Protein markers and methods and systems for predicting iodine uptake capacity of metastatic lesions of thyroid cancer

Abstract

The application discloses a protein marker for predicting iodine uptake capacity of a thyroid cancer metastatic focus, a prediction method and a system, and belongs to the technical field of tumor medicine. The application provides 23 protein markers capable of being used for predicting iodine uptake capacity of a thyroid cancer metastatic focus, and the specific prediction method is as follows: the expression level of the protein marker in the metastatic focus tissue is obtained; the protein expression amount or abundance value is subjected to data preprocessing and standardization; the standardized protein expression amount is input into a pre-trained logistic regression model to obtain an iodine uptake positive probability value of the thyroid cancer metastatic focus, and the iodine uptake capacity of the sample is judged based on pre-determined classification threshold comparison. The method can depict iodine uptake related characteristics of the metastatic focus at a molecular level, provides an objective, quantifiable and reproducible iodine uptake capacity prediction scheme, provides an effective auxiliary basis for radioiodine treatment decision and patient risk stratification, and has clear application value and popularization significance.

Description

Technical Field

[0001] This invention belongs to the field of oncology medical technology, specifically relating to a protein biomarker, prediction method, and system for predicting the iodine uptake capacity of thyroid cancer metastases. Background Technology

[0002] Differentiated thyroid carcinoma (DTC) is the most common malignant tumor of the endocrine system, with papillary thyroid carcinoma (PTC) accounting for more than 80% and being the most prevalent pathological type. For patients with metastases, radioactive iodine (I-131) therapy can achieve clinical cure and prolong life expectancy. However, 30-60% of these patients still experience radioactive iodine resistance, characterized by a lack of iodine uptake or gradual loss of iodine uptake as the disease progresses, leading to a poor prognosis. Therefore, accurately assessing the iodine uptake capacity of the lesions before treatment is crucial for avoiding ineffective treatment and improving patient prognosis.

[0003] Current methods for predicting response to radioactive iodine therapy or iodine uptake capacity mainly rely on clinicopathological features, imaging findings, serological indicators, and gene mutations. Overall accuracy and reproducibility are limited, and most methods are based on primary lesion information, making it difficult to accurately reflect the biological state of metastatic lesions. Metastatic lesions (such as lymph node metastases) may differ from primary lesions in their microenvironment and degree of differentiation, leading to heterogeneity in iodine uptake capacity among different lesions in the same patient.

[0004] As direct executors of cellular functions, proteins, through changes in their expression, modification state, and subcellular localization, are closely related to numerous cellular processes such as tumor differentiation, cellular stress responses, receptor tyrosine kinase pathways, apoptosis, and protein transport. Therefore, protein biomarkers may more directly reflect the iodine uptake characteristics of metastatic lesions. In recent years, with the development of proteomics technologies, especially the improved compatibility of data-independent acquisition (DIA) strategies with formalin-fixed and paraffin-embedded (FFPE) samples, high-throughput, reproducible protein analysis based on clinical samples has become possible.

[0005] However, for the specific clinical question of iodine uptake capacity in thyroid cancer metastases, there is still a lack of a method that can integrate proteomics features at the metastatic lesion sample level and construct a verifiable predictive model through machine learning to achieve objective identification of iodine uptake capacity in metastatic lesions and support treatment decisions. Therefore, developing a combination of biomarkers based on proteomics data and combining it with machine learning algorithms to predict iodine uptake capacity in thyroid cancer metastases and its application is of significant clinical importance and practical value. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the prior art and to provide a protein biomarker, prediction method, and system for predicting the iodine uptake capacity of thyroid cancer metastases.

[0007] The specific technical solution adopted in this invention is as follows:

[0008] In a first aspect, the present invention provides a protein biomarker for predicting the iodine uptake capacity of thyroid cancer metastases, wherein the protein biomarker is one or more of the following proteins: CDKN2C, CACTIN, AHCYL2, SORBS2, CENPV, LAD1, IQGAP2, URGCP, SDF2, TNFSF13, INPP5J, RAB7B, GOLM1, ARMCX3, ORMDL2, IGHA2, ENO3, IGKC, CNN1, CILP2, ZNF511, NGFR, and TNRC18.

[0009] Preferably, the protein biomarker includes at least one protein downregulated in iodine-free metastatic foci and at least one protein upregulated in iodine-free metastatic foci; wherein the downregulated protein is CDKN2C, CACTIN, AHCYL2, SORBS2, CENPV, IQGAP2, URGCP, TNFSF13, INPP5J, RAB7B, CNN1, CILP2, ZNF511, or TNRC18; and the upregulated protein is LAD1, SDF2, GOLM1, ARMCX3, ORMDL2, IGHA2, ENO3, IGKC, or NGFR.

[0010] Preferably, the protein biomarkers include at least the downregulated proteins CACTIN, CENPV, RAB7B, and ZNF511, and the upregulated proteins GOLM1 and ENO3.

[0011] Secondly, the present invention provides a method for predicting the iodine uptake capacity of thyroid cancer metastases based on the protein biomarkers described in the first aspect, the specific steps of which are as follows:

[0012] S1: Detect the expression levels of protein markers in tissue samples from metastatic papillary thyroid carcinoma to obtain the expression level or abundance value of each protein;

[0013] S2: Perform data preprocessing and standardization on the expression levels or abundance values ​​of each protein obtained in step S1 to obtain standardized expression levels;

[0014] S3: Input the standardized expression level obtained in step S2 into the pre-trained logistic regression model to obtain the iodine uptake positive probability value of thyroid cancer metastases. Compare the output iodine uptake positive probability value with the pre-determined classification threshold to determine the iodine uptake capacity of the sample.

[0015] Preferably, the expression level of the protein markers in step S1 is detected by mass spectrometry proteomics.

[0016] Preferably, the preprocessing in step S2 employs one or more of the following: missing value handling, normalization or log2 transformation for preliminary standardization, batch effect correction, outlier handling, or feature scaling.

[0017] Preferably, the standardization in step S2 adopts the Z-score method.

[0018] Preferably, the logistic regression model described in step S3 is a regularized logistic regression model for elastic networks, and the specific formula is as follows:

[0019]

[0020] In the formula: denoted as iodine uptake positive probability value; i represents the i-th sample, j represents the j-th protein, and x represents the positive probability value. ij βi represents the standardized expression level of the j-th protein in the i-th sample; β0 is the intercept; βj is the normalized expression level of the j-th protein in the i-th sample. j is the weighting coefficient of the j-th protein.

[0021] Preferably, if the output iodine uptake positive probability value is greater than or equal to a predetermined classification threshold, the metastatic lesion sample is determined to be iodine uptake positive; otherwise, it is determined to be iodine uptake negative.

[0022] Thirdly, the present invention provides a prediction system for the iodine uptake capacity of thyroid cancer metastases for implementing the prediction method described in the second aspect, comprising:

[0023] The data acquisition module is used to acquire the expression level or abundance value of protein markers in tissue samples of metastatic thyroid papillary carcinoma.

[0024] The data processing module is used to preprocess and standardize the expression levels or abundance values ​​of the obtained protein markers to generate standardized expression levels.

[0025] The model prediction module outputs the iodine uptake positivity probability value of the metastatic lesion tissue sample through a pre-trained logistic regression model, and generates a classification label of the iodine uptake capacity of the sample based on a preset classification threshold.

[0026] Compared with the prior art, the present invention has the following advantages:

[0027] (1) This invention provides a protein biomarker composition that can be used to predict the iodine uptake capacity of thyroid cancer metastases. By detecting these specific protein biomarkers and inputting them into a machine learning model, the probability of iodine uptake positivity and iodine uptake capacity stratification labels can be output before treatment, thereby providing an objective molecular-level predictive basis for the iodine uptake capacity of lesions. This effectively overcomes the limitations of relying solely on imaging experience interpretation or traditional pathological indicators, which are difficult to fully reflect the functional heterogeneity of tumors.

[0028] (2) In the prediction method provided by the present invention, the Elastic Net regularized logistic regression model is preferred. It can automatically screen key markers in complex data with high protein feature dimensions and strong correlations, and obtain a simple and robust combination of protein features. While ensuring high prediction performance, it enhances the interpretability of the model and effectively reduces the risk of overfitting.

[0029] (3) The prediction method provided by the present invention adopts a standardized sample processing, mass spectrometry acquisition and data preprocessing process (including quality control, missing value processing, normalization / standardization and batch effect assessment, etc.), which improves the stability and comparability of protein quantification data, makes the model input features more reliable, and thus improves the repeatability and generalizability of the prediction results.

[0030] (4) The protein biomarker combination, prediction method and related system provided by the present invention can be used to identify metastatic lesions that may not take in iodine or have insufficient iodine intake, thereby assisting clinicians in formulating more precise radioactive iodine treatment strategies, avoiding ineffective or inefficient treatment, improving the efficiency of individualized diagnosis and treatment, and providing important reference for subsequent treatment pathway selection and follow-up management. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the overall process of protein quantification and model training in this invention.

[0032] Figure 2 This is a schematic diagram of proteomics data quality control and batch consistency assessment in Example 2;

[0033] Figure 3 The heatmap shown in Example 2 illustrates the correlation between samples in the mixed pool.

[0034] Figure 4 This is a schematic diagram comparing tumor purity at different clinical groups and different sampling time points in Example 2. (a) is a comparison of tumor purity between the iodine-sensitive group and the iodine-refractory group; (b) is a comparison of tumor purity between the three subgroups in the iodine-refractory group and the iodine-sensitive group; and (c) is a comparison of tumor purity between lymph nodes before and after I-131 treatment.

[0035] Figure 5 This is a differential protein volcano plot of the training set in Example 3;

[0036] Figure 6 This is a heatmap of differentially expressed proteins in the training set of Example 3;

[0037] Figure 7 This is a diagram of the coefficient paths in the elastic network model of Example 3;

[0038] Figure 8 This is a cross-validation curve of the elastic network model in Example 3;

[0039] Figure 9 This is a schematic diagram of the regression coefficients of the 23 proteins screened in Example 3, used to show the direction and magnitude of the contribution of each feature protein to the prediction of iodine uptake capacity. The horizontal axis is the weight coefficient of each protein.

[0040] Figure 10 Box plot showing the expression differences of 14 upregulated proteins among the key characteristic proteins screened in Example 3 in iodine-positive and iodine-negative metastatic lesions;

[0041] Figure 11 Box plot showing the expression differences of nine downregulated proteins among the key characteristic proteins screened in Example 3 in iodine-positive and iodine-negative metastatic lesions;

[0042] Figure 12 The graph shows the receiver operating characteristic (ROC) curves of the elastic network model in Example 3 on the training set (a) and the validation set (b), where AUC is the area under the curve.

[0043] Figure 13 This is the confusion matrix diagram of the elastic network model in Example 3 on the validation set;

[0044] Figure 14 This is a calibration curve of the elastic network model in Example 3 on the validation set;

[0045] Figure 15 The image shown is a dimensionality reduction visualization from Example 3, where (a) is the actual iodine uptake state and (b) is the model-predicted iodine uptake state. Detailed Implementation

[0046] The present invention will be further described and illustrated below with reference to the accompanying drawings and specific embodiments. The technical features of each embodiment in the present invention can be combined accordingly without conflict. Those skilled in the art should understand that these specific embodiments and examples are for illustrative purposes only, and not for limiting the present invention.

[0047] Throughout this specification, unless otherwise specified, the terminology used herein should be understood as having the meaning commonly used in the art. Therefore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. In the event of any conflict, this specification shall prevail.

[0048] Unless otherwise specified, all non-patient-derived raw materials, reagents, instruments and equipment used in this invention can be obtained through commercial purchase or by existing methods.

[0049] A flowchart of the detailed implementation method can be found here. Figure 1 .

[0050] Example 1

[0051] (1) Sample selection

[0052] The initial sample set consisted of 11,821 patients diagnosed with thyroid cancer at our center. The samples were then screened sequentially for initial screening, metastatic lesion screening, and final screening to determine the patients who met the study criteria.

[0053] (a) Initial screening: 9,122 patients with incomplete medical records or who did not undergo I-131-related examinations / treatments were excluded, leaving 2,699 patients.

[0054] (b) Screening for metastatic lesions: Based on the results of postoperative pathological and imaging examinations (including but not limited to neck MRI, chest CT, whole-body bone scintigraphy, and / or PET, etc.), patients were assessed for the presence of local or distant metastases at the time of their initial surgery. 2626 patients without evidence of metastasis and those whose I-131 uptake was limited to the thyroid bed and who had achieved disease remission were excluded, leaving 73 patients.

[0055] (c) Final screening: The iodine uptake status of metastatic lesions was finally determined based on I-131 whole-body scintigraphy (WBS) and / or I-131 SPECT / CT, and 20 patients with non-papillary thyroid carcinoma (PTC) pathological types, patients with no positive metastatic tissue available for analysis, and patients with insufficient metastatic tissue for proteomics testing were excluded; in addition, 3 patients who did not receive I-131 treatment before resection of recurrent metastatic lesions were excluded.

[0056] After completing all screenings, 50 patients with papillary thyroid carcinoma (PTC) and clear metastatic lesions were finally included for subsequent studies.

[0057] (2) Iodine intake related grouping and labeling definitions

[0058] Based on the response to radioactive iodine therapy, the 50 included patients were further divided into two main groups: the iodine-sensitive group and the iodine-refractory group. The specific definitions are as follows: the iodine-sensitive group refers to metastatic lesions that show positive iodine uptake and achieve disease remission after radioactive iodine therapy; the iodine-refractory group refers to patients with persistent disease or poor treatment response after treatment.

[0059] The criteria for iodine refractory are based on the definition framework of the 2015 ATA guidelines, which states that metastatic lesions can be classified as such if any of the following conditions are met: (a) the metastatic lesion does not take up radioactive iodine from the outset; (b) the lesion previously had the ability to take up iodine, but lost this ability as the disease progresses; (c) there is significant heterogeneity in iodine uptake among different lesions; (d) the metastatic lesion is iodine-positive, but there is still metastatic progression / persistent disease, which can be classified as "iodine-positive but poorly treated".

[0060] Given the high heterogeneity of iodine-refractory papillary thyroid carcinoma, this embodiment further subdivides iodine-refractory patients into different subtypes to refine the study. Ultimately, the specific groupings and sample sizes for the 50 patients are as follows:

[0061] Iodine-sensitive group (Id): 7 cases.

[0062] Iodine-resistant group: a total of 43 cases, further subdivided into: 12 cases of iodine-positive but persistent disease group (ID); 8 cases of initial iodine-intolerant persistent disease group (iDF); and 23 cases of gradually losing iodine uptake capacity group (iDG).

[0063] (3) Tissue sample acquisition and pathological confirmation

[0064] All excised metastatic lesions were prepared and preserved as formalin-fixed paraffin-embedded (FFPE) blocks to obtain FFPE tissue samples of papillary thyroid carcinoma metastases for subsequent quantitative proteomics analysis.

[0065] For each patient, HE-stained sections corresponding to the metastatic lesions were obtained, and two thyroid pathologists independently reviewed and confirmed the nature of the metastatic lesions and tumor components. For the same patient, FFPE paraffin blocks with the highest tumor content were selected for subsequent testing. If metastatic tissue could be obtained for the same patient before and after I-131 treatment, one sample of metastatic tissue from before treatment and one sample from after treatment were selected for analysis. For each patient, only one sample of metastatic tissue with the highest tumor content was selected from before treatment and / or after treatment for analysis. The pathologists marked the boundaries of the tumor regions for subsequent proteomics sampling and analysis. Based on the results of I-131 whole-body scintigraphy (WBS) and / or I-131 single-photon emission computed tomography / computed tomography (SPECT / CT), all initial persistent iodine-deprived disease (iDF) metastases and progressive loss of iodine uptake (iDG) metastases after I-131 treatment were considered negative uptake metastases, while the remaining metastases were considered positive uptake metastases.

[0066] Example 2

[0067] This embodiment uses FFPE tissue samples from papillary thyroid carcinoma metastases obtained in Example 1 to perform quantitative proteomics detection, and conducts strict data quality control and preprocessing to construct a protein feature matrix for subsequent machine learning modeling.

[0068] (1) Experimental batch design and quality control

[0069] Considering the large number of samples and the long cycle of quantitative proteomics detection, to reduce the batch effect caused by instrument fluctuations, this embodiment allocates the samples to be tested by batch and performs the tests in batches. Specifically, all tumor tissue samples to be tested are allocated to 10 batches, and one sample is randomly selected from each batch to set up a technical replicate to evaluate the stability of sample preparation and mass spectrometry acquisition within the batch. At the same time, a homogeneous mixed thyroid tissue peptide pool is prepared and equally distributed to each batch for testing to evaluate the stability of the mass spectrometry platform between batches.

[0070] (2) Protein extraction and peptide preparation

[0071] (a) Tissue sampling: The FFPE tissue section of the papillary thyroid carcinoma metastasis was cut along the tumor boundary marked by the pathologist, and the tumor tissue was scraped for proteomics sample preparation.

[0072] (b) Dewaxing and rehydration: Heptane was used for dewaxing, followed by rehydration at room temperature in a gradient of 100%, 90%, and 75% ethanol.

[0073] (c) Decrosslinking: After rehydration, the protein was treated at 95°C for 30 min with 100 mM Tris-HCl buffer (pH=10.0) to reverse formalin-induced protein crosslinking, and then cooled to 4°C.

[0074] (d) Pressure Cycling Technology (PCT) Lysis and Enzymatic Digestion: The lysis buffer was a 100 mM triethylamine bicarbonate (TEAB) system containing urea, thiourea, tris(2-carboxyethyl)phosphonic acid hydrochloride (TCEP), and iodoacetamide (IAA). Lysis was performed using a Barocycler instrument at 45,000 psi and 30°C for 90 cycles (30 s of high pressure followed by 10 s of atmospheric pressure per cycle). Enzymatic digestion was then performed under PCT conditions: first with Lys-C (enzyme to substrate ratio 1:100), then with trypsin (enzyme to substrate ratio 1:50), for 120 cycles at 20,000 psi and 30°C (50 s of high pressure followed by 10 s of atmospheric pressure per cycle). The reaction was terminated with a final concentration of 1% trifluoroacetic acid.

[0075] (e) Peptide purification and quantification: The obtained peptide solution was desalted and purified using SOLA μ solid-phase extraction plates, and then purified using A... 280 The concentration was determined by absorbance method, and the peptide solution was stored at 4°C until analysis.

[0076] (3) DIA mass spectrometry data acquisition and quantification

[0077] Each injection yielded 200 ng of peptide, and DIA (dissociative intracellular emission) was acquired using a liquid chromatography-mass spectrometry (LC-MS) platform coupled with ion mobility separation. The peptide was loaded onto a C18 column (15 cm × 75 μm) for separation, with an effective chromatographic gradient of 60 min and a flow rate of 300 nL / min. Mobile phase B linearly increased from 5% to 27%, then increased from 27% to 40% within 10 min, and further increased to 80%. The ion mobility scan range was set to 0.7–1.3 Vs / cm. 2 The ion acquisition mass-to-charge ratio (m / z) range for both the first-stage mass spectrometer (MS1) and the second-stage mass spectrometer (MS2) was set to 100–1700 Th.

[0078] Raw data were retrieved and quantified using DIA-NN (v1.8.1) combined with a thyroid tissue-specific spectral library. Fixed modification was set to cysteine ​​carbamide methylation, and variable modification was set to methionine oxidation; peptide length ranged from 7 to 30 μm, precursor mass-to-charge ratio ranged from 300 to 1800, and fragment ion mass-to-charge ratio ranged from 200 to 1800; the precursor-level FDR was set to 1%, and other parameters were set to default.

[0079] (4) Data quality control and preprocessing

[0080] (a) Quality control and batch effect assessment of proteomic expression profiles. Pearson correlation coefficients (r) were calculated between technical replicates and their original samples, and between samples in inter-batch pools. Intra-batch and inter-batch stability were systematically assessed. Correlation results are available in [reference needed]. Figure 2 The nine scatter plots are correlation scatter plots of technical duplicate samples for each batch. Due to the small sample size, no technical duplicate samples were included in the tenth batch. r is the correlation coefficient.

[0081] according to Figure 2 The batch effect assessment results show that all Pearson correlation coefficients are not lower than 0.97, indicating good intra- and inter-batch stability, high data repeatability, and no significant batch effect was detected. Therefore, no batch correction is performed in this embodiment. If a batch effect is detected in subsequent analyses, correction methods such as ComBat will be used.

[0082] (b) To obtain a reliable and complete protein feature matrix, the following preprocessing was performed:

[0083] Proteins with a missing value ratio of less than 50% were defined as quantifiable proteins and then proceeded to the initial standardization and normalization process. Missing value imputation was performed using the K-Nearest Neighbor (KNN) algorithm (implemented using the SeqKnn tool). Subsequently, based on the UniProt database (v2023.03), protein identifiers (UniProtKB AC / ID) were uniformly converted to standard gene symbols. If multiple UniProt entries correspond to the same gene symbol, the median is taken as the protein abundance value corresponding to that gene symbol, forming a stable protein feature matrix.

[0084] (c) Analysis and validation of tumor purity in samples

[0085] To confirm sample quality and exclude potential confounding factors, the ESTIMATE algorithm was used to estimate the tumor purity of all samples. Results are shown below. Figure 4 The median tumor purity was 75.4%, with no significant differences in tumor purity across clinical groups and at different sampling time points, and all exceeding 50%. This confirms good sample quality, high tumor cell enrichment, and that the subsequent observed differences in protein expression are unlikely to be due to differences in the proportion of tumor cells between groups.

[0086] Example 3

[0087] This embodiment, based on the protein feature matrix obtained in Example 2 and the true labels of clinical iodine uptake status, constructs a machine learning classification model for predicting the iodine uptake capacity of thyroid cancer metastases, and evaluates the model's feature selection process and predictive performance. The overall modeling and evaluation process can be found in [link to example]. Figure 1 .

[0088] (1) Division of training set and validation set and determination of candidate features

[0089] All samples were randomly divided into training and validation sets in an 8:2 ratio. Protein differential expression analysis was performed only within the training set, using iodine uptake status (positive / negative) as the grouping variable. The selection criteria were p-value < 0.05 and |log2(fold change)| > 0.5. In this embodiment, 62 significantly differentially expressed proteins were selected as candidate features for subsequent modeling. The differential distribution and clustering patterns are described in [reference needed]. Figure 5 and Figure 6 .

[0090] Before model fitting, the protein expression levels or abundance values ​​of candidate features are Z-score standardized to reduce the impact of dimensional differences on model coefficient estimation.

[0091] (2) Elastic Net model training and feature selection

[0092] The logistic regression model (implemented using the glmnet tool) is trained using Elastic Net regularization. The optimal mixing parameters are determined through network search combined with five-fold cross-validation; in this embodiment, the optimal mixing parameter α is 0.4. Furthermore, to obtain a more concise and robust feature set, a λ value (λ_1se, 0.174 in this embodiment) within one standard error of the cross-validation error is selected as the final regularization strength. Figure 7 This is a schematic diagram illustrating the path of the coefficients of the elastic network model as a function of regularization, used to show the changing trends of the regression coefficients of each feature protein under different regularization strengths. Figure 7 In the diagram, the upper horizontal axis indicates the number of non-zero coefficients in the model, the vertical axis indicates the value of the coefficients, and the lower horizontal axis indicates the standardized coefficient vector. Each colored line represents one of the 23 proteins.

[0093] Figure 8 This is a cross-validation curve for the elastic network model, used to determine the optimal regularization parameter λ and select the final model. Figure 8 In the figure, the upper horizontal axis represents the number of variables corresponding to different λ values, the lower horizontal axis represents the logarithm of the λ penalty coefficient, and the vertical axis represents the cross-validation error. The upper horizontal axis value corresponding to the dashed line on the left side of the figure determines how many variables can be used for analysis.

[0094] Under optimal α and λ_1se, the model ultimately screened 23 proteins with non-zero coefficients, forming the predicted protein signatures. These proteins exhibited significantly different expression patterns between iodine-positive and iodine-negative metastatic lesions. Table 1 shows the names and corresponding weight coefficients of the 23 screened proteins. Figure 9 This diagram illustrates the weighting coefficients of 23 proteins, showing the direction and magnitude of each characteristic protein's contribution to the prediction of iodine uptake capacity. Protein signatures show significant expression differences between iodine-positive and iodine-negative metastatic lesions; representative differential expression details can be found in [link to relevant documentation]. Figure 10 .

[0095] Table 1. Names of the 23 selected characteristic proteins and their corresponding weight coefficients

[0096]

[0097] (3) Model prediction output and performance evaluation

[0098] The trained elastic network model is used to output the iodine uptake positive prediction probability for each sample, and further provides stratified labels for iodine uptake capacity.

[0099] (a) Elastic network model training: prediction probability calculation and prediction label generation

[0100] The protein expression levels of the training set samples were standardized using the method described in step (1) of Example 3, and then the iodine uptake positive prediction probability was calculated using the following formula. :

[0101]

[0102] In the formula, i represents the i-th sample, j represents the j-th protein, and x ij βi represents the standardized expression level of the j-th protein in the i-th sample; β0 is the intercept; βj is the normalized expression level of the j-th protein in the i-th sample. j is the weighting coefficient of the j-th protein.

[0103] In this embodiment, the intercept β0 of the logistic regression model obtained after training with the above training set samples is 0.32207, and the weight coefficient of each protein is shown in the rightmost column of Table 1.

[0104] In this embodiment, the classification threshold is preset to 0.5. The positive prediction probability is calculated for each sample. The predicted label is determined according to the following rules:

[0105] ;

[0106] Where Y i To predict label status.

[0107] (b) Model discrimination performance evaluation:

[0108] Based on the positive prediction probability, predicted label, and true label of the above training set samples, receiver operating characteristic (ROC) curves were plotted. Figure 12 In the figure, (a) and (b) are the ROC curves of the training set samples and the validation set samples, respectively. In this embodiment, the area under the curve (AUC) of the training set is 0.98 and the AUC of the validation set is 0.90, indicating that the model has excellent discrimination and good generalization ability.

[0109] (c) Assessment of classification accuracy and calibration:

[0110] Figure 13 This is a confusion matrix diagram of the elastic network model on the validation set, used to demonstrate the model's classification results and prediction accuracy. The calculated positive and negative prediction values ​​are both no less than 80%, confirming the accuracy and reliability of the classification results. Figure 14 The calibration curve of the elastic network model on the validation set is shown below. Figure 14 It can be seen that the model's predicted probability is consistent with the actual positive rate, and the probability output is accurate.

[0111] (d) Visualization of the discriminative power of protein signatures:

[0112] To visually demonstrate the discriminative power of the 23 protein signatures selected by the model, this embodiment employs Unified Manifold Approximation and Projection (UMAP) dimensionality reduction visualization analysis, and colors samples according to their actual iodine uptake state and the model's predicted labels. The results show that samples can form relatively clear separations in the low-dimensional space, and the predicted labels exhibit high consistency with the actual states. (See [link to relevant documentation]). Figure 15 .

[0113] In summary, this invention addresses the clinical challenge of assessing iodine uptake capacity in thyroid cancer metastases by establishing a predictive method and model framework centered on a combination of specific protein biomarkers. Through standardized proteomics analysis of metastatic tissue samples, coupled with rigorous quality control and data preprocessing, stable and reliable quantitative protein data can be obtained, providing consistent input features for subsequent machine learning modeling. Based on the construction of training and validation sets, and feature selection using a resilient network regularized logistic regression model, this invention generates a protein signature that can be used to predict iodine uptake status, outputting the prediction results in the form of iodine uptake positive probability values ​​and hierarchical labels. Performance evaluation shows that this predictive model exhibits good discriminative ability and predictive consistency in both training and validation data.

[0114] The results of the above embodiments show that the present invention can characterize the iodine uptake-related features of metastatic lesions at the molecular level, and provide an objective, quantifiable and reproducible iodine uptake capacity prediction scheme, thereby providing an effective auxiliary basis for radioactive iodine therapy decision-making and patient risk stratification, and has clear application value and promotion significance.

[0115] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A protein biomarker for predicting iodine uptake capacity in thyroid cancer metastases, characterized in that, The protein biomarkers are a combination of CDKN2C, CACTIN, AHCYL2, SORBS2, CENPV, LAD1, IQGAP2, URGCP, SDF2, TNFSF13, INPP5J, RAB7B, GOLM1, ARMCX3, ORMDL2, IGHA2, ENO3, IGKC, CNN1, CILP2, ZNF511, NGFR, and TNRC18 proteins.

2. A method for predicting iodine uptake capacity of thyroid cancer metastases using the protein biomarker according to claim 1, characterized in that, The specific steps are as follows: S1: Detect the expression levels of protein markers in tissue samples from metastatic papillary thyroid carcinoma to obtain the expression level or abundance value of each protein; S2: Perform data preprocessing and standardization on the expression levels or abundance values ​​of each protein obtained in step S1 to obtain standardized expression levels; S3: Input the standardized expression level obtained in step S2 into the pre-trained logistic regression model to obtain the iodine uptake positive probability value of thyroid cancer metastases. Compare the output iodine uptake positive probability value with the pre-determined classification threshold to determine the iodine uptake capacity of the sample.

3. The method for predicting iodine uptake capacity of thyroid cancer metastases using protein biomarkers according to claim 2, characterized in that, The expression levels of the protein markers described in step S1 were detected using mass spectrometry proteomics.

4. The method for predicting iodine uptake capacity of thyroid cancer metastases using protein biomarkers according to claim 2, characterized in that, The preprocessing described in step S2 employs one or more of the following: missing value handling, normalization or log2 transformation for preliminary standardization, batch effect correction, outlier handling, or feature scaling.

5. The method for predicting iodine uptake capacity of thyroid cancer metastases using protein biomarkers according to claim 2, characterized in that, The standardization described in step S2 uses the Z-score method.

6. The method for predicting iodine uptake capacity of thyroid cancer metastases using protein biomarkers according to claim 2, characterized in that, The specific formula for the pre-trained logistic regression model mentioned in step S3 is as follows: ； In the formula: denoted as iodine uptake positive probability value; i represents the i-th sample, j represents the j-th protein, and x represents the positive probability value. ij Let be the standardized expression level of the j-th protein in the i-th sample; β0 is the intercept; β j is the weighting coefficient of the j-th protein.

7. The method for predicting iodine uptake capacity of thyroid cancer metastases using protein biomarkers according to claim 2, characterized in that, If the output iodine uptake positive probability value is greater than or equal to the predetermined classification threshold, the metastatic lesion sample is determined to be iodine uptake positive; otherwise, it is determined to be iodine uptake negative.

8. A prediction system for the iodine uptake capacity of thyroid cancer metastases to implement the prediction method according to any one of claims 2 to 7, characterized in that, include: The data acquisition module is used to acquire the expression level or abundance value of protein markers in tissue samples of metastatic thyroid papillary carcinoma. The data processing module is used to preprocess and standardize the expression levels or abundance values ​​of the obtained protein markers to generate standardized expression levels. The model prediction module outputs the iodine uptake positivity probability value of the metastatic lesion tissue sample through a pre-trained logistic regression model, and generates a classification label of the iodine uptake capacity of the sample based on a preset classification threshold.

Images