Method for developing and evaluating lung cancer detection and diagnosis by using serum proteome data
By integrating database data and utilizing various algorithms, this technique addresses the invasiveness and subjectivity issues of existing lung cancer diagnosis methods, thereby improving the accuracy and reliability of lung cancer diagnosis.
Patent Information
- Application Number
- CN202511165634.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-20
- Publication Date
- 2025-12-09
AI Technical Summary
In existing technologies, early diagnosis methods for lung cancer are highly invasive, time-consuming, and rely on subjective interpretation. The complexity and high-dimensionality of serum proteomics data lead to the instability of biomarker identification and diagnostic models.
Data from the GEO and TCGA databases were integrated, and differential expression analysis was performed using the limma package. Models were constructed using the nnet, glmnet, sgboost, and svm algorithms. The common protein SLC16A4, which is consistent across models, was screened as a core biomarker for lung cancer diagnosis. Performance indicators were evaluated by combining multiple models.
This improved the accuracy and reliability of lung cancer diagnosis and ensured the effectiveness and generalization ability of the model on different validation sets.
Smart Images
Figure CN121096441A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of computational models, and in particular to a method for developing and evaluating lung cancer detection and diagnosis using serum proteomic data. BACKGROUND
[0002] Lung cancer is one of the most common and deadly cancers worldwide, causing a significant impact on global health. Despite advances in early detection and treatment strategies, patients often have poor prognosis due to the fact that lung cancer is often diagnosed at an advanced stage. Traditional diagnostic methods, such as imaging and biopsy, while valuable, are often invasive, time-consuming, and subject to subjective interpretation. Therefore, there is an increasing demand for non-invasive, efficient, and reliable diagnostic methods.
[0003] In recent years, proteomics, particularly serum proteomics, has emerged as a promising tool for early detection of various cancers, including lung cancer. Serum proteomic profiles provide important insights into molecular changes associated with tumor development, providing potential biomarkers for cancer diagnosis. However, the complexity and high-dimensional nature of proteomic data pose significant challenges in accurately identifying biomarkers and developing robust diagnostic models, requiring improvement. SUMMARY
[0004] The present application aims to provide a method for developing and evaluating lung cancer detection and diagnosis using serum proteomic data to address the problems raised in the background.
[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:
[0006] A method for developing and evaluating lung cancer detection and diagnosis using serum proteomic data, comprising the following steps:
[0007] Step S1: Integrate proteomic samples from the GEO database (Gene Expression Omnibus) and lung cancer tissue samples from the TCGA database (Cancer Genome Atlas) to construct a dataset, wherein the proteomic samples from the GEO database serve as the healthy control group and the lung cancer tissue samples from the TCGA database serve as the cancer group.
[0008] Step S2: Perform differential expression analysis on the dataset using the limma package in R language to identify significantly differentially expressed proteins between the cancer group and the healthy control group, forming a differential protein feature set.
[0009] Step S3: Construct models (based on nnet / glmnet / sgboost / svm algorithms) using different algorithms to achieve classification of the cancer group and the control group, and train the models based on the differential protein feature set.
[0010] Step S4, extract the common proteins consistent with the models constructed by different algorithms (such as SLC16A4) as the core biomarkers for lung cancer diagnosis.
[0011] As a further scheme of the present application: in step S1, the GSE168198 proteomics samples are obtained from the GEO database, the first phase is based on the GPL29809 platform, and the second phase is based on the GPL29810 platform. Meanwhile, the lung adenocarcinoma (LUAD) and lung squamous carcinoma (LUSC) tissue samples are extracted from the TCGA database to construct a data set, wherein the proteomics samples of the GEO database are used as the healthy control group, and the lung adenocarcinoma and lung squamous carcinoma tissue samples of the TCGA database are used as the cancer group.
[0012] As a further scheme of the present application: in step S2, the limma package in R language is used to perform differential expression analysis on the cancer group and the healthy control group: first, the protein expression values are logarithmically transformed (log2(x+1)). Second, the data is filtered according to the disease types of the samples, including LungCancer and NegativeControl. Further, the limma package is used for differential analysis and the differential expression results are extracted, and the significantly different proteins are identified based on the preset threshold (logFC>0.3 and adj.P.Val<0.05). Finally, the up-regulated and down-regulated proteins are extracted to form a differential protein feature set.
[0013] As a further scheme of the present application: in step S3, based on the differential protein feature set, four algorithms nnet (neural network), glmnet (elastic network), sgboost (gradient boosting), and svm (support vector machine) are applied to construct models, and the differential protein feature set is divided into training set and test set according to 7:3, ensuring that the training set contains the diagnosis label, and the target variable is converted into factor type to adapt to the model. The training set and the test set are standardized, including centering and scaling, to eliminate the dimensional difference between the features.
[0014] As a further scheme of the present application: in step S3, the train function in the caret package is used for training in the model training stage.
[0015] As a further scheme of the present application: in step S4, the Top5 high importance proteins of the models (nnet / glmnet / sgboost / svm algorithms) constructed by different algorithms are extracted through feature importance analysis (varImp function), and the common proteins (such as SLC16A4) consistent across models are screened as the core biomarkers for lung cancer diagnosis.
[0016] As a further scheme of the present application: the method for developing and evaluating lung cancer detection and diagnosis by using serum proteome data further comprises:
[0017] Step S5, the model performance is evaluated on three types of independent verification sets, including a healthy control group (only healthy samples), a disease control group (only other disease samples), and a mixed control group (healthy + disease samples). The three types of independent verification sets are uniformly preprocessed to ensure data standardization and comparability. The trained models constructed by different algorithms are used to analyze the three types of independent verification sets. During the evaluation process, the performance measurement indicators of the model include accuracy, sensitivity, specificity, precision, F1-score (a comprehensive indicator for evaluating the performance of a classification model), and AUC (area under the curve, a core indicator for evaluating the performance of a binary classification model).
[0018] Compared with the prior art, the beneficial effects of the present application are: the present application uses models constructed by different algorithms to identify the differential proteins in healthy data and lung cancer data, and ranks them. Based on the final Top5 high importance differential proteins of different models, common proteins (such as SLC16A4) consistent across models are screened as core biomarkers for lung cancer diagnosis. Through comprehensive judgment of multiple models, the accuracy of the core biomarkers for lung cancer diagnosis is ensured. BRIEF DESCRIPTION OF DRAWINGS
[0019] Figure 1 A schematic diagram for analyzing the differentially expressed proteins in the plasma proteins of lung cancer patients in the GSE168198 data set.
[0020] Figure 2 A performance curve diagram of the diagnostic models constructed by four machine learning algorithms is shown in the ROC.
[0021] Figure 3 A performance curve diagram of the four models in the internal test set is shown in the ROC.
[0022] Figure 4 A prediction performance curve diagram of the four models in the different control data.
[0023] Figure 5 A prediction performance curve diagram of the four models in the external test set. DETAILED DESCRIPTION
[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0025] A method for developing and evaluating lung cancer detection and diagnosis using serum proteome data, comprising the following steps:
[0026] Step S1, integrate proteomics samples of GEO database (public gene expression database) and lung cancer tissue samples of TCGA database (cancer genome database) to construct a dataset, wherein the proteomics samples of GEO database are used as the healthy control group and the lung cancer tissue samples of TCGA database are used as the cancer group.
[0027] Step S2, perform differential expression analysis on the dataset by using the limma package in R language to identify significantly differentially expressed proteins between the cancer group and the healthy control group, and form a differential protein feature set.
[0028] Step S3, construct models (based on nnet / glmnet / sgboost / svm algorithms) by different algorithms to realize the classification of the cancer group and the control group, and train the models based on the differential protein feature set.
[0029] Step S4, extract common proteins (such as SLC16A4) consistent in the models constructed by different algorithms as the core biomarkers for lung cancer diagnosis.
[0030] In this embodiment: in step S1, GSE168198 proteomics samples are obtained from the GEO database, the first phase is based on GPL29809 platform, and the second phase is based on GPL29810 platform. At the same time, lung adenocarcinoma (LUAD) and lung squamous carcinoma (LUSC) tissue samples are extracted from the TCGA database to construct a dataset, wherein the proteomics samples of the GEO database are used as the healthy control group and the lung adenocarcinoma and lung squamous carcinoma tissue samples of the TCGA database are used as the cancer group.
[0031] By selecting a variety of materials, a multi-dimensional and multi-source dataset is constructed to ensure the comprehensiveness and reliability of the analysis.
[0032] In this embodiment: in step S2, the limma package in R language is used to perform differential expression analysis on the cancer group and the healthy control group: first, the protein expression values are logarithmically transformed (log2(x+1)). Second, the data is filtered according to the disease types of the samples, including LungCancer, NegativeControl. Further, the limma package is used for differential analysis and the differential expression results are extracted, and the significantly differentially expressed proteins are identified based on the preset threshold (logFC>0.3 and adj.P.Val<0.05). Finally, the up-regulated and down-regulated proteins are extracted to form a differential protein feature set.
[0033] Significantly changed biomarkers, i.e. differential proteins, are identified to provide feature input for the model.
[0034] In this embodiment: in step S3, based on the differential protein feature set, four algorithms nnet (neural network), glmnet (elastic network), sgboost (gradient boosting), and svm (support vector machine) are applied to construct models, the differential protein feature set is divided into a training set and a test set at a ratio of 7:3, the diagnostic label is included in the training set, and the target variable is converted into a factor type to adapt to the model. The training set and the test set are standardized, including centering and scaling, to eliminate the dimensional difference between features.
[0035] In this embodiment: in step S3, in the model training stage, the train function in the caret package is used for training.
[0036] A high-precision diagnostic model is developed to capture complex data patterns by taking advantage of different algorithms.
[0037] In this embodiment: in step S4, the Top5 high importance proteins of the models (nnet / glmnet / sgboost / svm algorithms) constructed by different algorithms are extracted through feature importance analysis (varImp function), and the common proteins (such as SLC16A4) consistent across models are screened as core biomarkers for lung cancer diagnosis.
[0038] Through the joint judgment of multiple models, the core biomarkers strongly related to lung cancer diagnosis are locked.
[0039] In this embodiment: the development and evaluation of the lung cancer detection and diagnosis method using serum proteome data further comprises:
[0040] Step S5: evaluate the model performance on three types of independent validation sets, including a healthy control group (only healthy samples), a disease control group (only other disease samples), and a mixed control group (healthy+disease samples). The three types of independent validation sets are uniformly preprocessed to ensure data standardization and comparability. The trained models constructed by different algorithms are used to analyze the three types of independent validation sets. In the evaluation process, the performance measurement indicators of the model include accuracy, sensitivity, specificity, precision, F1-score, and AUC (area under the curve).
[0041] Evaluate the model performance (AUC / accuracy and other indicators) in multiple independent validation sets (healthy / disease / mixed) to test the generalization ability and clinical applicability of the model.
[0042] Actual verification:
[0043] Serum proteomic data and corresponding clinical data of lung cancer patients and normal and other disease control volunteers were obtained from GEO database (GSE168194). The study was divided into two phases, phase one was based on GPL29809 annotated protein microarray platform, which could detect 16048 proteins, 16 samples based on this platform were used as external validation set. Phase two was based on GPL29810 annotated protein microarray platform, which could detect 89 proteins, 180 samples based on this platform were split into training set (n=144) and internal test set (n=36) according to 7:3. The basic demographic and clinical data of training set, internal test set and validation set were summarized in Table 1. There were no significant differences in major clinicopathological features between training set, test set and the whole TCGA-LUAD dataset (Table 1) as expected.
[0044] Table 1. Distribution table of samples in training set, test set and validation set
[0045]
[0046] See Figure 1 , significantly changed proteins between cancer patients and negative control group. Based on these proteins, diagnostic models of four machine learning algorithms were constructed: nnet, glmnet, sgboost and svm. The performance of these models was evaluated by receiver operating characteristic (ROC) curve.
[0047] See Figure 2 , XGBoost model showed the highest diagnostic performance with an AUC value of 0.985, followed by nnet with an AUC value of 0.970. The AUC values of svm and glmnet models were 0.897 and 0.861, respectively. These models were further validated on the internal test set, and the ROC curves are shown in Figure 3 . The nnet model maintained an AUC value of 0.886, while the AUC values of glmnet, XGBoost and svm models were 0.827, 0.840 and 0.830, respectively.
[0048] See Figure 4 , the predictive performance of four machine learning models (XGBoost, glmnet, nnet and svm) on different control datasets was analyzed. The results showed that these models performed best on the negative control dataset, with nnet and XGBoost achieving the highest AUC values of 0.923 and 0.918, respectively. In contrast, when tested on a dataset containing only disease controls, the AUC values of these two algorithms were lower, at 0.756 and 0.774, respectively. When both negative and disease controls were included, the AUC values of nnet and XGBoost increased to 0.839 and 0.846, respectively.
[0049] See Figure 5 Further evaluation of the performance of the four diagnostic models in a small external cohort consisting of 5 healthy individuals and 5 lung cancer patients was performed. The results showed that all models performed well on the external dataset, with the svm algorithm showing the highest AUC of 0.922. This was followed by XGBoost with an AUC of 0.906, while the AUC values for glmnet and nnet were 0.875 and 0.867, respectively. These findings indicate that all four models exhibit strong predictive performance, with svm and XGBoost providing the most accurate classification results.
[0050] It is apparent for the person skilled in the art that the present application is not limited to the details of the above-described exemplary embodiments, but that it can be implemented in other concrete forms without departing from the spirit or the essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and not restrictive.
[0051] Furthermore, it should be understood that although the present specification is described in terms of embodiments, not every implementation embodies only one independent technical solution, and the description of the specification is only for the sake of clarity, and the person skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that the person skilled in the art can understand.
Claims
1. A method for developing an evaluation method for lung cancer detection diagnosis using serum proteome data, characterized by, The method for developing and evaluating lung cancer detection and diagnosis methods using serum proteome data comprises the following steps: Step S1, integrate the proteomics samples of the GEO database and the lung cancer tissue samples of the TCGA database, and construct a dataset, wherein the proteomics samples of the GEO database are used as the healthy control group, and the lung cancer tissue samples of the TCGA database are used as the cancer group; Step S2, use the limma package in R language to perform differential expression analysis on the dataset, identify the significantly differentially expressed proteins between the cancer group and the healthy control group, and form a differential protein feature set; Step S3, construct a model by different algorithms, realize the classification of the cancer group and the control group, and train the model based on the differential protein feature set; Step S4, extract the common proteins consistent in the models constructed by different algorithms as the core biomarkers for lung cancer diagnosis.
2. The method for developing an evaluation method for lung cancer detection diagnosis using serum proteome data according to claim 1, characterized in that, In step S1, the GSE168198 proteomics samples are obtained from the GEO database, the first phase is based on the GPL29809 platform, and the second phase is based on the GPL29810 platform; at the same time, the lung adenocarcinoma and lung squamous carcinoma tissue samples are extracted from the TCGA database, and a dataset is constructed, wherein the proteomics samples of the GEO database are used as the healthy control group, and the lung adenocarcinoma and lung squamous carcinoma tissue samples of the TCGA database are used as the cancer group.
3. The method for developing an evaluation method for lung cancer detection diagnosis using serum proteome data according to claim 1, characterized in that, In step S2, the limma package in R language is used to perform differential expression analysis on the cancer group and the healthy control group: first, the protein expression values are logarithmically transformed; second, the data is filtered according to the disease types of the samples, including LungCancer, NegativeControl; further, the limma package is used for differential analysis and the differential expression results are extracted, and the significantly differentially expressed proteins are identified based on the preset threshold; finally, the up-regulated and down-regulated proteins are extracted to form a differential protein feature set.
4. The method for developing an evaluation method for lung cancer detection diagnosis using serum proteome data according to claim 1, characterized in that, In step S3, based on the differential protein feature set, four algorithms nnet, glmnet, sgboost and svm are applied to construct a model, the differential protein feature set is divided into a training set and a test set according to a ratio of 7:3, the training set contains diagnostic labels, the target variable is converted to a factor type to adapt to the model; the training set and the test set are standardized, including centering and scaling, to eliminate the dimensional differences between the features.
5. The method for developing an evaluation method for lung cancer detection diagnosis using serum proteome data according to claim 4, characterized in that, In step S3, the train function in the caret package is used for training in the model training stage.
6. The method for developing an evaluation method for lung cancer detection diagnosis using serum proteome data according to claim 1, wherein, In step S4, the Top5 high importance proteins of the model constructed by different algorithms are extracted through feature importance analysis, the common proteins consistent across models are screened as the core biomarkers for lung cancer diagnosis.
7. The method for developing an evaluation method for lung cancer detection diagnosis using serum proteome data according to any one of claims 1 to 6, characterized by, The method for developing and evaluating lung cancer detection and diagnosis methods using serum proteome data further comprises: Step S5, evaluating the model performance on three types of independent verification sets, including the healthy control group, the disease control group and the mixed control group, uniformly preprocessing the three types of independent verification sets to ensure the standardization and comparability of the data; using the models constructed by the trained different algorithms to analyze the three types of independent verification sets respectively, and in the evaluation process, the performance measurement indexes of the models include the accuracy, the sensitivity, the specificity, the precision, the F1-score and the AUC.