Application of plasma protein markers in pan-cancer auxiliary diagnosis products

Abstract

The application discloses application of plasma protein markers in a pan-cancer auxiliary diagnosis product and belongs to the field of biomedical technology.The plasma protein markers are a combination of CYP1A1, ITGB5, MAP3K11, MMP7 and TIMP4. The application adopts a SomaScan high-throughput proteomics platform to systematically screen five plasma proteins from thousands of plasma proteins, and combines a LightGBM machine learning algorithm to construct a diagnosis model, so that efficient differential diagnosis (AUC=0.943) of pan-cancer is realized, and objective basis is provided for clinical large-scale early screening.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of biomedical technology, specifically to the application of plasma protein biomarkers in pan-cancer auxiliary diagnostic products. Background Technology

[0002] Cancer is one of the leading causes of death worldwide. Early detection, diagnosis, and treatment are the most effective means to reduce cancer mortality and improve patient prognosis. However, due to the high heterogeneity of tumors occurring in different tissues and organs, there is currently a lack of highly efficient methods in clinical practice that can simultaneously screen for a wide range of malignant tumors.

[0003] The current conventional cancer screening and auxiliary diagnosis system has the following pain points: (1) Single organ screening mode is time-consuming, laborious and has poor compliance: The clinically recommended early screening methods are mostly single examinations for specific organs, such as low-dose spiral CT (lung cancer), gastroscopy (digestive tract tumors), mammography (breast cancer), abdominal ultrasound (liver / kidney / pancreatic tumors), etc. If the general population wants to complete a full set of cancer screening, they not only face the problems of multiple appointments and long queues and high costs, but also have to endure the pain brought by endoscopy and the cumulative radiation risk brought by multiple CT scans, resulting in extremely low screening compliance in a large population. (2) Traditional broad-spectrum tumor markers have poor performance: The so-called "broad-spectrum" blood tumor markers currently available in clinical practice (such as CEA, CA125, CA19-9, etc.) have extremely low sensitivity in most early solid tumors and are easily interfered with by benign diseases such as inflammation, with high false positive and false negative rates, and cannot be competent for early warning of pan-cancer types. (3) Emerging multi-cancer early detection (MCED) technology has high threshold: In recent years, liquid biopsy technology based on cell-free DNA (cfDNA) methylation or gene mutation has shown potential in pan-cancer screening, but these technologies often rely on extremely high-depth next-generation sequencing (NGS), with long detection cycles and extremely high costs, and the ability to capture trace amounts of DNA fragments released by very early small tumors is still limited.

[0004] In summary, existing technologies suffer from limitations such as fragmented screening methods, low sensitivity of traditional blood indicators, and high costs of emerging gene testing. There is an urgent need for a non-invasive, cost-effective, and broad-spectrum combination of efficient plasma protein biomarkers for the auxiliary diagnosis of pan-cancer. Summary of the Invention

[0005] Therefore, embodiments of the present invention provide the application of plasma protein biomarkers in pan-cancer auxiliary diagnostic products.

[0006] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0007] According to a first aspect of the present invention, the present invention provides the application of plasma protein biomarkers in pan-cancer auxiliary diagnostic products, wherein the plasma protein biomarkers are a combination of CYP1A1, ITGB5, MAP3K11, MMP7 and TIMP4.

[0008] Furthermore, the pan-cancer includes bladder cancer, breast cancer, esophageal cancer, lung cancer, liver cancer, osteosarcoma, pancreatic cancer, and kidney cancer.

[0009] According to a second aspect of the present invention, the present invention provides a pan-cancer auxiliary diagnostic product, the product comprising a reagent for detecting plasma protein markers, wherein the plasma protein markers are a combination of CYP1A1, ITGB5, MAP3K11, MMP7 and TIMP4.

[0010] Furthermore, the product also includes reagents for detecting the expression levels of CYP1A1, ITGB5, MAP3K11, MMP7, and TIMP4 proteins.

[0011] Furthermore, the product is a reagent kit.

[0012] According to a third aspect of the present invention, the present invention provides a method for constructing a pan-cancer auxiliary diagnostic model, comprising the following steps:

[0013] (1) Plasma samples from the case group and the control group were subjected to proteomics detection and quality control using the SomaScan 11k platform. The dataset was randomly divided into training set and validation set in a 7:3 ratio.

[0014] (2) The machine learning model is trained on the training set using all protein features, and the top 5 proteins are selected as the final modeling features based on the importance of the Gain feature.

[0015] (3) Using the final modeling features, age, education level, smoking status, and BMI as covariates, a machine learning model is constructed on the training set, the model is evaluated on the validation set, and the optimal classification threshold is determined by the Youden index.

[0016] Furthermore, the machine learning model is the LightGBM model.

[0017] Furthermore, in steps (2) and (3), the training parameters of the LightGBM model are set as follows: learning rate 0.05, maximum number of leaf nodes per tree 31, maximum number of iterations 500, and number of early stopping rounds 20.

[0018] Furthermore, the final modeling features are a combination of CYP1A1, ITGB5, MAP3K11, MMP7, and TIMP4.

[0019] The embodiments of the present invention have the following advantages:

[0020] This invention utilizes the SomaScan high-throughput proteomics platform to systematically screen five plasma proteins (CYP1A1, ITGB5, MAP3K11, MMP7, and TIMP4) from thousands of plasma proteins. It then combines these proteins with the LightGBM machine learning algorithm to construct a diagnostic model, enabling efficient differential diagnosis of pan-cancer. This provides objective evidence for large-scale clinical cancer early screening. Furthermore, the detection cost of this method is low, making it feasible for large-scale promotion in primary healthcare systems. Attached Figure Description

[0021] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0022] Figure 1 The beehive plot of Shapley and Explanation Values ​​(SHAP) of the feature variables included in the diagnostic model is shown. The horizontal axis represents the SHAP value of each sample feature (i.e., the marginal contribution of the feature to the model's predicted output). Each point represents a sample, and the color from blue-purple to orange indicates that the feature value is from low to high.

[0023] Figure 2 The ROC curve and AUC value of the diagnostic model provided by this invention in the validation set. Detailed Implementation

[0024] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0025] Example 1: Screening and Model Building of Marker Combinations

[0026] (1) Sample collection

[0027] Fasting plasma samples were collected from 2074 patients with pathologically confirmed bladder cancer, breast cancer, esophageal cancer, lung cancer, liver cancer, osteosarcoma, pancreatic cancer, and kidney cancer (case group) and 981 healthy controls (control group). Among them: 278 cases of bladder cancer, with a median age of 69 years, and 83.1% male; TNM staging distribution was as follows: stage 0 166 cases (59.7%), stage I 56 cases (20.1%), stage II 29 cases (10.4%), stage III 18 cases (6.5%), and stage IV 9 cases (3.2%); 102 cases of breast cancer, with a median age of 53.5 years; TNM staging distribution was as follows: stage 0 13 cases (16.7%), stage I 27 cases (34.6%), stage II 30 cases (38.5%), and stage III 8 cases (10.3%); 75 cases of esophageal cancer... The median age of the patients was 67 years, with males accounting for 88.0%. The TNM stage distribution was as follows: Stage 0 (3 cases, 4%), Stage I (18 cases, 24.0%), Stage II (16 cases, 21.3%), Stage III (27 cases, 36.0%), and Stage IV (11 cases, 14.7%). There were 274 patients with liver cancer, with a median age of 64.5 years, and males accounting for 75.2%. The TNM stage distribution was as follows: Stage 0 (2 cases, 0.7%), Stage I (113 cases, 41.2%), Stage II (32 cases, 11.7%), Stage III (43 cases, 15.7%), and Stage IV (84 cases, 30.7%). Lung cancer... Of the 486 patients, the median age was 62 years, with males accounting for 51.2%. The TNM staging distribution was as follows: Stage 0 (6 cases, 1.2%), Stage I (323 cases, 66.5%), Stage II (63 cases, 13%), Stage III (83 cases, 17.1%), and Stage IV (11 cases, 2.3%). Of the 103 patients with osteosarcoma, the median age was 27 years, with males accounting for 72.8%. The TNM staging distribution was as follows: Stage I (13 cases, 12.7%), Stage II (69 cases, 67.6%), Stage III (9 cases, 8.8%), and Stage IV (11 cases, 10.8%). There were also 255 patients with pancreatic cancer. For example, the median age of the patients was 69 years, with males accounting for 58.4%. The TNM stage distribution was as follows: stage 0 4 cases (1.6%), stage I 60 cases (23.5%), stage II 49 cases (19.2%), stage III 53 cases (20.8%), and stage IV 89 cases (34.9%). There were 501 patients with renal cell carcinoma, with a median age of 60 years, with males accounting for 67.1%. The TNM stage distribution was as follows: stage 0 1 case (0.2%), stage I 442 cases (88.2%), stage II 23 cases (4.6%), stage III 24 cases (4.8%), and stage IV 11 cases (2.2%).

[0028] (2) High-throughput proteomics detection (SomaScan platform)

[0029] All samples were analyzed using the SomaScan 11k proteomics platform. This platform contains approximately 11,000 specific slow dissociation rate modified aptamers (SOMAmers) and can simultaneously quantify the expression levels of over 10,000 proteins in human plasma. The main detection process includes incubating plasma samples with a SOMAmer mixture to achieve specific binding; converting the protein binding signal into an aptamer quantitative signal through affinity capture and ultraviolet cleavage; quantifying using a microarray chip to obtain the relative fluorescence intensity (RFU) of each protein; and obtaining valid protein expression data after quality control (QC) filtering.

[0030] (3) Screening of key proteins

[0031] The raw data were stratified into case and control groups and randomly divided into training and validation sets in a 7:3 ratio. In the training set, all proteins were used as features input into the LightGBM model for training. The model was set with a learning rate of 0.05 and 31 leaves. The model underwent a maximum of 500 improvement iterations. Early termination was triggered if the AUC on the validation set did not improve within 20 consecutive iterations. The Gain metric was used to evaluate the importance of each feature (protein), and the top 5 proteins with the highest Gain values ​​were selected as the final modeling features.

[0032] The information on the top 5 proteins by Gain value is shown in Table 1 below.

[0033] Table 1

[0034]

[0035] Cytochrome P450 1A1: catalyzes the metabolism of exogenous substances and the carcinogenic activation of polycyclic aromatic hydrocarbons, while mediating the epoxidation of arachidonic acid and retinol oxidation, and participating in the regulation of lipid and vitamin metabolism.

[0036] Matrix metalloproteinase-7: degrades casein, type I, III, IV, and V gelatin and fibronectin, and activates procollagenase.

[0037] Integrin β5: Involved in cell-matrix adhesion, cell-cell adhesion, cell migration and TGF-β receptor signaling.

[0038] Mitogen-activated protein kinase kinase 11: drives cell proliferation and stress response by stimulating the JNK, p38 and ERK pathways, and indirectly supports BRAF activation in mitotic signaling.

[0039] Tissue inhibitor of metalloproteinases 4: It binds to the catalytic zinc cofactor of metalloproteinases (such as collagenase) and irreversibly inhibits their activity.

[0040] (4) Construction of auxiliary diagnostic model

[0041] A LightGBM model was constructed on the training set. The feature variables included the five proteins selected earlier, as well as covariates (age, education level, smoking status, and BMI). The SHAP values ​​of each variable are distributed as follows: Figure 1 As shown. The model performance was evaluated on the validation set, and the Youden index was used to determine the optimal classification threshold.

[0042] The validation set results show that the AUC of the model in this invention is 0.943 (95% confidence interval: 0.93, 0.96). Figure 2 The optimal threshold is 0.81; at this threshold, the model sensitivity is 0.86 and the specificity is 0.90. The above results demonstrate that the LightGBM model based on five plasma proteins constructed in this invention exhibits excellent pan-cancer discrimination ability, along with high sensitivity and specificity.

[0043] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. The application of plasma protein biomarkers in pan-cancer auxiliary diagnostic products, characterized in that, The plasma protein markers are a combination of CYP1A1, ITGB5, MAP3K11, MMP7, and TIMP4.

2. The application according to claim 1, characterized in that, The term "pan-cancer" includes bladder cancer, breast cancer, esophageal cancer, lung cancer, liver cancer, osteosarcoma, pancreatic cancer, and kidney cancer.

3. A pan-cancer auxiliary diagnostic product, characterized in that, The product includes reagents for detecting plasma protein markers, which are combinations of CYP1A1, ITGB5, MAP3K11, MMP7, and TIMP4.

4. The product according to claim 3, characterized in that, The product also includes reagents for detecting the expression levels of CYP1A1, ITGB5, MAP3K11, MMP7, and TIMP4 proteins.

5. The product according to claim 3, characterized in that, The product in question is a reagent kit.

6. A method for constructing a pan-cancer auxiliary diagnostic model, characterized in that, Includes the following steps: (1) Plasma samples from the case group and the control group were subjected to proteomics detection and quality control using the SomaScan 11k platform. The dataset was randomly divided into training set and validation set in a 7:3 ratio. (2) The machine learning model is trained on the training set using all protein features, and the top 5 proteins are selected as the final modeling features based on the importance of the Gain feature. (3) Using the final modeling features, age, education level, smoking status, and BMI as covariates, a machine learning model is constructed on the training set, the model is evaluated on the validation set, and the optimal classification threshold is determined by the Youden index.

7. The method for constructing a pan-cancer auxiliary diagnostic model according to claim 6, characterized in that, The machine learning model is the LightGBM model.

8. The method for constructing a pan-cancer auxiliary diagnostic model according to claim 7, characterized in that, In steps (2) and (3), the training parameters of the LightGBM model are set as follows: learning rate 0.05, maximum number of leaf nodes per tree 31, maximum number of iterations 500, and number of early stopping rounds 20.

9. The method for constructing a pan-cancer auxiliary diagnostic model according to claim 6, characterized in that, The final modeling features are a combination of CYP1A1, ITGB5, MAP3K11, MMP7, and TIMP4.

Images