Target tumor DNA fragment screening method based on deep learning

Abstract

The present invention relates to a method for screening target tumor DNA fragments based on deep learning, and belongs to the technical field of screening target tumor DNA fragments. The present invention constructs a tumor DNA fragment screening model through a deep neural network, and the training set is used to establish the model, and the test set is used to test the model performance. The DNA sequencing data of the positive tumor sample to be identified is obtained, and the DNA sequencing data of the positive tumor sample to be identified is located and analyzed according to the trained tumor DNA fragment screening model, the location of the target tumor DNA fragment is obtained, and it is displayed in a preset manner. The present invention can minimize the interference of multiple different types of tumor DNA fragments in the same tumor training sample data on the model training, thereby improving the prediction accuracy of the model and improving the screening accuracy of tumor DNA fragments.

Description

Technical Field

[0001] The present invention relates to the technical field of DNA detection and screening, and in particular to a method for screening target tumor DNA fragments based on deep learning. Background Art

[0002] Sequencing technology is also known as massive parallel sequencing. Its core idea is to sequence while synthesizing. It can sequence millions or even billions of DNA molecules at the same time, achieving the goal of large-scale and high-throughput sequencing. It is a revolutionary progress after Sanger sequencing. In recent years, with the rapid development of second-generation sequencing technology, it has gradually been applied to clinical medical detection and scientific research in the direction of blood tumors. Target sequence capture can selectively separate or enrich specific fragments of the genome, so it can use lower costs to obtain higher sequencing depth, laying a good foundation for low-frequency detection and big data accumulation. At present, in clinical tumor practice, second-generation sequencing is mainly used for driver gene sequencing, which is an important part of detection. Deep learning technology is a multi-disciplinary cross-disciplinary subject involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It mainly studies how to enable computers to simulate or realize human learning behavior, thereby acquiring new knowledge or skills, and reorganizing existing knowledge structures to continuously improve their own performance. Screening massive amounts of tumor DNA fragments through deep learning technology can greatly improve the speed of DNA fragment detection. However, in reality, in existing technologies, there may be more than one type of tumor DNA fragments, and since mutations are non-directional, and a free DNA fragment or ctDNA (free DNA from tumors) fragment may contain multiple different types of tumor DNA fragments, this problem is not considered in existing technologies. This will cause interference from multiple different types of tumor DNA fragments during model training, greatly reducing the prediction accuracy of the model. Summary of the invention

[0003] The present invention overcomes the shortcomings of the prior art and provides a method for screening target tumor DNA fragments based on deep learning.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] The first aspect of the present invention provides a method for screening target tumor DNA fragments based on deep learning, which specifically includes:

[0006] Obtaining a preset number of positive tumor sample DNA sequencing data and control sample free DNA sequencing data, and screening out specific features of ctDNA fragments based on the positive tumor sample DNA sequencing data and control sample free DNA sequencing data;

[0007] Constructing a training sample data set and a test sample data set according to the preset number of positive tumor sample DNA sequencing data and the specific features of the ctDNA fragments, and obtaining a processed training data set and a test data set by performing feature processing on the training sample data set and the test sample data set;

[0008] Build a tumor DNA fragment screening model based on a deep neural network, use the neural network algorithm to build the model with the processed training set, and use the test set to test the model performance to obtain the model with the best performance;

[0009] The DNA sequencing data of the positive tumor sample to be identified is obtained, and the DNA sequencing data of the positive tumor sample to be identified is positioned and analyzed according to the trained tumor DNA fragment screening model to obtain the location of the target tumor DNA fragment and display it in a preset manner.

[0010] Furthermore, in this method, a preset number of positive tumor sample DNA sequencing data and control sample free DNA sequencing data are obtained, specifically:

[0011] Obtaining index data of initial sample data of positive tumors, and setting a sample evaluation index range, to determine whether the index data of the initial sample data of positive tumors is within the sample evaluation index range;

[0012] If the index data of the initial sample data of the positive tumor is within the sample evaluation index range, extracting the positive tumor sample DNA sequencing data from the corresponding initial sample data of the positive tumor by using gene detection technology;

[0013] A preset number of positive tumor sample DNA sequencing data is counted and acquired, and free DNA sequencing data of control samples is collected, and the preset number of positive tumor sample DNA sequencing data and free DNA sequencing data of control samples are output.

[0014] Furthermore, in this method, the specific characteristics of ctDNA fragments are screened out according to the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample, specifically including:

[0015] Introducing a graph traversal algorithm, and performing a first traversal on the DNA sequencing data of the positive tumor sample by using the graph traversal algorithm;

[0016] Performing a second traversal of the cell-free DNA sequencing data of the control sample using the traversal algorithm of the graph;

[0017] By traversing, the identical parts and the non-identical parts between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample are obtained, and the DNA sequencing data of the non-identical parts between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample are extracted;

[0018] The DNA sequencing data of the non-identical part between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample is used as the specific feature of the ctDNA fragment, and the specific feature of the ctDNA fragment is output.

[0019] Furthermore, in the present method, a training sample data set and a test sample data set are constructed according to the preset number of positive tumor sample DNA sequencing data and the specific features of the ctDNA fragments, and the processed training sample data set and test sample data set are obtained by feature processing the training sample data set and the test sample data set, specifically including:

[0020] Selecting a plurality of training sample data from the preset number of positive tumor sample DNA sequencing data, and determining whether each of the training sample data contains only one type of ctDNA fragment-specific feature data;

[0021] If each of the training sample data contains only one type of ctDNA fragment-specific feature data, the corresponding training sample data is used as a training sample data set and a test sample data set, and the training sample data set and the test sample data set are output;

[0022] If each of the training sample data contains more than one specific feature data of the ctDNA fragment, the type of the specific feature data of the ctDNA fragment will be obtained, and the type of the specific feature data of each ctDNA fragment will be used as an independent ctDNA fragment type;

[0023] Taking the independent ctDNA fragment type as a benchmark, traversing the preset number of positive tumor sample DNA sequencing data to find sample data with only one independent ctDNA fragment type;

[0024] A preset number of sample data having only one independent ctDNA fragment type are selected to construct a training sample data set and a test sample data set, and the training sample data set and the test sample data set are output.

[0025] Furthermore, in this method, a tumor DNA fragment screening model is constructed based on a deep neural network, and the processed training data set and test data set are input into the tumor DNA fragment screening model for training and testing to obtain a trained tumor DNA fragment screening model, specifically:

[0026] A tumor DNA fragment screening model was constructed based on a deep neural network. The learning rate, mean square error of the stopping condition, and the maximum number of training neurons were set. The sine function was used as the activation function and training parameters. Through the attention module, the attention was focused on the specific feature data of the ctDNA fragments.

[0027] Keeping the maximum number of input neurons of the deep neural network consistent with the dimension of the learning sample, passing the processed training data set to the units in the hidden layer, and performing weighted summation on all neurons in the last layer of the hidden layer;

[0028] Dividing the sum of the outputs of the hidden layer by the output neuron to obtain a probability value of the output state, sorting the probability values ​​of the output state to obtain a sorting result, and obtaining a dynamic data state with a maximum probability value of the output state through the sorting result;

[0029] The dynamic data state with the maximum probability value of the output state is taken as the output result. After continuous training and testing through the test data set, when the training parameters reach the mean square error of the stopping condition, the trained tumor DNA fragment screening model is output.

[0030] Furthermore, in the present method, the DNA sequencing data of the positive tumor sample to be identified is obtained, and the DNA sequencing data of the positive tumor sample to be identified is positioned and analyzed according to the trained tumor DNA fragment screening model, the location of the target tumor DNA fragment is obtained, and it is displayed in a preset manner, specifically including:

[0031] Acquire DNA sequencing data of a plurality of positive tumor samples to be identified, and input the DNA sequencing data of the positive tumor samples to be identified into the trained tumor DNA fragment screening model for analysis;

[0032] Through analysis, DNA sequencing data of positive tumor samples containing target tumor DNA fragments are obtained, and the target tumor DNA fragments are located in the DNA sequencing data of positive tumor samples containing target tumor DNA fragments, the location of the target tumor DNA fragments is obtained, and displayed in a preset manner.

[0033] The present invention solves the defects existing in the background technology and has the following beneficial effects:

[0034] The present invention obtains a preset number of positive tumor sample DNA sequencing data and free DNA sequencing data of control samples, and screens out specific features of ctDNA fragments according to the positive tumor sample DNA sequencing data and the free DNA sequencing data of control samples, and then constructs a training sample data set and a test sample data set according to the preset number of positive tumor sample DNA sequencing data and the specific features of ctDNA fragments, and performs feature processing on the training sample data set and the test sample data set to obtain the processed training data set and the test data set, thereby constructing a tumor DNA fragment screening model based on a deep neural network, inputting the processed training data set and the test data set into the tumor DNA fragment screening model for training and testing, obtaining the trained tumor DNA fragment screening model, and finally obtaining the positive tumor sample DNA sequencing data to be identified, and positioning and analyzing the positive tumor sample DNA sequencing data to be identified according to the trained tumor DNA fragment screening model, obtaining the location of the target tumor DNA fragment, and displaying it in a preset manner. The present invention can minimize the interference of multiple different types of tumor DNA fragments in the same tumor training sample data on the model training by processing the training sample data in which there are multiple different types of tumor DNA fragments, thereby improving the prediction accuracy of the model and improving the screening accuracy of tumor DNA fragments. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, drawings of other embodiments can be obtained based on these drawings without paying creative work.

[0036] Figure 1 The overall flow chart of the method for screening target tumor DNA fragments based on deep learning is shown;

[0037] Figure 2 A partial flowchart of the target tumor DNA fragment screening method based on deep learning is shown. DETAILED DESCRIPTION

[0038] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0039] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0040] like Figure 1 As shown, the first aspect of the present invention provides a method for screening target tumor DNA fragments based on deep learning, specifically comprising:

[0041] S102: Obtain a preset number of positive tumor sample DNA sequencing data and control sample free DNA sequencing data, and screen out specific features of ctDNA fragments based on the positive tumor sample DNA sequencing data and the control sample free DNA sequencing data;

[0042] S104: constructing a training sample data set and a test sample data set according to a preset number of positive tumor sample DNA sequencing data and specific features of ctDNA fragments, and performing feature processing on the training sample data set and the test sample data set to obtain a processed training data set and a test data set;

[0043] S106: constructing a tumor DNA fragment screening model based on a deep neural network, using the processed training set to build a model using a neural network algorithm, and using the test set to test the model performance to obtain the model with the best performance;

[0044] S108: Obtain the DNA sequencing data of the positive tumor sample to be identified, and perform positioning analysis on the DNA sequencing data of the positive tumor sample to be identified according to the trained tumor DNA fragment screening model, obtain the location of the target tumor DNA fragment, and display it in a preset manner.

[0045] It should be noted that the present invention can minimize the interference of multiple different types of tumor DNA fragments in the same tumor training sample data on model training by processing the training sample data in which there are multiple different types of tumor DNA fragments, thereby improving the prediction accuracy of the model and improving the screening accuracy of tumor DNA fragments.

[0046] Furthermore, in this method, a preset number of positive tumor sample DNA sequencing data and control sample free DNA sequencing data are obtained, specifically:

[0047] Obtaining index data of initial sample data of positive tumors, and setting a sample evaluation index range, to determine whether the index data of initial sample data of positive tumors is within the sample evaluation index range;

[0048] If the index data of the initial sample data of the positive tumor is within the sample evaluation index range, the DNA sequencing data of the positive tumor sample is extracted from the corresponding initial sample data of the positive tumor by using the gene detection technology;

[0049] Count and obtain a preset number of positive tumor sample DNA sequencing data, collect free DNA sequencing data of control samples, and output the preset number of positive tumor sample DNA sequencing data and free DNA sequencing data of control samples.

[0050] It should be noted that the index data of the initial sample data of positive tumors include the pH value, osmotic pressure, sterility, and pollution-freeness of the test samples. This method can eliminate abnormal indicators and leave normal test data.

[0051] Furthermore, in this method, the specific characteristics of ctDNA fragments are screened out based on the DNA sequencing data of positive tumor samples and the free DNA sequencing data of control samples, specifically including:

[0052] Introducing a graph traversal algorithm, and performing a first traversal on the DNA sequencing data of the positive tumor sample by using the graph traversal algorithm;

[0053] Performing a second traversal of the cell-free DNA sequencing data of the control sample using the traversal algorithm of the graph;

[0054] By traversing, the identical parts and the non-identical parts between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample are obtained, and the DNA sequencing data of the non-identical parts between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample are extracted;

[0055] The DNA sequencing data of the non-identical part between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample is used as the specific feature of the ctDNA fragment, and the specific feature of the ctDNA fragment is output.

[0056] It should be noted that by traversing the DNA sequencing data of positive tumor samples and the free DNA sequencing data of control samples through a graph traversal algorithm, the DNA sequencing data of non-identical parts can be extracted, thereby extracting the specific characteristics of ctDNA fragments.

[0057] Furthermore, in this method, a training sample data set and a test sample data set are constructed according to a preset number of positive tumor sample DNA sequencing data and specific features of ctDNA fragments, and feature processing is performed on the training sample data set and the test sample data set to obtain the processed training data set and test data set, specifically including:

[0058] Selecting a number of training sample data from a preset number of positive tumor sample DNA sequencing data, and determining whether each training sample data contains only one type of ctDNA fragment-specific feature data;

[0059] If each training sample data has only one specific feature data of ctDNA fragment, the corresponding training sample data is used as a training sample data set and a test sample data set, and the training sample data set and the test sample data set are output;

[0060] If each training sample data contains more than one specific feature data of ctDNA fragment, the type of occurrence of the specific feature data of the ctDNA fragment will be obtained, and the type of occurrence of the specific feature data of each ctDNA fragment will be used as an independent ctDNA fragment type;

[0061] Taking the independent ctDNA fragment type as a benchmark, traverse the preset number of positive tumor sample DNA sequencing data to find sample data with only one independent ctDNA fragment type;

[0062] A preset number of sample data having only one independent ctDNA fragment type are selected to construct a training sample data set and a test sample data set, and then the training sample data set and the test sample data set are output.

[0063] It should be noted that, since there may be one or more ctDNA fragment types in a training sample data, by traversing a preset number of positive tumor sample DNA sequencing data, sample data with only one independent ctDNA fragment type is found, thereby forming a training sample data set corresponding to several independent ctDNA fragment types, and then the model is trained separately with several training sample data sets. This can avoid the interference of multiple ctDNA fragment types in model training, thereby improving the recognition and positioning accuracy of ctDNA fragment types.

[0064] like Figure 2 As shown, further, in this method, a tumor DNA fragment screening model is constructed based on a deep neural network, and the processed training data set and test data set are input into the tumor DNA fragment screening model for training and testing to obtain a trained tumor DNA fragment screening model, specifically:

[0065] S202: constructing a tumor DNA fragment screening model based on a deep neural network, setting a learning rate, a mean square error of a stop condition, a maximum number of training neurons, and using a sine function as an activation function and a training parameter, and focusing attention on the specific feature data of ctDNA fragments through the action of an attention module;

[0066] S204: Keep the maximum number of input neurons of the deep neural network consistent with the dimension of the learning sample, pass the processed training data set to the units in the hidden layer, and perform weighted summation on all neurons in the last layer of the hidden layer;

[0067] S206: Dividing the summed value of the hidden layer output by the output neuron to obtain the probability value of the output state, sorting the probability values ​​of the output state to obtain a sorting result, and obtaining the dynamic data state with the maximum probability value of the output state through the sorting result;

[0068] S208: The dynamic data state with the maximum probability value of the output state is taken as the output result, and after continuous training and testing through the test data set, when the training parameters reach the mean square error of the stopping condition, the trained tumor DNA fragment screening model is output.

[0069] It should be noted that this method can focus on the specific feature data of ctDNA fragments through the attention module and the interaction of each layer, thereby improving the prediction accuracy of the model.

[0070] Furthermore, in this method, DNA sequencing data of a positive tumor sample to be identified is obtained, and the DNA sequencing data of the positive tumor sample to be identified is positioned and analyzed according to the trained tumor DNA fragment screening model, the location of the target tumor DNA fragment is obtained, and displayed in a preset manner, specifically including:

[0071] Acquire DNA sequencing data of several positive tumor samples to be identified, and input the DNA sequencing data of the positive tumor samples to be identified into the trained tumor DNA fragment screening model for analysis;

[0072] Through analysis, DNA sequencing data of positive tumor samples containing target tumor DNA fragments are obtained, and the target tumor DNA fragments are located in the DNA sequencing data of positive tumor samples containing target tumor DNA fragments, the location of the target tumor DNA fragments is obtained, and displayed in a preset manner.

[0073] In addition, the method further comprises:

[0074] The maximum data processing capacity of the data processing system per unit time under various working temperature environments is obtained through big data, and a maximum data processing capacity prediction model of the processing system is constructed based on a deep neural network;

[0075] Introducing a graph neural network, inputting the maximum data processing amount of the data processing system within a unit time under each working temperature environment into the graph neural network, taking the working temperature environment as the first node and the maximum data processing amount as the second node;

[0076] Constructing a directed description relationship, connecting the first node with the second node based on the directed description relationship, constructing a topological structure graph, acquiring a related adjacency matrix based on the topological structure graph, and inputting the adjacency matrix into a maximum data processing volume prediction model of the processing system for training;

[0077] Through training, a maximum data processing capacity prediction model of a processing system that has been trained is obtained, and the operating temperature information of the data processing system within a preset time is obtained, and the operating temperature information of the data processing system within the preset time is input into the maximum data processing capacity prediction model of the processing system for prediction, so as to obtain the maximum data processing capacity of the data processing system within a unit time under the current operating temperature environment;

[0078] A preset number of training sample data is set according to the maximum data processing volume per unit time of the data processing system under the current working temperature environment, and the number of training sample data per unit time is dynamically adjusted according to the maximum data processing volume per unit time of the data processing system under the current working temperature environment.

[0079] It should be noted that, in fact, the maximum data processing capacity of the data processing system at different operating temperatures is different, and the appropriate training amount can make the data processing system in a normal working state. This method can set the correct data training amount, thereby improving the rationality of model training.

[0080] In addition, the amount of training sample data per unit time is dynamically adjusted according to the maximum data processing volume per unit time of the data processing system under the current working temperature environment, specifically:

[0081] Introducing a genetic algorithm, setting a genetic generation based on the genetic algorithm, and obtaining a maximum data processing amount per unit time of the data processing system under the current working temperature environment, and initializing the training quantity of training sample data;

[0082] Obtain the data processing volume occupied by each training quantity during training in unit time, and calculate the real-time total data processing volume based on the data processing volume occupied by each training quantity during training in unit time and the training quantity of training sample data;

[0083] Determine whether the real-time total data processing volume is greater than the maximum data processing volume of the data processing system within a unit time under the current working temperature environment; if not, output the training quantity of the training sample data, and perform training according to the training quantity of the training sample data;

[0084] If the real-time total data processing volume is greater than the maximum data processing volume of the data processing system within a unit time under the current operating temperature environment, inheritance is performed based on the genetic algebra, and the training quantity of the training sample data is replanned until it is no greater than the maximum data processing volume of the data processing system within a unit time under the current operating temperature environment.

[0085] It should be noted that, through this method, the training quantity of training sample data can be dynamically adjusted according to the change of the operating temperature of the data processing system, thereby improving the rationality of training during training.

[0086] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0087] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0088] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0089] Those skilled in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0090] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0091] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for screening target tumor DNA fragments based on deep learning, characterized in that: Specifically include: Obtaining a preset number of positive tumor sample DNA sequencing data and control sample free DNA sequencing data, and screening out specific features of ctDNA fragments based on the positive tumor sample DNA sequencing data and control sample free DNA sequencing data; Constructing a training sample data set and a test sample data set according to the preset number of positive tumor sample DNA sequencing data and the specific features of the ctDNA fragments, and obtaining a processed training data set and a test data set by performing feature processing on the training sample data set and the test sample data set; Build a tumor DNA fragment screening model based on a deep neural network, use the neural network algorithm to build the model with the processed training set, and use the test set to test the model performance to obtain the model with the best performance; Obtain the DNA sequencing data of the positive tumor sample to be identified, and perform positioning analysis on the DNA sequencing data of the positive tumor sample to be identified according to the trained tumor DNA fragment screening model, obtain the location of the target tumor DNA fragment, and display it in a preset manner; Constructing a training sample data set and a test sample data set according to the preset number of positive tumor sample DNA sequencing data and the specific features of the ctDNA fragments, and performing feature processing on the training sample data set and the test sample data set to obtain the processed training data set and test data set, specifically including: Selecting a plurality of training sample data from the preset number of positive tumor sample DNA sequencing data, and determining whether each of the training sample data contains only one type of ctDNA fragment-specific feature data; If each of the training sample data has only one specific feature data of a ctDNA fragment, the corresponding training sample data is used as a training sample data set and a test sample data set, and the training sample data set and the test sample data set are output; If each of the training sample data contains more than one specific feature data of the ctDNA fragment, the type of occurrence of the specific feature data of the ctDNA fragment is obtained, and the type of occurrence of the specific feature data of each ctDNA fragment is used as an independent ctDNA fragment type; Taking the independent ctDNA fragment type as a benchmark, traversing the preset number of positive tumor sample DNA sequencing data to find sample data with only one independent ctDNA fragment type; Selecting a preset number of sample data of only one independent ctDNA fragment type to construct a training sample data set and a test sample data set, and then outputting the training sample data set and the test sample data set; The maximum data processing capacity of the data processing system per unit time under various working temperature environments is obtained through big data, and a maximum data processing capacity prediction model of the processing system is constructed based on a deep neural network; Introducing a graph neural network, inputting the maximum data processing amount of the data processing system within a unit time under each working temperature environment into the graph neural network, taking the working temperature environment as the first node and the maximum data processing amount as the second node; Constructing a directed description relationship, connecting the first node with the second node based on the directed description relationship, constructing a topological structure graph, acquiring a related adjacency matrix based on the topological structure graph, and inputting the adjacency matrix into a maximum data processing volume prediction model of the processing system for training; Through training, a maximum data processing capacity prediction model of a processing system that has been trained is obtained, and the operating temperature information of the data processing system within a preset time is obtained, and the operating temperature information of the data processing system within the preset time is input into the maximum data processing capacity prediction model of the processing system for prediction, so as to obtain the maximum data processing capacity of the data processing system within a unit time under the current operating temperature environment; A preset number of training sample data is set according to the maximum data processing volume per unit time of the data processing system under the current working temperature environment, and the number of training sample data per unit time is dynamically adjusted according to the maximum data processing volume per unit time of the data processing system under the current working temperature environment.

2. The method for screening target tumor DNA fragments based on deep learning according to claim 1, characterized in that Obtain a preset number of positive tumor sample DNA sequencing data and control sample free DNA sequencing data, specifically: Obtaining index data of initial sample data of positive tumors, and setting a sample evaluation index range, to determine whether the index data of the initial sample data of positive tumors is within the sample evaluation index range; If the index data of the initial sample data of the positive tumor is within the sample evaluation index range, extracting the positive tumor sample DNA sequencing data from the corresponding initial sample data of the positive tumor by using gene detection technology; A preset number of positive tumor sample DNA sequencing data is counted and acquired, and free DNA sequencing data of control samples is collected, and the preset number of positive tumor sample DNA sequencing data and free DNA sequencing data of control samples are output.

3. The method for screening target tumor DNA fragments based on deep learning according to claim 1, characterized in that: The specific features of ctDNA fragments are screened out based on the DNA sequencing data of the positive tumor samples and the free DNA sequencing data of the control samples, specifically including: Introducing a graph traversal algorithm, and performing a first traversal on the DNA sequencing data of the positive tumor sample by using the graph traversal algorithm; Performing a second traversal of the cell-free DNA sequencing data of the control sample using the traversal algorithm of the graph; By traversing, the identical parts and the non-identical parts between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample are obtained, and the DNA sequencing data of the non-identical parts between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample are extracted; The DNA sequencing data of the non-identical part between the DNA sequencing data of the positive tumor sample and the free DNA sequencing data of the control sample is used as the specific feature of the ctDNA fragment, and the specific feature of the ctDNA fragment is output.

4. The method for screening target tumor DNA fragments based on deep learning according to claim 1, characterized in that: A tumor DNA fragment screening model was constructed based on a deep neural network. The processed training set was used to build the model using a neural network algorithm. The model performance was tested using a test set to obtain the optimal performance model. Specifically: A tumor DNA fragment screening model was constructed based on a deep neural network. The learning rate, mean square error of the stopping condition, and the maximum number of training neurons were set. The sine function was used as the activation function and training parameters. Through the attention module, the attention was focused on the specific feature data of the ctDNA fragments. Keeping the maximum number of input neurons of the deep neural network consistent with the dimension of the learning sample, passing the processed training data set to the units in the hidden layer, and performing weighted summation on all neurons in the last layer of the hidden layer; Dividing the sum of the hidden layer outputs by the output neuron to obtain the probability value of the output state, sorting the probability values ​​of the output state to obtain a sorting result, and obtaining the dynamic data state with the maximum probability value of the output state through the sorting result; The dynamic data state with the maximum probability value of the output state is taken as the output result. After continuous training and testing through the test data set, when the training parameters reach the mean square error of the stopping condition, the trained tumor DNA fragment screening model is output.

5. The method for screening target tumor DNA fragments based on deep learning according to claim 1, characterized in that: Obtain the DNA sequencing data of the positive tumor sample to be identified, and perform positioning analysis on the DNA sequencing data of the positive tumor sample to be identified according to the trained tumor DNA fragment screening model, obtain the location of the target tumor DNA fragment, and display it in a preset manner, including: Acquire DNA sequencing data of a plurality of positive tumor samples to be identified, and input the DNA sequencing data of the positive tumor samples to be identified into the trained tumor DNA fragment screening model for analysis; Through analysis, DNA sequencing data of positive tumor samples containing target tumor DNA fragments are obtained, and the target tumor DNA fragments are located in the DNA sequencing data of positive tumor samples containing target tumor DNA fragments, the location of the target tumor DNA fragments is obtained, and displayed in a preset manner.

Images