Pancreatic cancer ferroptosis related gene prognosis model construction method and device

Abstract

The embodiment of the invention provides a pancreatic cancer ferroptosis related gene prognosis model construction method and device. The method comprises the following steps: acquiring a gene expression data set and a clinical information data set of a pancreatic cancer patient; performing data preprocessing on the gene expression data set and the clinical information data set to generate a standardized gene expression matrix; performing differential expression analysis and survival analysis according to the standardized gene expression matrix in combination with a ferroptosis pathway gene set, and screening out a ferroptosis feature gene set significantly related to prognosis; constructing a deep learning prognosis prediction model based on the ferroptosis feature gene set; and evaluating the performance of the deep learning prognosis prediction model by using the independent verification set to obtain a model evaluation index. The problem that in the prior art, a pancreatic cancer prognosis prediction model is insufficient in accuracy and interpretability can be solved.

Description

Technical Field

[0001] The present application relates to the technical field of bioinformatics, and in particular to a method and device for constructing a prognostic model of ferroptosis-related genes in pancreatic cancer. Background Art

[0002] Pancreatic cancer is a malignant tumor with a very poor prognosis, with a 5-year survival rate of less than 10%. Accurately predicting the prognosis of pancreatic cancer patients is of great significance for developing personalized treatment plans and improving patient survival. Studies in recent years have shown that ferroptosis, as a new type of programmed cell death, plays an important role in the occurrence and development of tumors. There are also methods for predicting the prognosis of pancreatic cancer through relevant prediction models. However, the current pancreatic cancer prognosis prediction model based on ferroptosis-related genes still has the following problems: 1) The existing model mainly relies on a single statistical method and fails to fully explore the complex nonlinear relationship in gene expression data; 2) There is a lack of interpretable analysis of the model prediction results; 3) The external validation of the model is not sufficient, and its prediction effect on different races and sequencing platforms has not been verified. Summary of the invention

[0003] In view of this, the present application provides a method and device for constructing a pancreatic cancer ferroptosis-related gene prognosis model based on gene expression data, which solves the problem of insufficient accuracy and interpretability of pancreatic cancer prognosis prediction models in the prior art.

[0004] The present application embodiment provides a method for constructing a prognostic model of pancreatic cancer ferroptosis-related genes based on gene expression data, comprising:

[0005] Obtain gene expression datasets and clinical information datasets of pancreatic cancer patients;

[0006] Performing data preprocessing on the gene expression dataset and the clinical information dataset to generate a standardized gene expression matrix;

[0007] According to the standardized gene expression matrix, combined with the ferroptosis pathway gene set, differential expression analysis and survival analysis are performed to screen out the ferroptosis characteristic gene set that is significantly correlated with prognosis;

[0008] Constructing a deep learning prognosis prediction model based on the ferroptosis characteristic gene set;

[0009] The performance of the deep learning prognosis prediction model was evaluated using an independent validation set to obtain model evaluation indicators.

[0010] Optionally, data preprocessing is performed on the gene expression dataset, including:

[0011] Removing gene expression data with a missing value ratio exceeding 20% ​​in the gene expression data set;

[0012] The KNN algorithm was used to fill the missing values ​​of the remaining gene expression data;

[0013] The ComBat algorithm was used to normalize the filled gene expression data;

[0014] The clinical information data set is preprocessed, including:

[0015] Unifying the calculation unit of survival time in the clinical information data in the clinical information dataset into months;

[0016] Encoding the survival status in the clinical information data;

[0017] Integrating the clinical characteristic information of the patient in the clinical information data;

[0018] Generate a normalized gene expression matrix including:

[0019] The processed gene expression data and clinical information data were matched according to the sample ID to generate a standardized gene expression matrix.

[0020] Optionally, performing differential expression analysis based on the standardized gene expression matrix in combination with the ferroptosis pathway gene set includes:

[0021] matching the normalized gene expression matrix with the ferroptosis pathway gene set;

[0022] Calculate the expression difference of each matched gene between tumor tissue and normal tissue to obtain the differential expression analysis results;

[0023] The Benjamini-Hochberg method was used to calculate the FDR value of the differential expression analysis results;

[0024] The results of differential expression analysis were corrected according to the FDR value.

[0025] Optionally, performing survival analysis based on the standardized gene expression matrix in combination with a ferroptosis pathway gene set comprises:

[0026] Perform univariate Cox regression analysis on each matched gene to obtain the Cox regression analysis results;

[0027] According to the results of Cox regression analysis, genes significantly associated with survival time were screened;

[0028] Kaplan-Meier survival analysis was performed on genes significantly correlated with survival time to obtain survival analysis results.

[0029] Optionally, the screening of ferroptosis characteristic gene sets significantly associated with prognosis includes:

[0030] According to the results of differential expression analysis and survival analysis, the basic framework of LASSO regression model was determined;

[0031] Set the candidate range of the penalty parameter λ of the basic framework;

[0032] For the ferroptosis pathway gene set, the 10-fold cross-validation method was used to select the optimal λ value;

[0033] The optimal λ value was used to retrain the LASSO regression model on the ferroptosis pathway gene set to obtain the regression coefficient of each gene;

[0034] The features of genes with non-zero regression coefficients were extracted to form a ferroptosis feature gene set.

[0035] Optionally, constructing a deep learning prognosis prediction model based on the ferroptosis characteristic gene set comprises:

[0036] Designing a deep learning model architecture, wherein the deep learning model architecture includes an input layer, a fully connected layer, a Dropout layer, and an output layer, and the dimension of the input layer is the same as the number of genes in the ferroptosis characteristic gene set;

[0037] Combine the Cox partial likelihood loss function and the ranking loss function to define the multi-task learning objective of the deep learning model architecture;

[0038] Add L1 and L2 regularization terms to the multi-task learning objective;

[0039] According to the multi-task learning objectives and the mini-batch stochastic gradient descent algorithm, the deep learning model architecture is trained and optimized to obtain a deep learning prognosis prediction model.

[0040] Optionally, the evaluating the performance of the deep learning prognosis prediction model using an independent validation set comprises:

[0041] Calculating the model performance indicators of the deep learning prognostic prediction model on an independent validation set, the model performance indicators including: C-index, time-dependent ROC curve and survival analysis;

[0042] The Cox regression model and random forest survival model were constructed using the data from the independent validation set.

[0043] Calculate the C-index of the Cox regression model on an independent validation set and the C-index of the random forest survival model on an independent validation set;

[0044] Comparing the C-index of the deep learning prognostic prediction model, the C-index of the Cox regression model, and the C-index of the random forest survival model;

[0045] The performance of the deep learning prognostic prediction model was determined based on the comparison results, time-dependent ROC curve and survival analysis.

[0046] Optionally, the method further comprises:

[0047] Acquire multiple external validation sets, wherein the multiple external validation sets include external validation sets of data from different ethnic groups and external validation sets of data from different sequencing platforms;

[0048] Evaluate the external prediction effect of the deep learning prognostic prediction model on data from different ethnic groups and different sequencing platforms based on multiple external validation sets;

[0049] According to the external prediction effect, the deep learning prognosis prediction model is retrained.

[0050] Optionally, the method further comprises:

[0051] Calculate the contribution of each gene to the prediction results;

[0052] Pathway enrichment analysis was performed on high-risk genes whose contribution was greater than the threshold;

[0053] Generate a personalized prognosis prediction report based on the results of pathway enrichment analysis.

[0054] The present application also provides a device for constructing a pancreatic cancer ferroptosis-related gene prognosis model, comprising:

[0055] A data acquisition module, used to obtain gene expression data sets and clinical information data sets of pancreatic cancer patients;

[0056] A preprocessing module, used to perform data preprocessing on the gene expression dataset and the clinical information dataset to generate a standardized gene expression matrix;

[0057] A feature selection module, for performing differential expression analysis and survival analysis based on the standardized gene expression matrix in combination with the ferroptosis pathway gene set, and screening out the ferroptosis feature gene set significantly associated with prognosis;

[0058] A model building module, used to build a deep learning prognosis prediction model based on the ferroptosis characteristic gene set;

[0059] The model evaluation module is used to evaluate the performance of the deep learning prognosis prediction model using an independent validation set to obtain model evaluation indicators.

[0060] The present application also provides a computer device, the computer device comprising:

[0061] at least one processor; and,

[0062] a memory communicatively connected to the at least one processor; wherein,

[0063] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned method for constructing a pancreatic cancer ferroptosis-related gene prognostic model based on gene expression data.

[0064] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the above-mentioned method for constructing a pancreatic cancer ferroptosis-related gene prognosis model based on gene expression data.

[0065] An embodiment of the present application also provides a computer program product, including computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the method for constructing a pancreatic cancer ferroptosis-related gene prognosis model based on gene expression data are implemented.

[0066] This application has the following technical effects:

[0067] 1. By combining deep learning methods, it can effectively capture the complex nonlinear relationships in gene expression data and improve the accuracy of prognosis prediction;

[0068] 2. Through the interpretability analysis of the model prediction results, identify key prognosis-related genes and their biological functions, and provide guidance for clinical personalized treatment;

[0069] 3. Through verification of multiple external validation sets, the prediction effect of the model on different ethnic groups and sequencing platforms is guaranteed, and the generalization ability of the model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0071] Figure 1 This is a flow chart of a method for constructing a prognostic model of pancreatic cancer ferroptosis-related genes based on gene expression data provided in an embodiment of the present application.

[0072] Figure 2 A schematic diagram of the network structure of the deep learning prognosis prediction model provided in an embodiment of the present application.

[0073] Figure 3The ROC curve diagram of the model provided in the embodiment of the present application on an independent validation set.

[0074] Figure 4 This is a diagram of the pathway enrichment analysis results of high-risk genes provided in the examples of this application.

[0075] Figure 5 This is a structural diagram of a device for constructing a prognostic model of pancreatic cancer ferroptosis-related genes based on gene expression data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0076] The embodiments of the present application are described below in conjunction with the drawings in the present application. It should be understood that the implementation methods described below in conjunction with the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0077] It will be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application refer to that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude the implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the technical field. It should be understood that when we say that an element is "connected" or "coupled" to another element, the one element may be directly connected or coupled to the other element, or it may refer to that the one element and the other element establish a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein indicates at least one of the items defined by the term, for example, "A and / or B" may be implemented as "A", or as "B", or as "A and B".

[0078] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0079] Embodiment 1

[0080] Reference Figure 1 , the embodiment of the present application provides a method for constructing a prognostic model of pancreatic cancer ferroptosis-related genes based on gene expression data. This method can accurately predict the prognosis of pancreatic cancer patients and provide a scientific basis for the formulation of clinical personalized treatment plans. The technical solution of the present application will be described in detail below in conjunction with specific embodiments. The method for constructing a prognostic model of pancreatic cancer ferroptosis-related genes based on gene expression data includes:

[0081] S1: Obtain gene expression datasets and clinical information datasets of pancreatic cancer patients

[0082] The gene expression data set contains the gene expression profile data of tumor samples and normal samples, which can be obtained from public databases (such as TCGA, GEO, etc.). The clinical information data set contains clinical characteristic information such as patient survival time, survival status, age, gender, tumor stage, etc. In this embodiment, RNA sequencing data of 178 pancreatic cancer patients can be obtained from the TCGA-PAAD data set as a training set, and gene expression data of 69 pancreatic cancer patients can be obtained from the GSE62452 data set of the GEO database as an independent validation set.

[0083] S2: Perform data preprocessing on gene expression dataset and clinical information dataset

[0084] The preprocessing of gene expression datasets includes the following steps:

[0085] S2.1.1: Perform quality control on gene expression data and remove gene expression data with missing value ratio exceeding 20%.

[0086] This step can be achieved by calculating the missing value ratio of each gene in all samples. If the missing value ratio of a gene exceeds the set threshold, the gene will be removed from the data set. This can ensure the data quality of subsequent analysis.

[0087] S2.1.2: For the missing values ​​in the remaining data, the K nearest neighbor (KNN) algorithm is used to fill them

[0088] The KNN algorithm estimates missing values ​​based on the similarity between samples. The specific steps are as follows: first, the Euclidean distance between samples is calculated, then the K nearest neighbor samples that are most similar to the missing value sample are selected (in this embodiment, the value of K is 10), and finally the average value of the corresponding gene expression values ​​of these K nearest neighbor samples is used to fill the missing value.

[0089] S2.1.3: Use ComBat algorithm to eliminate batch effects and normalize data after filling gene expression data

[0090] The ComBat algorithm is a data normalization method based on the empirical Bayesian method, which can effectively remove the systematic differences between data from different batches or platforms. The normalized expression value range is usually between [-1,1], which is convenient for subsequent model training.

[0091] S2.2: Preprocessing of clinical information dataset

[0092] Preprocessing clinical information datasets is an important step to ensure the availability of clinical data, including:

[0093] S2.2.1: The calculation unit of survival time in clinical information data should be unified as month.

[0094] Since different data sets may use different time units (such as days, months, and years) to record the patient's survival time, unit conversion is required. For example, the survival time in days is divided by 30.44 to convert to months, and the survival time in years is multiplied by 12 to convert to months. This unified processing can ensure the accuracy of subsequent survival analysis.

[0095] S2.2.2: Encoding survival status

[0096] The survival status of patients is usually coded in binary format: 0 indicates survival or loss of follow-up, and 1 indicates death. For patients who are lost to follow-up, their survival time is recorded as the last follow-up time. This processing method is in line with the principle of censored data processing in survival analysis.

[0097] S2.2.3: Integrate information on the patient’s clinical characteristics

[0098] The integration of patients' clinical characteristic information includes: standardizing demographic characteristics such as age and gender; converting TNM staging into the unified AJCC staging standard; encoding categorical variables such as treatment options and pathological types; deleting clinical characteristics with too high a missing ratio; and for missing values ​​of existing clinical characteristics, selecting appropriate filling strategies based on the feature type, such as filling categorical variables with the mode and filling continuous variables with the median.

[0099] S2.3: Generate a normalized gene expression matrix

[0100] The processed gene expression data and clinical information data are matched according to the sample ID to generate a standardized gene expression matrix. The specific implementation process includes:

[0101] First, extract the sample IDs from the gene expression dataset and the clinical information dataset; determine the sample IDs shared by the two datasets; extract the corresponding gene expression data from the gene expression dataset based on the shared sample ID; extract the corresponding clinical feature data from the clinical information dataset; merge the gene expression data and clinical feature data according to the sample ID to form a complete data matrix; check the merged data matrix to ensure the integrity and consistency of the data; normalize the format of the merged data matrix to ensure that the data format meets the requirements of subsequent analysis.

[0102] The rows of the final standardized gene expression matrix represent samples, and the columns include gene expression values ​​and clinical characteristics. Each element in the matrix has been standardized and is suitable for subsequent analysis.

[0103] S3: Perform differential expression analysis and survival analysis

[0104] S3 aims to screen ferroptosis genes related to pancreatic cancer prognosis from multiple dimensions. Specifically, differential expression analysis and survival analysis include:

[0105] S3.1: Perform differential expression analysis

[0106] S3.1.1: Matching the normalized gene expression matrix to the ferroptosis pathway gene set

[0107] Exemplarily, matching the standardized gene expression matrix with the ferroptosis pathway gene set specifically includes: uniformly converting the gene names in the standardized gene expression matrix into official gene symbols; converting the gene names in the ferroptosis pathway gene set into the same naming system; using the gene symbols as indexes to locate the expression data of ferroptosis-related genes in the expression matrix.

[0108] S3.1.2: Calculate the expression difference of each gene between tumor tissue and normal tissue

[0109] Exemplarily, calculating the expression difference of each gene between tumor tissue and normal tissue specifically includes: using the limma package to perform differential expression analysis; calculating the average expression level difference (logFC value) of each gene; using moderatedt-test to calculate the difference significance level (p value); considering the expression difference multiple and statistical significance at the same time, and screening differentially expressed genes.

[0110] S3.1.3: Calculate FDR value using Benjamini-Hochberg method

[0111] Exemplarily, using the Benjamini-Hochberg method to calculate the FDR value specifically includes: sorting the p values ​​of all genes from small to large; calculating the FDR value corresponding to each p value; and setting the FDR threshold (usually 0.05) for significance judgment.

[0112] S3.1.4: Correct the results of differential expression analysis based on FDR value

[0113] Exemplarily, correcting the differential expression analysis results according to the FDR value specifically includes: screening genes with FDR<0.05 as significantly differentially expressed genes; visualizing the differential expression analysis results to generate volcano plots and heat maps; and recording the specific expression change characteristics of each significantly differentially expressed gene.

[0114] S3.2: Construction of a knowledge base of ferroptosis-related gene sets

[0115] The construction process of this knowledge base obtains information from multiple authoritative databases to ensure the integrity and reliability of the gene set. Specifically, the construction of the ferroptosis-related gene set knowledge base includes:

[0116] S3.2.1: Extracting ferroptosis-related pathway genes from the KEGG database

[0117] From the KEGG database, we can extract gene information in related pathways such as "ferroptosis pathway" and "iron metabolism pathway". These pathways include important biological processes such as iron ion transport, ferritin metabolism, and oxidative stress response.

[0118] S3.2.2: Obtain iron metabolism pathway genes from the Reactome database

[0119] By accessing the Reactome database, more detailed information on iron metabolism-related pathways can be obtained, including genes involved in processes such as intracellular iron homeostasis regulation and iron-sulfur cluster assembly.

[0120] S3.2.3: Extract ferroptosis-related gene ontology annotations from the Gene Ontology database

[0121] In the Gene Ontology database, annotation information on biological processes, molecular functions, and cellular components related to ferroptosis can be extracted. This annotation information can help understand the functional attributes and cellular localization characteristics of ferroptosis-related genes.

[0122] S3.2.4: Integration of ferroptosis genes verified by literature

[0123] Through literature mining, we integrated experimentally verified key ferroptosis genes, such as GPX4, SLC7A11, ACSL4, etc. These genes play a core regulatory role in the ferroptosis process, and their expression levels are closely related to tumor prognosis.

[0124] S3.2.5: Establishing gene-gene interaction networks

[0125] Based on the gene information collected above, the String database and biological knowledge were used to construct a gene-gene interaction network. This network not only reflects the physical interactions between genes, but also includes functional associations such as transcriptional regulation and signal pathway cascades. Through network topology analysis, key node genes and functional modules in the ferroptosis pathway can be identified, providing an important basis for subsequent feature selection.

[0126] S3.3: Perform survival analysis

[0127] After completing differential expression analysis and knowledge base construction, survival analysis is needed to evaluate the correlation between ferroptosis-related genes and patient prognosis. Specifically, performing survival analysis includes:

[0128] S3.3.1: Perform univariate Cox regression analysis

[0129] Univariate Cox proportional hazard regression analysis was performed for each ferroptosis-related gene. This analysis method can evaluate the strength of the association between the expression level of a single gene and the patient's survival time, and quantify this association by calculating the hazard ratio (HR) and the corresponding p value. For example, if the HR value of a gene is greater than 1 and the p value is less than 0.05, it indicates that high expression of the gene is associated with a poor prognosis.

[0130] S3.3.2: Screening for genes significantly associated with survival time

[0131] After obtaining the results of Cox regression analysis, we focused on those genes with statistical significance. The expression levels of these genes showed a significant correlation with the patient's survival time and may play an important role in the process of tumor progression.

[0132] S3.3.3: Perform Kaplan-Meier survival analysis

[0133] In order to further verify the prognostic value of these genes, Kaplan-Meier survival analysis is required. According to the median of gene expression level, patients are divided into high expression group and low expression group, survival curves are drawn, and the log-rank test is used to evaluate the significance of survival difference between the two groups. The results of survival analysis can intuitively show the relationship between gene expression level and patient prognosis, and provide important biological basis for subsequent model construction.

[0134] By combining the results of differential expression analysis and survival analysis, the potential value of each ferroptosis-related gene in predicting the prognosis of pancreatic cancer can be comprehensively evaluated. This multidimensional analysis strategy not only takes into account the changes in gene expression levels, but also focuses on the actual impact of these changes on patient prognosis, thus providing a reliable scientific basis for subsequent feature selection.

[0135] S4: Screening for prognosis-related ferroptosis signature gene sets

[0136] Specifically, the ferroptosis signature gene set screened for prognosis-related features includes:

[0137] S4.1: Determine the basic framework of the LASSO regression model

[0138] In order to screen out the characteristic gene set with the most prognostic prediction value from the numerous ferroptosis-related genes, this application uses the LASSO (Least Absolute Shrinkage and Selection Operator) regression method for feature selection. This method can achieve model sparsification while performing feature selection by introducing the L1 regularization term to avoid overfitting problems. The basic framework of the LASSO regression model is based on the Cox proportional hazard model, taking the patient's survival time as the dependent variable and the expression value of the ferroptosis-related genes screened in the early stage as the independent variable.

[0139] S4.2: Setting the candidate range of penalty parameter λ

[0140] In the model, the strictness of feature selection is controlled by introducing a penalty term λ. The candidate range of the penalty parameter λ is usually set to a logarithmic sequence, for example, from 10 -4 to 10 1 , 100 values ​​are taken at equal intervals as candidates.

[0141] S4.3: Select the optimal lambda value using 10-fold cross validation

[0142] In order to select the optimal λ value, a 10-fold cross validation method is used. Specifically, the training set is randomly divided into 10 subsets of similar size, 9 subsets are used to train the model each time, and the remaining 1 subset is used for validation. By calculating the cross validation error of the model under different λ values ​​(partial likelihood deviation is usually used as an evaluation indicator), the λ value that minimizes the cross validation error is selected as the optimal parameter.

[0143] S4.4: Retrain the LASSO regression model using the optimal lambda value

[0144] Retraining the LASSO regression model with the optimal λ value can obtain the regression coefficient of each gene. Due to the characteristics of LASSO, the regression coefficients of some genes will be compressed to 0, and these genes will be excluded from the final feature set.

[0145] S4.5: Extract gene features with non-zero regression coefficients

[0146] Genes with non-zero regression coefficients were extracted to form a ferroptosis feature gene set. These genes are not only significantly correlated with patient prognosis, but also have strong predictive ability and can be used as input features for subsequent deep learning models.

[0147] In this embodiment, the characteristic gene set screened out by LASSO regression usually includes 10-30 genes. The expression pattern of these genes can better reflect the patient's prognosis risk, and is also convenient for the construction and clinical application of subsequent models. Through this feature selection method, the predictive performance of the model is guaranteed and the interpretability of the model is improved.

[0148] S5: Building a deep learning prognostic prediction model

[0149] Specifically, building a deep learning prognosis prediction model includes:

[0150] S5.1: Designing Deep Learning Model Architecture

[0151] like Figure 2 After obtaining the ferroptosis characteristic gene set, this application uses a deep learning method to build a prognosis prediction model. The design of the deep learning model fully considers the characteristics of gene expression data and the requirements of the prognosis prediction task, and captures the complex nonlinear relationship between genes through a multi-layer neural network structure.

[0152] The architecture design of the deep learning model adopts a multi-layer feedforward neural network structure. The dimension of the input layer is the same as the number of genes in the ferroptosis feature gene set, and each neuron corresponds to the expression value of a gene. After the input layer, multiple fully connected layers are set for feature conversion and information extraction. To prevent overfitting, a Dropout layer is added between adjacent fully connected layers, and the dropout rate is set to 0.3. The last layer of the network is the output layer, which is used to predict the patient's prognostic risk score. The entire network uses ReLU as the activation function. This choice can effectively alleviate the gradient vanishing problem and speed up the model convergence.

[0153] S5.2: Define multi-task learning objectives

[0154] The learning objective of the model adopts a multi-task learning framework, taking into account Cox partial likelihood loss and ranking loss. Cox partial likelihood loss is used to describe the relationship between patient survival time and predicted risk, while ranking loss ensures that the model can correctly distinguish between high-risk and low-risk patients. Specifically, for any two samples, if the actual survival time of one of the samples is shorter, then the model should predict a higher risk score for this sample. This multi-task learning strategy can improve the prediction accuracy and generalization ability of the model.

[0155] S5.3: Add L1 and L2 regularization terms

[0156] In order to further improve the generalization ability of the model, L1 and L2 regularization terms are added to the loss function. L1 regularization can achieve feature selection, while L2 regularization can prevent the model from over-relying on certain features. The weight coefficients of the two regularization terms are determined by cross-validation. The model is trained using the mini-batch stochastic gradient descent algorithm, the batch size is set to 32, the initial value of the learning rate is set to 0.001, and the learning rate decay strategy is used.

[0157] S5.4: Training and optimization using mini-batch stochastic gradient descent

[0158] Mini-batch Stochastic Gradient Descent (SGD) is an algorithm for optimizing machine learning models, especially neural networks. It is a variant of Stochastic Gradient Descent (SGD) that aims to balance computational efficiency and convergence speed. In stochastic gradient descent, each parameter update is based on only one training sample (or one data point). This means that the algorithm updates the weights of the model very frequently, which may cause the training process to be unstable and may oscillate at local minima instead of converging to the global minimum. The mini-batch stochastic gradient descent algorithm solves this problem by using a small batch (mini-batch) of training samples each time the parameters are updated. This mini-batch usually contains dozens to hundreds of samples.

[0159] During the training process, an early stopping strategy can also be used to prevent overfitting. The specific approach is to monitor the loss function value on the validation set, and stop the training process when the validation set loss does not improve for multiple consecutive epochs. This allows the training to be terminated in time when the model reaches the best performance, avoiding overfitting of the training data. The above design ensures that the deep learning model can effectively learn prognosis-related feature patterns from gene expression data.

[0160] S6: Evaluate model performance

[0161] The evaluation of deep learning prognostic prediction models is a key step in verifying the effectiveness of the models. This application uses multiple evaluation indicators and multiple verification methods to comprehensively evaluate the predictive performance of the model. Specifically, the evaluation of model performance includes:

[0162] S6.1: Calculate model performance metrics

[0163] On the independent validation set, the C-index (consistency index) of the model is first calculated. The C-index reflects the accuracy of the model in ranking the patient's prognostic risk, and the value range is between 0.5 and 1. The larger the value, the stronger the model's ability to distinguish. A good prognostic prediction model can usually achieve a C-index value of more than 0.7.

[0164] In order to evaluate the prediction accuracy of the model at different time points, the time-dependent ROC curve and its area under it (AUC value) were calculated, as Figure 3 As shown. Usually, three time points of 1 year, 3 years and 5 years are selected to analyze the short-term, medium-term and long-term forecasting capabilities of the model.

[0165] In addition, survival analysis was used to evaluate the risk stratification effect of the model. For example, patients were divided into high-risk and low-risk groups according to the risk scores predicted by the model, and the Kaplan-Meier method was used to draw survival curves, and the log-rank test was used to evaluate the statistical significance of the survival difference between the two groups of patients.

[0166] In addition, the robustness of the model can be evaluated by analyzing its predictive effect in different subgroups. For example, patients can be divided into different subgroups according to their age, gender, tumor stage and other clinical characteristics, and the prediction accuracy of the model can be calculated separately. This stratified analysis can help understand the applicability of the model in different patient groups and provide more targeted guidance for clinical application.

[0167] S6.2: Build comparison models and calculate performance

[0168] In order to demonstrate the superiority of the deep learning model, this application constructs a traditional Cox regression model and a random forest survival model as controls. These models use the same feature set and training data to calculate their C-index values ​​on an independent validation set.

[0169] S6.3: Comparison of performance metrics of different models

[0170] By comparing the performance indicators of different models, the advantages of deep learning models can be intuitively demonstrated. Practice shows that deep learning models can usually obtain higher C-index values ​​and AUC values, thanks to their ability to capture complex nonlinear relationships in gene expression data.

[0171] S7: Perform external verification

[0172] Specifically, external verification includes:

[0173] S7.1: Obtain multiple external validation sets

[0174] External validation of the model is an important part of evaluating its practical application value. In order to ensure that the model has good generalization ability, this application uses multiple independent external validation sets for comprehensive evaluation. These validation sets include data from different ethnic groups, different regions, and different sequencing platforms, which can fully verify the scope of application and stability of the model.

[0175] For example, first, multiple external validation sets are obtained from the GEO database and the ICGC database. These data sets include pancreatic cancer data sets for Asian populations, pancreatic cancer data sets for European and American populations, and data from different sequencing platforms (such as chip platforms and high-throughput sequencing platforms). Then, each external validation set needs to undergo the same data preprocessing process as the training set, including steps such as missing value processing, batch effect elimination, and data standardization. This ensures that the external validation data is comparable to the training data.

[0176] S7.2: Evaluate the model's predictive performance on different data

[0177] When evaluating the prediction effect of the model on different external validation sets, focus on the consistency and stability of the model performance. By calculating the C-index, time-dependent ROC curve, and survival analysis results on each validation set, you can fully understand the performance of the model on different data sets. If you find that the performance of the model on some validation sets has dropped significantly, you need to analyze the reasons and perform targeted optimizations. For example, you may need to consider the heterogeneity of the data set or adjust the parameter settings of the model.

[0178] S7.3: Retraining based on external prediction results

[0179] Depending on the results of external validation, the model may need to be retrained or fine-tuned. For example, if the model's performance on the Asian population dataset is found to be less than ideal, you can consider incorporating some Asian population data into the training set to update the model. During the retraining process, it is necessary to pay attention to maintaining the generalization ability of the model and avoid overfitting the data of a specific population or platform. Through this iterative optimization process, a model with good cross-population and cross-platform prediction capabilities is eventually obtained.

[0180] The external validation results of the model not only reflect its predictive performance, but also provide an important reference for clinical application. Detailed records of the model's performance on different validation sets can help clinicians understand the applicability of the model in their specific population, thereby better guiding individualized treatment decisions.

[0181] S8: Generate forecast report

[0182] like Figure 4 ,The generation of personalized prognosis prediction reports aims to provide intuitive and explainable prediction results, and provide valuable reference information for clinical decision-making.

[0183] Specifically, generating a forecast report includes:

[0184] S8.1: Calculate gene contribution

[0185] This can be achieved by analyzing the weight parameters and activation values ​​in the deep learning model. Specifically, the Integrated Gradients method is used to calculate the importance score of each input feature to the final prediction result. This method can quantitatively evaluate the role of each gene in the prediction process and help understand the decision-making basis of the model. For genes with a large contribution, special attention should be paid to their expression in patient samples, as these genes are often key factors affecting prognosis.

[0186] S8.2: Pathway enrichment analysis of high-risk genes

[0187] For those high-risk genes whose contribution exceeds the preset threshold, pathway enrichment analysis is performed. This step uses multiple bioinformatics databases, such as KEGG, Reactome, and Gene Ontology, to analyze the biological pathways and functions involved in these genes. The enrichment significance of each pathway is calculated by hypergeometric test to screen out significantly enriched biological processes. These enrichment analysis results help explain the molecular mechanism of the patient's prognostic risk and provide a theoretical basis for the formulation of subsequent treatment plans.

[0188] S8.3: Generate personalized prognostic prediction reports

[0189] Based on the above analysis results, a comprehensive personalized prognosis prediction report is generated. The report contains the following core contents: the patient's predicted survival risk score, survival curve prediction graph, expression level analysis of key prognosis-related genes, enrichment results of important biological pathways, and personalized treatment recommendations based on these findings. The report is presented in a combination of text and graphics, which ensures professionalism and is easy for clinicians to understand and use. For example, a heat map can be used to display the expression pattern of key genes, a bar chart can be used to display the pathway enrichment results, and a survival curve chart can be used to display the prognosis prediction results.

[0190] This personalized prognosis prediction report can not only help doctors better understand the prognosis risk of each patient, but also provide data support for the development of personalized treatment plans. This precision medicine program based on multi-omics data is expected to improve the treatment effect and quality of life of pancreatic cancer patients.

[0191] Embodiment 2

[0192] like Figure 5 As shown, the present application embodiment also provides a device for constructing a pancreatic cancer ferroptosis-related gene prognosis model, comprising:

[0193] A data acquisition module 10 is used to acquire a gene expression dataset and a clinical information dataset of pancreatic cancer patients;

[0194] A preprocessing module 20, used to perform data preprocessing on the gene expression data set and the clinical information data set to generate a standardized gene expression matrix;

[0195] A feature selection module 30 is used to perform differential expression analysis and survival analysis based on the standardized gene expression matrix and the ferroptosis pathway gene set, and screen out the ferroptosis feature gene set that is significantly correlated with prognosis;

[0196] A model building module 40 is used to build a deep learning prognosis prediction model based on the ferroptosis characteristic gene set;

[0197] The model evaluation module 50 is used to evaluate the performance of the deep learning prognosis prediction model using an independent validation set to obtain a model evaluation index.

[0198] The present application also provides a computer device, the computer device comprising:

[0199] at least one processor; and,

[0200] a memory communicatively connected to the at least one processor; wherein,

[0201] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned method for constructing a pancreatic cancer ferroptosis-related gene prognostic model based on gene expression data.

[0202] An embodiment of the present application also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the above-mentioned method for constructing a pancreatic cancer ferroptosis-related gene prognosis model based on gene expression data.

[0203] An embodiment of the present application also provides a computer program product, including computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the method for constructing a pancreatic cancer ferroptosis-related gene prognosis model based on gene expression data are implemented.

[0204] The terms "first", "second", "third", "fourth", "1", "2", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than that shown or described in the drawings.

[0205] It should be understood that, although each operation step is indicated by arrows in the flowchart of the embodiment of the present application, the implementation order of these steps is not limited to the order indicated by the arrows. Unless clearly stated herein, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be performed in other orders according to demand. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on actual implementation scenarios. Some or all of these sub-steps or stages may be executed at the same time, and each sub-step or stage in these sub-steps or stages may also be executed at different times respectively. In different scenarios of execution time, the execution order of these sub-steps or stages may be flexibly configured according to demand, and the embodiment of the present application does not limit this.

[0206] The above is only an optional implementation method for some implementation scenarios of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of the present application, other similar implementation methods based on the technical ideas of the present application are also within the protection scope of the embodiments of the present application.

Claims

1. A method for constructing a prognostic model of ferroptosis-related genes in pancreatic cancer, characterized in that: include: Obtain gene expression datasets and clinical information datasets of pancreatic cancer patients; Performing data preprocessing on the gene expression dataset and the clinical information dataset to generate a standardized gene expression matrix; According to the standardized gene expression matrix, combined with the ferroptosis pathway gene set, differential expression analysis and survival analysis are performed to screen out the ferroptosis characteristic gene set that is significantly correlated with prognosis; Constructing a deep learning prognosis prediction model based on the ferroptosis characteristic gene set; The performance of the deep learning prognosis prediction model was evaluated using an independent validation set to obtain model evaluation indicators.

2. The method according to claim 1, characterized in that The gene expression data set is subjected to data preprocessing, including: Removing gene expression data with a missing value ratio exceeding 20% ​​in the gene expression data set; The KNN algorithm was used to fill the missing values ​​of the remaining gene expression data; The ComBat algorithm was used to normalize the filled gene expression data; The clinical information data set is preprocessed, including: Unifying the calculation unit of survival time in the clinical information data in the clinical information dataset into months; Encoding the survival status in the clinical information data; Integrating the clinical characteristic information of the patient in the clinical information data; Generate a normalized gene expression matrix including: The processed gene expression data and clinical information data were matched according to the sample ID to generate a standardized gene expression matrix.

3. The method according to claim 2, characterized in that The step of performing differential expression analysis based on the standardized gene expression matrix and combining the ferroptosis pathway gene set includes: matching the normalized gene expression matrix with the ferroptosis pathway gene set; Calculate the expression difference of each matched gene between tumor tissue and normal tissue to obtain the differential expression analysis results; The Benjamini-Hochberg method was used to calculate the FDR value of the differential expression analysis results; The results of differential expression analysis were corrected according to the FDR value.

4. The method according to claim 3, characterized in that The method of performing survival analysis based on the standardized gene expression matrix and combining the ferroptosis pathway gene set includes: Perform univariate Cox regression analysis on each matched gene to obtain the Cox regression analysis results; According to the results of Cox regression analysis, genes significantly associated with survival time were screened; Kaplan-Meier survival analysis was performed on genes significantly correlated with survival time to obtain survival analysis results.

5. The method according to claim 4, characterized in that The ferroptosis characteristic gene set screened out and significantly associated with prognosis includes: According to the results of differential expression analysis and survival analysis, the basic framework of LASSO regression model was determined; Set the candidate range of the penalty parameter λ of the basic framework; For the ferroptosis pathway gene set, the 10-fold cross-validation method was used to select the optimal λ value; The optimal λ value was used to retrain the LASSO regression model on the ferroptosis pathway gene set to obtain the regression coefficient of each gene; The features of genes with non-zero regression coefficients were extracted to form a ferroptosis feature gene set.

6. The method according to claim 1, characterized in that The deep learning prognosis prediction model is constructed based on the ferroptosis characteristic gene set, comprising: Designing a deep learning model architecture, wherein the deep learning model architecture includes an input layer, a fully connected layer, a Dropout layer, and an output layer, and the dimension of the input layer is the same as the number of genes in the ferroptosis characteristic gene set; Combine the Cox partial likelihood loss function and the ranking loss function to define the multi-task learning objective of the deep learning model architecture; Add L1 and L2 regularization terms to the multi-task learning objective; According to the multi-task learning objectives and the mini-batch stochastic gradient descent algorithm, the deep learning model architecture is trained and optimized to obtain a deep learning prognosis prediction model.

7. The method according to claim 1, characterized in that The use of an independent validation set to evaluate the performance of the deep learning prognosis prediction model includes: Calculating the model performance indicators of the deep learning prognostic prediction model on an independent validation set, the model performance indicators including: C-index, time-dependent ROC curve and survival analysis; The Cox regression model and random forest survival model were constructed using the data from the independent validation set. Calculate the C-index of the Cox regression model on an independent validation set and the C-index of the random forest survival model on an independent validation set; Comparing the C-index of the deep learning prognostic prediction model, the C-index of the Cox regression model, and the C-index of the random forest survival model; The performance of the deep learning prognostic prediction model was determined based on the comparison results, time-dependent ROC curve and survival analysis.

8. The method according to claim 7, characterized in that Also includes: Acquire multiple external validation sets, wherein the multiple external validation sets include external validation sets of data from different ethnic groups and external validation sets of data from different sequencing platforms; Evaluate the external prediction effect of the deep learning prognostic prediction model on data from different ethnic groups and different sequencing platforms based on multiple external validation sets; According to the external prediction effect, the deep learning prognosis prediction model is retrained.

9. The method according to claim 1, characterized in that: Also includes: Calculate the contribution of each gene to the prediction results; Pathway enrichment analysis was performed on high-risk genes whose contribution was greater than the threshold; Generate a personalized prognosis prediction report based on the results of pathway enrichment analysis.

10. A device for constructing a prognostic model of ferroptosis-related genes in pancreatic cancer, characterized in that: include: A data acquisition module, used to obtain gene expression data sets and clinical information data sets of pancreatic cancer patients; A preprocessing module, used to perform data preprocessing on the gene expression dataset and the clinical information dataset to generate a standardized gene expression matrix; A feature selection module, for performing differential expression analysis and survival analysis based on the standardized gene expression matrix in combination with the ferroptosis pathway gene set, and screening out the ferroptosis feature gene set significantly associated with prognosis; A model building module, used to build a deep learning prognosis prediction model based on the ferroptosis characteristic gene set; The model evaluation module is used to evaluate the performance of the deep learning prognosis prediction model using an independent validation set to obtain model evaluation indicators.

Images