A blood pressure data prediction method and system based on an FCDN neural network and cfRNA

Abstract

The application discloses a blood pressure data prediction method and system based on an FCDN neural network and cfRNA, cfRNA sequencing is performed on a blood sample of an early pregnancy object, and cfRNA sequencing expression of the blood sample is extracted; data preprocessing is performed on the cfRNA sequencing expression of the blood sample, cfRNA sequencing data indexes that are different and represent an increase in a blood pressure value during pregnancy are screened, a training data set and a verification data set are simultaneously established; an FCDN neural network model is established; the FCDN neural network model is trained by using the training data set, the trained FCDN neural network model is verified by using the verification data set to obtain an optimal FCDN neural network model, and a probability result of an increase in a blood pressure value during pregnancy is obtained by using the optimal FCDN neural network model.

Description

Technical Field

[0001] This invention belongs to the field of biological sequencing data processing technology, specifically relating to a blood pressure data prediction method and system based on FCDN neural network and cfRNA. Background Technology

[0002] Blood, urine, saliva, and other bodily fluids contain a large number of cell-free DNA and RNA biomarkers, which have important applications in early cancer screening, cancer medication, efficacy evaluation, and recurrence monitoring. Among them, circulating cell-free RNA (cfRNA) has attracted the most attention, including almost all types of RNA such as microRNA (miRNA), long non-coding RNA (lncRNA), circular RNA (circRNA), transfer RNA, and messenger RNA (mRNA).

[0003] Previous studies have reported the use of cfRNA in the early diagnosis of some solid tumors, including breast cancer, colorectal cancer, follicular lymphoma, and hepatocellular carcinoma. cfRNA is less affected by DNA released from senescent cells, has high abundance in body fluids, and has important applications in disease prediction, diagnosis, and prognosis.

[0004] In medical research, clinical specimens are typically commissioned to companies with sequencing equipment for cfRNA sequencing. The delivered results usually contain variations in at least three thousand cfRNAs. Due to varying professional backgrounds, after obtaining the sequencing data, statistical software such as SPSS is typically used to screen for specific cfRNAs based on significant differences, and then experiments are conducted on that single cfRNA. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a blood pressure data prediction method and system based on FCDN neural network and cfRNA to address the shortcomings of the prior art. The method and system utilize deep neural network for data learning and fitting to achieve overall interpretation of cfRNA sequencing data, thereby solving the technical problem that PE phenomenon cannot be quantitatively predicted in early pregnancy.

[0006] The present invention adopts the following technical solution:

[0007] A blood pressure data prediction method based on FCDN neural network and cfRNA includes the following steps:

[0008] S1. Perform cfRNA sequencing on blood samples from individuals in early pregnancy and extract the cfRNA expression levels from the blood samples;

[0009] S2. Perform data preprocessing on the cfRNA sequencing expression levels of blood samples obtained in step S1, screen cfRNA sequencing data indicators that characterize elevated blood pressure during pregnancy, and establish training and validation datasets.

[0010] S3. Establish the FCDN neural network model;

[0011] S4. Use the training dataset obtained in step S2 to train the FCDN neural network model established in step S3. Use the validation dataset obtained in step S2 to validate the trained FCDN neural network model to obtain the optimal FCDN neural network model. Use the optimal FCDN neural network model to obtain the probability result of elevated blood pressure during pregnancy.

[0012] Specifically, in step S1, the expression level of cfRNA in the blood sample during sequencing is as follows:

[0013]

[0014] Where x_train[i] is the training dataset for cfRNA expression level, x_test[i] is the validation dataset for cfRNA expression level, EF[i] is the cfRNA expression level after preprocessing of actual sequencing data from the gestational hypertension sample group, RF[i] is the cfRNA expression level after preprocessing of actual sequencing data from the normal group, M is the number of samples generated for the training dataset, and Q is the number of samples generated for the validation dataset.

[0015] Specifically, in step S2, data preprocessing includes:

[0016] Term with expression level of 0 in both PE and non-PE groups was removed; terms with 100% overlap between the cfRNA distribution intervals of the PE group and the non-PE group were excluded; the mean deviation between the cfRNA distribution intervals of the PE group and the non-PE group was calculated; and sequencing data indicators of significant differences in cfRNA were screened according to the overlap ratio threshold of the distribution intervals.

[0017] Furthermore, the specific indicators for screening cfRNA significantly different sequencing data are as follows:

[0018] Remove cfRNAs with 0 expression levels in both the PE and non-PE groups from the sequencing data; remove cfRNAs with 100% overlap in sequencing data between the PE and non-PE groups; remove cfRNAs with a mean deviation of less than 10% in sequencing data between the PE and non-PE groups.

[0019] Specifically, in step S2, the training dataset and validation dataset corresponding to the training and validation cfRNA datasets are as follows:

[0020] Based on the amplitude distribution of the selected cfRNA sequencing vectors, simulated sequencing samples were generated using a Gaussian random number function to obtain training and validation datasets. The training and validation datasets respectively include cfRNA and blood pressure elevation probability contribution data; the blood pressure elevation probability contribution data is specifically as follows:

[0021]

[0022]

[0023] Where i∈[1,2,3,…N], y_train[i] is the training dataset for the probability contribution data of elevated blood pressure, y_test[i] is the validation dataset for the probability contribution data of elevated blood pressure, EF[i] is the expression level of cfRNA after preprocessing of the actual sequencing data of the sample group with elevated blood pressure during pregnancy, RF[i] is the expression level of cfRNA after preprocessing of the actual sequencing data of the normal group, and avg() is the operation to calculate the mean.

[0024] Specifically, in step S3, the FCDN neural network model includes:

[0025] Convolutional layer (Conv): Performs coarse extraction of previous features;

[0026] The fully connected layer Dense, together with the parallel branch add operation, forms a residual network structure that integrates and reorganizes the multidimensional feature maps from the previous stage.

[0027] Feature extraction module TU: composed of batch regularization operation BN;

[0028] Convolutional layer Conv: with ReLU as a non-linear activation function;

[0029] DP: This refers to the proportional discard operation and the maximum pooling operation (MP).

[0030] The convolutional layer Conv and the fully connected layer Dense are combined to form the previous feature map tensor, which is then used as the output layer to output a (1,N,M)-dimensional probability vector representing the risk of PE.

[0031] Specifically, in step S4, training the FCDN neural network model using the training dataset involves the following steps:

[0032] The loss function loss represents the deviation between the training prediction output value y_pred and the standard value y_train. The cfRNA expression dataset x_train is input into the FCDN neural network model in batches, and the corresponding predicted probability vector y_pred and the true value of the PE incidence probability y_train are used to calculate the MAE loss value. Then, the optimizer Adam is used to optimize and update the parameters of the FCDN neural network model. Through multiple loop iterations, the loss value Loss between the predicted value y_pred and the labeled value y_train is made less than T, where the threshold T = 0.02, and the training is completed.

[0033] Furthermore, the loss function loss uses the mean absolute error MAE, and the optimizer Adam uses a learning rate decay mechanism to dynamically adjust the model learning rate parameter LR, specifically:

[0034]

[0035]

[0036] where M is the total number of samples, y[i] is the probability vector of increased gestational blood pressure values in the training dataset, x[i] is the cfRNA expression vector in the training dataset, FCDN() is the prediction function of the deep neural network model, loss is the loss value, and a, b, c are the decay coefficients.

[0037] Specifically, in step S4, the verification of the trained FCDN neural network model using the verification dataset is as follows:

[0038] The verification data x_test outside the training dataset or the cfRNA sequencing sample data of other early pregnancy objects is input into the trained FCDN neural network model for prediction, and the obtained y_pred is used as the final probability vector of increased gestational blood pressure values. The average value of the probability values contributed by each cfRNA in the probability vector is calculated to obtain the final probability of increased gestational blood pressure value for an individual. The probability of increased gestational blood pressure value is compared with the preset true value y_test, and the training dataset is corrected in combination with the clinical diagnosis conclusion of PE, and the FCDN neural network model is trained again to obtain the optimal FCDN neural network model.

[0039] In a second aspect, an embodiment of the present invention provides a data regression prediction system based on a fully convolutional deep neural network, which is characterized by including:

[0040] A sequencing module that performs cfRNA sequencing on the blood samples of early pregnancy objects and extracts the cfRNA sequencing expression levels of the blood samples;

[0041] The processing module preprocesses the cfRNA sequencing expression levels of blood samples obtained from the sequencing module, screens cfRNA sequencing data indicators that characterize elevated blood pressure during pregnancy, and establishes training and validation datasets.

[0042] Network module, establishes the FCDN neural network model;

[0043] The prediction module uses the training dataset obtained from the processing module to train the FCDN neural network model established by the network module, and uses the validation dataset obtained from the processing module to validate the trained FCDN neural network model to obtain the optimal FCDN neural network model. The optimal FCDN neural network model is then used to obtain the probability result of elevated blood pressure during pregnancy.

[0044] Compared with the prior art, the present invention has at least the following beneficial effects:

[0045] This invention discloses a blood pressure data prediction method based on FCDN neural network and cfRNA. It collects peripheral blood samples from pregnant women and performs sequencing to obtain their cfRNA expression levels. This data can be used for quantitative analysis, and based on this information, the probability of elevated blood pressure in pregnant women during late pregnancy can be predicted.

[0046] Furthermore, peripheral blood samples from pregnant women are collected and sequenced to obtain their cfRNA expression levels. This information can be used for quantitative analysis, and ultimately, the probability of elevated blood pressure in late pregnancy can be predicted.

[0047] Furthermore, peripheral blood samples from pregnant women are collected and sequenced to obtain their cfRNA expression levels. This information can be used for quantitative analysis, and ultimately, the probability of elevated blood pressure in late pregnancy can be predicted.

[0048] Furthermore, based on the variation patterns of cfRNA data from actual collected samples, a dataset was established to train the neural network model to optimize its performance and improve accuracy.

[0049] Furthermore, based on deep learning technology, designing the deep neural network model into an FCDN model structure can effectively extract the variation patterns of cfRNA data and accurately predict the contribution of each cfRNA index to PE based on these patterns. This model has a simple structure, strong feature extraction capability, low computational resource consumption, and high computational efficiency, giving it significant advantages in this invention.

[0050] Furthermore, training neural network models can optimize their performance and improve accuracy.

[0051] Furthermore, the mean absolute error (MAE) is set as the loss function to measure the gap between the current model's predicted output and the target result. Setting an appropriate model learning rate can stably and effectively train the neural network, update its intrinsic parameters, and prevent overfitting and underfitting.

[0052] Furthermore, through the model training process, the optimal FCDN model was obtained for data prediction. A validation set was set primarily for evaluating the network model's performance.

[0053] It is understandable that the beneficial effects of the second aspect mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0054] In summary, this invention can accurately predict the probability of elevated blood pressure in late pregnancy using sequencing data from peripheral blood samples taken by pregnant women in the early stages (12 weeks), and has the advantages of early prediction, high prediction accuracy, and high computational efficiency.

[0055] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the process of the present invention;

[0057] Figure 2 This is a schematic diagram of the data preprocessing principle of the present invention, wherein (a) all expression levels contain 0; (b) expression levels overlap 100%; (c) expression intervals have large overlap and small mean deviation; and (d) significant difference distribution.

[0058] Figure 3 This is a model diagram of the FCDN neural network model of the present invention;

[0059] Figure 4 This is a flowchart of the FCDN neural network for predicting PE onset in this invention;

[0060] Figure 5 This is a diagram illustrating the training strategy of the FCDN neural network model of this invention.

[0061] Figure 6 This is a diagram illustrating the FCDN neural network model verification strategy of this invention.

[0062] Figure 7 This is a comparison chart of the neural network prediction results and the calculated values ​​of sequencing data in this invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0065] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0066] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" relationship.

[0067] It should be understood that although terms such as first, second, third, etc., may be used in the embodiments of the present invention to describe the preset range, these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0068] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0069] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0070] This invention provides a blood pressure data prediction method based on FCDN neural network and cfRNA. A Full Convolution DenseNet FCDN is constructed, which has excellent data fitting and multi-objective classification capabilities. Predicting elevated blood pressure during pregnancy is treated as a data regression problem. A vector mapping relationship is established between the cfRNA sequencing expression level of blood samples from early pregnancy subjects and the probability of elevated blood pressure during pregnancy. Before prediction, the initial cfRNA data needs to be preprocessed to screen out cfRNA indicators with significant differences between the PE and non-PE groups. Then, training and validation datasets are generated based on the screened cfRNA indicators to train the neural network. Through training, only cfRNA sequencing data from blood samples from early pregnancy subjects needs to be input into the network to predict the probability of elevated blood pressure during pregnancy based on its changes. Experiments have shown that the accuracy of the prediction model proposed in this invention reaches over 95%.

[0071] Please see Figure 1 This invention discloses a blood pressure data prediction method based on FCDN neural network and cfRNA. It processes data based on the cfRNA sequencing expression levels of blood samples from early pregnancy subjects, establishes a mapping relationship between cfRNA expression levels and the probability of elevated blood pressure during pregnancy, and screens cfRNA expression levels with significant differences. Then, it establishes training and validation datasets based on this. By constructing an FCDN neural network model, it fits the data based on the abundance of cfRNA sequencing expression levels, establishes a multidimensional mapping relationship between cfRNA expression levels and the probability of elevated blood pressure during pregnancy, obtains the contribution of each cfRNA to the probability of elevated blood pressure during pregnancy, and ultimately realizes a prediction mechanism for elevated blood pressure during individual pregnancy. The method includes the following steps:

[0072] S1. Perform cfRNA sequencing on blood samples from individuals in early pregnancy and extract the cfRNA expression levels from the blood samples;

[0073] Biological samples with completed RNA library construction were entrusted to a sequencing company to determine cfRNA expression levels, and cfRNA expression levels were extracted and sequenced.

[0074] This invention is based on the raw data of cfRNA sequencing expression levels from blood samples of 199 individuals in early pregnancy. The cfRNA expression level information was obtained by performing DEseq on the data.

[0075] The key cfRNA screening method and dataset establishment involved screening cfRNA indicators with significant differences in the initial sequencing data and using these changes as a basis to build a dataset for training the neural network.

[0076] Initial sequencing was performed on blood samples from pregnant women in early pregnancy, yielding expression levels of up to 7160 cfRNA sequencing indicators. To predict the risk of elevated blood pressure during pregnancy based on cfRNA sequencing expression levels, it is necessary to establish the relationship between the two and screen for cfRNA indicators that have a key impact on PE pathogenesis.

[0077] S2. Perform data preprocessing on the cfRNA sequencing expression levels of blood samples obtained in step S1, screen cfRNA sequencing data indicators that characterize elevated blood pressure during pregnancy, and establish neural network training dataset and validation dataset; S201. Perform data preprocessing on the cfRNA sequencing expression levels of blood samples obtained in step S1.

[0078] Please see Figure 2 The specific steps of the cfRNA significant difference data preprocessing method are as follows:

[0079] 1. The expression levels of both the PE and non-PE groups were 0.

[0080] 2. Exclude terms where the cfRNA distribution intervals of the PE group and the non-PE group overlap 100%;

[0081] 3. Calculate the mean deviation between the distribution intervals of cfRNA in the PE group and those in the non-PE group;

[0082] 4. Screen for significant differences in cfRNA based on the overlap ratio threshold of distribution intervals.

[0083] The initial cfRNA expression levels obtained by sequencing before screening are denoted as PE[r, p], and non-PE levels are denoted as [r, q], where p and q are the number of samples in the blood pressure deviation group and the control group, respectively, and r is the number of cfRNA indicators, here r = 7160. After screening, the number of samples in the blood pressure deviation group and the control group are denoted as EF[s, p] and RF[s, q], respectively, where s is related to the screening algorithm parameters.

[0084] Subsequently, based on the expression levels of the selected cfRNA indicators, datasets were created for training / validating the FCDN neural network model. The datasets include training datasets x_train and y_train, and validation datasets x_test and y_test.

[0085] S202. The screening principle for significant differential expression levels consists of three steps. Please refer to [link / reference]. Figure 2 ,

[0086] 1) Remove sequencing data containing cfRNA with 0 expression levels in both the PE and non-PE groups;

[0087] 2) Remove cfRNAs from the sequencing data where the sequencing volume of the PE group and the non-PE group overlaps by 100%;

[0088] 3) Remove cfRNAs from the sequencing data where the mean difference in sequencing volume between the PE group and the non-PE group is less than 10%.

[0089] Therefore, before performing neural network-based disease prediction, this invention proposes a method for preprocessing data on significant differences in cfRNA.

[0090] S203. Based on the selected cfRNA sequencing vectors, simulated sequencing samples are generated using a Gaussian random number function according to their amplitude distribution, with 8000 samples used as the training dataset and 1000 samples used as the validation dataset.

[0091] Where x_train and x_test are cfRNA expression abundance matrices, and y_train and y_test are contribution vectors of the probability of elevated blood pressure values ​​during pregnancy for each cfRNA. The training dataset consists of actual sequencing values ​​of subjects in early pregnancy and values ​​randomly generated by a Gaussian function. x_train and x_test are cfRNA expression levels, and their value range follows the actual sequencing value range. The training dataset and validation dataset are created using a Gaussian random number generator, as shown in equation (1):

[0092]

[0093] Where i∈[1,2,3,…,N], rands() is a Gaussian random function, Max() is the maximum value operation of a vector, Min() is the minimum value operation of a vector, and M and Q are the number of generated values, respectively.

[0094] Based on the actual sequencing samples EF[s, p] and RF[s, q] of subjects in early pregnancy, and combined with prior knowledge of human judgment, the contribution of different cfRNA indicators to the probability of elevated blood pressure during pregnancy is calculated (for example, the expression level of cfRNA ENSG00000000460 in those with higher blood pressure during pregnancy is much higher than that in the control group and is the highest value among all samples, so the contribution to the probability of disease can be considered to be 1, and vice versa). Therefore, the vector set of probability of elevated blood pressure during pregnancy y_train and y_test is obtained from x_train and x_test, as well as clinical diagnostic information, as shown in equations (2) and (3):

[0095]

[0096]

[0097] Where i∈[1,2,3,…N], and avg() is the operation to calculate the mean.

[0098] The final results are x_train and y_train with dimensions [M, N], and x_test and y_test with dimensions [Q, N]. These are then transformed into [1, N, M] and [1, N, Q] samples respectively. Here, N = s, M = 8000, and Q = 1000, representing 8000 1xN training samples and 1000 1xN validation samples. See Table 1 for details.

[0099] Table 1. Training and Validation Datasets for Neural Networks

[0100]

[0101] S3. Establish a Full convolution DenseNet (FCDN) deep neural network model to implement data regression;

[0102] Using cfRNA sequencing data as input and elevated blood pressure during pregnancy as output, an end-to-end mapping relationship is established through data fitting of the FCDN neural network. Based on the variation characteristics of cfRNA sequencing data in a large dataset, feature values ​​are extracted and integrated using a large number of convolution and pooling operations in deep neural networks, and finally, the predicted probability is output to achieve data regression.

[0103] Full convolutional DenseNet (FCDN) is a variant of the fully convolutional deep neural network, possessing excellent data fitting and multi-object classification capabilities, and showing strong advantages in applications such as image segmentation and data prediction. In this invention, the FCDN neural network is mainly used for data regression tasks to establish a mapping relationship between cfRNA expression levels and the probability of elevated blood pressure during pregnancy, thereby achieving risk prediction.

[0104] Please see Figure 3 The FCDN neural network structure consists of three parts:

[0105] 1) Input layer;

[0106] 2) Serial residual convolution module;

[0107] 3) Output layer.

[0108] In the FCDN neural network, Batch Normalization (BN) represents batch regularization; Convolution2D (Conv) is a two-dimensional convolution operation; Convolution Transpose (ConvT) is a two-dimensional deconvolution operation; ReLU is a non-linear activation function running in the convolutional layer; Max pooling (MP) is the maximum pooling operation; Dense is a fully connected layer; and Dropout (DP) is the proportional dropout operation for neurons.

[0109] The input data dimension of the network model is (1, N, M), which is a 1xN dimensional cfRNA expression vector of M subjects in early pregnancy. The corresponding probability vector is a (1, N, M) dimensional probability vector, which is a 1xN dimensional probability vector of elevated blood pressure during pregnancy for the M samples.

[0110] Working principle of S302 and FCDN neural network models:

[0111] The cfRNA datasets x_train / x_test are fed into the FCDN neural network model in batches from the input layer. A coarse feature extraction process is performed using a convolutional layer (Conv). Subsequently, a residual network structure is formed by a fully connected layer (Dense) and parallel branch add operations, integrating and recombining the multi-dimensional feature maps from the previous stage for subsequent precise feature extraction. The feature extraction module (TU) in the network consists of batch regularization (BN), a convolutional layer (Conv) with ReLU as a non-linear activation function, dropout (DP) for proportional neuron discarding, and max pooling (MP). Its main function is to accurately extract the expression level variation features in the cfRNA dataset, while simultaneously performing spatial downsampling. Then, two-level integration and two-dimensional inverse convolutional transform are performed on the feature maps to achieve feature restoration. Based on this, concatenation operations are used to integrate the low-dimensional features from the previous stage and the high-dimensional features from the subsequent stage, improving the fitting ability of the network model. It can both macroscopically fit the relationship between the expression levels of cfRNA and the probability of disease occurrence and extract detailed information about the changes in the data.

[0112] Finally, the convolutional layer (Conv) and the fully connected layer (Dense) are combined to form the previous-stage feature map tensor, which is then used as the output layer to output a (1,N,M)-dimensional probability vector representing the risk of PE. 1xN represents the contribution of N cfRNA indicators to the probability of PE in a single sample, and M is the number of samples.

[0113] S4. Use the training dataset and validation dataset obtained in step S2 to train and validate the deep neural network model established in step S3, evaluate its accuracy, and explain how to achieve data regression prediction in the corresponding topic.

[0114] Please refer to Figure 4 , the workflow for predicting elevated blood pressure values during pregnancy using the FCDN neural network is as follows:

[0115] 1) Read the significantly different cfRNA training dataset obtained after preprocessing and screening;

[0116] 2) Establish an FCDN neural network model and use the dataset to train the network model;

[0117] 3) After multiple rounds of repeated training, make the loss value Loss of the predicted value y_pred and the labeled value y_train less than T (threshold T = 0.02) to complete the training;

[0118] 4) Use the trained FCDN neural network model to predict the sample data of the validation dataset x_test (non-overlapping with the training dataset data x_train) to obtain the final result of the probability of elevated blood pressure values during pregnancy. This result can be compared with the manual judgment conclusion in the later stage for accuracy evaluation.

[0119] If the prediction result is consistent with the clinical diagnosis result, it can be considered a credible conclusion. If there are discrepancies, the training dataset needs to be adjusted, and the FCDN neural network needs to be retrained to update its parameters.

[0120] S401. Training of the FCDN neural network model;

[0121] To achieve efficient and accurate prediction of the risk of elevated blood pressure values during pregnancy, it is necessary to train the created initial FCDN neural network model to obtain the optimal model parameters. In the training process, the loss function loss represents the deviation between the training prediction output value y_pred and the standard value y_train. The mean absolute error (MAE) is used as the loss function, as shown in Equation 4:

[0122]

[0123] where M is the total number of samples, y[i] is the vector of the probability of elevated blood pressure values in the training dataset, x[i] is the vector of cfRNA expression levels in the training dataset, and FCDN() is the prediction function of the deep neural network model. The model optimizer in the training process uses the adaptive motion estimation algorithm (Adam optimization algorithm), as Figure 5 shown.

[0124] During training, the cfRNA expression dataset x_train is input into the FCDN neural network model in batches (Batch), where Batch = 64. The corresponding predicted probability vector y_pred and the true value of the PE incidence probability y_train are obtained to calculate the MAE loss value. Then, the optimizer Adam is used to optimize and update the parameters of the FCDN neural network model. Through multiple loop iterations, the training is finally terminated when loss < T (threshold T = 0.02).

[0125] During the training process, a learning rate decay mechanism is adopted to dynamically adjust the learning parameters, as shown in Equation 5:

[0126]

[0127] where LR is the model learning rate parameter, and loss is the MAE loss value. Here, the decay coefficients a = 1, b = 0, and c = 0.5.

[0128] It can be seen that as the Loss value continuously decreases, the learning rate also continuously decreases. This strategy has the function of improving the training stability of the model and reducing the oscillation amplitude of Loss in the later stage of training.

[0129] S402. Verification strategy of the FCDN neural network model;

[0130] To verify the effectiveness of the model, a verification strategy for the FCDN neural network model is also proposed, as Figure 6 shown.

[0131] The validation data x_test outside the training dataset or the cfRNA sequencing sample data of other early pregnancy subjects are input into the trained FCDN neural network model for prediction. Again, y_pred is the final probability vector of the increased gestational blood pressure value. By averaging the probability values contributed by each cfRNA in the probability vector, the final probability of the increased gestational blood pressure value of an individual can be obtained. Comparing this probability with the preset true value y_test can evaluate the prediction error of the network model. In addition, in order to apply the invention in practice, in the later stage, the dataset will be corrected in combination with the clinical diagnosis conclusion of PE, and the FCDN neural network model will be retrained to obtain the FCDN neural network model with the optimal accuracy.

[0132] S403. Using the optimal FCDN neural network model obtained after completing the training process of the neural network in step S402, directly predict the probability of increased gestational blood pressure value based on the cfRNA expression of individual samples.

[0133] In another embodiment of the present invention, a data regression prediction system based on a fully convolutional deep neural network is provided. This system can be used to implement the above-mentioned blood pressure data prediction method based on FCDN neural network and cfRNA. Specifically, the data regression prediction system based on a fully convolutional deep neural network includes a sequencing module, a processing module, a network module, and a prediction module.

[0134] The sequencing module performs cfRNA sequencing on blood samples from individuals in early pregnancy and extracts the cfRNA expression levels from the blood samples.

[0135] The processing module preprocesses the cfRNA sequencing expression levels of blood samples obtained from the sequencing module, screens cfRNA sequencing data indicators that show significant differences in characterizing elevated blood pressure during pregnancy, and simultaneously establishes training and validation datasets.

[0136] Network module, establishes the FCDN neural network model;

[0137] The prediction module uses the training dataset obtained from the processing module to train the FCDN neural network model established by the network module, and uses the validation dataset obtained from the processing module to validate the trained FCDN neural network model to obtain the optimal FCDN neural network model. The optimal FCDN neural network model is then used to obtain the probability result of elevated blood pressure during pregnancy.

[0138] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0139] Please see Figure 7 The experimental data in the figure show a high degree of overlap between the neural network prediction results and the sequencing value calculation results. Calculations show that the accuracy of predicting elevated blood pressure using neural networks can reach 95%. Figure 7 The data shows that the PE prediction results obtained by the FCDN deep neural network are highly consistent with the results calculated from the actual data, with only a small error in some samples. By calculating the average of all results, the final result error of this method is only 5%.

[0140] In summary, this invention provides a blood pressure data prediction method and system based on FCDN neural network and cfRNA, which can accurately predict the probability of elevated blood pressure in late pregnancy using sequencing data from peripheral blood samples taken by pregnant women in the early stages (12 weeks). This method has advantages such as early prediction period, high prediction accuracy, and high computational efficiency.

[0141] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0143] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0144] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0145] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for predicting blood pressure data based on FCDN neural network and cfRNA, characterized in that, Includes the following steps: S1. Perform cfRNA sequencing on blood samples and extract the cfRNA expression levels from the blood samples; S2. Perform data preprocessing on the cfRNA sequencing expression levels of blood samples obtained in step S1, screen for cfRNA sequencing data indicators that characterize elevated blood pressure values, and establish training and validation datasets. The data preprocessing specifically includes: Terms with zero expression levels in both the PE and non-PE groups were removed; terms with 100% overlap between the cfRNA distribution intervals of the PE and non-PE groups were excluded; the mean deviation between the cfRNA distribution intervals of the PE and non-PE groups was calculated; and sequencing data indicators showing significant differences in cfRNA expression were screened based on the overlap ratio threshold. Remove cfRNAs with 0 expression levels in both the PE and non-PE groups from the sequencing data; remove cfRNAs with 100% overlap in sequencing data between the PE and non-PE groups; remove cfRNAs with a mean deviation of less than 10% in sequencing data between the PE and non-PE groups. S3. Establish the FCDN neural network model; S4. Use the training dataset obtained in step S2 to train the FCDN neural network model established in step S3. Use the validation dataset obtained in step S2 to validate the trained FCDN neural network model. Combine the clinical PE diagnosis conclusion to correct the training dataset and train again to obtain the optimal FCDN neural network model. Use the optimal FCDN neural network model to obtain the probability result of elevated blood pressure value.

2. The blood pressure data prediction method based on FCDN neural network and cfRNA according to claim 1, characterized in that, In step S1, the specific expression levels of cfRNA in the blood sample during sequencing are as follows: in, The training dataset is for cfRNA expression levels. This is a dataset for validating cfRNA expression levels. The expression level of cfRNA after preprocessing of actual sequencing data from the elevated blood pressure sample group. This represents the cfRNA expression level after preprocessing of actual sequencing data from the normal group. The number of samples generated for the training dataset. To verify the number of samples generated in the dataset.

3. The blood pressure data prediction method based on FCDN neural network and cfRNA according to claim 1, characterized in that, In step S2, the training dataset and validation dataset corresponding to the training and validation cfRNA datasets are specifically as follows: Based on the amplitude distribution of the selected cfRNA sequencing vectors, simulated sequencing samples were generated using a Gaussian random number function to obtain training and validation datasets. The training and validation datasets respectively include cfRNA and blood pressure elevation probability contribution data; the blood pressure elevation probability contribution data is specifically as follows: in, , To contribute data to the training dataset for the probability of elevated blood pressure, This dataset serves as a validation source for the probability of elevated blood pressure. The expression level of cfRNA after preprocessing of actual sequencing data from the sample group with elevated blood pressure during pregnancy. This represents the cfRNA expression level after preprocessing of actual sequencing data from the normal group. To calculate the mean.

4. The blood pressure data prediction method based on FCDN neural network and cfRNA according to claim 1, characterized in that, In step S3, the FCDN neural network model includes: Convolutional layer (Conv): Performs coarse extraction of previous features; The fully connected layer Dense, together with the parallel branch add operation, forms a residual network structure that integrates and reorganizes the multidimensional feature maps from the previous stage. Feature extraction module TU: composed of batch regularization operation BN; Convolutional layer Conv: with ReLU as a non-linear activation function; DP: This refers to the proportional discard operation and the maximum pooling operation (MP). The convolutional layer Conv and the fully connected layer Dense are combined to form the previous feature map tensor, which is then used as the output layer to output a (1,N,M)-dimensional probability vector representing the risk of PE.

5. The blood pressure data prediction method based on FCDN neural network and cfRNA according to claim 1, characterized in that, In step S4, training the FCDN neural network model using the training dataset specifically involves: The loss function loss represents the deviation between the training prediction output value y_pred and the standard value y_train. The cfRNA expression dataset x_train is input into the FCDN neural network model in batches, and the corresponding predicted probability vector y_pred and the true value of the PE incidence probability y_train are used to calculate the MAE loss value. Then, the optimizer Adam is used to optimize and update the parameters of the FCDN neural network model. Through multiple cyclic iterations, the loss value Loss < T of the predicted value y_pred and the labeled value y_train is achieved, where the threshold T = 0.02, and the training is completed.

6. The blood pressure data prediction method based on FCDN neural network and cfRNA according to claim 5, characterized in that, The loss function uses the mean absolute error. The optimizer Adam uses a learning rate decay mechanism to dynamically adjust the model's learning rate parameter. Specifically: Where M is the total number of samples, y[i] is the probability vector of elevated blood pressure during pregnancy in the training dataset, and x[i] is the vector of cfRNA expression levels in the training dataset. For deep neural network model prediction functions, Let be the loss value, and a, b, and c be the attenuation coefficients.

7. The blood pressure data prediction method based on FCDN neural network and cfRNA according to claim 1, characterized in that, In step S4, the specific method for validating the trained FCDN neural network model using the validation dataset is as follows: The validation data x_test outside the training dataset or the cfRNA sequencing sample data of other early pregnancy subjects are input into the trained FCDN neural network model for prediction, and the obtained y_pred is used as the final probability vector of the increased gestational blood pressure value. The average value of the probability values contributed by each cfRNA in the probability vector is calculated to obtain the final probability of the increased gestational blood pressure value of the individual. The probability of the increased gestational blood pressure value is compared with the preset true value y_test, and the training dataset is corrected in combination with the clinical diagnosis conclusion of PE, and the FCDN neural network model is trained again to obtain the optimal FCDN neural network model.

8. A data regression prediction system based on a fully convolutional deep neural network, characterized in that, It includes: A sequencing module that performs cfRNA sequencing on the blood sample and extracts the cfRNA sequencing expression level of the blood sample. A processing module that preprocesses the cfRNA sequencing expression level of the blood sample obtained by the sequencing module, screens the cfRNA sequencing data indicators that show differences in the increased blood pressure value, and simultaneously establishes a training dataset and a validation dataset. The specific data preprocessing is as follows: Eliminate the items with 0 expression levels in both the PE and non-PE groups; exclude the items with 100% overlapping distribution intervals between the PE group and the non-PE group of cfRNA; calculate the mean deviation between the cfRNA distribution intervals of the PE group and the non-PE group; screen the cfRNA significant difference sequencing data indicators according to the distribution interval overlap ratio threshold, specifically: Eliminate the cfRNA sequencing amounts with 0 expression levels in both the PE and non-PE groups in the sequencing data; eliminate the cfRNA with 100% interval overlap of the sequencing amounts between the PE and non-PE groups in the sequencing data; eliminate the cfRNA with a mean deviation of less than 10% between the PE and non-PE groups in the sequencing data. A network module that establishes an FCDN neural network model. A prediction module that uses the training dataset obtained by the processing module to train the FCDN neural network model established by the network module, uses the validation dataset obtained by the processing module to validate the trained FCDN neural network model, corrects the training dataset in combination with the clinical PE diagnosis conclusion and trains again to obtain the optimal FCDN neural network model, and uses the optimal FCDN neural network model to obtain the probability result of the increased blood pressure value.

Images