Preeclampsia noninvasive screening method based on deep sequencing 8bp oligonucleotide double-fragment characteristics

Through high-throughput sequencing and machine learning technology, a model was constructed using 8bp oligonucleotide double-fragment features, which solved the problem of non-invasive screening for preeclampsia, achieved efficient and accurate diagnosis, simplified feature requirements, and is suitable for early intervention and treatment.

CN120727103AActive Publication Date: 2025-09-30INNER MONGOLIA UNIVERSITY

Patent Information

Application Number
CN202510922605.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-07-04
Publication Date
2025-09-30
Estimated Expiration
2045-07-04

AI Technical Summary

Technical Problem

Existing technologies make it difficult to achieve efficient, non-invasive and specific preeclampsia screening, and traditional methods have poor predictive effects or invasive risks.

Method used

High-throughput sequencing technology is used to obtain free DNA from the peripheral blood of pregnant women. Combined with bioinformatics analysis and machine learning technology, a model is constructed using 8bp oligonucleotide double-fragment features for diagnosis, including feature extraction, screening and model construction, and predictions are made using models such as support vector machines, K-nearest neighbors, and extreme gradient boosting trees.

Benefits of technology

It achieves efficient and accurate diagnosis of preeclampsia with an AUC value exceeding 0.95, simplifies feature complexity, requires only peripheral blood collection and no invasive operation, is highly safe, and is suitable for early intervention and treatment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120727103A_ABST
    Figure CN120727103A_ABST
Patent Text Reader

Abstract

The invention relates to the field of noninvasive prenatal diagnosis, and particularly discloses a preeclampsia noninvasive screening method based on deep sequencing 8bp oligonucleotide double-fragment characteristics, which comprises the following steps: collecting preeclampsia and healthy pregnant woman peripheral blood samples, and extracting free DNA for high-throughput sequencing; the method comprises the following steps: extracting core 8-mer sequences' GTGCGCCC 'and' GATGGGGT 'in a long fragment of 150-200bp through bioinformatics analysis; an integrated support vector machine, K-nearest neighbor, extreme gradient lifting, a random forest and a multi-layer perceptron are combined with a logistic regression element classifier to construct a stacking model, the frequency of a core sequence is normalized, machine learning analysis is carried out, and the preeclampsia risk is predicted. According to the invention, two 8bp oligonucleotide characteristic fragments are specifically screened, and a deep learning architecture of multi-model fusion is combined, so that the limitations of low specificity and invasive detection of a traditional screening method are effectively broken through.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of non-invasive prenatal diagnosis, and specifically to a non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features. Background Art

[0002] Preeclampsia (PE) is a complex pregnancy disorder that typically develops after 20 weeks of gestation and presents with hypertension and at least one complication (such as proteinuria, maternal organ dysfunction, or placental dysfunction). It affects approximately 5% to 9% of pregnant women worldwide and results in 10% to 15% of maternal deaths. The etiology of PE is complex, with the placenta playing a key role in its pathogenesis. During early pregnancy (10-12 weeks), gene expression changes are significant in the placenta of patients with PE. After delivery, defective spiral artery remodeling and insufficient trophoblast invasion lead to hypoperfusion of the placenta, which in turn causes placental hypoxia and ischemia. Although chorionic villus sampling (CVS) can be used for early placental phenotyping, its invasive nature and risk of miscarriage limit its use in screening. With delayed marriage and childbearing and changes in modern lifestyles, the incidence of PE has increased annually, becoming a significant complication threatening the health of both mothers and fetuses. PE causes over 70,000 maternal deaths and 500,000 fetal deaths annually. Therefore, early screening and prevention of PE are crucial.

[0003] Currently, the diagnosis of PE relies primarily on clinical indicators such as urine protein and blood pressure, or through invasive methods such as placental single-cell sequencing and amniocentesis. However, these methods have limitations: urine protein is poorly predictive, and invasive methods carry the risk of miscarriage. In recent years, non-invasive prenatal testing based on cfDNA has gradually become a research hotspot. Existing studies have analyzed PE using cfDNA, focusing on methylation, fragment information, or multi-omics combined analysis, but there is a lack of efficient and specific non-invasive markers. Therefore, the development of an efficient, non-invasive method for PE diagnosis is of great clinical significance. Summary of the Invention

[0004] In response to the above-mentioned problems of the prior art, the present invention provides a non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment characteristics. The free DNA in the peripheral blood of pregnant women, namely cfDNA sequence data, is obtained through high-throughput sequencing technology, and combined with bioinformatics analysis and machine learning technology to achieve efficient and accurate PE diagnosis.

[0005] To achieve the above objectives, the present invention proposes a non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features, comprising:

[0006] S1. Sample collection, extraction, and sequencing: Peripheral blood samples were collected from pregnant women with preeclampsia (PE) and healthy pregnant women, cfDNA was extracted, and high-throughput whole-genome sequencing was performed at a sequencing depth of 20-30X to obtain raw sequencing data;

[0007] S2. Align and sort the data: Align the original sequencing reads with the human reference genome hg38, generate an alignment result file, and sort and remove duplicates to obtain tens of billions of cfDNA fragments;

[0008] S3. Dataset division: The preprocessed data is divided into a training set and a test set in a ratio of 7:3. The training set is used for model training and parameter tuning, and the test set is used for independent verification and performance evaluation of the final trained model.

[0009] S4. Extract features from sequencing data: extract the frequencies of kmer features of 150-200 bp fragments in the training set and the test set respectively, and perform normalization on them respectively;

[0010] S5. Feature screening: LASSO regression was used for preliminary feature screening in the training set, and TURF plus single-factor incremental feature screening method was used to screen out two key 8-mer sequences;

[0011] S6. Analyze regulatory function and detect cfDNA frequency changes: Analyze the conservation and transcriptional regulatory functions of the core 8-mer sequences GTGCGCCC and GATGGGGT to obtain the specificity of abnormal frequency changes of the core 8-mer sequences in peripheral blood cfDNA, which can be used to reflect the placental dysfunction status of preeclampsia;

[0012] S7. Build models and adjust parameters: Build five basic models and use the values ​​of the training set to adjust the parameters of the five basic models, and build a stacked model based on the optimal parameters;

[0013] S8, model validation and performance evaluation;

[0014] S9. Calculate and output the predicted probability of pre-eclampsia.

[0015] Preferably, in S4, the specific steps of extracting features from sequencing data are:

[0016] S41, extracting fragments with a sequence length of 150-200bp from tens of billions of sequences;

[0017] S42, extract the frequencies corresponding to the kmer features from the training set and the test set respectively, where k is 8, and obtain 65536 subsequences;

[0018] S43. Normalize the frequencies of the subsequences in the training set and the test set respectively.

[0019] Preferably, in S5, the specific steps of screening two key 8-mer sequences are:

[0020] S51. Perform preliminary feature screening on the 8-mer subsequences in training using LASSO, and obtain 80 features;

[0021] S52. TURF combined with single-factor incremental feature screening method was used to screen out the 80 features of 8-mer, and two core 8-mer sequences 'GTGCGCCC' and 'GATGGGGT' were screened out.

[0022] Preferably, in S6, the steps of analyzing the sequence conservation and transcriptional regulatory function are:

[0023] S61. Sequence evolutionary conservation: A position-specific base frequency matrix of the core 8-mer sequences GTGCGCCC and GATGGGGT was constructed and compared with the genomes of multiple vertebrate species. The maximum expectation algorithm was used to calculate the evolutionary conservation score of each base position.

[0024] S62, transcription factor binding function: The base frequency matrix generated by S61 was compared with the public database of vertebrate transcription factor binding sites for homology, and the motif matching algorithm was used to screen the high-ranking results.

[0025] Preferably, in S7, the specific process of building the model and adjusting the parameters is:

[0026] S71, label the pre-eclampsia patients as 1 and the healthy pregnancy population as 0;

[0027] S72. Extract the normalized values ​​of the frequencies of the two 8-mer sequences 'GTGCGCCC' and 'GATGGGGT' in the 150-200 bp long fragments of the training set and the test set respectively;

[0028] S73. Based on the core features 'GTGCGCCC' and 'GATGGGGT' of the frequency-normalized values ​​of the screened 8-mer sequences, and on data from preeclampsia samples and healthy pregnant populations, we constructed support vector machine (SVM), K-nearest neighbor (KNN), extreme gradient boosting tree (XGBoost), random forest (RF), and multilayer perceptron (MLP) basic models.

[0029] S74. Perform hyperparameter tuning on the above basic models respectively, and select the optimal parameters to obtain 5 optimal basic models;

[0030] S75. The five optimal basic models obtained in S74 are used as the basic classifier. Among them, the multi-layer perceptron MLP is built based on Keras and contains three hidden layers with ReLU activation with L2 regularization and Dropout, an output layer that uses the Sigmoid function to process the binary classification task, and a classifier that is custom encapsulated and compatible with the scikit-learn interface. The logistic regression with default parameters is used as the meta-classifier to build a stacked model to obtain the trained prediction model.

[0031] Preferably, in S8, the specific steps of model verification and performance evaluation are: using a strictly isolated test set to evaluate the generalization performance and diagnostic ability of the trained prediction model obtained in step S75, wherein the indicators of generalization performance and diagnostic ability include calculating accuracy, sensitivity, specificity, and AUC.

[0032] Preferably, in S9, the specific steps of calculating and outputting the predicted probability of pre-eclampsia for the screened new pregnant woman sample are:

[0033] S91. Collect peripheral blood samples according to S1-S2, extract cfDNA and perform high-throughput sequencing at a depth of 20-30X, and align to hg38;

[0034] S92. Following steps S3-S5, extract 150-200 bp fragments, calculate the frequencies of the two core 8-mer sequences 'GTGCGCCC' and 'GATGGGGT', and perform normalization to obtain the feature vector of the sample;

[0035] S93, inputting the feature vector obtained in S92 into the prediction model trained in S75;

[0036] S94. Call the predict_proba method of the model to obtain the predicted probability P that the sample belongs to the pre-eclampsia category label 1;

[0037] S95. Output the predicted probability P as the predicted value of the risk of pre-eclampsia for the pregnant woman.

[0038] Therefore, the present invention proposes a non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features, which has the following beneficial effects:

[0039] (1) The present invention only requires two key sequences to achieve efficient differentiation of preeclampsia, significantly simplifying the feature complexity, and can diagnose preeclampsia through non-invasive detection at an early stage, which is conducive to early intervention and treatment.

[0040] (2) The machine learning model is used for diagnosis. The AUC values ​​of both the training set and the test set are over 0.95, which is highly accurate. It only requires the collection of peripheral blood from pregnant women, without the need for invasive procedures. It is highly safe, easy to apply clinically, and has greater practicality.

[0041] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of a non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features of the present invention;

[0043] Figure 2 This is an evolutionary conserved motif map based on the characteristics of deep sequencing 8bp oligonucleotide double fragments of the present invention;

[0044] Figure 3 The present invention is a transcription factor FOXO1 enriched by sequence based on the double fragment characteristics of 8bp oligonucleotides through deep sequencing;

[0045] Figure 4 The present invention is a transcription factor EGR1 based on sequence enrichment of 8bp oligonucleotide double fragment characteristics;

[0046] Figure 5 The present invention is a transcription factor KLF4 based on sequence enrichment of 8bp oligonucleotide double fragment characteristics;

[0047] Figure 6 It is a transcription factor GCMI of the present invention based on sequence enrichment of 8bp oligonucleotide double fragment characteristics;

[0048] Figure 7 The ROC curves of the five basic models of the non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features in the training set and test set of the present invention are as follows;

[0049] Figure 8 This is the ROC curve of the stacking model of the non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features in the training set and test set of the present invention;

[0050] Figure 9 This is a flowchart illustrating an implementation example of a non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features of the present invention;

[0051] Figure 10 This is a graph showing the independent validation results of real clinical samples for a non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features of the present invention. DETAILED DESCRIPTION

[0052] To make the technical solutions, advantages, and purposes of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below. The described embodiments are part of the embodiments of the present invention, not all of them. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0054] like Figures 1-8 As shown, a non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features provided by the present invention specifically comprises the following steps:

[0055] S1. Sample collection, extraction, and sequencing: Peripheral blood samples were collected from pregnant women with preeclampsia (PE) and healthy pregnant women, cfDNA was extracted, and high-throughput whole-genome sequencing was performed at a sequencing depth of 20-30X to obtain raw sequencing data;

[0056] S2. Align and sort the data: Align the original sequencing reads with the human reference genome hg38, generate an alignment result file, and sort and remove duplicates to obtain tens of billions of cfDNA fragments;

[0057] S3. Dataset division: The preprocessed data is divided into a training set and a test set in a ratio of 7:3. The training set is used for model training and parameter tuning, and the test set is used for independent verification and performance evaluation of the final trained model.

[0058] S4. Extracting features from sequencing data: Extract the frequencies of kmer features of 150-200 bp fragments in the training set and the test set, and perform normalization on each. The specific steps for extracting features from sequencing data are as follows:

[0059] S41, extracting fragments with a sequence length of 150-200bp from tens of billions of sequences;

[0060] S42, extract the frequencies corresponding to the kmer features from the training set and the test set respectively, where k is 8, and obtain 65536 subsequences;

[0061] S43. Normalize the frequencies of the subsequences in the training set and the test set respectively.

[0062] S5. Feature screening: LASSO regression was used to perform preliminary feature screening in the training set, and TURF plus single-factor incremental feature screening method was used to screen out two key 8-mer sequences. The specific steps for screening the two key 8-mer sequences were as follows:

[0063] S51. Perform preliminary feature screening on the 8-mer subsequences in training using LASSO, and obtain 80 features;

[0064] S52. TURF combined with single-factor incremental feature screening method was used to screen out the 80 features of 8-mer, and two core 8-mer sequences 'GTGCGCCC' and 'GATGGGGT' were screened out.

[0065] S6. Analyze regulatory function and detect cfDNA frequency changes: Perform conservation and transcriptional regulatory function analysis on the core 8-mer sequences GTGCGCCC and GATGGGGT to obtain the specificity of abnormal frequency changes of the core 8-mer sequences in peripheral blood cfDNA, which can be used to reflect the placental dysfunction status of preeclampsia. The steps for sequence conservation and transcriptional regulatory function analysis are as follows:

[0066] S61. Sequence evolutionary conservation: construct a position-specific base frequency matrix of the core 8-mer sequences GTGCGCCC and GATGGGGT, align them with the genomes of multiple species of vertebrates, and use the maximum expectation algorithm to calculate the evolutionary conservation score of each base position; Figure 2 As shown, the results showed strong conservation of thymine (T) and guanine (G) bases at positions 2, 3, and 5, suggesting that they have key functions in mammalian placental development;

[0067] S62, transcription factor binding function: The base frequency matrix generated by S61 was compared with the public database of vertebrate transcription factor binding sites for homology, and the motif matching algorithm was used to screen the high-ranking results.

[0068] like Figure 3-Figure 6 As shown, the results showed that the sequences were significantly enriched in the transcription factor binding sites that regulate placental angiogenesis and trophoblast differentiation, including but not limited to: FOXO1 (regulates placental cell apoptosis and oxidative stress response)

[0069] Among them, transcription factor dysfunction is directly related to the placental hypoxia and ischemia pathological characteristics of preeclampsia.

[0070] S7. Build models and adjust parameters: Build five basic models and use the values ​​of the training set to adjust the parameters of the five basic models, and build a stacked model based on the optimal parameters;

[0071] The specific process of building a model and adjusting parameters is as follows:

[0072] S71, label the pre-eclampsia patients as 1 and the healthy pregnancy population as 0;

[0073] S72. Extract the normalized values ​​of the frequencies of the two 8-mer sequences 'GTGCGCCC' and 'GATGGGGT' in the 150-200 bp long fragments of the training set and the test set respectively;

[0074] S73. Based on the core features 'GTGCGCCC' and 'GATGGGGT' of the frequency-normalized values ​​of the screened 8-mer sequences, and on data from preeclampsia samples and healthy pregnant populations, we constructed support vector machine (SVM), K-nearest neighbor (KNN), extreme gradient boosting tree (XGBoost), random forest (RF), and multilayer perceptron (MLP) basic models.

[0075] S74. Perform hyperparameter tuning on the above basic models respectively, and select the optimal parameters to obtain 5 optimal basic models;

[0076] S75. The five optimal basic models obtained in S74 are used as the basic classifier. Among them, the multi-layer perceptron MLP is built based on Keras and contains three hidden layers with ReLU activation with L2 regularization and Dropout, an output layer that uses the Sigmoid function to process the binary classification task, and a classifier that is custom encapsulated and compatible with the scikit-learn interface. The logistic regression with default parameters is used as the meta-classifier to build a stacked model to obtain the trained prediction model.

[0077] S8, model validation and performance evaluation;

[0078] The specific steps of model validation and performance evaluation are: using a strictly isolated test set to evaluate the generalization performance and diagnostic ability of the trained prediction model obtained in step S75, wherein the indicators of generalization performance and diagnostic ability include calculation accuracy, sensitivity, specificity, and AUC.

[0079] S9. Calculate and output the predicted probability of pre-eclampsia.

[0080] The specific steps for calculating and outputting the predicted probability of preeclampsia for the newly screened pregnant sample are:

[0081] S91. Collect peripheral blood samples according to S1-S2, extract cfDNA and perform high-throughput sequencing at a depth of 20-30X, and align to hg38;

[0082] S92. Following steps S3-S5, extract 150-200 bp fragments, calculate the frequencies of the two core 8-mer sequences 'GTGCGCCC' and 'GATGGGGT', and perform normalization to obtain the feature vector of the sample;

[0083] S93, inputting the feature vector obtained in S92 into the prediction model trained in S75;

[0084] S94. Call the predict_proba method of the model to obtain the predicted probability P that the sample belongs to the pre-eclampsia category label 1;

[0085] S95. Output the predicted probability P as the predicted value of the risk of pre-eclampsia for the pregnant woman.

[0086] In S7, the five basic models are:

[0087] K-nearest neighbor model: The model was hyperparameter tuned using GridSearchCV. 5-fold cross-validation was used to evaluate the performance of different parameter combinations on the training set. Accuracy was used as the evaluation metric. The final parameters were determined to be n_neighbors = 5 and uniform weighting, meaning that all neighbors had the same voting weight.

[0088] Random Forest Model: We optimized key hyperparameters using grid search (GridSearchCV) and 5-fold cross-validation, using accuracy as the evaluation metric. The final parameters were the number of decision trees (n_estimators = 200), maximum tree depth (max_depth = None), minimum number of samples per internal node split (min_samples_split = 2), and minimum number of samples per leaf node (min_samples_leaf = 2).

[0089] Model Support Vector Machine: GridSearchCV was used to tune its hyperparameters. The model used 5-fold cross-validation, with accuracy as the scoring criterion. The parameters were finally set to a regularization strength parameter C = 0.1 and a radial basis kernel (RBF).

[0090] Model extreme gradient boosting: Grid search (GridSearchCV) was used to tune the key hyperparameters. Through 5-fold cross validation, the parameters were finally determined to be learning rate (learning_rate=0.1), number of trees (n_estimators=500) and maximum tree depth (max_depth=6).

[0091] Model Multilayer Perceptron: The model was built using Keras and encapsulated using the KerasClassifier for use with GridSearchCV to optimize hyperparameters. The grid search used 5-fold cross-validation, with accuracy as the scoring metric. The final parameters were the number of hidden layer neurons (units = 8), dropout rate (dropout_rate = 0.3), L2 regularization strength (l2_reg = 0.01), optimizer (Adam), number of training epochs (epochs = 256), and batch size (batch_size = 64).

[0092] The model's multilayer perceptron architecture consists of three hidden layers with ReLU activation functions. Each layer applies L2 regularization and dropout to prevent overfitting. The output layer consists of a single neuron with a sigmoid activation function for binary classification. The model is optimized using binary cross entropy as the loss function.

[0093] The stacking model is constructed by integrating five base classifiers with excellent performance, including support vector machine (SVM), K-nearest neighbor (KNN), random forest (RF), extreme gradient boosting tree (XGBoost) and multi-layer perceptron (MLP);

[0094] The MLP is built on Keras and consists of three hidden layers activated by ReLU with L2 regularization and dropout. The output layer uses a sigmoid function for binary classification. This is then custom-encapsulated into a classifier compatible with the scikit-learn interface. Each base classifier is initialized based on the optimal hyperparameters obtained through previous parameter tuning, and the stacking framework uses logistic regression as the meta-learner.

[0095] The specific process of building a stacking model is as follows:

[0096] (1) Basic classifier and hyperparameter tuning

[0097] KNN: Use GridSearchCV to perform 5-fold CV on n_neighbor and weight, and finally select n_neighbors=16 and weights='uniform';

[0098] RF: Perform 5-fold CV on n_estimator, max_depth, min_samples_split, min_samples_leaf, etc., and finally select n_estimators = 4, max_depth = 5, min_samples_split = 2, min_samples_leaf = 2;

[0099] SVM: Grid search the regularization parameter C and kernel parameter γ, select C = 1 and kernel = 'rbf' after 5-fold CV;

[0100] XGBoost: Perform 5-fold CV on learning_rate, n_estimators, and max_depth, and finally determine learning_rate = 1, n_estimators = 10, and max_depth = 2;

[0101] MLP: Construct three hidden layers (ReLU + L2 regularization + Dropout) based on Keras, and use KerasClassifier with GridSearchCV. 5-fold CV is scored by accuracy. The parameter space includes units, dropout_rate, l2_reg, epochs, and batch_size. Finally, units = 16, dropout_rate = 0.1, l2_reg = 0.01, epochs = 32, and batch_size = 16 are selected.

[0102] (2) Cross-validation generates first-level features

[0103] A 5-fold cross-validation was used for the training set: 4 folds were used each time to train a basic model, and the remaining 1 fold was used to predict the probability that the sample belonged to preeclampsia. After repeating this 5 times, the predicted probabilities of all folds were collected to obtain a first-level probability feature vector corresponding one-to-one to the training samples, with a total of 5 dimensions, each dimension corresponding to a basic model.

[0104] (3) Stacking structure construction

[0105] The above-mentioned first-level probability feature matrix (number of samples × 5) and the true label are used as input, and the meta-classifier is trained using logistic regression (C = 1.0) to form the final StackingClassifier.

[0106] After training, the stacked model can perform weighted fusion of the prediction probabilities of the five basic models to output more accurate classification results.

[0107] (4) Final prediction process

[0108] For test or validation set samples: extract two 8-mer normalized frequency features; input the features into the trained StackingClassifier at once; call stacking_classifier.predict_proba(X) to directly obtain the predicted probability P of label 1 (preeclampsia); use P as the "predicted probability of preeclampsia risk" for this sample for clinical high-risk screening and early intervention.

[0109] After building the stacking model, the stacking framework uses logistic regression as a meta-learner to relearn and synthesize the predicted probabilities output by each base model. The entire model training is fitted using the training set, and the generalization performance is evaluated using the test set. The training set is used to build the model, and the test set is used to evaluate the model.

[0110] In S3, the 7:3 ratio of sample division into training set and test set is a key step to achieve model reliability and generalization ability:

[0111] 1) Training set (involved in model building): This set is used for parameter learning and hyperparameter tuning of five basic classifiers (SVM, KNN, RF, XGBoost, and MLP) and a stacked model, as well as 5-fold cross-validation. By learning the normalized frequency features of two key 8-mers ('GTGCGCCC' and 'GATGGGGT') in 150–200 bp fragments, the model acquires the mapping relationship from "feature to preeclampsia risk."

[0112] 2) Test set (completely independent of the training process): This set was used to objectively evaluate the predictive performance of the trained model on unseen data. On 570 test cases (246 cases of preeclampsia and 324 cases of normal pregnancy), the stacked model achieved excellent performance with an AUC of 0.983 (95% CI 0.902–0.983), an accuracy of 0.96, and a precision of 0.96, demonstrating that the model is not simply a "memorization" of the training data but rather possesses the ability to generalize to real clinical samples.

[0113] Therefore, sample partitioning is not only a universal best practice in the field of machine learning, but also ensures that the prediction model of the present invention is both effective and reliable in real clinical applications: the training set is responsible for "learning" the classification rules, and the test set is responsible for "testing" whether the rules can be promoted and applied.

[0114] like Figure 9 As shown, an embodiment of the present invention provides a non-invasive screening method for pre-eclampsia based on deep sequencing of 8bp oligonucleotide double fragment features, which uses a pre-eclampsia probability assessment system based on the frequencies of two 8-mer sequences in the subject to be screened and a stacking model to predict the probability of the subject to be screened having pre-eclampsia. The system is divided into seven steps:

[0115] Step 1: Obtain peripheral blood samples from pregnant women, extract cfDNA, and perform quantification, library construction, and sequencing;

[0116] Step 2: Perform pre-processing such as alignment, sorting, and deduplication of the sequencing data with the human reference genome hg38;

[0117] Step 3: Screening of sequences with a fragment length of 150-200 bp in cfDNA;

[0118] Step 4: Calculate the frequency of 'GTGCGCCC' and 'GATGGGGT' in these sequences;

[0119] Step 5: Normalize the frequency;

[0120] Step 6: Use the frequency-normalized values ​​of the two sequences as features and input them into the stacking model;

[0121] Step 7: Obtain the probability of preeclampsia in the subject to be screened through model calculation.

[0122] Among them, as shown in Table 1, the diagnostic performance of six models, including support vector machine (SVM), K-nearest neighbor (KNN), random forest (RF), extreme gradient boosting tree (XGBoost), multi-layer perceptron model, and the generated stacking model (Stacking), were verified on the training set and test set. The results showed that the area under the curve (AUC) of each model in the test set exceeded 0.95, which can accurately identify diseased patients and healthy people.

[0123] Table 1 Evaluation index table of six models in training set / test set

[0124]

[0125] like Figure 10 To verify the feasibility and practicality of the method of the present invention, we conducted independent validation based on real clinical samples, mainly including the following two data sets:

[0126] Test set (570 cases): During the model development phase, an independent test set (30% of the total data, including 246 preeclampsia samples and 324 normal pregnancy samples) was reserved. This set was not used in model training and was used to objectively evaluate the model's predictive performance in distinguishing preeclampsia from normal pregnancy. The results showed that the five basic models and the final stacked model all had good generalization capabilities. Among them, the AUC of the stacked model on the test set reached 0.983 (95% CI: 0.902–0.983) (see Figure 3 ), the accuracy rate was 0.96, the precision rate was 0.96, and other evaluation indicators also performed well (see Table 1), reflecting the significant recognition ability of preeclampsia samples.

[0127] Validation set (55 cases): We also collected 55 real-world clinical samples (28 cases of preeclampsia and 27 cases of normal pregnancy) to further evaluate the model's applicability in clinical scenarios. The results showed that the predicted probability of preeclampsia was concentrated around 0.9, while the predicted probability of preeclampsia for normal pregnancy was concentrated around 0.05, indicating that the model was able to effectively distinguish between the two groups.

[0128] In summary, through the evaluation results of two independent sample sets, it was preliminarily verified that the modeling method proposed in this project has good feasibility and application prospects based on cfDNA data.

[0129] Therefore, the present invention provides a non-invasive screening method for pre-eclampsia based on deep sequencing 8bp oligonucleotide double fragment features, which achieves high specificity and high accuracy early diagnosis by analyzing the specific 8bp oligonucleotide double fragment features in the free DNA of the peripheral blood of pregnant women. The method only needs to collect peripheral blood, does not require invasive procedures, is highly safe, and significantly improves the diagnostic accuracy through machine learning models, with the area under the test set curve exceeding 0.95. In addition, the method simplifies feature complexity and only requires two key sequences to achieve efficient differentiation, providing strong support for early intervention and treatment. At the same time, it is highly practical and has laid the foundation for the production of directional detection chips, with broad clinical application prospects.

[0130] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features, characterized in that: include: S1. Sample collection, extraction, and sequencing: Peripheral blood samples were collected from pregnant women with preeclampsia (PE) and healthy pregnant women, cfDNA was extracted, and high-throughput whole-genome sequencing was performed at a sequencing depth of 20-30X to obtain raw sequencing data; S2. Align and sort the data: Align the original sequencing reads with the human reference genome hg38, generate an alignment result file, and sort and remove duplicates to obtain tens of billions of cfDNA fragments; S3. Dataset division: The preprocessed data is divided into a training set and a test set in a ratio of 7:

3. The training set is used for model training and parameter tuning, and the test set is used for independent verification and performance evaluation of the final trained model. S4. Extract features from sequencing data: extract the frequencies of kmer features of 150-200 bp fragments in the training set and the test set respectively, and perform normalization on them respectively; S5. Feature screening: LASSO regression was used for preliminary feature screening in the training set, and TURF plus single-factor incremental feature screening method was used to screen out two key 8-mer sequences; S6. Analyze regulatory function and detect cfDNA frequency changes: Analyze the conservation and transcriptional regulatory functions of the core 8-mer sequences GTGCGCCC and GATGGGGT to obtain the specificity of abnormal frequency changes of the core 8-mer sequences in peripheral blood cfDNA, which can be used to reflect the placental dysfunction status of preeclampsia; S7. Build models and adjust parameters: Build five basic models and use the values ​​of the training set to adjust the parameters of the five basic models, and build a stacked model based on the optimal parameters; S8, model validation and performance evaluation; S9. Calculate and output the predicted probability of pre-eclampsia.

2. The non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features according to claim 1, characterized in that: In S4, the specific steps for extracting features from sequencing data are as follows: S41, extracting fragments with a sequence length of 150-200bp from tens of billions of sequences; S42, extract the frequencies corresponding to the kmer features from the training set and the test set respectively, where k is 8, and obtain 65536 subsequences; S43. Normalize the frequencies of the subsequences in the training set and the test set respectively.

3. The non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features according to claim 1, characterized in that: In S5, the specific steps for screening two key 8-mer sequences are: S51. Perform preliminary feature screening on the 8-mer subsequences in training using LASSO, and obtain 80 features; S52. TURF combined with single-factor incremental feature screening method was used to screen out the 80 features of 8-mer, and two core 8-mer sequences 'GTGCGCCC' and 'GATGGGGT' were screened out.

4. The non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features according to claim 1, characterized in that: In S6, the steps for analyzing the sequence conservation and transcriptional regulatory function are as follows: S61. Sequence evolutionary conservation: A position-specific base frequency matrix of the core 8-mer sequences GTGCGCCC and GATGGGGT was constructed and compared with the genomes of multiple vertebrate species. The maximum expectation algorithm was used to calculate the evolutionary conservation score of each base position. S62, transcription factor binding function: The base frequency matrix generated by S61 was compared with the public database of vertebrate transcription factor binding sites for homology, and the motif matching algorithm was used to screen the high-ranking results.

5. The non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features according to claim 1, characterized in that: In S7, the specific process of building the model and adjusting the parameters is as follows: S71, label the pre-eclampsia patients as 1 and the healthy pregnancy population as 0; S72. Extract the normalized values ​​of the frequencies of the two 8-mer sequences 'GTGCGCCC' and 'GATGGGGT' in the 150-200 bp long fragments of the training set and the test set respectively; S73. Based on the core features 'GTGCGCCC' and 'GATGGGGT' of the frequency-normalized values ​​of the screened 8-mer sequences, and on data from preeclampsia samples and healthy pregnant populations, we constructed support vector machine (SVM), K-nearest neighbor (KNN), extreme gradient boosting tree (XGBoost), random forest (RF), and multilayer perceptron (MLP) basic models. S74. Perform hyperparameter tuning on the above basic models respectively, and select the optimal parameters to obtain 5 optimal basic models; S75. The five optimal basic models obtained in S74 are used as the basic classifier. Among them, the multi-layer perceptron MLP is built based on Keras and contains three hidden layers with ReLU activation with L2 regularization and Dropout, an output layer that uses the Sigmoid function to process the binary classification task, and a classifier that is custom encapsulated and compatible with the scikit-learn interface. The logistic regression with default parameters is used as the meta-classifier to build a stacked model to obtain the trained prediction model.

6. The non-invasive screening method for preeclampsia based on deep sequencing of 8bp oligonucleotide double fragment features according to claim 5, characterized in that: In S8, the specific steps of model verification and performance evaluation are: using a strictly isolated test set to evaluate the generalization performance and diagnostic ability of the trained prediction model obtained in step S75, where the indicators of generalization performance and diagnostic ability include calculating accuracy, sensitivity, specificity, and AUC.

7. A non-invasive screening method for pre-eclampsia based on deep sequencing of 8bp oligonucleotide double fragment features as described in claims 1-6, characterized in that: In S9, the specific steps for calculating and outputting the predicted probability of preeclampsia for the newly screened pregnant sample are: S91. Collect peripheral blood samples according to S1-S2, extract cfDNA and perform high-throughput sequencing at a depth of 20-30X, and align to hg38; S92. Following steps S3-S5, extract 150-200 bp fragments, calculate the frequencies of the two core 8-mer sequences 'GTGCGCCC' and 'GATGGGGT', and perform normalization to obtain the feature vector of the sample; S93, inputting the feature vector obtained in S92 into the prediction model trained in S75; S94. Call the predict_proba method of the model to obtain the predicted probability P that the sample belongs to the pre-eclampsia category label 1; S95. Output the predicted probability P as the predicted value of the risk of pre-eclampsia for the pregnant woman.

Citation Information

Patent Citations

  • Methods and systems for differentiating somatic genomic sequences from germline genomic sequences

    CN115698323A

  • Fetal chromosomal abnormality detection method and system

    US20240013859A1

Cited By

  • Preeclampsia noninvasive screening method based on deep sequencing of 4-mer terminal motif spectrum characteristics

    CN121380322A

  • Probiotic tolerance characteristic prediction method and device, electronic equipment and storage medium

    CN122117042A