Parkinson's disease prediction method and system based on intestinal flora data and deep learning

Abstract

The present invention provides a Parkinson's disease prediction method and system based on intestinal flora data and deep learning. The method includes: collecting a first stool sample of several Parkinson's disease patients and a second stool sample of a healthy control, obtaining a first flora abundance matrix and a second flora abundance matrix through flora determination, adding patient labels and health labels to the first flora abundance matrix and the second flora abundance matrix respectively; dividing each first flora abundance matrix and the second flora abundance matrix with labels into a training set and a test set, training and testing a deep neural network model; using the trained deep neural network model to analyze and process the third flora abundance matrix of the target object, obtain a Parkinson's disease risk score and output it. The Parkinson's disease prediction scheme of the present invention has the advantages of non-invasiveness, low cost, and easy operation, and realizes automated analysis of the entire process.

Description

Technical Field

[0001] The present invention relates to the technical field of Parkinson's disease prediction, and in particular to a method and system for predicting Parkinson's disease based on intestinal flora data and deep learning. Background Art

[0002] Parkinson's Disease (PD) is a common neurodegenerative disease that causes motor dysfunction due to the degeneration of dopaminergic neurons. Currently, the diagnosis of Parkinson's disease mainly relies on clinical symptom assessment and neuroimaging examinations, such as brain magnetic resonance imaging (MRI) and positron emission tomography (PET). However, these diagnostic methods have limitations such as high cost, long time consumption, reliance on professional equipment and physician experience, and are difficult to perform large-scale screening and early detection. In addition, symptoms usually appear only at a later stage of the disease progression, limiting the opportunity for early intervention.

[0003] In recent years, studies have shown that intestinal flora plays an important role in the pathogenesis of Parkinson's disease, and flora imbalance may be associated with neuroinflammation and neuronal degeneration. Therefore, developing a Parkinson's disease prediction scheme based on intestinal flora characteristics can provide a non-invasive, low-cost, and easy-to-operate screening method, providing a potential solution for early diagnosis and intervention, significantly reducing diagnostic costs, and expanding the accessibility and convenience of disease screening. However, there is currently no mature Parkinson's disease prediction scheme based on intestinal flora characteristics in the existing technology. Summary of the invention

[0004] In order to solve at least one of the above technical problems, the present invention specifically provides a Parkinson's disease prediction method, system, electronic device, computer storage medium and computer program product based on intestinal flora data and deep learning.

[0005] The present invention provides a method for predicting Parkinson's disease based on intestinal flora data and deep learning, comprising the following steps: Collecting first stool samples from several Parkinson's disease patients and second stool samples from healthy controls, performing flora measurement on each of the first stool samples and the second stool samples, and performing data standardization and flora abundance calculation on the flora measurement data to obtain a first flora abundance matrix and a second flora abundance matrix, respectively, and adding patient labels and health labels to the first flora abundance matrix and the second flora abundance matrix, respectively; Dividing the labeled first flora abundance matrix and the second flora abundance matrix into a training set and a test set, and training and testing the constructed deep neural network model until a training end condition is met; Collecting a third stool sample from the target object, performing a flora measurement on the third stool sample, performing data standardization and flora abundance calculation on the flora measurement data, obtaining a third flora abundance matrix, and analyzing and processing the third flora abundance matrix using the trained deep neural network model; The deep neural network model outputs a Parkinson's disease risk score, and outputs the Parkinson's disease risk score.

[0006] Furthermore, the collecting of first stool samples from a number of Parkinson's disease patients and second stool samples from healthy controls, and performing bacterial flora determination on each of the first stool samples and the second stool samples, comprises: First stool samples from several Parkinson's disease patients and second stool samples from healthy controls were collected, and the stool samples were stored in sterile sampling tubes and immediately frozen; The 16S rRNA high-throughput sequencing technology is used to measure the flora composition of each of the first stool sample and the second stool sample to obtain corresponding flora measurement data.

[0007] Furthermore, the deep neural network model includes an input layer, a convolutional layer, an activation function, a pooling layer, and a fully connected layer; And, the deep neural network model adopts the following cross entropy loss function during training: ;

[0008] in, is the cross entropy loss function value, is the true label, is the Parkinson's disease risk score predicted by the model, and N is the total number of training samples; Use the Adam optimizer to update the network weights. The update rule of the Adam optimizer is: ; ; ; ;

[0009] in, and are the first-order momentum and second-order momentum estimates of the gradient at the t-th time step, , are the first-order momentum and second-order momentum estimates for the t-1th time step, , are the deviations of the first-order momentum and the second-order momentum at the t-th time step, and is the momentum parameter, , They are the gradient updates of the t-th time step and the t-1-th time step, respectively. is the learning rate, is a small constant to avoid zero denominator; Dropout regularization is used to randomly discard a certain proportion of neurons in the fully connected layer to prevent the model from overfitting; Use grid search to optimize hyperparameters and select the configuration that gives the best model performance.

[0010] Furthermore, the constructed deep neural network model is trained and tested until the training end condition is reached, and the method also includes: The trained deep neural network model is tested using a test set. If the deviation between the predicted data obtained by the test and the label data is lower than the deviation threshold, the trained deep neural network model is copied into multiple copies, and each of the copied deep neural network models is published to multiple collaborative nodes via the blockchain network; Each of the collaborative nodes collects and uses the training data to perform fine-tuning training on the received deep neural network model, and feeds back the second key model parameters of the deep neural network model obtained through the fine-tuning training to the local node, and the local node fuses each second key model parameter with the first key model parameter of the deep neural network model in the local node to obtain the target key model parameter; The target key model parameters are imported into the deep neural network model in the local node, and the deep neural network model in the local node is trained a second time until the training end condition is met.

[0011] Furthermore, the first number of the collaborative nodes is determined in the following manner: Determine the total number of the collected first stool samples and the second stool samples, i.e., the second number, and the cluster number of the disease severity levels of Parkinson's disease patients corresponding to the first stool samples; Substituting the second number and the cluster number into the following calculation formula, the first number of the collaborative nodes is calculated: ;

[0012] In the formula, is the first number of the collaborative nodes; is the minimum number of the collaborative nodes, which is a fixed value determined based on experience; is the total number of the first stool samples and the second stool samples, i.e., the second number; is the cluster number of the disease severity level of the Parkinson's disease patients corresponding to each of the first stool samples; A and B are dimensionless adjustment parameters, both of which are constant values; To round up.

[0013] Furthermore, each of the copied deep neural network models is published to multiple collaborative nodes via the blockchain network, including: Sending a collaborative training task to each of the collaborative nodes, wherein the collaborative training task includes a copy of the deep neural network model and an amount of fine-tuning training data; The amount of fine-tuning training data is determined in the following manner: Determine a total amount of fine-tuning training data according to the first number of the collaborative nodes and the empirical amount of fine-tuning training data, and randomly divide the total amount of fine-tuning training data based on a constraint condition to obtain each amount of the fine-tuning training data corresponding to the first number; wherein the constraint condition is used to ensure that each amount of the fine-tuning training data is greater than a data amount threshold; The amount of each fine-tuning training data is randomly allocated to each of the collaborative nodes.

[0014] The present invention also provides a Parkinson's disease prediction system based on intestinal flora data and deep learning, the system comprising a processing module and a storage module; the processing module is electrically connected to the storage module; the processing module calls the executable computer program code stored in the storage module to perform the following steps: Collecting first stool samples from several Parkinson's disease patients and second stool samples from healthy controls, performing flora measurement on each of the first stool samples and the second stool samples, and performing data standardization and flora abundance calculation on the flora measurement data to obtain a first flora abundance matrix and a second flora abundance matrix, respectively, and adding patient labels and health labels to the first flora abundance matrix and the second flora abundance matrix, respectively; Dividing the labeled first flora abundance matrix and the second flora abundance matrix into a training set and a test set, and training and testing the constructed deep neural network model until a training end condition is met; Collecting a third stool sample from the target object, performing a flora measurement on the third stool sample, performing data standardization and flora abundance calculation on the flora measurement data, obtaining a third flora abundance matrix, and analyzing and processing the third flora abundance matrix using the trained deep neural network model; The deep neural network model outputs a Parkinson's disease risk score, and outputs the Parkinson's disease risk score.

[0015] The present invention also provides an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the method as described in any of the preceding items.

[0016] The present invention also provides a computer storage medium, on which a computer program is stored. When the computer program is executed by a processor, any of the above methods is executed.

[0017] The present invention also provides a computer program product, which is used to be called and executed by a processor of an electronic device to implement any of the methods described above.

[0018] The beneficial effects of the present invention are: The solution of the present invention combines intestinal flora data and deep learning technology to achieve accurate prediction of the risk of Parkinson's disease, and has the advantages of non-invasiveness, low cost, and easy operation. It also realizes automated analysis of the entire process, which not only improves the analysis efficiency, but also reduces the errors caused by human operation, ensuring the stability and consistency of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 It is a flow chart of a method for predicting Parkinson's disease based on intestinal flora data and deep learning disclosed in an embodiment of the present invention.

[0021] Figure 2 It is a schematic diagram of the structure of the deep neural network model disclosed in an embodiment of the present invention.

[0022] Figure 3 It is a structural schematic diagram of a Parkinson's disease prediction system based on intestinal flora data and deep learning disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0024] See also Figure 1 As shown in the flowchart, an embodiment of the present invention provides a method for predicting Parkinson's disease based on intestinal flora data and deep learning, comprising the following steps: Collecting first stool samples from several Parkinson's disease patients and second stool samples from healthy controls, performing flora measurement on each of the first stool samples and the second stool samples, and performing data standardization and flora abundance calculation on the flora measurement data to obtain a first flora abundance matrix and a second flora abundance matrix, respectively, and adding patient labels and health labels to the first flora abundance matrix and the second flora abundance matrix, respectively; Dividing the labeled first flora abundance matrix and the second flora abundance matrix into a training set and a test set, and training and testing the constructed deep neural network model until a training end condition is met; Collecting a third stool sample from the target object, performing a flora measurement on the third stool sample, performing data standardization and flora abundance calculation on the flora measurement data, obtaining a third flora abundance matrix, and analyzing and processing the third flora abundance matrix using the trained deep neural network model; The deep neural network model outputs a Parkinson's disease risk score, and outputs the Parkinson's disease risk score.

[0025] The present invention constructs a deep neural network model, and uses the deep neural network model to perform in-depth analysis on the stool sample data of the target patient, thereby obtaining the corresponding Parkinson's disease risk score to assist doctors in the diagnosis and treatment of Parkinson's disease. Specifically: In the model training stage: first collect the first stool samples of several Parkinson's patients and the second stool samples of healthy controls, perform flora measurement on each first stool sample and each second stool sample, perform data standardization and flora abundance calculation on the flora measurement data, thereby obtaining the first flora abundance matrix and the second flora abundance matrix respectively, and then add patient labels and health labels to the first flora abundance matrix and the second flora abundance matrix respectively. Then, use the labeled first flora abundance matrix and the second flora abundance matrix as training data, and the training data is divided into a training set and a test set, for example, 80% as a training set and 20% as a test set, use the training set to train the constructed deep neural network model, and use the test set to test the deep neural network model until the training end condition is reached, so that a deep neural network model with the ability to predict and evaluate the risk of Parkinson's disease is obtained.

[0026] In the application stage: a third stool sample of the target object is collected, and the third flora abundance matrix is ​​obtained by the same sampling method. The third flora abundance matrix is ​​analyzed and processed using the trained deep neural network model mentioned above, that is, the Parkinson's disease risk score of the target object predicted by the model is obtained, and the Parkinson's disease risk score is output to relevant personnel including doctors to assist doctors in conducting relevant diagnosis and treatment of the target object.

[0027] The solution of the present invention combines intestinal flora data with deep learning technology to accurately predict the risk of Parkinson's disease, and has the advantages of non-invasiveness, low cost, and easy operation. It also realizes full-process automated analysis, which not only improves the analysis efficiency, but also reduces the errors caused by human operation, ensuring the stability and consistency of the results. The solution of the present invention is suitable for large-scale health screening and scientific research projects.

[0028] Furthermore, the collecting of first stool samples from a number of Parkinson's disease patients and second stool samples from healthy controls, and performing bacterial flora determination on each of the first stool samples and the second stool samples, comprises: First stool samples from several Parkinson's disease patients and second stool samples from healthy controls were collected, and the stool samples were stored in sterile sampling tubes and immediately frozen; The 16S rRNA high-throughput sequencing technology is used to measure the flora composition of each of the first stool sample and the second stool sample to obtain corresponding flora measurement data.

[0029] In this example, stool samples can be collected in advance from, for example, 80 Parkinson's disease patients and 100 healthy controls, and the stool samples can be stored in sterile sampling tubes and immediately frozen to maximize the integrity and activity of the flora. Subsequently, the flora composition is determined using 16S rRNA high-throughput sequencing technology, that is, multiple groups of corresponding flora measurement data are obtained, and then each group of flora measurement data is standardized and the flora abundance is calculated, that is, the types of microorganisms in each stool sample and their relative abundance, that is, the first flora abundance matrix and the second flora abundance matrix are obtained.

[0030] Furthermore, the deep neural network model includes an input layer, a convolutional layer, an activation function, a pooling layer, and a fully connected layer; And, the deep neural network model adopts the following cross entropy loss function during training: ;

[0031] in, is the cross entropy loss function value, is the true label, is the Parkinson's disease risk score predicted by the model, and N is the total number of training samples; Use the Adam optimizer to update the network weights. The update rule of the Adam optimizer is: ; ; ; ;

[0032] in, and are the first-order and second-order momentum estimates of the gradient, and is the momentum parameter, is the learning rate, is a small constant to avoid zero denominator; Dropout regularization is used to randomly discard a certain proportion of neurons in the fully connected layer to prevent the model from overfitting; Use grid search to optimize hyperparameters and select the configuration that gives the best model performance.

[0033] In this example, if Figure 2 As shown, the deep neural network model in the present invention includes an input layer, a convolutional layer, an activation function, a pooling layer, and a fully connected layer; the functional configuration of each layer is as follows: Input layer: The standardized bacterial abundance matrix (dimension: 100 x N, where N is the number of screened microbial species) is input into the model.

[0034] Convolutional Layer: Use multiple convolution kernels (filters) of size kxk to extract features. The convolution operation is defined as follows:

[0035] Activation function (ReLU, Rectified Linear Unit): Apply the ReLU activation function to the convolution output, defined as , the ReLU function can effectively accelerate model training and reduce the gradient disappearance problem.

[0036] Pooling Layer: Use Max Pooling to reduce feature dimensions and model complexity. The pooling operation formula is:

[0037] Here, p is the size of the pooling window.

[0038] Fully Connected Layer: The pooled features are expanded into a one-dimensional vector, input into the fully connected layer, and the Softmax classifier is used to output the Parkinson's disease risk probability.

[0039] And, during the training process of the deep neural network model, the following configuration is used: The loss function uses the cross-entropy loss function (Cross-Entropy Loss): ;

[0040] in, is the cross entropy loss function value, is the true label, is the model prediction probability, N is the total number of training samples; Use the Adam optimizer to update the network weights. The update rule of the Adam optimizer is: ; ; ; ;

[0041] in, and are the first-order momentum and second-order momentum estimates of the gradient at the t-th time step, , are the first-order momentum and second-order momentum estimates for the t-1th time step, , are the deviations of the first-order momentum and the second-order momentum at the t-th time step, and is the momentum parameter, , They are the gradient updates of the t-th time step and the t-1-th time step, respectively. is the learning rate, is a small constant to avoid zero denominator; Dropout regularization is used to randomly discard a certain proportion of neurons in the fully connected layer to prevent the model from overfitting; Use grid search to optimize hyperparameters (learning rate, kernel size, etc.) and choose the configuration that gives the best model performance.

[0042] Furthermore, the constructed deep neural network model is trained and tested until the training end condition is reached, and the method also includes: The trained deep neural network model is tested using a test set. If the deviation between the predicted data obtained by the test and the label data is lower than the deviation threshold, the trained deep neural network model is copied into multiple copies, and each of the copied deep neural network models is published to multiple collaborative nodes via the blockchain network; Each of the collaborative nodes collects and uses the training data to perform fine-tuning training on the received deep neural network model, and feeds back the second key model parameters of the deep neural network model obtained through the fine-tuning training to the local node, and the local node fuses each second key model parameter with the first key model parameter of the deep neural network model in the local node to obtain the target key model parameter; The target key model parameters are imported into the deep neural network model in the local node, and the deep neural network model in the local node is trained a second time until the training end condition is met.

[0043] In this example, the present invention uses a combination of local node training and collaborative node distributed training to train the constructed deep neural network model, which can overcome the defect of insufficient training data. Specifically: First, the local node uses the test set to test the trained deep neural network model. If the deviation between the predicted data and the label data obtained in the test is lower than the deviation threshold, it means that the local training has initially reached the training end conditions.

[0044] Then, the trained deep neural network model is copied multiple times, and each copy of the deep neural network model is published to multiple collaborative nodes via the blockchain network. After each collaborative node receives a copy of the deep neural network model, it collects training data on its own and performs secondary training on the copy of the deep neural network model. After the training meets the standard, the second key model parameters of the deep neural network model are fed back to the local node. Among them, the secondary training of the copy of the deep neural network model by the collaborative node is different from the training of the local node, which is mainly reflected in that the collaborative node uses small sample training data for fine-tuning training, while the local node uses large sample training data for full training. The combination of full training and fine-tuning training is conducive to improving the analytical ability of the deep neural network model while improving the training efficiency.

[0045] Finally, the local node fuses the received second key model parameters with the first key model parameters in the deep neural network model after sufficient training of the local node, such as taking the average value, to obtain the target key model parameters. The target key model parameters are imported into the deep neural network model that has completed local training, and then the deep neural network model in the local node is trained again until the training end condition is met. The secondary training here can use another set of different training sets and test sets. Among them, the first key model parameters and the second key model parameters include but are not limited to the number of network structure layers (number of layers, convolution kernel size, number of convolution kernels), learning rate, momentum, etc.

[0046] Furthermore, the first number of the collaborative nodes is determined in the following manner: Determine the total number of the collected first stool samples and the second stool samples, i.e., the second number, and the cluster number of the disease severity levels of Parkinson's disease patients corresponding to the first stool samples; Substituting the second number and the cluster number into the following calculation formula, the first number of the collaborative nodes is calculated:

[0047] In the formula, is the first number of the collaborative nodes; is the minimum number of the collaborative nodes, which is a fixed value determined based on experience; is the total number of the first stool samples and the second stool samples, i.e., the second number; is the cluster number of the disease severity level of the Parkinson's disease patients corresponding to each of the first stool samples; A and B are dimensionless adjustment parameters, both of which are constant values; To round up.

[0048] In this example, the number of collaborative nodes in the present invention is determined according to the training data used when the local node is trained, specifically including: the total number of the collected first stool samples and second stool samples, i.e., the second number, which is the total data volume of the training set and the test set; the clustering number of the disease severity levels of Parkinson's disease patients corresponding to each first stool sample, each Parkinson's disease patient corresponding to the first stool sample is associated with a corresponding disease severity level, and the disease severity levels are divided into levels 1-5 according to the severity. The disease severity levels of all Parkinson's disease patients are clustered, so as to obtain the number of disease severity levels involved in the collected sample data, i.e., the total number of first stool samples of Parkinson's disease patients of how many levels are involved.

[0049] Among them, when the total number of samples of training data, i.e., the second number, is larger, the local node has more fully trained the deep neural network model, and the higher the probability that the deep neural network model after preliminary training is truly "up to standard", fewer collaborative nodes can be set to participate in collaborative training at this time, and vice versa, more collaborative nodes can be set to participate in collaborative training. The more clusters of disease severity levels of Parkinson's disease patients corresponding to each first fecal sample, the higher the sample richness of the training data, and the more comprehensive the local node training of the deep neural network model. Therefore, the higher the probability that the deep neural network model that has met the preliminary training standards is truly "up to standard", fewer collaborative nodes can be set to participate in collaborative training at this time, and vice versa, more collaborative nodes can be set to participate in collaborative training.

[0050] Under the above principle, the present invention substitutes the obtained second number and cluster number into the above calculation formula, that is, the first number of collaborative nodes required for this collaborative training is obtained. Then, the first number of collaborative nodes can be screened in the blockchain network.

[0051] Furthermore, each of the copied deep neural network models is published to multiple collaborative nodes via the blockchain network, including: Sending a collaborative training task to each of the collaborative nodes, wherein the collaborative training task includes a copy of the deep neural network model and an amount of fine-tuning training data; The amount of fine-tuning training data is determined in the following manner: Determine a total amount of fine-tuning training data according to the first number of the collaborative nodes and the empirical amount of fine-tuning training data, and randomly divide the total amount of fine-tuning training data based on a constraint condition to obtain each amount of the fine-tuning training data corresponding to the first number; wherein the constraint condition is used to ensure that each amount of the fine-tuning training data is greater than a data amount threshold; The amount of each fine-tuning training data is randomly allocated to each of the collaborative nodes.

[0052] In this example, the collaborative training task issued by the local node to the collaborative node through the blockchain network includes a copy of the deep neural network model and the amount of fine-tuning training data. The amount of fine-tuning training data is used to indicate how much training data the corresponding collaborative node should use to train the copy of the deep neural network model.

[0053] The method for determining the amount of fine-tuning training data is as follows: the total amount of fine-tuning training data is determined based on the first number of collaborative nodes determined above and the amount of experience in fine-tuning training data, and the total amount of fine-tuning training data = the amount of experience in fine-tuning training data * the first number * the adjustment coefficient, and the adjustment coefficient is greater than 1. The determined total amount of fine-tuning training data is randomly divided to obtain each amount of fine-tuning training data corresponding to the first number, and each amount of fine-tuning training data is randomly allocated to each selected collaborative node. In order to ensure that each amount of fine-tuning training data is greater than the data amount threshold, so as to ensure that the degree of fine-tuning training is also sufficient, it is also necessary to configure constraints to limit the random division method of the total amount of fine-tuning training data.

[0054] like Figure 3 As shown, an embodiment of the present invention further discloses a Parkinson's disease prediction system based on intestinal flora data and deep learning, the system comprising a processing module and a storage module; the processing module is electrically connected to the storage module; the processing module calls the executable computer program code stored in the storage module to perform the following steps: Collecting first stool samples from several Parkinson's disease patients and second stool samples from healthy controls, performing flora measurement on each of the first stool samples and the second stool samples, and performing data standardization and flora abundance calculation on the flora measurement data to obtain a first flora abundance matrix and a second flora abundance matrix, respectively, and adding patient labels and health labels to the first flora abundance matrix and the second flora abundance matrix, respectively; Dividing the labeled first flora abundance matrix and the second flora abundance matrix into a training set and a test set, and training and testing the constructed deep neural network model until a training end condition is met; Collecting a third stool sample from the target object, performing a flora measurement on the third stool sample, performing data standardization and flora abundance calculation on the flora measurement data, obtaining a third flora abundance matrix, and analyzing and processing the third flora abundance matrix using the trained deep neural network model; The deep neural network model outputs a Parkinson's disease risk score, and outputs the Parkinson's disease risk score.

[0055] An embodiment of the present invention further discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the method described in the above embodiment.

[0056] The embodiment of the present invention further discloses a computer storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in the above embodiment is executed.

[0057] An embodiment of the present invention further discloses a computer program product, which is used to be called and executed by a processor of an electronic device to implement any of the above methods.

[0058] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0059] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable load balancing device, so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0060] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0061] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0062] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A Parkinson's disease prediction method based on intestinal flora data and deep learning, characterized in that: The steps include: Collecting first stool samples from several Parkinson's disease patients and second stool samples from healthy controls, performing flora measurement on each of the first stool samples and the second stool samples, and performing data standardization and flora abundance calculation on the flora measurement data to obtain a first flora abundance matrix and a second flora abundance matrix, respectively, and adding patient labels and health labels to the first flora abundance matrix and the second flora abundance matrix, respectively; Dividing the labeled first flora abundance matrix and the second flora abundance matrix into a training set and a test set, and training and testing the constructed deep neural network model until a training end condition is met; Collecting a third stool sample from the target object, performing a flora measurement on the third stool sample, performing data standardization and flora abundance calculation on the flora measurement data, obtaining a third flora abundance matrix, and analyzing and processing the third flora abundance matrix using the trained deep neural network model; The deep neural network model outputs a Parkinson's disease risk score, and outputs the Parkinson's disease risk score.

2. The method for predicting Parkinson's disease based on intestinal flora data and deep learning according to claim 1, characterized in that: Collecting first stool samples from a number of Parkinson's disease patients and second stool samples from healthy controls, and performing bacterial flora determination on each of the first stool samples and the second stool samples, including: First stool samples from several Parkinson's disease patients and second stool samples from healthy controls were collected, and the stool samples were stored in sterile sampling tubes and immediately frozen; The 16S rRNA high-throughput sequencing technology is used to measure the flora composition of each of the first stool sample and the second stool sample to obtain corresponding flora measurement data.

3. The method for predicting Parkinson's disease based on intestinal flora data and deep learning according to claim 2, characterized in that: The deep neural network model includes an input layer, a convolution layer, an activation function, a pooling layer, and a fully connected layer; And, the deep neural network model adopts the following cross entropy loss function during training: ； in, is the cross entropy loss function value, is the true label, is the Parkinson's disease risk score predicted by the model, and N is the total number of training samples; Use the Adam optimizer to update the network weights. The update rule of the Adam optimizer is: ； ； ； ； in, and are the first-order momentum and second-order momentum estimates of the gradient at the t-th time step, , are the first-order momentum and second-order momentum estimates for the t-1th time step, , are the deviations of the first-order momentum and the second-order momentum at the t-th time step, and is the momentum parameter, , They are the gradient updates at the t-th time step and the t-1-th time step, respectively. is the learning rate, is a small constant to avoid zero denominator; Dropout regularization is used to randomly discard a certain proportion of neurons in the fully connected layer to prevent the model from overfitting; Use grid search to optimize hyperparameters and select the configuration that gives the best model performance.

4. The method for predicting Parkinson's disease based on intestinal flora data and deep learning according to claim 3, characterized in that: The constructed deep neural network model is trained and tested until the training end condition is reached, and the method further includes: The trained deep neural network model is tested using a test set. If the deviation between the predicted data obtained by the test and the label data is lower than the deviation threshold, the trained deep neural network model is copied into multiple copies, and each of the copied deep neural network models is published to multiple collaborative nodes via the blockchain network; Each of the collaborative nodes collects and uses the training data to perform fine-tuning training on the received deep neural network model, and feeds back the second key model parameters of the deep neural network model obtained through the fine-tuning training to the local node, and the local node fuses each second key model parameter with the first key model parameter of the deep neural network model in the local node to obtain the target key model parameter; The target key model parameters are imported into the deep neural network model in the local node, and the deep neural network model in the local node is trained a second time until the training end condition is met.

5. The method for predicting Parkinson's disease based on intestinal flora data and deep learning according to claim 4, characterized in that: The first number of the collaborative nodes is determined by: Determine the total number of the collected first stool samples and the second stool samples, i.e., the second number, and the cluster number of the disease severity levels of Parkinson's disease patients corresponding to the first stool samples; Substituting the second number and the cluster number into the following calculation formula, the first number of the collaborative nodes is calculated: ； In the formula, is the first number of the collaborative nodes; is the minimum number of the collaborative nodes, which is a fixed value determined based on experience; is the total number of the first stool samples and the second stool samples, i.e., the second number; The cluster number of the disease severity level of the Parkinson's disease patients corresponding to each of the first stool samples; A and B are dimensionless adjustment parameters, both of which are constant values; To round up.

6. The method for predicting Parkinson's disease based on intestinal flora data and deep learning according to claim 5, characterized in that: Publishing the copied deep neural network models to multiple collaborative nodes via the blockchain network, including: Sending a collaborative training task to each of the collaborative nodes, wherein the collaborative training task includes a copy of the deep neural network model and an amount of fine-tuning training data; The amount of fine-tuning training data is determined in the following manner: Determine a total amount of fine-tuning training data according to the first number of the collaborative nodes and the empirical amount of fine-tuning training data, and randomly divide the total amount of fine-tuning training data based on a constraint condition to obtain each amount of the fine-tuning training data corresponding to the first number; wherein the constraint condition is used to ensure that each amount of the fine-tuning training data is greater than a data amount threshold; The amount of each fine-tuning training data is randomly allocated to each of the collaborative nodes.

7. A Parkinson's disease prediction system based on intestinal flora data and deep learning, the system comprising a processing module and a storage module; the processing module is electrically connected to the storage module; characterized in that: The processing module calls the executable computer program code stored in the storage module to perform the following steps: Collecting first stool samples from several Parkinson's disease patients and second stool samples from healthy controls, performing flora measurement on each of the first stool samples and the second stool samples, and performing data standardization and flora abundance calculation on the flora measurement data to obtain a first flora abundance matrix and a second flora abundance matrix, respectively, and adding patient labels and health labels to the first flora abundance matrix and the second flora abundance matrix, respectively; Dividing the labeled first flora abundance matrix and the second flora abundance matrix into a training set and a test set, and training and testing the constructed deep neural network model until a training end condition is met; Collecting a third stool sample from the target object, performing a flora measurement on the third stool sample, performing data standardization and flora abundance calculation on the flora measurement data, obtaining a third flora abundance matrix, and analyzing and processing the third flora abundance matrix using the trained deep neural network model; The deep neural network model outputs a Parkinson's disease risk score, and outputs the Parkinson's disease risk score.

8. An electronic device, comprising: A memory storing executable program code; A processor coupled to the memory; characterized in that: the processor calls the executable program code stored in the memory to execute the method according to any one of claims 1-6.

9. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is executed.

10. A computer program product, characterized in that: The computer program product is used to be called and executed by a processor of an electronic device to implement the method according to any one of claims 1 to 6.

Images