Diabetes prediction method, electronic equipment and storage medium

Abstract

The invention provides a diabetes prediction method, electronic equipment and a storage medium. The method comprises the following steps: acquiring a characteristic parameter set of biomolecules in multi-omics data of an organism; selecting target characteristic parameters related to diabetes from the characteristic parameter set, wherein the target characteristic parameters comprise a first nucleic acid characteristic parameter, a second nucleic acid characteristic parameter, a lipid characteristic parameter and a microorganism characteristic parameter; inputting the first nucleic acid characteristic parameter, the lipid characteristic parameter and the microorganism characteristic parameter into a neural network model to obtain a first characteristic vector, a second characteristic vector and a third characteristic vector; inputting the second nucleic acid characteristic parameter into a convolutional layer to obtain a fourth characteristic vector; generating a target vector based on the first feature vector, the second feature vector, the third feature vector and the fourth feature vector; and calling the classification model to predict the target vector to obtain a probability value of the organism suffering from diabetes. By using the method, the prediction accuracy of diabetes mellitus can be improved.

Description

Technical Field

[0001] The present application relates to the biomedical field, and in particular to a diabetes prediction method, electronic device and storage medium. Background Art

[0002] In the biomedical field, the prediction of specific diseases can help effectively manage the patient's body. For example, the occurrence of diabetes is affected by a combination of complex factors within the biological system. Therefore, if the probability of an organism suffering from diabetes is predicted based on a single omics data, such as proteomic data, the prediction results may be inaccurate due to the inability to fully consider the complexity of the biological system. In addition, single omics data are often easily affected by factors such as sample contamination or noise generated during the analysis process, which further reduces the accuracy of the prediction results.

[0003] In addition, due to the complexity of biological systems, the models trained with single omics data may have adaptability issues, making it difficult to accurately predict diabetes. Summary of the invention

[0004] In view of the above, it is necessary to provide a diabetes prediction method, electronic device and storage medium that can solve the technical problem of poor prediction accuracy of diabetes.

[0005] On the one hand, the present application provides a diabetes prediction method, the method comprising: obtaining multi-omics data of an organism, the multi-omics data comprising a characteristic parameter set of biological molecules, selecting a target characteristic parameter related to diabetes from the characteristic parameter set, wherein the target characteristic parameter comprises a first nucleic acid characteristic parameter, a second nucleic acid characteristic parameter, a lipid characteristic parameter and a microbial characteristic parameter, calling a trained neural network model to predict the target characteristic parameter, and obtaining a probability value of the organism suffering from diabetes, wherein the neural network model comprises a fully connected network model, a convolutional layer and a classification model, comprising: inputting the first nucleic acid characteristic parameter, the lipid characteristic parameter and the microbial characteristic parameter into the fully connected network model to obtain a first feature vector of the first nucleic acid characteristic parameter, a second feature vector of the lipid characteristic parameter and a third feature vector of the microbial characteristic parameter, inputting the second nucleic acid characteristic parameter into the convolutional layer to obtain a fourth feature vector of the second nucleic acid characteristic parameter, generating a target vector based on the first feature vector, the second feature vector, the third feature vector and the fourth feature vector, calling the classification model to predict the target vector, and obtaining a probability value of the organism suffering from diabetes.

[0006] In some embodiments of the present application, the selecting of target feature parameters related to diabetes from the feature parameter set includes: performing at least one iterative update on a preset parameter model based on the feature parameter set until a target parameter model that meets the requirements is obtained, wherein any one of the at least one iterative update includes: selecting a feature parameter from the feature parameter set, and updating the parameter model obtained last time according to the selected feature parameter and the weight corresponding to the selected feature parameter to obtain a current updated model, calculating the current updated model according to a likelihood ratio chi-square test method to obtain a significance value of the selected feature parameter, selecting an initial parameter model from the current updated model and the parameter model obtained last time according to the significance value and the preset value, if there is at least one unselected feature parameter in the feature parameter set, updating the initial parameter model using the unselected feature parameter, or, if all feature parameters in the feature parameter set are selected, determining the initial parameter model as the target parameter model, and determining all feature parameters corresponding to the target parameter model as the target feature parameters.

[0007] In some embodiments of the present application, the parameter model obtained last time is updated according to the selected feature parameters and the weights corresponding to the selected feature parameters to obtain the current updated model, including: performing operations on the selected feature parameters and the weights corresponding to the selected feature parameters to obtain operation results, and adding the operation results to the parameter model obtained last time to obtain the current updated model.

[0008] In some embodiments of the present application, selecting the initial parameter model from the current update model and the last obtained parameter model based on the significance value and the preset value includes: if the significance value is less than the preset value, determining the current update model as the initial parameter model, or, if the significance value is greater than or equal to the preset value, determining the last obtained parameter model as the initial parameter model.

[0009] In some embodiments of the present application, the fully connected network model includes multiple fully connected neural networks, each fully connected neural network includes multiple fully connected layers, the first nucleic acid feature parameters are multiple, and the method for generating the first feature vector includes: vectorizing the multiple first nucleic acid parameters to obtain a nucleic acid parameter vector, inputting the nucleic acid parameter vector into any fully connected neural network, and obtaining a feature vector output by each fully connected layer in any fully connected neural network, wherein the feature vector output by the previous fully connected layer is used as input data of the subsequent fully connected layer, and the feature vector output by the last fully connected layer in any fully connected neural network is determined as the first feature vector.

[0010] In some embodiments of the present application, there are multiple second nucleic acid feature parameters, and the second nucleic acid feature parameters are input into the convolution layer to obtain a fourth feature vector of the second nucleic acid feature parameters, including: generating a first feature matrix according to the multiple second nucleic acid feature parameters, performing convolution processing on the feature matrix to obtain a second feature matrix, and performing pooling operation on the second feature matrix to obtain the fourth feature vector.

[0011] In some embodiments of the present application, if the diabetes is gestational diabetes, the biomolecules include free deoxyribonucleic acid, free ribonucleic acid, lipids and microorganisms.

[0012] In some embodiments of the present application, the method further includes: controlling a sequencer to sequence the deoxyribonucleic acid in a stool sample of the organism to obtain characteristic parameters of the microorganism.

[0013] On the other hand, the present application provides an electronic device, comprising: a memory storing at least one instruction; and a processor executing the at least one instruction to implement the diabetes prediction method.

[0014] On the other hand, the present application provides a computer-readable storage medium, wherein at least one instruction is stored in the computer-readable storage medium, and the at least one instruction is executed by a processor in an electronic device to implement the diabetes prediction method.

[0015] Through the above-mentioned implementation mode, since the occurrence of diabetes is affected by the combined influence of various complex factors in the biological system, the present application predicts diabetes through the characteristic parameter set of biological molecules (the present application predicts diabetes through various biological data of the organism), which can avoid the problem of not being able to fully consider the complexity of the biological system, thereby improving the prediction accuracy. The target characteristic parameters related to diabetes can be selected from the characteristic parameter set by the likelihood ratio chi-square test method, thereby reducing the interference of redundant characteristic parameters. The first nucleic acid characteristic parameter, lipid characteristic parameter and microbial characteristic parameter are feature extracted by the neural network model, which can make the features in the first characteristic vector, the second characteristic vector and the third characteristic vector more accurate. Therefore, when predicting the target vector corresponding to the first characteristic vector, the second characteristic vector, the third characteristic vector and the fourth characteristic vector, the prediction accuracy can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a structural diagram of an electronic device provided in one embodiment of the present application.

[0017] Figure 2 This is a flowchart of a diabetes prediction method provided in one embodiment of the present application.

[0018] Figure 3 This is the basic information of the healthy control sample and the gestational diabetes sample provided in one embodiment of the present application.

[0019] Figure 4 This is a prediction effect diagram of gestational diabetes provided in one embodiment of the present application.

[0020] Figure 5 This is a prediction effect diagram of gestational diabetes provided by another embodiment of the present application.

[0021] Figure 6 It is a flowchart of a method for generating a target parameter model provided in one embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] It should be noted that in this application, "at least one" means one or more, and "more than one" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and drawings of this application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0024] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0025] In the biomedical field, the prediction of specific diseases can help effectively manage the patient's body. For example, the occurrence of diabetes is affected by a combination of complex factors within the biological system. Therefore, if the probability of an organism suffering from diabetes is predicted based on a single omics data, such as proteomic data, the prediction results may be inaccurate due to the inability to fully consider the complexity of the biological system. In addition, single omics data are often easily affected by factors such as sample contamination or noise generated during the analysis process, which further reduces the accuracy of the prediction results.

[0026] In addition, due to the complexity of biological systems, the models trained with single omics data may have adaptability issues, making it difficult to accurately predict diabetes.

[0027] In order to solve the above technical problems, the present application provides a diabetes prediction method, an electronic device and a storage medium, which can improve the accuracy of diabetes prediction. The diabetes prediction method provided in the embodiment of the present application can be applied to one or more electronic devices.

[0028] like Figure 1 1 is a block diagram of an electronic device provided in an embodiment of the present application. The electronic device 10 may be a mobile phone, a tablet computer, a laptop computer, or a computer. The present embodiment of the present application does not impose any restrictions on the specific type of the electronic device.

[0029] like Figure 1 As shown, the electronic device 10 may include a communication module 101, a memory 102, a processor 103, an input / output (I / O) interface 104 and a bus 105. The processor 103 is coupled to the communication module 101, the memory 102 and the input / output interface 104 through the bus 105.

[0030] The communication module 101 may include a wired communication module and / or a wireless communication module. The wired communication module may provide one or more wired communication solutions such as universal serial bus (USB), controller area network bus (CAN, Controller Area Network). The wireless communication module may provide one or more wireless communication solutions such as wireless fidelity (Wi-Fi), Bluetooth (BT), mobile communication network, frequency modulation (FM), near field communication technology (NFC), infrared technology (IR), etc.

[0031] The memory 102 may include one or more random access memories (RAM) and one or more non-volatile memories (NVM). The random access memory can be directly read and written by the processor 103, and can be used to store executable programs (such as machine instructions) of other running programs, and can also be used to store user and application data. The random access memory may include static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), etc.

[0032] The non-volatile memory may also store executable programs and user and application data, etc., and may be loaded into the random access memory in advance for direct reading and writing by the processor 110. The non-volatile memory may include a disk storage device and a flash memory.

[0033] The memory 102 is used to store one or more computer programs. The one or more computer programs are configured to be executed by the processor 103. The one or more computer programs include a plurality of instructions, and when the plurality of instructions are executed by the processor 103, the diabetes prediction method executed on the electronic device 10 can be implemented.

[0034] In other embodiments, Figure 1 The electronic device 10 shown also includes an external memory interface for connecting to an external memory to expand the storage capacity of the electronic device 10 .

[0035] The processor 103 may include one or more processing units, for example, the processor 103 may include an application processor (application processor, AP), a modem processor, a graphics processor (graphics processing unit, GPU), an image signal processor (image signal processor, ISP), a controller, a video codec, a digital signal processor (digital signal processor, DSP), and / or a neural network processor (neural-network processing unit, NPU), etc. Among them, different processing units may be independent devices or integrated in one or more processors.

[0036] The processor 103 provides computing and control capabilities. For example, the processor 103 is used to execute a computer program stored in the memory 102 to implement the above-mentioned diabetes prediction method.

[0037] The input / output interface 104 is used to provide a channel for user input or output. For example, the input / output interface 104 can be used to connect various input and output devices, such as a mouse, keyboard, touch device, display screen, etc., so that the user can enter information or visualize information.

[0038] The bus 105 is at least used to provide a channel for mutual communication among the communication module 101 , the memory 102 , the processor 103 , and the input / output interface 104 in the electronic device 10 .

[0039] It is to be understood that the structure illustrated in the embodiment of the present application does not constitute a specific limitation on the electronic device 10. In other embodiments of the present application, the electronic device 10 may include more or fewer components than shown in the figure, or combine some components, or split some components, or arrange the components differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0040] like Figure 2 FIG. 1 is a flowchart of a diabetes prediction method provided by an embodiment of the present application. According to different requirements, the order of each step in the flowchart can be adjusted according to actual requirements, and some steps can be omitted. The execution subject of the method is an electronic device, such as Figure 1 An electronic device 1 is shown.

[0041] S11, obtaining multi-omics data of an organism, where the multi-omics data includes a characteristic parameter set of biological molecules.

[0042] In some embodiments of the present application, the organism may be a human, a kitten, a puppy or other animal, and the present application does not limit the organism. Multi-omics data refers to data obtained by comprehensively analyzing biological samples of an organism from multiple levels using different technologies and methods. Multi-omics data may include information from multiple levels such as genomics, transcriptomics, proteomics, metabolomics, microbiome, lipidomics, etc. Multi-omics data can be used to comprehensively understand the complexity and function of biological systems in organisms. The characteristic parameter set includes characteristic parameters of multiple biological molecules, and each characteristic parameter of a biological molecule may be multiple.

[0043] Diabetes includes type 1 diabetes, type 2 diabetes and gestational diabetes (GDM). Gestational diabetes is a metabolic disorder that occurs during pregnancy, mainly manifested by elevated blood sugar levels. Gestational diabetes not only has adverse effects on the health of pregnant women themselves, but also increases the risk of pregnancy complications (such as gestational hypertension) in pregnant women. Therefore, it is very necessary to predict gestational diabetes.

[0044] In order to more clearly explain the present solution, the following will introduce the organism, biological sample, biological molecule and characteristic parameters by taking the prediction of gestational diabetes as an example. The biological sample corresponds to the organism. For example, if the organism is a pregnant woman in early pregnancy, the biological sample may include, but is not limited to: blood (plasma and serum), urine, feces, tissue, saliva, amniotic fluid and maternal peripheral blood, etc., and the present application does not limit this.

[0045] Biomolecules refer to chemical substances that exist in organisms and participate in their life activities. Biomolecules in maternal peripheral plasma include, but are not limited to, nucleic acids and lipids (lipidics). Nucleic acids include free deoxyribonucleic acid (cfDNA) and free ribonucleic acid (cfRNA). Biomolecules in feces can be microorganisms (microbiome).

[0046] The characteristic parameters correspond to biological molecules. For example, the characteristic parameters of lipids include, but are not limited to: the abundance of various types of lipids (such as phospholipids, hydroxy fatty acids, acetylcholine, triglycerides, etc.). The characteristic parameters of cfRNA can be the expression value of highly expressed messenger ribonucleic acid (mRNA) (the transcription level or translation level of messenger ribonucleic acid in cells). The characteristic parameters of microorganisms include, but are not limited to: the types and abundance of microbial communities. The characteristic parameters of cfDNA include, but are not limited to: methylation analysis parameters (MA), transcription start site score (TSS score), motif diversity score (MDS), end motif (End Motif) and fetal concentration (fetal fraction).

[0047] For example, the acquisition of characteristic parameters of lipids includes: performing lipid mass spectrometry analysis on maternal peripheral plasma after freezing treatment (for example, it can be stored in an environment of -80°C) to measure and identify the composition of lipid molecules, thereby obtaining characteristic parameters such as the abundance of various types of lipids (such as phospholipids, hydroxy fatty acids, acetylcholine, triglycerides, etc.). Among them, the maternal peripheral plasma after freezing treatment is obtained by plasma separation of maternal peripheral blood. In order to avoid the inactivation or damage of biologically active substances (such as enzymes, hormones, antibodies, etc.) in plasma due to excessively high temperature or excessively long time, which affects the accuracy of the prediction of gestational diabetes, it is necessary to perform plasma separation on the collected maternal peripheral blood as soon as possible (for example, plasma separation can be performed within 8 hours after the collection of maternal peripheral blood is completed) and the maternal peripheral blood should also be frozen after collection (for example, it can be stored in an environment of 4°C).

[0048] For example, the process of generating characteristic parameters of cfRNA includes: adding TrizolLS reagent to maternal peripheral plasma after freezing treatment according to the ratio of TrizolLS1: maternal peripheral plasma 3 to obtain a plasma mixed reagent, extracting cfRNA from the plasma mixed reagent, and sequencing the extracted cfRNA to obtain cfRNA sequencing data. Low-quality RNA sequences (such as RNA with a length less than 17bp read length), rRNA sequences, valueRNA sequences and YRNA sequences are filtered out from the cfRNA sequencing data, wherein the read length is the length of a single sequence. The remaining RNA sequences in the cfRNA sequencing data are aligned to the human transcriptome to obtain an alignment result, and each RNA is quantified according to the alignment result to obtain the expression value of each mRNA, wherein the alignment order of the remaining RNA sequences in the cfRNA sequencing data to the human transcriptome is miRNA sequence→tRNA sequence→piRNA sequence→mRNA sequence→lncRNA sequence→other RNA sequences. Among them, the method of extracting cfRNA from the plasma mixed reagent can refer to the relevant technology, and this application will not be described in detail here.

[0049] For example, the process of obtaining the characteristic parameters of cfDNA includes: isolating cfDNA from maternal peripheral plasma after freezing by using the DNAMiniKF kit, preparing the extracted cfDNA by using the MGIEasyDNA library preparation kit (MGI) protocol to obtain the prepared cfDNA library, and quantitatively processing the cfDNA library to obtain the quality and concentration of the cfDNA library. The cfDNA is circularized to obtain a single-stranded DNA ring (ssDNA), and the circularized ssDNA is quantified to obtain the concentration and quality of the ssDNA. DNA nanoballs (DNBs) are obtained by amplifying ssDNA, and the nanoballs are quantified by using the QubitssDNA detection kit, and the nanoballs are multiplexed using a double-end 100 base pair strategy to obtain cfDNA sequencing data. The methylation analysis parameters, transcription start site scores, terminal sequence diversity scores, terminal motifs, and fetal concentrations in the characteristic parameters of cfDNA can all be obtained by analyzing the cfDNA sequencing data. For example, methylation analysis is performed on cfDNA sequencing data to obtain methylation analysis parameters. The fetal concentration is obtained by analyzing the cfDNA sequencing data through a deep learning algorithm. The terminal sequence diversity score can be obtained by determining the coverage of the starting transcription region corresponding to the starting transcription point in the cfDNA sequencing data by a preset length (e.g., 500 bp). The calculation method of the terminal sequence diversity score can refer to formula (1):

[0050]

[0051] Among them, MDS represents the terminal sequence diversity score, P {i} Represents the probability of the terminal motif appearing in cfDNA.

[0052] For example, the process of obtaining the characteristic parameters of microorganisms includes: extracting DNA from fecal samples through a nucleic acid extractor (such as KingFisherML, KingFisherFlex, KingFisherDuo) and a nucleic acid kit (such as MagpureStoolDNAKFKitB), purifying the extracted DNA through a purification kit (such as AxygenTMAxyPrepTMMagPCRClean-UpKit), and performing end-repair and A-base ligation treatment on the purified DNA through an enzyme mixture composed of multiple enzymes (inactivating enzymes) to obtain repaired DNA, wherein the end-repair and A-base ligation treatment of the purified DNA needs to be inactivated at 65°C, and the repaired DNA is treated with an enzyme with a specification of TRANSGEN600000U / mL to obtain connected DNA, and the connected DNA is sequenced by a sequencer (such as DNBSEQ-T10, PE100), thereby obtaining parameters such as the abundance of various types of lipids (such as phospholipids, hydroxy fatty acids, acetylcholine, triglycerides, etc.).

[0053] The characteristic parameter set of this embodiment includes 61,165 characteristic parameters, of which 39,129 are characteristic parameters of cfDNA, 19,610 are characteristic parameters of cfRNA, 1,442 are characteristic parameters of lipids, and 984 are characteristic parameters of microorganisms.

[0054] In this embodiment, since multi-omics data can comprehensively understand the complexity and function of biological systems in an organism, obtaining multi-omics data can provide a basis for accurately predicting diabetes (gestational diabetes).

[0055] S12, selecting target characteristic parameters related to diabetes from the characteristic parameter set, wherein the target characteristic parameters include a first nucleic acid characteristic parameter, a second nucleic acid characteristic parameter, a lipid characteristic parameter, and a microbial characteristic parameter.

[0056] In some embodiments of the present application, the electronic device performs at least one iterative update on the preset parameter model based on the feature parameter set by the likelihood ratio chi-square test method, so that the updated model is different from the original model, until a target parameter model that meets the requirements is obtained, and all feature parameters corresponding to the target parameter model are determined as target feature parameters. Among them, the preset parameter model is an empty model that only includes initial weights, and the initial weights can be preset vectors. The preset parameter model can be set by itself, and the present application does not limit this.

[0057] For example, the preset parameter model provided in the embodiment of the present application can refer to formula (2):

[0058] Model empty :

[0059] Among them, Model empty represents the preset parameter model, y represents the output of the preset parameter model, β 0 represents the initial weight.

[0060] In some embodiments of the present application, the first nucleic acid characteristic parameter, the second nucleic acid characteristic parameter, the lipid characteristic parameter and the microorganism characteristic parameter may be multiple.

[0061] In this embodiment, the first nucleic acid characteristic parameter may be a characteristic parameter of cfDNA, and the second nucleic acid characteristic parameter may be a characteristic parameter of cfRNA. 198 target characteristic parameters were selected from 61,165 characteristic parameters by the likelihood ratio chi-square test method. Among the 198 target characteristic parameters, 46 characteristic parameters of cfDNA, 25 characteristic parameters of cfRNA, 106 characteristic parameters of lipids, and 22 characteristic parameters of microorganisms.

[0062] In this embodiment, since the feature parameter set of multi-omics data may contain feature parameters (noise) that are not related to the generation of diabetes, the forward selection method and the likelihood ratio chi-square test method of continuously iteratively updating the model are used to select target feature parameters related to diabetes from the feature parameter set, which can filter out the noise and reduce the impact of the noise on the prediction results, thereby improving the prediction accuracy of diabetes.

[0063] In other embodiments of the present application, after obtaining the target characteristic parameters, the trained neural network model can be called to predict the target characteristic parameters to obtain the probability value of the organism suffering from gestational diabetes.

[0064] Among them, the neural network model can be a pulse neural network model (Feature-Specific Neural Network, SNN), and the neural network model includes a fully connected network model, a convolutional layer and a classification model. Before calling the neural network model for prediction, it is necessary to obtain the data set and the learner corresponding to the neural network model, train the learner with the training data in the data set, and test the trained learner with the test data in the data set. When the trained learner passes the test of the test data, the trained learner is determined as a neural network model. Among them, the data set can be divided according to a preset ratio to obtain training data and test data. The preset ratio can be set by yourself, and this application does not limit this. For example, the preset ratio can be: training data 7: test data 3.

[0065] The data set can be the characteristic parameters of healthy control samples and gestational diabetes samples. Figure 3 Shown is the basic information of the healthy control sample and the gestational diabetes sample provided in one embodiment of the present application. Figure 3 The number of healthy control samples and gestational diabetes samples in N is 299. Figure 3 The distribution of age, pre-pregnancy body mass index (Body Mass Index, BMI), whether it is primiparous, and whether there is a family history of diabetes for 299 healthy control samples and 299 gestational diabetes samples are shown respectively. BMI values ​​between 18.5-24.9 are normal, BMI values ​​between 25-29.9 are overweight, and BMI values ​​of 30 and above are obese. Since the average BMI value of gestational diabetes samples of 4.49 is greater than the BMI value of healthy control samples of 3.32, it can be seen that the average weight of gestational diabetes samples is greater than that of healthy control samples. The proportion of gestational diabetes samples with a family history of diabetes, 23%, is also higher than the proportion of healthy control samples with a family history of diabetes, 16%. In addition, Figure 3 The significance values ​​corresponding to information such as age, pre-pregnancy body mass index (BMI), whether or not a primiparous woman has a family history of diabetes, also known as p-values, are also introduced. Among them, the p-value corresponding to age is the largest, the p-value corresponding to pre-pregnancy body mass index is the smallest, the p-value corresponding to whether or not a family history of diabetes is greater than the p-value corresponding to pre-pregnancy body mass index, and the p-value corresponding to whether or not a primiparous woman has a smaller p-value than the p-value corresponding to whether or not a family history of diabetes exists. Since the smaller the p-value, the greater the impact of the information corresponding to whether or not a woman has gestational diabetes, the impact of age, pre-pregnancy body mass index, whether or not a primiparous woman has a family history of diabetes on whether or not a woman has diabetes is, from largest to smallest, BMI, whether or not a family history of diabetes exists, whether or not a primiparous woman has a family history of diabetes, and age.

[0066] In some embodiments of the present application, the training data and the test data both include characteristic parameters of healthy control samples and gestational diabetes samples. During training, the cross entropy loss can be used as the loss function of the learner. When the learner is trained with the training data and the trained learner is tested with the test data, the prediction performance of the learner for the training data and the prediction performance of the trained learner for the test data can be evaluated by a preset indicator, thereby determining whether the trained learner passes the test of the test data. The preset indicator can be the area under the curve.

[0067] (AreaUnderCurve, AUC), where the area under the curve can be the area enclosed by the ROC curve and the coordinate axis (horizontal axis), ROC is a curve with the true positive rate (TruePositiveRate, TPR) as the vertical axis and the false positive rate (FalsePositiveRate, FPR) as the horizontal axis, where TPR represents the proportion of actual positive samples (healthy control samples) that are correctly predicted as positive samples (healthy control samples), and FPR represents the proportion of actual negative samples (gestational diabetes samples) that are correctly predicted as negative samples (gestational diabetes samples). The larger the AUC value, the higher the prediction performance of the corresponding learner. When the AUC value of the trained learner for the test data meets the preset conditions (for example, the AUC value is greater than 95%), the trained learner can be determined as a neural network model.

[0068] In other embodiments of the present application, the prediction performance of the learner for training data and the prediction performance of the trained learner for test data can be evaluated by other preset indicators. For example, other preset indicators can be the area under the ROC curve (Area Under Curve, AUC) and the Kolmogorov-Smirnov distance (Kolmogorov-Smirnov, KS), etc.

[0069] In addition, the present application embodiment studies have found that the number of categories of feature parameters in the training data and test data and the missing of feature parameters of samples affect the prediction performance of the learner.

[0070] For example, Figure 4 , which is a diagram showing the prediction effect of gestational diabetes provided by an embodiment of the present application. Figure 4In the figure, the curve corresponding to AUC=0.57 is the ROC curve obtained by predicting the first test data by the learner trained by the first training data, wherein the first training data and the first test data only include the characteristic parameters of cfDNA. The curve corresponding to AUC=0.70 is the ROC curve obtained by predicting the second test data by the learner trained by the second training data, wherein the second training data and the second test data both include the characteristic parameters of cfDNA and the characteristic parameters of microorganisms. The curve corresponding to AUC=0.72 is the ROC curve obtained by predicting the third test data by the learner trained by the third training data, wherein the third training data and the third test data both include the characteristic parameters of cfDNA, the characteristic parameters of microorganisms, and the characteristic parameters of lipids. The curve corresponding to AUC=0.76 is the ROC curve obtained by predicting the fourth test data by the learner trained by the fourth training data, wherein the fourth training data and the fourth test data both include the characteristic parameters of cfDNA, the characteristic parameters of cfRNA, the characteristic parameters of microorganisms, and the characteristic parameters of lipids. The first training data, the first test data, the second training data, the second test data, the third training data, the third test data, the fourth training data, and some samples of the fourth test data are missing feature parameters. Figure 4 It can be seen that the more categories of feature parameters there are in the training data and test data, the larger the AUC value of the trained learner is and the higher the prediction performance is.

[0071] In another embodiment of the present application, characteristic parameters of samples with missing parameters are deleted in the first training data, the first test data, the second training data, the second test data, the third training data, the third test data, the fourth training data, and the fourth test data, to obtain characteristic parameters of samples with missing parameters deleted in the first target training data, the first target test data, the second target training data, the second target test data, the third target training data, the third target test data, the fourth target training data, and the fourth target test data, and the learner is trained and tested by each target training data and each target test data after deleting the missing parameters. Figure 5 , is a diagram showing the prediction effect of gestational diabetes provided by another embodiment of the present application. Figure 5In the figure, the curve corresponding to AUC=0.77 is the ROC curve obtained by predicting the first target test data using the learner trained with the first target training data. The curve corresponding to AUC=0.85 is the ROC curve obtained by predicting the second target test data using the learner trained with the second target training data. The curve corresponding to AUC=0.94 is the ROC curve obtained by predicting the third target test data using the learner trained with the third target training data. The curve corresponding to AUC=0.97 is the ROC curve obtained by predicting the fourth target test data using the learner trained with the fourth target training data. Figure 5 It can be seen that the more categories of feature parameters there are in the target training data and the target test data, the larger the AUC value of the trained learner is and the higher the prediction performance is.

[0072] By Figure 4 and Figure 5 By comparison, we can find that Figure 5 The AUC value corresponding to each ROC curve is greater than Figure 4 The AUC value corresponding to each ROC curve in , so after deleting the feature parameters of samples with missing parameters, the prediction performance of the trained learner is significantly improved.

[0073] S13, inputting the first nucleic acid characteristic parameter, the lipid characteristic parameter and the microorganism characteristic parameter into the fully connected network model to obtain a first characteristic vector of the first nucleic acid characteristic parameter, a second characteristic vector of the lipid characteristic parameter and a third characteristic vector of the microorganism characteristic parameter.

[0074] In one embodiment of the present application, the fully connected network model includes multiple fully connected neural networks, and the number of fully connected layers in each fully connected neural network can be set by itself. For example, the number of fully connected layers in each fully connected neural network can be 4 layers.

[0075] In some embodiments of the present application, the electronic device inputs a first nucleic acid feature parameter into a fully connected network model to obtain a first feature vector of the first nucleic acid feature parameter, including: the electronic device vectorizes multiple first nucleic acid parameters to obtain a nucleic acid parameter vector, and inputs the nucleic acid parameter vector into any fully connected neural network to obtain a feature vector output by each fully connected layer in any fully connected neural network, wherein the feature vector output by the previous fully connected layer can be used as input data for the subsequent fully connected layer, and the electronic device determines the feature vector output by the last fully connected layer in any fully connected neural network as the first feature vector.

[0076] Each fully connected layer includes weights, biases, and activation functions, and the activation function may be ReLu. The method for generating the first eigenvector may refer to formula (3):

[0077] L1=[Relu(W m l…lRelu(W j [Relu(W i ·x i +b i )]+b j )]…]+b m )]; (3)

[0078] Among them, L1 represents the first eigenvector, x i represents the nucleic acid parameter vector, i, j, ..., m are the layer indices, starting from the input layer to the output layer, W i , W j , ..., W m denote the weights of the i, j, ..., m fully connected layers, respectively, and b i , b j , .., b m They represent the bias of the i-th, j-th, ..., m-th fully connected layers respectively, and Relu represents the activation function.

[0079] In some embodiments of the present application, due to multiple fully connected neural networks, the lipid characteristic parameters and the microbial characteristic parameters can be input into each of the remaining fully connected neural networks to obtain the second eigenvector of the lipid characteristic parameters and the third eigenvector of the microbial characteristic parameters. The generation process of the second eigenvector and the third eigenvector is basically the same as the generation process of the first eigenvector, so this application will not repeat the description.

[0080] In this embodiment, since the characteristic parameters of cfDNA, the characteristic parameters of lipids, and the characteristic parameters of microorganisms are usually uniform, a fully connected neural network is used to process the characteristic parameters of cfDNA, the characteristic parameters of lipids, and the characteristic parameters of microorganisms. Since the fully connected neural network can automatically capture the complex relationship between the characteristic parameters, the accuracy of the first feature vector, the second feature vector, and the third feature vector can be improved.

[0081] S14, input the second nucleic acid feature parameter into the convolution layer to obtain a fourth feature vector of the second nucleic acid feature parameter.

[0082] In some embodiments, the electronic device inputs the second nucleic acid feature parameter into the convolution layer to obtain a fourth feature vector of the second nucleic acid feature parameter, including: the electronic device generates a first feature matrix based on multiple second nucleic acid feature parameters, performs convolution processing on the feature matrix to obtain a second feature matrix, and performs pooling operation on the second feature matrix to obtain a fourth feature vector.

[0083] The pooling operation may be a maximum pooling operation, and the rows and columns of the first feature matrix may be set arbitrarily, and this application does not impose any restrictions on this. The method for generating the fourth feature vector may refer to formula (4):

[0084] R=Maxpool(∑ j ∑ k filter[j,k]cfRNA[mj,nk]); (4)

[0085] Among them, R represents the fourth eigenvector, Maxpool represents the maximum pooling operation, which is used to reduce the dimension and highlight important features. Filter represents the convolution kernel (filter), j represents the row in the convolution kernel, and k represents the column in the convolution kernel. Filter[j, k] represents the value corresponding to the jth row and the kth column in the convolution kernel, cfRNA represents the first feature matrix, m represents the row in the first feature matrix, n represents the column in the first feature matrix, and cfRNA[mj, nk] represents the value corresponding to the mjth row and the nkth column in the first feature matrix.

[0086] In this embodiment, since the convolutional layer can achieve weight sharing and regularization, the complex relationship between genes in the characteristic parameters of cfRNA is effectively learned and extracted.

[0087] S15, generating a target vector based on the first eigenvector, the second eigenvector, the third eigenvector and the fourth eigenvector.

[0088] In some embodiments of the present application, the electronic device concatenates the first feature vector, the second feature vector, the third feature vector, and the fourth feature vector (for example, by fusion) to obtain a target vector. 1 =

[12] , the second eigenvector is L 2 =

[34] , the third eigenvector is L 3 =

[56] , the fourth eigenvector is L 4 =

[78] , then the target vector obtained after splicing is R=[1 2 3 4 5 6 7 8].

[0089] In this embodiment, the first eigenvector, the second eigenvector, the third eigenvector and the fourth eigenvector are concatenated so that the target vector can incorporate information of multiple feature parameters.

[0090] S16, calling the classification model to predict the target vector, and obtaining the probability value of the organism suffering from diabetes.

[0091] In some embodiments of the present application, the classification model includes a classification layer and a fully connected layer, the classification layer includes weights and biases, and the fully connected layer can be a softmax activation function. The calculation method of the probability value can refer to formula (5):

[0092] class=softmax(p i )=softmax(R·W c +b c ); (5)

[0093] Among them, class represents the probability value, softmax represents the activation function of the fully connected layer, R represents the target vector, and W c represents the weight of the classification layer, b c Represents the bias of the classification layer.

[0094] Through the above-mentioned implementation mode, since the occurrence of diabetes is affected by the combined influence of various complex factors in the biological system, the present application predicts diabetes through the characteristic parameter set of biological molecules (the present application predicts diabetes through various biological data of the organism), which can avoid the problem of not being able to fully consider the complexity of the biological system, thereby improving the prediction accuracy. The target characteristic parameters related to diabetes can be selected from the characteristic parameter set by the likelihood ratio chi-square test method, thereby reducing the interference of redundant characteristic parameters. The first nucleic acid characteristic parameter, lipid characteristic parameter and microbial characteristic parameter are feature extracted by the neural network model, which can make the features in the first characteristic vector, the second characteristic vector and the third characteristic vector more accurate. Therefore, when predicting the target vector corresponding to the first characteristic vector, the second characteristic vector, the third characteristic vector and the fourth characteristic vector, the prediction accuracy can be improved.

[0095] In some embodiments of the present application, Figure 6 FIG. 1 is a flow chart of a method for generating a target parameter model provided by an embodiment of the present application, comprising the following steps:

[0096] S121, selecting a feature parameter from the feature parameter set, and updating the parameter model obtained last time according to the selected feature parameter and the weight corresponding to the selected feature parameter to obtain a current updated model.

[0097] In some embodiments, the electronic device may select any characteristic parameter from the characteristic parameter set until all characteristic parameters are selected to execute subsequent processes, or may select characteristic parameters one by one according to a certain order. The electronic device updates the parameter model obtained last time according to the selected characteristic parameters and the weights corresponding to the selected characteristic parameters to obtain the current updated model, including: the electronic device operates the selected characteristic parameters and the weights corresponding to the selected characteristic parameters to obtain the operation results, and adds the operation results to the parameter model obtained last time to obtain the current updated model.

[0098] Each feature parameter in the feature parameter set has a corresponding weight, and each feature parameter has a corresponding weight that can be set by itself to an initial random value, and then the initial random value is updated through training. The electronic device can multiply the selected feature parameter with the weight corresponding to the selected feature parameter to obtain a calculation result.

[0099] In other embodiments of the present application, the electronic device may also perform a division / addition / subtraction operation on the selected characteristic parameter and the weight corresponding to the selected characteristic parameter to obtain an operation result. The present application does not limit this.

[0100] For example, following the above embodiment, if the model obtained last time is a preset parameter model (empty model) Model empty : The characteristic parameter selected is β 1 , the weight corresponding to the selected feature parameter is f 1 , if the selected feature parameter is multiplied by the corresponding weight, the result is β 1 f 1 .Currently updated modelModel new The generation method of can refer to formula (6):

[0101] Model new :

[0102] Among them, Model new represents the current update model, y′ represents the output of the current update model, β 0 represents the initial weight in the preset parameter model, β 1 f 1 Indicates the result of an operation.

[0103] In this embodiment, the electronic device selects different characteristic parameters each time. The parameter model obtained last time is updated by the selected characteristic parameters and the weights corresponding to the selected characteristic parameters, so that the current updated model can be different from the parameter model obtained last time.

[0104] S122, calculating the current updated model according to the likelihood ratio chi-square test method to obtain the significance value of the selected characteristic parameter.

[0105] In some embodiments, the significance value is also called the p-value. The significance value can represent the difference between the performance of the current updated model and the model obtained last time. When the significance value is less than the preset value, it means that the performance of the current updated model is higher than the performance of the model obtained last time (it means that the newly added feature parameters in the current updated model are key parameters that affect the prediction effect of diabetes). When the significance value is equal to or greater than the preset value, it means that the performance of the current updated model is lower than the performance of the model obtained last time (the newly added feature parameters in the current updated model have a small impact on the prediction effect of diabetes, and the newly added feature parameters in the current updated model can be determined as noise).

[0106] In this embodiment, the current updated model that adds new feature parameters and is different from the parameter model obtained last time is calculated through the likelihood ratio chi-square test method. The method for obtaining the significance value of the selected feature parameter can refer to the relevant technology and will not be described in detail in this application.

[0107] S123, selecting an initial parameter model from the current update model and the last obtained parameter model according to the significance value and the preset value.

[0108] In some embodiments, the electronic device selects an initial parameter model from a current update model and a last obtained parameter model based on a significance value and a preset value, including: if the significance value is less than the preset value, determining the current update model as the initial parameter model; or, if the significance value is greater than or equal to the preset value, determining the last obtained parameter model as the initial parameter model.

[0109] The preset value can be set by yourself, and this application does not limit this. For example, the preset value can be 0.05.

[0110] In this embodiment, when the significance value is less than the preset value, it represents that the newly added characteristic parameter in the current updated model is the key parameter affecting the prediction effect of diabetes, and the current updated model is determined as the initial parameter model, so that the key parameters affecting the prediction effect of diabetes can be retained.

[0111] S124, determining whether there are any unselected feature parameters in the feature parameter set.

[0112] In some embodiments of the present application, if all feature parameters in the feature parameter set are selected, step S125 is executed. If there is at least one unselected feature parameter in the feature parameter set, the process returns to step S121, and another unselected feature parameter is selected from the feature parameter set to update the initial parameter model.

[0113] Among them, the process of updating the initial parameter model is basically the same as the process of generating the current update model, so this application will not repeat the description.

[0114] S125, stop the iterative update, and determine the model obtained by the iterative update as the target parameter model.

[0115] In this embodiment, when all the characteristic parameters in the characteristic parameter set are selected, it means that all the key parameters that affect the prediction effect of diabetes have been determined from the characteristic parameter set, and therefore the iterative update can be stopped.

[0116] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions. The method implemented when the program instructions are executed can refer to the methods in the above-mentioned embodiments of the present application.

[0117] The computer-readable storage medium may be an internal memory of the electronic device described in the above embodiment, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), etc. equipped on the electronic device.

[0118] In some embodiments, the computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the electronic device, etc.

[0119] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation.

[0120] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0121] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0122] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present application is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present application. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0123] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in this application can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, a person of ordinary skill in the art should understand that the technical solution of the present application may be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present application.

Claims

1. A diabetes prediction method, characterized in that: The method comprises: Acquiring multi-omics data of an organism, wherein the multi-omics data includes a characteristic parameter set of biological molecules; Selecting a target characteristic parameter related to diabetes from the characteristic parameter set, wherein the target characteristic parameter includes a first nucleic acid characteristic parameter, a second nucleic acid characteristic parameter, a lipid characteristic parameter, and a microbial characteristic parameter; The trained neural network model is called to predict the target characteristic parameter to obtain the probability value of the organism suffering from diabetes, wherein the neural network model includes a fully connected network model, a convolutional layer and a classification model, including: Inputting the first nucleic acid characteristic parameter, the lipid characteristic parameter and the microorganism characteristic parameter into the fully connected network model to obtain a first characteristic vector of the first nucleic acid characteristic parameter, a second characteristic vector of the lipid characteristic parameter and a third characteristic vector of the microorganism characteristic parameter; Inputting the second nucleic acid feature parameter into the convolution layer to obtain a fourth feature vector of the second nucleic acid feature parameter; Generate a target vector based on the first feature vector, the second feature vector, the third feature vector and the fourth feature vector; The classification model is called to predict the target vector to obtain a probability value of the organism suffering from diabetes.

2. The diabetes prediction method according to claim 1, characterized in that: The selecting of target characteristic parameters related to diabetes from the characteristic parameter set comprises: Perform at least one iterative update on the preset parameter model based on the characteristic parameter set until a target parameter model that meets the requirements is obtained, wherein any one of the at least one iterative update includes: Selecting a feature parameter from the feature parameter set, and updating the parameter model obtained last time according to the selected feature parameter and the weight corresponding to the selected feature parameter to obtain a current updated model; Calculating the current updated model according to the likelihood ratio chi-square test method to obtain the significance value of the selected characteristic parameter; According to the significance value and the preset value, selecting an initial parameter model from the current update model and the parameter model obtained last time; If the feature parameter set has at least one unselected feature parameter, the initial parameter model is updated using the unselected feature parameter, or if all feature parameters in the feature parameter set are selected, the initial parameter model is determined as the target parameter model; All characteristic parameters corresponding to the target parameter model are determined as the target characteristic parameters.

3. The diabetes prediction method according to claim 2, characterized in that: The updating of the parameter model obtained last time according to the selected feature parameters and the weights corresponding to the selected feature parameters to obtain the current updated model includes: The selected characteristic parameters and the weights corresponding to the selected characteristic parameters are operated to obtain operation results, and the operation results are added to the parameter model obtained last time to obtain the current updated model.

4. The diabetes prediction method according to claim 2, characterized in that: The selecting the initial parameter model from the current update model and the last obtained parameter model according to the significance value and the preset value comprises: If the significance value is less than the preset value, determining the current update model as the initial parameter model; or If the significance value is greater than or equal to the preset value, the parameter model obtained last time is determined as the initial parameter model.

5. The diabetes prediction method according to claim 1, characterized in that: The fully connected network model includes multiple fully connected neural networks, each fully connected neural network includes multiple fully connected layers, the first nucleic acid feature parameter is multiple, and the method for generating the first feature vector includes: Performing vectorization processing on the plurality of first nucleic acid parameters to obtain a nucleic acid parameter vector; Inputting the nucleic acid parameter vector into any fully connected neural network to obtain a feature vector output by each fully connected layer in any fully connected neural network, wherein the feature vector output by the first fully connected layer serves as input data for the second fully connected layer; The feature vector output by the last fully connected layer in any one of the fully connected neural networks is determined as the first feature vector.

6. The diabetes prediction method according to claim 1, characterized in that: There are multiple second nucleic acid feature parameters, and inputting the second nucleic acid feature parameters into the convolution layer to obtain a fourth feature vector of the second nucleic acid feature parameters includes: Generating a first characteristic matrix according to a plurality of second nucleic acid characteristic parameters, and performing convolution processing on the characteristic matrix to obtain a second characteristic matrix; A pooling operation is performed on the second feature matrix to obtain the fourth feature vector.

7. The diabetes prediction method according to claim 1, characterized in that: If the diabetes is gestational diabetes, the biomolecules include free deoxyribonucleic acid, free ribonucleic acid, lipids and microorganisms.

8. The diabetes prediction method according to claim 7, characterized in that: The method further comprises: The sequencer is controlled to sequence the deoxyribonucleic acid in the stool sample of the organism to obtain characteristic parameters of the microorganism.

9. An electronic device, characterized in that: The electronic device comprises: a memory storing at least one instruction; and A processor executes the at least one instruction to implement the diabetes prediction method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the diabetes prediction method according to any one of claims 1 to 8.

Images