Cancer prognosis evaluation and target recognition method and system based on deep neural network

Abstract

This invention discloses a cancer prognostic assessment and target identification method based on deep neural networks. The method includes: acquiring gene expression data and preprocessing the gene expression data to obtain initial data; using a deep neural network combining a diffusion model and an autoencoder to perform feature extraction processing on the initial data to obtain a low-dimensional representation of gene expression features; inputting the low-dimensional representation of the gene expression features into an XGBoost survival analysis model to obtain a risk value, and obtaining a cancer prognostic assessment result based on the risk value; and performing differential expression analysis based on the risk value and patient gene expression to obtain cancer prognostic targets. Therefore, this invention, by combining a deep neural network with a diffusion model and an autoencoder, and an XGBoost survival analysis model, achieves accurate assessment of prognostic risk for cancer patients and identification of relevant targets, thus providing a scientific basis for the design of personalized treatment and targeted therapies.

Description

Technical Field

[0001] This invention relates to the field of cancer prognostic analysis technology, and in particular to a method and system for cancer prognostic assessment and target identification based on deep neural networks. Background Technology

[0002] Cancer is a complex and deadly malignant disease whose pathogenesis is influenced by the interaction of genomic characteristics and environmental factors. Despite progress in cancer research and treatment, prognostic outcomes remain significantly variable among patients due to the biological characteristics of individual tumors. Therefore, accurately assessing cancer prognostic risk remains a major challenge in clinical practice. In recent years, advanced deep learning technologies have enabled the extraction of crucial information from relevant biological data to improve the accuracy of prognostic assessment. Furthermore, accurate assessment of cancer prognostic risk helps in the precise identification of potential targets, aiding in the realization of personalized treatment and the design of targeted therapies.

[0003] However, extracting key tumor characteristic information from complex gene expression data is currently very difficult, and data shortages or sample scarcity are common in the biomedical field. Obtaining sufficient data is often a challenge. Clearly, existing technologies have shortcomings that urgently need to be addressed. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method and system for cancer prognosis assessment and target identification based on deep neural networks. By combining a diffusion model and an autoencoder deep neural network with an XGBoost survival analysis model, it achieves accurate assessment of prognostic risk of cancer patients and identification of relevant targets, thereby providing a scientific basis for the design of personalized treatment and targeted therapy.

[0005] To address the aforementioned technical problems, the first aspect of this invention discloses a method for cancer prognostic assessment and target identification based on deep neural networks, the method comprising:

[0006] A method for cancer prognostic assessment and target identification based on deep neural networks, characterized in that the method includes:

[0007] Gene expression data is acquired and preprocessed to obtain initial data;

[0008] The initial data is processed by a deep neural network combining a diffusion model and an autoencoder to obtain a low-dimensional expression of gene expression features.

[0009] The low-dimensional expression of the gene expression characteristics is input into the XGBoost survival analysis model to obtain a risk value, and the cancer prognosis assessment result is obtained based on the risk value.

[0010] Differential expression analysis was performed based on the risk values ​​and patient gene expression to obtain cancer prognostic targets.

[0011] As an optional implementation, in the first aspect of the present invention, the step of acquiring gene expression data and preprocessing the gene expression data to obtain initial data includes: acquiring gene expression data of patients and filling in missing values ​​in the gene expression data by averaging to obtain initial data.

[0012] As an optional implementation, in the first aspect of the invention, the step of using a deep neural network combining a diffusion model and an autoencoder to perform feature extraction processing on the initial data to obtain low-dimensional expression of gene expression features includes:

[0013] The initial data is mapped to a low-dimensional space, key features are extracted, and low-dimensional expression of gene expression features is obtained.

[0014] As an optional implementation, in the first aspect of the present invention, the training process of the deep neural network combining the diffusion model and the autoencoder includes the following steps:

[0015] Acquire experimental data and preprocess the experimental data to obtain training data;

[0016] Noise is added to the training data to perform the diffusion step, and the noisy data and the generated noise are returned to obtain the sample data;

[0017] The sample data is input into a deep neural network that combines a diffusion model and an autoencoder. The deep neural network that combines a diffusion model and an autoencoder includes an encoder and a decoder. The encoder maps the sample data to a latent space, and the decoder restores the compressed representation back to the original dimension to obtain a low-dimensional expression of the training gene expression features.

[0018] Backpropagation is performed using an optimizer to compute the gradient of the loss function with respect to the parameters of the deep neural network that combines the diffusion model and the autoencoder, minimize the reconstruction error, and update the parameters of the encoder and the decoder.

[0019] The deep neural network combining the diffusion model and the autoencoder is trained based on the number of iterations, with each batch performing the diffusion step, forward propagation, loss calculation, backpropagation, and optimization.

[0020] As an optional implementation, in the first aspect of the invention, the optimizer is the Adam optimizer.

[0021] As an optional implementation, in the first aspect of the present invention, the step of acquiring experimental data and preprocessing the experimental data to obtain training data includes:

[0022] Gene expression and survival data of cancer patients were obtained from publicly available datasets such as TCGA and GEO. Missing values ​​in the obtained data were imputed using the mean to obtain training data.

[0023] As an optional implementation, in the first aspect of the present invention, the training process of the XGBoost survival analysis model includes the following steps:

[0024] The low-dimensional expression of the training gene expression features is used as input to construct a data format suitable for survival analysis;

[0025] Initialize the XGBoost survival analysis model and train the XGBoost survival analysis model using the low-dimensional expression and survival time of the training gene expression features as labels.

[0026] A second aspect of this invention discloses a cancer prognostic assessment and target identification system based on deep neural networks, the system comprising:

[0027] A data acquisition module is used to acquire gene expression data and preprocess the gene expression data to obtain initial data.

[0028] The feature acquisition module is used to perform feature extraction processing on the initial data using a deep neural network combining a diffusion model and an autoencoder to obtain a low-dimensional expression of gene expression features.

[0029] The evaluation result acquisition module is used to input the low-dimensional expression of the gene expression characteristics into the XGBoost survival analysis model, obtain the risk value, and obtain the cancer prognosis assessment result based on the risk value.

[0030] The target acquisition module is used to perform differential expression analysis based on the risk value and patient gene expression to obtain cancer prognostic targets.

[0031] As an optional implementation, in a second aspect of the present invention, the data acquisition module acquires gene expression data and preprocesses the gene expression data to obtain initial data, including: acquiring gene expression data of patients and filling in missing values ​​in the gene expression data by averaging to obtain initial data.

[0032] As an optional implementation, in a second aspect of the invention, the feature acquisition module uses a deep neural network combining a diffusion model and an autoencoder to perform feature extraction processing on the initial data to obtain a low-dimensional expression of gene expression features, including:

[0033] The initial data is mapped to a low-dimensional space, key features are extracted, and a compressed representation of gene expression features is obtained.

[0034] As an optional implementation, in a second aspect of the invention, the training process of the deep neural network combining the diffusion model and the autoencoder includes the following steps:

[0035] Acquire experimental data and preprocess the experimental data to obtain training data;

[0036] Noise is added to the training data to perform the diffusion step, and the noisy data and the generated noise are returned to obtain the sample data;

[0037] The sample data is input into a deep neural network that combines a diffusion model and an autoencoder. The deep neural network that combines a diffusion model and an autoencoder includes an encoder and a decoder. The encoder maps the sample data to a latent space, and the decoder restores the compressed representation back to the original dimension to obtain a low-dimensional expression of the training gene expression features.

[0038] Backpropagation is performed using an optimizer to compute the gradient of the loss function with respect to the parameters of the deep neural network that combines the diffusion model and the autoencoder, minimize the reconstruction error, and update the parameters of the encoder and the decoder.

[0039] The deep neural network combining the diffusion model and the autoencoder is trained based on the number of iterations, with each batch performing the diffusion step, forward propagation, loss calculation, backpropagation, and optimization.

[0040] As an optional implementation, in a second aspect of the invention, the optimizer is the Adam optimizer.

[0041] As an optional implementation, in a second aspect of the present invention, the step of acquiring experimental data and preprocessing the experimental data to obtain training data includes:

[0042] Gene expression and survival data of cancer patients were obtained from publicly available datasets such as TCGA and GEO. Missing values ​​in the obtained data were imputed using the mean to obtain training data.

[0043] As an optional implementation, in a second aspect of the present invention, the training process of the XGBoost survival analysis model includes the following steps:

[0044] The low-dimensional expression of the training gene expression features is used as input to construct a data format suitable for survival analysis;

[0045] Initialize the XGBoost survival analysis model and train the XGBoost survival analysis model using the low-dimensional expression and survival time of the training gene expression features as labels.

[0046] A third aspect of this invention discloses another cancer prognostic assessment and target identification device based on a deep neural network, the device comprising:

[0047] Memory containing executable program code;

[0048] A processor coupled to the memory;

[0049] The processor calls the executable program code stored in the memory to execute some or all of the steps in the cancer prognosis assessment and target identification method based on deep neural networks disclosed in the first aspect of the present invention.

[0050] The fourth aspect of this invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the cancer prognosis assessment and target identification method based on deep neural networks disclosed in the first aspect of this invention.

[0051] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0052] In this embodiment of the invention, gene expression data is acquired and preprocessed to obtain initial data. A deep neural network combining a diffusion model and an autoencoder is used to extract features from the initial data, obtaining a low-dimensional representation of gene expression features. These gene expression features are then input into an XGBoost survival analysis model to obtain cancer prognostic assessment results. Based on the cancer prognostic assessment results, differential gene expression analysis is performed to obtain cancer prognostic targets. Therefore, this invention can effectively extract key tumor feature information from complex gene expression data by combining a diffusion model and an autoencoder, thereby generating high-quality synthetic data. Furthermore, by introducing the XGBoost model, the accuracy of cancer prognostic assessment can be effectively improved. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart illustrating a cancer prognostic assessment and target identification method based on deep neural networks disclosed in an embodiment of the present invention.

[0055] Figure 2 This is a schematic diagram of the structure of a cancer prognostic assessment and target identification system based on a deep neural network disclosed in an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of another cancer prognostic assessment and target identification device based on deep neural networks disclosed in an embodiment of the present invention;

[0057] Figure 4 This is a graph showing the results of differential expression of prognostic targets in skin cancer. Detailed Implementation

[0058] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0060] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0061] This invention discloses a method and system for cancer prognostic assessment and target identification based on deep neural networks. It effectively extracts key tumor feature information from complex gene expression data by combining a diffusion model with an autoencoder, thereby generating high-quality synthetic data. Furthermore, by introducing an XGBoost model, the accuracy of cancer prognostic assessment is effectively improved. These are described in detail below.

[0062] Example 1

[0063] Please see Figure 1 , Figure 1 This is a flowchart illustrating a cancer prognostic assessment and target identification method based on deep neural networks disclosed in an embodiment of the present invention. Wherein, Figure 1 The described method is applied to a cancer prognostic assessment and target identification system based on deep neural networks. This identification system can be a corresponding identification terminal, identification device, or server, and the server can be a local server or a cloud server; the embodiments of this invention are not limited thereto. Figure 1 As shown, this deep neural network-based cancer prognostic assessment and target identification method may include the following operations:

[0064] 101. Obtain gene expression data and preprocess the gene expression data to obtain initial data.

[0065] 102. Use a deep neural network combining a diffusion model and an autoencoder to perform feature extraction processing on the initial data to obtain low-dimensional expression of gene expression features.

[0066] 103. Input the low-dimensional expression of the gene expression characteristics into the XGBoost survival analysis model to obtain the risk value, and obtain the cancer prognosis assessment result based on the risk value; wherein, the risk value is divided into high-risk group and low-risk group according to the median risk value.

[0067] 104. Based on the stated risk values ​​and patient gene expression, differential expression analysis was performed to obtain cancer prognostic targets.

[0068] As can be seen, the method described in the embodiments of this invention can accurately assess the prognostic risk of cancer patients and identify relevant targets by combining a diffusion model and an autoencoder in a deep neural network with an XGBoost survival analysis model, thus providing a scientific basis for the design of personalized treatment and targeted therapies. Furthermore, by utilizing a deep learning network combining a diffusion model and an autoencoder to effectively extract key tumor feature information from complex gene expression data, high-quality synthetic data can be generated. Based on this, the introduction of the XGBoost model can effectively improve the accuracy of cancer prognostic assessment.

[0069] In an optional implementation, step 101, which involves acquiring gene expression data and preprocessing the gene expression data to obtain initial data, includes: acquiring the patient's gene expression data and filling in the missing values ​​in the gene expression data by averaging them to obtain initial data.

[0070] In an optional implementation, step 102, which involves using a deep neural network combining a diffusion model and an autoencoder to extract features from the initial data to obtain low-dimensional expressions of gene expression features, includes:

[0071] The initial data is mapped to a low-dimensional space, key features are extracted, and low-dimensional expression of gene expression features is obtained.

[0072] In this embodiment of the invention, a deep neural network combining a diffusion model and an autoencoder can reconstruct gene expression data of cancer cells, obtain higher quality data, and more effectively extract discriminative features for risk assessment of cancer prognosis.

[0073] In an optional implementation, the training process of the deep neural network combining the diffusion model and the autoencoder in step 102 includes the following steps:

[0074] Acquire experimental data and preprocess the experimental data to obtain training data;

[0075] Noise is added to the training data to perform the diffusion step, and the noisy data and the generated noise are returned to obtain the sample data;

[0076] The sample data is input into a deep neural network that combines a diffusion model and an autoencoder. The deep neural network that combines a diffusion model and an autoencoder includes an encoder and a decoder. The encoder maps the sample data to a latent space, and the decoder restores the compressed representation back to the original dimension to obtain a low-dimensional expression of the training gene expression features.

[0077] Backpropagation is performed using the Adam optimizer to compute the gradient of the loss function with respect to the parameters of the deep neural network that combines the diffusion model and the autoencoder, minimize the reconstruction error, and update the parameters of the encoder and the decoder.

[0078] The deep neural network combining the diffusion model and the autoencoder is trained based on the number of iterations, with each batch performing the diffusion step, forward propagation, loss calculation, backpropagation, and optimization.

[0079] In this embodiment of the invention, training data is input into a deep neural network combining a diffusion model and an autoencoder, wherein a custom autoencoder structure is defined, including an encoder and a decoder; the diffusion step is implemented using noise, returning noisy data and generated noise. Next, the input dimension (i.e., the number of genes) is obtained, and an autoencoder model is created. The encoder processes the training data... Mapping to the latent space, encoder Output latent variables decoder The compressed representation is then restored to its original dimensions to obtain the reconstruction result. The algorithm defines mean squared error loss and the Adam optimizer, and calculates loss functions such as the loss between the reconstructed output and the original output, and the denoising loss. Backpropagation calculates the gradient of the loss function with respect to the model parameters, minimizes the loss function, and updates the encoder and decoder parameters. The model is trained based on the number of iterations, with each batch performing a diffusion step, forward propagation, loss calculation, backpropagation, and optimization. The autoencoder loss is the sum of the reconstruction loss, denoising loss, and regularization loss. The expression for the reconstruction loss is as follows:

[0080]

[0081] The reconstruction loss is the mean squared error, where It is the first input data One sample, The first one reconstructed by the decoder One sample, It represents the total number of samples.

[0082] The expression for the denoising loss is as follows:

[0083]

[0084] The denoising loss is the mean squared error used for denoising, where It is a true, noise-free latent representation. It is the denoised result predicted by the model.

[0085] The regularization loss is the L2 regularization loss, and its expression is as follows:

[0086]

[0087] in It is the model's first One parameter, It is a hyperparameter of regularization strength.

[0088] Therefore, the loss of the autoencoder is ,in It is a weighting parameter used to balance the impact of the loss of each part.

[0089] In an optional implementation, the above steps, specifically acquiring experimental data and preprocessing the experimental data to obtain training data, include:

[0090] Gene expression and survival data of cancer patients were obtained from publicly available datasets such as TCGA and GEO. Missing values ​​in the obtained data were imputed using the mean to obtain training data.

[0091] In an optional implementation, the training process of the XGBoost survival analysis model in step 103 includes the following steps:

[0092] The low-dimensional expression of the training gene expression features is used as input to construct a data format suitable for survival analysis;

[0093] Initialize the XGBoost survival analysis model and train the XGBoost survival analysis model using the low-dimensional expression and survival time of the training gene expression features as labels.

[0094] In this embodiment of the invention, the input is a low-dimensional expression of the training gene expression features reconstructed by a deep neural network combining a diffusion model and an autoencoder. A data format suitable for survival analysis is constructed, the model is initialized, and model training is performed using feature data and survival time. The objective is to minimize the sum of the difficulty of setting the loss function and the regularization term, i.e., to minimize the objective function, thereby continuously optimizing the model. The objective function of the XBboost survival analysis model can be expressed as: .in, For loss function, For predicted values, The value represents the true value; the setting of the regularization term relates to the complexity of the model, and can be expressed as... It contains two important parts. One is the number of leaf nodes in the tree. One is the score of the leaf nodes on the tree. (right conduct Regularization is equivalent to increasing the score for each leaf node. (Smoothing). Therefore, the objective function of XGBoost is to define the complexity of the loss function by adding regularization to the training error, where the objective is to minimize the objective function. .

[0095] Based on the low-dimensional expression features of the obtained training genes, the loss function and regularization are calculated using the XGBoost ensemble learning algorithm to define the complexity. This is achieved by minimizing their sum (i.e., the objective function). ) Continuously optimize performance to predict the survival risk of each sample in the input data.

[0096] Finally, the accuracy of the cancer prognosis prediction model constructed by our algorithm was evaluated by calculating the C-index.

[0097] In this step, the low-dimensional expression data of the trained gene expression features are input into four commonly used methods: Cox proportional hazards model, RSF random survival forest, Cox-elastic neural network, and DNN deep neural network, for computational evaluation and comparison of their accuracy. Specific results are shown in Table 1. Our method achieves higher confidence indices than the compared methods, demonstrating the superiority of our approach.

[0098] Table 1. Confidence C-index scores of public cancer datasets

[0099]

[0100] Regarding differential expression analysis, the results of differential expression analysis in skin cancer, based on obtained prognostic risk values ​​and gene expression in cancer patients, are as follows: Figure 4 As shown in Table 2, differential expression analysis is an analytical method that identifies genes that show significant changes in a given state by comparing differences in gene expression levels between different sample groups (such as cancerous tissue and normal tissue). Differential expression analysis is of great significance in cancer target identification. It can not only help discover new cancer biomarkers for early cancer diagnosis, prognostic assessment, and treatment efficacy monitoring, but also reveal potential drug targets by identifying significantly upregulated or downregulated genes, especially key genes and pathways that play an important role in tumorigenesis and progression. Furthermore, differential expression analysis can deepen our understanding of cancer-related molecular mechanisms, helping researchers identify core factors driving malignant transformation and providing strong scientific evidence for targeted therapy. Therefore, this technology has irreplaceable application value in cancer research and therapeutic target development. The top 10 potential targets affecting the prognosis of skin cancer are shown in Table 2.

[0101] Table 2. Top 10 potential targets affecting skin cancer prognosis

[0102]

[0103] Training data is input into a deep neural network incorporating a diffusion autoencoder (DAO), which is then trained to output features. Based on these features, an XGBoost survival analysis model is trained, constructing decision trees over multiple iterations. Each tree is improved based on the prediction errors of all previous trees, thus determining the model settings for the XGBoost survival analysis model. Finally, cancer prognosis is assessed based on genetic data.

[0104] Deep neural networks based on a combination of diffusion models and autoencoders represent a strategy that combines the generative power of diffusion models with the efficient encoding power of autoencoders. A traditional autoencoder consists of an encoder and a decoder; the encoder compresses the input data into a latent representation, and the decoder reconstructs the original data from the latent representation. In contrast, the diffusion model is a generative model based on Markov chains that maps data to a simple prior distribution (such as a Gaussian distribution) by progressively adding noise, and then generates data from the noise through the inverse process.

[0105] Combining the advantages of diffusion models and autoencoders can significantly enhance the ability to learn and generate data features. Diffusion models learn the latent distribution of data by progressively adding noise. This method excels in generating high-quality new samples, but it is extremely sensitive to noise management and parameter settings. Autoencoders, on the other hand, focus on effectively compressing data and extracting key features. While they can learn quickly, they may sometimes overlook subtle structural information in the data, leading to a decline in reconstruction quality. Combining these two approaches to form a diffusion autoencoder not only improves the model's robustness to noise through the diffusion step but also enhances reconstruction quality, thereby extracting stronger feature representations. This combination is particularly important in cancer prognostic assessment and target identification applications. First, diffusion autoencoders can effectively extract key tumor feature information from complex gene expression data, thereby generating high-quality synthetic data. This is crucial for situations with insufficient data or scarce samples, especially in the biomedical field where obtaining sufficient data is often a challenge. Finally, the dimensionality-reduced feature representation helps improve the predictive accuracy of prognostic models, making them more practical and reliable, thus supporting the identification of potential therapeutic targets. The integrated application of this method can significantly advance cancer research and improve treatment outcomes.

[0106] Meanwhile, XGBoost has demonstrated numerous significant advantages in cancer prognostic assessment and target identification. As an advanced ensemble learning method, XGBoost possesses powerful predictive capabilities, effectively handling high-dimensional features and complex data patterns, thus providing highly accurate results in survival analysis and risk assessment. During training, it automatically performs feature selection, and based on feature importance scoring, XGBoost can identify key biomarkers closely related to cancer occurrence and development, providing a solid foundation for target identification. Furthermore, XGBoost employs parallel computing and optimization algorithms, exhibiting superior computational speed when processing large-scale clinical data, making it suitable for fast-response medical environments. XGBoost also effectively prevents overfitting and enhances the model's generalization ability by introducing L1 and L2 regularization, which is particularly important when processing biological data, as biological data is often complex and susceptible to noise. Therefore, the application of XGBoost not only improves the accuracy of cancer prognostic assessment and helps doctors develop personalized treatment plans, but also effectively identifies key cancer-related genes and pathways, providing potential targets for new drug development and treatment strategies, thereby promoting the in-depth development of cancer research.

[0107] Example 2

[0108] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a cancer prognostic assessment and target identification system based on a deep neural network, as disclosed in an embodiment of the present invention. Figure 2 The described system can be applied to corresponding identification terminals, identification devices, or servers, and the server can be a local server or a cloud server; this embodiment of the invention does not limit the application. Figure 2 As shown, the system may include:

[0109] The data acquisition module 100 is used to acquire gene expression data and preprocess the gene expression data to obtain initial data.

[0110] The feature acquisition module 200 is used to perform feature extraction processing on the initial data using a deep neural network combining a diffusion model and an autoencoder to obtain a low-dimensional expression of gene expression features.

[0111] The evaluation result acquisition module 300 is used to input the gene expression characteristics into the XGBoost survival analysis model, obtain a risk value, and obtain a cancer prognosis assessment result based on the risk value.

[0112] The target acquisition module 400 is used to perform differential expression analysis based on the risk value and patient gene expression to obtain cancer prognostic targets.

[0113] As an optional implementation, in a second aspect of the present invention, the data acquisition module 100 acquires gene expression data and preprocesses the gene expression data to obtain initial data, including: acquiring the patient's gene expression data and filling in the missing values ​​in the gene expression data by averaging to obtain initial data.

[0114] As an optional implementation, in a second aspect of the invention, the feature acquisition module 200 uses a deep neural network combining a diffusion model and an autoencoder to perform feature extraction processing on the initial data to obtain a low-dimensional expression of gene expression features, including:

[0115] The initial data is mapped to a low-dimensional space, key features are extracted, and gene expression features are obtained.

[0116] As an optional implementation, in a second aspect of the invention, the training process of the deep neural network combining the diffusion model and the autoencoder includes the following steps:

[0117] Acquire experimental data and preprocess the experimental data to obtain training data;

[0118] Noise is added to the training data to perform the diffusion step, and the noisy data and the generated noise are returned to obtain the sample data;

[0119] The sample data is input into a deep neural network that combines a diffusion model and an autoencoder. The deep neural network that combines a diffusion model and an autoencoder includes an encoder and a decoder. The encoder maps the sample data to a latent space, and the decoder restores the compressed representation back to the original dimension to obtain a low-dimensional expression of the training gene expression features.

[0120] Backpropagation is performed using an optimizer to compute the gradient of the loss function with respect to the parameters of the deep neural network that combines the diffusion model and the autoencoder, minimize the reconstruction error, and update the parameters of the encoder and the decoder.

[0121] The deep neural network combining the diffusion model and the autoencoder is trained based on the number of iterations, with each batch performing the diffusion step, forward propagation, loss calculation, backpropagation, and optimization.

[0122] As an optional implementation, in a second aspect of the invention, the optimizer is the Adam optimizer.

[0123] As an optional implementation, in a second aspect of the present invention, the step of acquiring experimental data and preprocessing the experimental data to obtain training data includes:

[0124] Gene expression and survival data of cancer patients were obtained from publicly available datasets such as TCGA and GEO. Missing values ​​in the obtained data were imputed using the mean to obtain training data.

[0125] As an optional implementation, in a second aspect of the present invention, the training process of the XGBoost survival analysis model includes the following steps:

[0126] The low-dimensional expression of the training gene expression features is used as input to construct a data format suitable for survival analysis;

[0127] Initialize the XGBoost survival analysis model and train the XGBoost survival analysis model using the low-dimensional expression and survival time of the training gene expression features as labels.

[0128] Example 3

[0129] Please see Figure 3 , Figure 3 This is a schematic diagram of another cancer prognostic assessment and target identification system based on deep neural networks disclosed in an embodiment of the present invention. Figure 3 As shown, the device may include:

[0130] Memory 301 storing executable program code;

[0131] Processor 302 coupled to memory 301;

[0132] The processor 302 calls the executable program code stored in the memory 301 to execute some or all of the steps in the cancer prognosis assessment and target identification method based on deep neural networks disclosed in Embodiment 1 of the present invention.

[0133] Example 4

[0134] This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute some or all of the steps in the deep neural network-based cancer prognosis assessment and target identification method disclosed in Embodiment 1 of this invention.

[0135] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0136] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0137] Finally, it should be noted that the cancer prognosis assessment and target identification method and system based on deep neural networks disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for cancer prognostic assessment and target identification based on deep neural networks, characterized in that, The method includes: Gene expression data is acquired and preprocessed to obtain initial data; The initial data is processed by a deep neural network combining a diffusion model and an autoencoder to obtain a low-dimensional expression of gene expression features. The low-dimensional expression of the gene expression characteristics is input into the XGBoost survival analysis model to obtain a risk value, and the cancer prognosis assessment result is obtained based on the risk value. Differential expression analysis was performed based on the risk values ​​and patient gene expression to obtain cancer prognostic targets. The step of using a deep neural network combining a diffusion model and an autoencoder to extract features from the initial data to obtain low-dimensional expressions of gene expression features includes: The initial data is mapped to a low-dimensional space, key features are extracted, and low-dimensional expression of gene expression features is obtained. The training process of the deep neural network combining the diffusion model and the autoencoder includes the following steps: Acquire experimental data and preprocess the experimental data to obtain training data; Noise is added to the training data to perform the diffusion step, and the noisy data and the generated noise are returned to obtain the sample data; The sample data is input into a deep neural network that combines a diffusion model and an autoencoder. The deep neural network that combines a diffusion model and an autoencoder includes an encoder and a decoder. The encoder maps the sample data to a latent space, and the decoder restores the compressed representation back to the original dimension to obtain a low-dimensional expression of the training gene expression features. Backpropagation is performed using an optimizer to compute the gradient of the loss function with respect to the parameters of the deep neural network that combines the diffusion model and the autoencoder, minimize the reconstruction error, and update the parameters of the encoder and the decoder. The deep neural network combining the diffusion model and the autoencoder is trained according to the number of iterations, and the diffusion step, forward propagation, loss calculation, backpropagation and optimization are performed in each batch; The autoencoder loss is the sum of the reconstruction loss, denoising loss, and regularization loss.

2. The cancer prognostic assessment and target identification method based on deep neural networks according to claim 1, characterized in that, The process of acquiring gene expression data and preprocessing the gene expression data to obtain initial data includes: acquiring the patient's gene expression data and filling in the missing values ​​in the gene expression data by averaging them to obtain initial data.

3. The cancer prognostic assessment and target identification method based on deep neural networks according to claim 1, characterized in that, The optimizer is the Adam optimizer.

4. The cancer prognostic assessment and target identification method based on deep neural networks according to claim 1, characterized in that, The process of acquiring experimental data and preprocessing the experimental data to obtain training data includes: Gene expression and survival data of cancer patients were obtained from publicly available datasets such as TCGA and GEO. Missing values ​​in the obtained data were imputed using the mean to obtain training data.

5. The cancer prognostic assessment and target identification method based on deep neural networks according to claim 1, characterized in that, The training process of the XGBoost survival analysis model includes the following steps: The low-dimensional expression of the training gene expression features is used as input to construct a data format suitable for survival analysis; Initialize the XGBoost survival analysis model and train the XGBoost survival analysis model using the low-dimensional expression and survival time of the training gene expression features as labels.

6. A cancer prognostic assessment and target identification system based on deep neural networks, characterized in that, The system includes: A data acquisition module is used to acquire gene expression data and preprocess the gene expression data to obtain initial data. The feature acquisition module is used to perform feature extraction processing on the initial data using a deep neural network combining a diffusion model and an autoencoder to obtain a low-dimensional expression of gene expression features. The evaluation result acquisition module is used to input the low-dimensional expression of the gene expression characteristics into the XGBoost survival analysis model, obtain the risk value, and obtain the cancer prognosis assessment result based on the risk value. The target acquisition module is used to perform differential expression analysis based on the risk value and patient gene expression to obtain cancer prognostic targets. The step of using a deep neural network combining a diffusion model and an autoencoder to extract features from the initial data to obtain low-dimensional expressions of gene expression features includes: The initial data is mapped to a low-dimensional space, key features are extracted, and low-dimensional expression of gene expression features is obtained. The training process of the deep neural network combining the diffusion model and the autoencoder includes the following steps: Acquire experimental data and preprocess the experimental data to obtain training data; Noise is added to the training data to perform the diffusion step, and the noisy data and the generated noise are returned to obtain the sample data; The sample data is input into a deep neural network that combines a diffusion model and an autoencoder. The deep neural network that combines a diffusion model and an autoencoder includes an encoder and a decoder. The encoder maps the sample data to a latent space, and the decoder restores the compressed representation back to the original dimension to obtain a low-dimensional expression of the training gene expression features. Backpropagation is performed using an optimizer to compute the gradient of the loss function with respect to the parameters of the deep neural network that combines the diffusion model and the autoencoder, minimize the reconstruction error, and update the parameters of the encoder and the decoder. The deep neural network combining the diffusion model and the autoencoder is trained according to the number of iterations, and the diffusion step, forward propagation, loss calculation, backpropagation and optimization are performed in each batch; The autoencoder loss is the sum of the reconstruction loss, denoising loss, and regularization loss.

7. A cancer prognostic assessment and target identification device based on deep neural networks, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the cancer prognostic assessment and target identification method based on deep neural networks as described in any one of claims 1-5.

8. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the deep neural network-based cancer prognostic assessment and target identification method as described in any one of claims 1-5.

Images