divided into a training set and a test set;
[0355] Machine learning modeling is performed based on the training set to obtain an initial model;
[0356] The test set is processed based on the initial model to obtain the estimated fetal concentration for each test sample in the test set.
[0357] The comparison results are obtained by comparing the estimated fetal concentration with the actual fetal concentration of each test sample in the test set.
[0358] Based on the comparison results, the model parameters of the initial model were adjusted to obtain the first fetal concentration quantitative model.
[0359] Specifically, the first determining unit is also used for:
[0360] The comparison results are input into the first preset model to obtain the initial fetal concentration. The first preset model is a linear relationship pattern established based on the comparison results in the second sample data and the initial fetal concentration corresponding to the comparison results. The second sample data is the comparison results obtained by comparing the sequencing sequence in the cfDNA sample with the reference genome, and the fetal concentration corresponding to the comparison results.
[0361] The initial fetal concentration is corrected according to the second preset model to obtain the fetal concentration. The second preset model is a model obtained by processing the first preset model with constants determined by linear fitting.
[0362] In another implementation, the first comparison unit is specifically used for:
[0363] The count of each base site in the reference genome that is the starting position for sequencing sequence alignment is determined as the second alignment number, and the nucleosome center fraction corresponding to each base site in the reference genome is calculated based on the second alignment number.
[0364] The nucleosome center location is determined based on the nucleosome center score and the center score screening threshold;
[0365] Based on the nucleosome center position, the second logarithms of corresponding positions in all nucleosome regions are aligned and summed to obtain the summed second logarithm;
[0366] The second logarithm after summation is subjected to dimensionality reduction to obtain the dimensionality-reduced normalized second logarithm, and the dimensionality-reduced normalized second logarithm is determined as the alignment result;
[0367] The calculation of the nucleosome center fraction corresponding to each base site in the reference genome includes:
[0368] Calculate the first average count of the first specific number of base ranges to the left and right of each base site in the reference genome;
[0369] Calculate the second average count of the second specific number of base ranges to the left and right of each base site in the reference genome;
[0370] The nucleosome center score for each site is determined based on the first average count and the second average count.
[0371] As one possible implementation of this application, the formula for calculating the nucleosome center fraction is as follows:
[0372]
[0373] In the formula, x is a base site, [x-93,x-74-n] represents the interval from 93 nucleotides away from x on one side to 74-n nucleotides away from x on the same side, [x+93,x+74-n] represents the interval from 93 nucleotides away from x on the other side to 74-n nucleotides away from x on the same side, and [x-73-n,x+73-n] represents the interval from 73-n nucleotides away from x on the other side to 73-n nucleotides away from x on the other side. n is a natural number less than or equal to 5.
[0374] As one possible implementation of this application, the nucleosome center score is determined based on the nucleosome center score and a center score screening threshold, including:
[0375] Based on the maximum value of the nucleosome center fraction, the position of the maximum value on the reference genome is determined, and the position is determined as the first position;
[0376] The nucleosome center scores of the bases on both sides of the first position are cleared to zero, and the maximum value is determined again based on the remaining nucleosome center scores after clearing. The position of the maximum value on the reference genome is determined until the nucleosome center score is less than the second target threshold. The selected positions are then determined as candidate nucleosome center positions.
[0377] The nucleosome center position is determined by using the base center score screening threshold and the nucleosome center score corresponding to the candidate nucleosome center position.
[0378] Specifically, in the step of aligning and summing the second logarithms of corresponding positions within all nucleosome regions based on the nucleosome center position, the corresponding positions within the nucleosome regions include:
[0379] The range of a specific number of bases to the left and right of the center of the nucleosome.
[0380] In one implementation, the first determining unit is further configured to:
[0381] Acquire the second training sample data. Each sample in the second training sample data is labeled with a feature value and a target value. The feature value is the normalized second logarithm after dimensionality reduction, and the target value is the actual fetal concentration.
[0382] A second fetal concentration quantification model was created based on the second training sample data;
[0383] The comparison results are input into the second fetal concentration quantification model to obtain the fetal concentration.
[0384] Based on the foregoing embodiments, see Figure 10In another embodiment of this application, another device for determining fetal concentration is provided, which may include:
[0385] The second acquisition unit 20 is used to acquire the sequencing sequence of the cfDNA sample to be tested;
[0386] The second determining unit 21 is used to segment the reference genome to obtain multiple segment intervals and determine the first alignment number of each sequencing sequence falling into each segment interval.
[0387] The third determining unit 22 is used to determine the first fetal concentration based on the first logarithm;
[0388] The fourth determining unit 23 is used to determine the nucleosome center position based on the second alignment number of each base site in the reference genome as the starting position of the sequencing sequence alignment;
[0389] Processing unit 24 is used to align and sum the second logarithms of corresponding positions in all nucleosome regions based on the nucleosome center position to obtain the summed second logarithm;
[0390] The fifth determining unit 25 is used to determine the comparison result based on the first fetal concentration and the second comparison number after summation.
[0391] The sixth determining unit 26 is used to determine the fetal concentration based on the comparison results.
[0392] In some embodiments, the second determining unit includes:
[0393] The second segmentation subunit is specifically used for:
[0394] The reference genome is segmented based on a preset segmentation length to obtain multiple initial segmentation intervals;
[0395] By removing the segment intervals corresponding to specific chromosomes from the initial segment intervals, multiple segment intervals are obtained;
[0396] or,
[0397] By removing specific chromosomes from the reference genome, a reference genome with the chromosomes removed is obtained;
[0398] The removed reference genome is segmented based on a preset segmentation length to obtain multiple segmentation intervals.
[0399] In some embodiments, the apparatus further includes:
[0400] The second correction unit is used to correct the first logarithm to obtain the corrected first logarithm, and to determine the first fetal concentration using the corrected first logarithm.
[0401] In some embodiments, the second correction unit includes:
[0402] The second processing subunit is used to normalize the first logarithm of each segment interval based on the first logarithm of each segment interval and the average first logarithm of all segment intervals, so as to obtain the normalized first logarithm of each segment interval.
[0403] The GC correction subunit is used to perform GC correction on the normalized first logarithm of each segment interval to obtain the GC-corrected first logarithm.
[0404] As one possible implementation of this application, the GC correction subunit is specifically used for:
[0405] Based on the GC content and the normalized first logarithm corresponding to each segment interval, a first relationship curve matching a specific coordinate system is generated.
[0406] Based on the first relationship curve, GC correction is performed on the normalized first logarithm to obtain the GC-corrected first logarithm.
[0407] In some embodiments, the second correction subunit is further configured to:
[0408] The reference genome is truncated based on a specific sliding window length. The truncated sequence is compared with the reference genome, and the first alignment number is counted for each segment of the reference genome. The first alignment number is determined as the sliding window alignment number.
[0409] Based on the sliding window logarithm, determine the normalized sliding window logarithm for each segment interval;
[0410] Based on the normalized sliding window logarithm and the normalized first logarithm for each segment interval, the alignment probability is corrected to determine the first logarithm after alignment probability correction.
[0411] In some embodiments, the third determining unit is specifically used for:
[0412] Obtain the first training sample data. Each sample in the first training sample data is labeled with a first feature value and a first target value. The first feature value is the log number and the first target value is the actual fetal concentration.
[0413] Based on a specific model structure, machine learning modeling is performed on the first training sample data to obtain the first fetal concentration quantification model;
[0414] The first comparison result is input into the first fetal concentration quantification model to obtain the first fetal concentration.
[0415] As one possible implementation of this application, the third determining unit is further configured to:
[0416] The first alignment number is input into the first preset model to obtain the initial fetal concentration. The first preset model is a linear relationship pattern established based on the comparison results in the second sample data and the initial fetal concentration corresponding to the comparison results. The second sample data is the alignment result obtained by comparing the sequencing sequence in the cfDNA sample with the reference genome and the fetal concentration corresponding to the alignment result.
[0417] The initial fetal concentration is corrected according to the second preset model to obtain the first fetal concentration. The second preset model is a model obtained by processing the first preset model with constants determined by linear fitting.
[0418] In some embodiments, the fourth determining unit includes:
[0419] The fraction calculation subunit is used to determine the count of each base site in the reference genome that is the start position of the sequencing sequence alignment as the second alignment number, and to calculate the nucleosome center fraction corresponding to each base site in the reference genome based on the second alignment number.
[0420] The center location determination subunit is used to determine the center location of nucleosomes based on the nucleosome center score and the center score screening threshold.
[0421] In some embodiments, the fifth determining unit is specifically used for:
[0422] The second logarithm, after summing the base sites in the nucleosome region, is dimensionality reduced to obtain the normalized second logarithm after dimensionality reduction.
[0423] The alignment result was determined based on the first fetal concentration and the normalized second logarithm after dimensionality reduction.
[0424] or,
[0425] The initial alignment results were determined based on the first fetal concentration and the second alignment number after summation.
[0426] The initial alignment results are dimensionality reduced, and the dimensionality-reduced initial alignment results are determined as the alignment results.
[0427] It should be noted that the execution process of each unit of the fetal concentration determination device in this embodiment can be found in [reference needed]. Figure 8 The method for determining the fetal concentration is not detailed here.
[0428] In another embodiment of this application, a readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method for determining fetal concentration as described above.
[0429] In another embodiment of this application, an electronic device is also provided, which may include:
[0430] Memory is used to store applications and the data generated during application execution.
[0431] A processor for executing applications to implement any of the methods for determining fetal concentrations as described above.
[0432] It should be noted that the specific implementation of the processor in this embodiment can be referred to the corresponding content above, and will not be described in detail here.
[0433] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.
[0434] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0435] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0436] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for determining fetal concentration, It is characterized in that include: Obtaining the sequencing sequence of the cfDNA sample to be tested; Comparing the sequencing sequence with the reference genome to obtain a comparison result includes: Determine the count of the base sites at which each base site of the reference genome is the starting position of the sequencing sequence alignment as a second alignment number, and calculate the nucleosome center score corresponding to each base site of the reference genome based on the second alignment number; Determining the nucleosome center position based on the nucleosome center score and the center score screening threshold; Based on the nucleosome center position, align and sum the second alignment numbers of corresponding positions in all nucleosome regions to obtain the summed second alignment numbers; Performing dimensionality reduction processing on the second alignment number after the addition processing to obtain a normalized second alignment number after dimensionality reduction, and determining the normalized second alignment number after dimensionality reduction as the alignment result; Based on the comparison results, the fetal concentration is determined.
2. The method according to claim 1, It is characterized in that The calculating the nucleosome center score corresponding to each base site of the reference genome comprises: Calculate a first average count of a first specific number of base ranges corresponding to each base site on the left and right of the reference genome; Calculate a second average count of a second specific number of base ranges corresponding to each base site on the left and right of the reference genome; The nucleosome center score corresponding to each site is determined according to the first average count and the second average count.
3. The method according to claim 2, Features ,, the calculation formula of the nucleosome center score is: In the formula, x is the base site, [x-93,x-74-n] represents the interval from 93 nucleotides on one side of x to 74-n nucleotides on the same side, [x+93,x+74-n] represents the interval from 93 nucleotides on the other side of x to 74-n nucleotides on the same side, [x-73-n,x+73-n] represents the interval from 73-n nucleotides on the other side of x to 73-n nucleotides on the other side of x, and n is a natural number less than or equal to 5.
4. The method according to any one of claims 1 to 3, It is characterized in that The determining of the nucleosome center score based on the nucleosome center score and the center score screening threshold comprises: Based on the maximum value of the nucleosome center score, determining the position of the maximum value on the reference genome, and determining the position as the first position; clearing the nucleosome center scores of the bases of the specific data on both sides of the first position, and determining the maximum value again based on the remaining nucleosome center scores after clearing, and determining the position of the maximum value on the reference genome until the nucleosome center score is less than a second target threshold, and determining the screened position as a candidate nucleosome center position; The nucleosome center position is determined based on the center score screening threshold and the nucleosome center score corresponding to the candidate nucleosome center position.
5. The method according to any one of claims 1 to 3, It is characterized in that In the step of aligning and summing the second alignment numbers of corresponding positions in all nucleosome regions based on the nucleosome center position, the corresponding positions in the nucleosome region include: The range of a specific number of bases is around the center of the nucleosome.
6. The method according to any one of claims 1 to 3, It is characterized in that Determining the fetal concentration based on the comparison result comprises: Acquire second training sample data, wherein each sample in the second training sample data is labeled with a characteristic value and a target value, the characteristic value is the normalized second logarithm after the dimension reduction, and the target value is the actual fetal concentration; Creating a second fetal concentration quantitative model based on the second training sample data; The comparison result is input into the second fetal concentration quantitative model to obtain the fetal concentration.
7. A device for determining fetal concentration, It is characterized in that include: A first acquisition unit is used to obtain a sequencing sequence of the cfDNA sample to be tested; The first comparison unit is used to compare the sequencing sequence with the reference genome to obtain a comparison result, including: Determine the count of the base sites at which each base site of the reference genome is the starting position of the sequencing sequence alignment as a second alignment number, and calculate the nucleosome center score corresponding to each base site of the reference genome based on the second alignment number; Determining the nucleosome center position based on the nucleosome center score and the center score screening threshold; Based on the nucleosome center position, align and sum the second alignment numbers of corresponding positions in all nucleosome regions to obtain the summed second alignment numbers; Performing dimensionality reduction processing on the second alignment number after the addition processing to obtain a normalized second alignment number after dimensionality reduction, and determining the normalized second alignment number after dimensionality reduction as the alignment result; The first determination unit is used to determine the fetal concentration based on the comparison result.
8. The device according to claim 7, It is characterized in that The first comparison unit is specifically used for: Determine the count of the base sites at which each base site of the reference genome is the starting position of the sequencing sequence alignment as a second alignment number, and calculate the nucleosome center score corresponding to each base site of the reference genome based on the second alignment number; Determining the nucleosome center position based on the nucleosome center score and the center score screening threshold; Based on the nucleosome center position, align and sum the second alignment numbers of corresponding positions in all nucleosome regions to obtain the summed second alignment numbers; The second alignment number after the addition process is subjected to dimensionality reduction process to obtain a normalized second alignment number after dimensionality reduction, and the normalized second alignment number after dimensionality reduction is determined as the alignment result.
9. The device according to claim 8, It is characterized in that The calculating the nucleosome center score corresponding to each base site of the reference genome comprises: Calculate a first average count of a first specific number of base ranges corresponding to each base site on the left and right of the reference genome; Calculate a second average count of a second specific number of base ranges corresponding to each base site on the left and right of the reference genome; The nucleosome center score corresponding to each site is determined according to the first average count and the second average count.
10. The device according to any one of claims 7 to 9, It is characterized in that The first determining unit is further configured to: Acquire second training sample data, wherein each sample in the second training sample data is labeled with a characteristic value and a target value, the characteristic value is the normalized second logarithm after the dimension reduction, and the target value is the actual fetal concentration; Creating a second fetal concentration quantitative model based on the second training sample data; The comparison result is input into the second fetal concentration quantitative model to obtain the fetal concentration.
Citation Information
Patent Citations
Fetal concentration quantification method and device and fetal genotyping method and device
CN110993025A
cffDNA aneuploid detection method and system, storage medium and detection equipment
CN112037846A
Transcription factor binding site analysis of nucleosome depleted circulating cell free chromatin fragments
WO2022144408A1
