A method for constructing a prediction model related to preeclampsia, a feature combination and application thereof

Abstract

The application provides a prediction model construction method, an analysis system, a target gene combination, a feature combination, a detection method, a detection model and related applications related to preeclampsia. Specifically, the application uses a machine learning method to combine cfDNA features and clinical indicators to construct a model for predicting preeclampsia. The application can make up for the volatility of prediction results between data when using clinical features alone for prediction. Moreover, the cfDNA features used for prediction come from the analysis results of low-depth whole genome sequencing data of pregnant women in the early / middle stage of pregnancy, so compared with other methods, it is more economical and convenient, and has better application prospects.

Description

Technical Field

[0001] This invention relates to the field of preeclampsia analysis, specifically to a method for constructing a prediction model related to preeclampsia, an analysis system, target gene combinations, feature combinations, detection methods, detection models, and related applications. Background Technology

[0002] Preeclampsia (PE) is a pregnancy disorder characterized by new-onset hypertension after 20 weeks of gestation, accompanied by proteinuria or damage to other organs (such as the kidneys, liver, or brain). When the condition progresses, it can lead to complications such as eclamptic seizures, HELLP syndrome, and intracranial hemorrhage, endangering the lives of both mother and fetus. Preeclampsia and its associated hypertension are the second leading cause of maternal mortality. Preeclampsia is classified into early-onset and late-onset preeclampsia based on the time of onset (34 weeks as the cutoff). Currently, the main treatment remains termination of pregnancy and placental delivery, primarily because the pathogenesis of preeclampsia is complex, thus lacking effective targets and treatments. Early-onset preeclampsia, due to its early onset and rapid progression, can lead to premature delivery and adverse consequences for fetal organ development caused by prematurity. Therefore, earlier and more accurate prediction and intervention (e.g., using low-dose aspirin) are particularly important for patients with early-onset preeclampsia.

[0003] To date, nearly 100 articles have evaluated the relationship between clinical risk factors in early pregnancy and the risk of preeclampsia. Significant clinical risk factors include a history of preeclampsia, chronic hypertension, age, BMI, mode of pregnancy, blood pressure, and history of diabetes. Although various professional institutions and organizations recommend multiple combinations of maternal risk factors for preeclampsia screening, using these combinations alone often results in low accuracy or a high false-positive rate. The preeclampsia risk assessment model from the UK Fetal Medicine Foundation, which incorporates maternal biochemical indicators, has been widely validated in different populations. The best predictive model combines maternal factors with mean arterial pressure, uterine artery pulsatility index, and serum placental growth factor (PLGF). With a false-positive rate of 10%, the detection rates for very early (less than 32 weeks of gestation), preterm, and term preeclampsia were 90%, 75%, and 41%, respectively.

[0004] In recent years, research has found that the whole-genome promoter profile of cell-free DNA (cfDNA) in maternal peripheral blood can reflect gene expression in cells and can be applied to predict pregnancy-related diseases. The invention "A Target Gene Combination Associated with Early-Onset Severe Preeclampsia and Its Application" (Application No.: CN110305954A) utilizes whole-genome sequencing of cell-free DNA in pregnant women's plasma (data volume: 4X) to discover significant differences in the abundance of transcription start sites in certain genes between patients with early-onset severe preeclampsia and healthy pregnant women. After standardizing and discretizing the abundance results of these differentially expressed transcription start sites, and combining them with gestational age information, a machine learning algorithm was used to construct a model that can effectively predict the risk of early-onset severe preeclampsia. The invention, "A Predictive Model for Early and Accurate Detection of Preeclampsia" (Application No.: CN114822682A), is based on the significant difference in serum cell-free DNA abundance in gene transcription start site regions between preeclampsia patients and healthy pregnant women. After normalizing and correcting the cell-free DNA abundance in pregnant women, a machine learning algorithm is used to construct a model that can effectively predict the onset of preeclampsia through the optimal combination of different differentially expressed genes. However, the above method still has various problems, such as only considering the overall expression profile of pregnant women's cfDNA in the promoter region, lacking comprehensive consideration of other features; the constructed model lacks universality; and the cost of 4X whole genome sequencing is high. Therefore, it is necessary to develop new, more accurate and effective detection methods. Summary of the Invention

[0005] The purpose of this invention is to obtain a series of cfDNA features for predicting preeclampsia in early pregnancy by screening and calculating various physiological and biochemical indicators of pregnant women. These cfDNA features are then combined with clinical indicators, and a machine learning method is used to provide an optimal feature combination and prediction model that can accurately and stably predict and differentiate patients with preeclampsia.

[0006] In a first aspect, the present invention provides a method for constructing a predictive model related to preeclampsia, comprising:

[0007] a. Sample Analysis Procedure: After aligning and deduplicating the whole-genome sequencing data of cfDNA from pregnant women's samples with a reference genome, several types of cfDNA feature analyses were performed on the same sample. cfDNA features included the expression profile and distribution profile of the transcript promoter region, as well as the frequency of the 6-base motif at the 5' end of the cfDNA. The pregnant women's samples included both diseased and healthy control groups.

[0008] Specifically, the analysis of several types of cfDNA features includes: analyzing the expression profile of each transcript promoter region in the same sample; dividing each promoter region of the reference genome into several equally divided subregions and analyzing the expression distribution trend of each transcript promoter region in the same sample; and calculating the frequency of the 5' end 6-base motif of cfDNA in the same sample.

[0009] b. cfDNA feature screening steps: Based on the analysis results of step a above for different samples, statistical screening is conducted for transcript promoter regions with significantly different coverage, significantly different distribution, and significantly different frequencies of cfDNA 5' end 6-base sequences that are significantly different between the disease group and the healthy control group, resulting in several candidate cfDNA features.

[0010] c. Predictive Model Construction Steps: Using the pregnant woman's clinical indicators and the aforementioned candidate cfDNA features, a predictive model is constructed using machine learning algorithms to obtain the optimal feature combination and the corresponding predictive model. This feature combination includes several cfDNA features and clinical indicators.

[0011] In other words, the model construction method of this invention involves: using the pregnant woman's clinical indicators and cfDNA characteristics as input, and whether the pregnant woman has preeclampsia as output, constructing a predictive model through machine learning algorithms, and obtaining the optimal feature combination and the corresponding predictive model. The aforementioned feature combination includes several cfDNA characteristics and clinical indicators, and the cfDNA characteristics include transcript promoter regions with significantly different coverage and distribution between the diseased group and the healthy control group, and cfDNA 5' terminal 6-base motifs with significantly different frequencies.

[0012] Secondly, corresponding to the aforementioned predictive model construction method related to preeclampsia, this invention provides an analysis system related to preeclampsia, which includes:

[0013] The sample analysis module is used to implement the sample analysis steps mentioned in the first part above, analyzing the expression profiles and distribution patterns of transcript promoter regions, as well as the frequency of the 6-base motif at the 5' end of cfDNA. Specifically, it aligns the whole-genome sequencing data of cfDNA from pregnant women's samples to a reference genome, analyzes the expression profile of each transcript promoter region, divides each promoter region into several equally divided subregions, and analyzes their expression distribution trends; and analyzes the frequency of several base motifs at the 5' end of the sequencing sequence aligned to the reference genome. The pregnant women's samples include diseased groups and healthy control groups.

[0014] Feature screening module: This module implements the cfDNA screening steps described in Part 1 above, obtaining several candidate cfDNA features for predicting preeclampsia models. Specifically, based on the results of the sample analysis module, it statistically screens transcript promoter regions and cfDNA 5' terminal 6-base sequences that show significant differences in coverage, distribution, and frequency between the diseased group and the healthy control group, obtaining several candidate cfDNA features.

[0015] Prediction Model Building Module: This module is used to implement the steps for building prediction models mentioned in the first part above. Specifically, it uses candidate cfDNA features obtained from the feature screening module and clinical indicators of pregnant women to build prediction models through machine learning algorithms, and obtains the optimal feature combination and the corresponding prediction model. This feature combination includes several cfDNA features and clinical indicators.

[0016] The aforementioned analysis system may further include a calculation module, which processes data of the combination of the aforementioned characteristics of the pregnant woman to be tested into the prediction model to obtain a judgment result on whether the pregnant woman to be tested will develop preeclampsia.

[0017] Part Three, the present invention provides a target gene combination associated with preeclampsia, characterized in that the target gene combination is any combination of KRTAP5-9, TMPRSS6, EFNA2, MOV10L1, FOSL2, PDE6G and GLTP.

[0018] Furthermore, this invention provides the application of the aforementioned target gene combination as a biomarker for screening early-onset preeclampsia. It should be noted that in this invention, the cfDNA characteristics of the aforementioned target genes are closely related to preeclampsia, especially early-onset preeclampsia, and can be used as a biomarker for preeclampsia in related fields. Simultaneously, the aforementioned target gene combination is closely related to the detection of preeclampsia, and therefore can be used to prepare preeclampsia detection products. The above applications include, but are not limited to, a series of primers, probes, and other related reagents that can detect the aforementioned target gene combination and amplify the aforementioned target gene fragments.

[0019] Part Four, this invention provides a combination of features associated with preeclampsia, which includes several cfDNA features and clinical indicators, wherein:

[0020] The cfDNA characteristics include: expression profiles of the promoter regions of the KRTAP5-9 gene transcripts, the TMPRSS6 gene transcripts, the EFNA2 gene transcripts, the MOV10L1 gene transcripts, the FOSL2 gene transcripts, the PDE6G gene transcripts, the GLTP gene transcripts, and the cfDNA 5' end ACCGGC motif frequency.

[0021] Clinical indicators include: maternal age, BMI, parity, past medical history, and mode of pregnancy.

[0022] Part Five, this invention provides information on the application of the above-mentioned target gene combinations or characteristic combinations in the preparation of detection products related to preeclampsia.

[0023] Part Six, based on the prediction model and feature combination obtained in the prediction model construction method in Part One above, this invention provides a detection method related to preeclampsia, which includes:

[0024] 1) Obtain data on the characteristic combination of the pregnant woman to be tested, which includes cfDNA characteristics and clinical indicators.

[0025] 2) Input the feature combination data obtained in 1) into the prediction model for processing to obtain the detection results.

[0026] Part VII, based on the detection method in Part VI above, discloses a detection system related to preeclampsia, the system comprising:

[0027] 1) A device for acquiring or inputting data on a combination of features associated with preeclampsia in a pregnant woman to be tested.

[0028] 2) A device for performing predictive model processing on data of preeclampsia-related feature combinations obtained in 1).

[0029] 3) Devices for outputting detection results.

[0030] Part 8, the present invention provides a detection product related to early-onset preeclampsia, the product comprising: a memory and a processor. The memory is used to store a program; the processor is used to execute the program stored in the memory to implement the detection method related to preeclampsia as mentioned in aspect 6 above.

[0031] In addition, the present invention also provides a computer-readable storage medium storing a program that can be executed by a processor to implement the detection method related to preeclampsia as mentioned in the sixth aspect above.

[0032] The beneficial effects of this invention are as follows: Based on low-depth whole-genome sequencing data of pregnant women's cfDNA in early pregnancy, this invention has found several differential cfDNA features in early pregnancy compared to healthy controls, especially in the early-onset preeclampsia group, across multiple datasets. These features include the coverage and relative distribution of transcript promoter regions, and the frequency of the 5'-terminal 6-base motif in cfDNA. These cfDNA features can serve as biomarkers for screening early-onset preeclampsia based on low-depth whole-genome sequencing data of pregnant women's cfDNA in early pregnancy. Furthermore, this study uses machine learning methods to construct a model for predicting early-onset preeclampsia by combining cfDNA features with clinical indicators, which can compensate for the variability in prediction results between data sets when using clinical features alone. Moreover, the cfDNA features used for prediction are derived from the analysis results of low-depth whole-genome sequencing data of pregnant women in early / mid pregnancy, making it more economical and convenient compared to other methods. The prepared early-onset preeclampsia screening product also has better application prospects. Attached Figure Description

[0033] Figure 1 This example illustrates the difference between the disease group and the control group in the dataset selected by the prediction model for individual cfDNA features;

[0034] Figure 2 This example illustrates the difference between the disease group and the control group in the dataset selected by the prediction model for cfDNA feature combinations.

[0035] Figure 3 This is a graph showing the performance of a model that uses clinical indicators alone in the dataset selected by model feature screening in the example.

[0036] Figure 4 The image shows the performance of the model combining cfDNA features and clinical indicators of pregnant women in the training and validation sets, as illustrated in the examples. Detailed Implementation

[0037] This invention, based on whole-genome sequencing data of cfDNA from early pregnancy, deeply mines multiple characteristics of peripheral blood cfDNA in pregnant women, creatively discovering an association between several cfDNA features and preeclampsia, showing significant differences compared to the healthy control group. Through detailed research on these cfDNA features, this invention finds that it can improve the predictive effect of using single clinical indicators of pregnant women for preeclampsia, especially for the prediction of early-onset preeclampsia at first diagnosis.

[0038] Accordingly, the present invention provides a method for constructing a predictive model related to preeclampsia mentioned in the first aspect of the foregoing invention, and an analysis system for the second aspect of the foregoing invention obtained based on the construction method.

[0039] Furthermore, in a specific embodiment of the present invention relating to early-onset preeclampsia, the sample analysis step in the preeclampsia-related prediction model construction method described above specifically includes:

[0040] a-1. Sample selection and sequencing: Pregnant women with early-onset preeclampsia were selected based on the diagnosis time of preeclampsia ≤34 weeks. The healthy control group consisted of randomly selected pregnant women and pregnant women with healthy fetuses during the same period. Low-depth whole-genome sequencing data of cfDNA in early pregnancy were retrieved from both groups of pregnant women.

[0041] a-2. Alignment and deduplication of sequencing data: The low-depth whole-genome sequencing data of the pregnant woman's early gestational cfDNA in a-1 is aligned with the reference genome hg38, and multiple matching results and PCR duplications are removed to obtain the effective sequencing sequence.

[0042] a-3.cfDNA feature 1 (expression profile of transcript promoter region) is obtained as follows: based on the total number of valid sequencing sequences obtained by screening (a-2) as a standardization parameter, the coverage of each transcript promoter region of the same sample is corrected for sequencing depth according to formula (1) to obtain the expression profile of each transcript promoter region of the same sample, and the median is corrected in the same batch of data.

[0043]

[0044] Here, total reads represents the total number of valid sequencing sequences obtained through (a-2) screening.

[0045] a-4.cfDNA Feature 2 (Distribution and Expression Profile of Transcript Promoter Regions), which is obtained as follows: Each transcript promoter region Xkb in (a-3) is divided into Y equal windows, each window being X / Ybp in size. The difference between the average coverage of the Z windows closest to the transcription start site and the average coverage of the W windows furthest from the transcription start site is calculated and standardized according to formula (2-1) to obtain the expression distribution trend of each transcript promoter region in the same sample, and then the median is corrected in the same batch of data.

[0046]

[0047] Wherein, depth of midBin and depth of sideBin represent the coverage of the Z windows closest to the transcription start site and the W windows furthest from the transcription start site, respectively; total reads represent the total number of valid sequencing sequences obtained by (a-2) screening.

[0048] Preferably, in one specific embodiment of the present invention, the above-mentioned transcript promoter regions are -1kb to +1kb upstream and downstream of the transcription start site of each transcript. For example, the 2kb promoter region of each transcript in (a-3) is divided into 20 equal windows, each window being 100bp. The difference between the average coverage of the 6 windows closest to the transcription start site and the average coverage of the 10 windows furthest from the transcription start site is calculated and standardized according to formula (2) to obtain the expression distribution trend of each transcript promoter region in the same sample, and the median is corrected in the same batch of data.

[0049]

[0050] Here, depth of midBin and depth of sideBin represent the coverage of the 6 windows closest to the transcription start site and the 10 windows furthest away, respectively; total reads represent the total number of valid sequencing sequences obtained by (a-2) screening.

[0051] It should be noted that when other division methods are used above, the specific data in formula (2-1) can be adjusted according to the above example.

[0052] a-5.cfDNA Feature 3 (Frequency of the 6-base sequence at the 5' end of cfDNA): Calculate the 4096 (4) bases at the 5' end of the effective sequencing sequence of cfDNA obtained from (a-2) screening. 6 The frequency of 6-base sequences was determined and corrected for median within the same batch of data.

[0053] Meanwhile, the sample analysis module of the analysis system related to preeclampsia in this application is preferably specifically used to implement the above steps a-1 to a-5.

[0054] As a preferred embodiment, the aforementioned low-depth whole-genome sequencing data of early pregnancy cfDNA refers to high-throughput sequencing data of single-end 35bp cell-free DNA from the peripheral blood of pregnant women obtained in early pregnancy using DNA nanosphere sequencing technology (DNBSEQ), with a data volume of 6M (0.1X). ​​This invention can also be achieved using other sequencing methods or with different data volumes.

[0055] In a specific embodiment of the present invention relating to early-onset preeclampsia, the cfDNA feature screening step in the preeclampsia-related prediction model construction method described above specifically includes:

[0056] b-1. Pregnant women from the same hospital in both the disease group and the healthy control group were used as a dataset. Data sets from different hospitals were processed separately. Wilcox.test was used to screen and identify promoter regions of transcripts with significantly different coverage, significantly different distribution, and significantly different frequencies of cfDNA 5' end 6-base motifs in the same dataset between the disease group and the healthy control group. The cfDNA feature analysis results were obtained.

[0057] It should be noted that when analyzing data from only one hospital, step b-1 above is sufficient to obtain candidate cfDNA features. Furthermore, when analyzing data from multiple hospitals, step b-2 is added. Different datasets correspond to different hospitals.

[0058] b-2. Based on the cfDNA feature results obtained by (b-1) for different datasets, cfDNA feature results that show the same trend of change in the diseased group compared with the healthy control group and have significant differences in at least multiple datasets are selected as candidate cfDNA features.

[0059] It should be noted that in this invention, during the model construction process, it is preferable to use data from at least four different hospitals for separate analysis. The advantage of this approach is that merging samples from different hospitals and redistributing the training set for feature selection can, to some extent, eliminate the inherent differences between the data source hospitals. These differences may stem from population distribution, the samples themselves, post-sampling sample processing, and biases introduced during the experimental process—differences that are unavoidable in practical applications. Therefore, to extract more universally applicable and valuable cfDNA features and to reflect the actual application effect of the model across different hospitals, it is preferable to divide the dataset based on hospitals. However, this does not affect the value of this invention in the individual application to a single hospital; rather, the combined application across multiple hospitals yields more stable cfDNA features under conditions of unknown biases. When selecting candidate cfDNA features, if the condition of significant differences across all four hospitals cannot be found or only 1-2 candidate cfDNA features are available, this can be appropriately relaxed to at least significant differences across three hospitals. Those skilled in the art can verify and debug this invention in conjunction with practical applications.

[0060] Furthermore, "same trend of change" includes the aforementioned cfDNA features 1-3. This trend refers to whether the disease group is upregulated or downregulated relative to the healthy control group. For example, if the average result of cfDNA feature 1 for a specific gene is 1 in the control group, while it is 2 in the disease group, it means that cfDNA feature 1 of that gene is upregulated in the disease group.

[0061] Meanwhile, the feature screening module of the analysis system related to preeclampsia in this application is preferably specifically used to implement the above steps b-1 to b-2.

[0062] In a specific embodiment of the present invention relating to early-onset preeclampsia, the step of constructing the prediction model in the above-described method for constructing a prediction model related to preeclampsia further includes:

[0063] c-1. Median correction for maternal age and BMI: The age of each maternal sample was corrected to the median of the age of all maternal samples in the same dataset, and the BMI of each maternal sample was corrected to the median of the BMI of all maternal samples in the same dataset.

[0064] c-2. The parity, medical history, and mode of pregnancy are discrete values. The medical history is the sum of the number of patients with a history of chronic hypertension, preeclampsia, systemic lupus erythematosus, and antiphospholipid syndrome. The mode of pregnancy is 1 if it is assisted reproductive technology and 0 if it is natural pregnancy.

[0065] c-3. Based on the trend of cfDNA feature changes in the disease group compared with the healthy control group, the multiplication and division pattern is used to traverse (b-2) to obtain a variety of candidate cfDNA feature combinations (the number of which is as shown in Formula 3).

[0066] Number of candidate cfDNA feature combinations

[0067] N and M represent the number of candidate cfDNA features that are upregulated and downregulated in the disease group compared to the healthy control group, respectively.

[0068] c-4. By combining candidate cfDNA features from pregnant women with clinical indicators, a machine learning model is constructed to predict early-onset preeclampsia. In the feature combination, cfDNA features are input into the prediction model in the form of a product.

[0069] It should be noted that the machine learning method described above is preferably a neural network. During the model building process, the results from the training set are used to optimize various parameters of the model. In one specific implementation, Python is preferably used for predictive model building.

[0070] It should be noted that, in the specific embodiments of this application, an optimal combination of features was obtained through the above method. This optimal combination of features includes cfDNA features related to several target genes and several clinical indicators. In this case, the cfDNA features of several target genes provided by the present invention are the optimal combination selected by the model from the candidate cfDNA feature combinations in c-3 above, which is the cfDNA feature input when the model is used for prediction. Specifically, the several clinical indicators include maternal BMI, age, parity, past medical history, and pregnancy method. The cfDNA features include: expression profiles of the KRTAP5-9 gene transcript promoter region, TMPRSS6 gene transcript promoter region, EFNA2 gene transcript promoter region, MOV10L1 gene transcript promoter region, FOSL2 gene transcript promoter region, PDE6G gene transcript promoter region, GLTP gene transcript promoter region, and the cfDNA 5' end ACCGGC motif frequency (e.g., ...). Figure 1 and Figure 2 (As shown). Meanwhile, the predictive model building module of the analysis system related to preeclampsia in this application is preferably specifically used to implement steps c-1 to c-4 above.

[0071] It should be noted that the methods for obtaining the cfDNA characteristics and clinical information of the pregnant woman to be tested in the calculation module, detection method, and detection system mentioned above all refer to the relevant content of the prediction model construction method mentioned above. In a preferred embodiment of this application, the cfDNA characteristics in the feature combination are input into the prediction model in the form of a product, specifically, ACCGGC motif frequency * KRTAP5-9 gene transcript promoter region expression spectrum * TMPRSS6 gene transcript promoter region expression spectrum * EFNA2 gene transcript promoter region expression spectrum * MOV10L1 gene transcript promoter region expression spectrum * FOSL2 gene transcript promoter region expression spectrum * PDE6G gene transcript promoter region expression spectrum * GLTP gene transcript promoter region expression distribution spectrum (which can also be written as ACCGGC * KRTAP5-9 * TMPRSS6 * EFNA2 * MOV10L1 * FOSL2 * PDE6G * GLTP). The specific calculations for the expression profiles of the KRTAP5-9 gene transcript promoter region, TMPRSS6 gene transcript promoter region, EFNA2 gene transcript promoter region, MOV10L1 gene transcript promoter region, FOSL2 gene transcript promoter region, and PDE6G gene transcript promoter region are as described in formula (1) above (a-3). The specific calculations for the expression distribution profile of the GLTP gene transcript promoter region are as described in formula (2) above (a-4). The specific frequency of the cfDNA 5' end ACCGGC motif is as described in formula (a-5) above (a-5). The clinical indicators in the mentioned feature combinations are converted into discrete data and input into the prediction model, referring to formulas (c-1) and (c-2) above (c-2). The data of the above feature combinations are input into the constructed prediction model to predict whether the sample has preeclampsia at its first diagnosis. If the model output is 0, the sample is judged to be low-risk for preeclampsia at its first diagnosis; if the model output is 1, the sample is judged to be high-risk for preeclampsia at its first diagnosis.

[0072] In the specific embodiments of the present invention, the sample used is the plasma of a pregnant woman, but other sample forms, such as whole blood, amniotic fluid, serum, urine, etc., may also be used.

[0073] The specific embodiments of the present invention are mainly aimed at early-onset preeclampsia, but it should be understood that the same method can be used to construct a prediction model for late-onset preeclampsia.

[0074] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0075] Example:

[0076] 1. Sample collection

[0077] 1.1 Inclusion criteria

[0078] Inclusion criteria for early-onset preeclampsia at first diagnosis: Pregnant women at 20-34 weeks of gestation with systolic blood pressure ≥140 mmHg and / or diastolic blood pressure ≥90 mmHg, accompanied by any one of the following: urinary protein quantification ≥0.3g / 24h, or urinary protein / creatinine ratio ≥0.3, or random urinary protein ≥(+) (testing method when quantification of protein is not available); no proteinuria but accompanied by involvement of any of the following organs or systems: vital organs such as the heart, lungs, liver, and kidneys, or abnormal changes in the blood system, digestive system, or nervous system, or placental-fetal involvement, etc.

[0079] Inclusion criteria for healthy controls: The sample consisted of full-term singleton pregnancies without pregnancy complications, with the fetus growing well at birth and no obstetric, medical, or surgical complications during pregnancy. Exclusion criteria: (1) other pregnancy complications; (2) severe heart, liver, or kidney dysfunction; (3) patients with autoimmune diseases or malignant tumors. Pregnant women with abnormalities due to chromosomal abnormalities, congenital abnormalities, premature birth, or multiple pregnancies were excluded.

[0080] 1.2 Participants: This invention employs a disease control method, using a unique information matching principle among pregnant women in four hospitals. Clinical data of pregnant women aged 24–45 years who underwent prenatal cfDNA low-depth whole-genome sequencing at BGI Genomics between 2018 and 2021 were exported from the HIS system. A total of 2,140 cases meeting the inclusion criteria in 1.1 were selected, as detailed in Table 1. This invention uses retrospective samples and data, and generalized informed consent has been obtained.

[0081] Table 1. Statistics on the number of samples in each group

[0082]

[0083] 2. Characterization of cell-free DNA in plasma

[0084] After obtaining low-depth whole-genome sequencing data of cfDNA from enrolled pregnant women in early pregnancy, the data was compared with the human reference genome hg38. After removing multiple matching results and PCR replicates, cfDNA characteristics were analyzed, including:

[0085] cfDNA Feature 1: The total number of valid sequencing sequences obtained after alignment and filtering is used as a standardized parameter. The coverage of the promoter region of each transcript is analyzed according to formula (1), and the median is corrected in the same batch of data.

[0086]

[0087] The total reads represent the total number of valid sequencing sequences obtained after alignment and filtering.

[0088] cfDNA Feature 2: The 2kb promoter region of each transcript is divided into 20 equal windows, each window being 100bp. The difference between the average coverage of the 6 windows closest to the transcription start site and the average coverage of the 10 windows furthest from the transcription start site is calculated and normalized according to Formula (2), and then corrected for median in the same batch of data.

[0089]

[0090] The depth of midBin and depth of sideBin represent the coverage of the 6 windows closest to the transcription start site and the 10 windows furthest from the transcription start site, respectively; total reads represent the total number of valid sequencing sequences obtained after alignment and filtering.

[0091] cfDNA Feature 3: 4096 (4) 5' ends of the effective sequencing sequences obtained after statistical alignment and filtering. 6 The frequency of 6-base sequences was determined and corrected for median within the same batch of data.

[0092] 3. Screening of cfDNA features for predicting early-onset preeclampsia at first diagnosis

[0093] The four sets of data used for model construction in Table 1 (from four hospitals) were analyzed using wilcox.test to identify three cfDNA features that showed significant differences between the first-diagnosis early-onset preeclampsia group and the healthy control group. By comparing the differences among the four sets of data, features that showed the same trend compared to the healthy control group and were significantly different in at least three hospitals were selected as candidate cfDNA features.

[0094] 4. Model construction for predicting early-onset preeclampsia at first visit

[0095] 4.1 Correction of Clinical Indicators for Pregnant Women: The median age and BMI of the pregnant women in the four groups of data in Table 1 were calculated respectively, and the median was used to correct for the age and BMI of each sample. Parity, medical history, and pregnancy method of the pregnant women are discrete values. Medical history is the sum of the number of women with a history of chronic hypertension, preeclampsia, systemic lupus erythematosus, and antiphospholipid syndrome. The pregnancy method is 1 if it is assisted reproductive technology and 0 if it is natural pregnancy.

[0096] 4.2 Combined cfDNA features: Based on the trend of change in the first-diagnosis early-onset preeclampsia group compared with the control group, the permutations and combinations of all cfDNA features were traversed using a multiplication and division pattern, and the number of features was referenced from formula (3).

[0097] Number of candidate cfDNA feature combinations

[0098] N and M represent the number of candidate cfDNA features that are upregulated and downregulated in the disease group compared to the healthy control group, respectively.

[0099] 4.3 Combining cfDNA features and maternal clinical information, the samples from dataset 1 in Table 1 (model construction) were used as the training set, and the other three datasets were used as internal validation sets. Machine learning methods were used to construct the model. Finally, based on the AUC (Areaunder curve) results of the four datasets calculated using the Receiver Operating Characteristic Curve (ROC), the optimal feature combination and model were selected. In the feature combination, the cfDNA features were input into the prediction model in the form of a product. The optimal features were: maternal BMI, age, parity, past medical history, pregnancy method, and ACCGGC*KRTAP5-9*TMPRSS6*EFNA2*MOV10L1*FOSL2*PDE6G*GLTP. The optimal prediction model was a neural network (parameters: activation = 'tanh', alpha = 0.05, 'hidden_layer_sizes': (50, 100, 50), 'learning_rate': 'adaptive', 'solver': 'adam'). The AUCs of the four hospitals participating in feature selection under the optimal prediction model were 0.8413, 0.9012, 0.8142, and 0.8341, respectively, improving the results of using clinical indicators alone in different datasets (e.g., Figure 3 and Figure 4 (As shown).

[0100] 5. External validation of a model for predicting early-onset preeclampsia at first diagnosis

[0101] Using two sets of data from the validation set in Table 1 (from two different hospitals), the maternal BMI, age, parity, medical history, pregnancy method, and the product of ACCGGC*KRTAP5-9*TMPRSS6*EFNA2*MOV10L1*FOSL2*PDE6G*GLTP obtained as described in 3 and 4 were used as input values ​​(input matrix) for model validation of the optimal neural network model (the optimal prediction model in 4.3). The AUC results of the two sets of data in the external validation set both reached above 0.8, specifically 0.8116 and 0.8024. Specific accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (PPV) are shown in Table 2. As shown in Table 2, the model of this invention can effectively predict preeclampsia.

[0102] Table 2 Evaluation Indicators for Predicting Early-Onset Preeclampsia at First Visit

[0103] Validation set 1 73.7 81.8 70.6 9.6 99 Validation set 2 72.2 71.4 72.3 13.5 97.7

[0104] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A method for constructing a predictive model related to preeclampsia, characterized in that, The prediction model construction method includes: a. Sample analysis steps: After comparing and deduplicating the whole genome sequencing data of the pregnant woman's cfDNA with the reference genome, several types of cfDNA feature analysis were performed on the same sample; the cfDNA features include the expression profile and expression distribution profile of the transcript promoter region and the frequency of the 6-base motif at the 5' end of the cfDNA; the pregnant woman's samples include the disease group and the healthy control group; b. cfDNA feature screening step: Based on the analysis results of step a for different samples, statistically screen the promoter regions of transcripts with significantly different coverage, significantly different distribution, and significantly different frequencies of cfDNA 5' end 6-base sequence between the disease group and the healthy control group to obtain several candidate cfDNA features. c. Predictive Model Construction Steps: A predictive model is constructed using machine learning algorithms based on the pregnant woman's clinical indicators and the candidate cfDNA features, yielding the optimal feature combination and the corresponding predictive model. The clinical indicators include: pregnant woman's age, BMI, parity, past medical history, and pregnancy mode. The candidate cfDNA features include: expression profiles of the KRTAP5-9 transcript promoter region, TMPRSS6 transcript promoter region, EFNA2 transcript promoter region, MOV10L1 transcript promoter region, FOSL2 transcript promoter region, PDE6G transcript promoter region, GLTP gene transcript promoter region, and cfDNA 5' terminal ACCGGC motif frequency. The feature combination includes several cfDNA features and clinical indicators.

2. An analytical system related to preeclampsia, characterized in that, The analysis system includes: Sample analysis module: Aligns the whole genome sequencing data of cfDNA from pregnant women's samples to a reference genome, analyzes the expression profile and distribution profile of transcript promoter regions, and the frequency of the 6-base motif at the 5' end of cfDNA. The pregnant women's samples include diseased groups and healthy control groups. Feature screening module: Based on the results of the sample analysis module, statistical screening is performed on promoter regions of transcripts with significantly different coverage and distribution and cfDNA 5' end 6-base motifs with significantly different frequencies between the disease group and the healthy control group, to obtain several candidate cfDNA features; Predictive Model Construction Module: Using the candidate cfDNA features obtained from the feature screening module, combined with clinical indicators of pregnant women, a predictive model is constructed through machine learning algorithms to obtain the optimal feature combination and the corresponding predictive model. The clinical indicators include: maternal age, BMI, parity, past medical history, and pregnancy mode. The candidate cfDNA features include: expression profiles of the KRTAP5-9 transcript promoter region, TMPRSS6 transcript promoter region, EFNA2 transcript promoter region, MOV10L1 transcript promoter region, FOSL2 transcript promoter region, PDE6G transcript promoter region, GLTP gene transcript promoter region, and cfDNA 5' end ACCGGC motif frequency. This feature combination includes several cfDNA features and clinical indicators.

3. The analysis system as described in claim 2, characterized in that, It also includes a calculation module, which processes the data of the combination of features of the pregnant woman to be tested into the prediction model to obtain a judgment result that the pregnant woman to be tested will have preeclampsia.

4. The application of a combination of features associated with preeclampsia in the preparation of a detection product associated with preeclampsia, said combination of features including several cfDNA features and clinical indicators; The aforementioned cfDNA features include: Expression profiles of the KRTAP5-9 transcript promoter region, TMPRSS6 transcript promoter region, EFNA2 transcript promoter region, MOV10L1 transcript promoter region, FOSL2 transcript promoter region, PDE6G transcript promoter region, GLTP gene transcript promoter region, and cfDNA 5' end ACCGGC motif frequency; The clinical indicators include: maternal age, BMI, parity, past medical history, and mode of pregnancy.

5. A detection method related to preeclampsia, characterized in that... include: Obtain data on the combination of features of the pregnant woman to be tested; The obtained feature combination data is input into the prediction model for processing to obtain the detection results; The feature combination and the prediction model are obtained by the prediction model construction method as described in claim 1.

6. A detection system associated with preeclampsia, characterized in that... include: A device for acquiring data on a combination of features associated with preeclampsia in a pregnant woman under investigation; A device for predictive modeling of data on feature combinations associated with preeclampsia; Device for outputting detection results; The feature combination and the prediction model are obtained by the prediction model construction method as described in claim 1.

7. A detection product related to preeclampsia, characterized in that... include: Memory, used to store programs; A processor for implementing the method as described in claim 5 by executing a program stored in the memory.

8. A computer-readable storage medium, characterized in that, The medium stores a program that can be executed by a processor to implement the method as described in claim 5.

Images