A prostate cancer biochemical recurrence prognosis risk prediction model based on fatty acid metabolism and cancer cell stemness genes and a construction method thereof
By constructing a biochemical recurrence prediction model for prostate cancer that combines fatty acid metabolism and cancer cell stem genes, the problem of neglecting tumor metabolic reprogramming in existing technologies has been solved, achieving highly accurate prediction of biochemical recurrence risk and providing important clinical prediction evidence.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANTONG UNIV
- Filing Date
- 2026-03-25
- Publication Date
- 2026-07-14
AI Technical Summary
In existing technologies, biochemical recurrence prediction models for prostate cancer only focus on stem genes in cancer cells, neglecting the crucial role of tumor metabolic reprogramming.
A prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells was constructed. Through data collection, stemness score analysis, fatty acid metabolism score analysis, co-expressed gene module identification, and machine learning algorithm combination, a prostate cancer BCR prognostic risk prediction model was constructed. The model was trained and tested using R package algorithm, which is combined with fatty acid metabolism and stemness-related gene set.
It achieved high predictive accuracy (AUC=0.779), providing important clues for the clinical prediction of biochemical recurrence of prostate cancer, and exploring the potential role of fatty acid metabolism and cancer cell stemness in prognostic risk prediction.
Smart Images
Figure CN122392918A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of biomedical testing technology, specifically to a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stem genes of cancer cells, and its construction method. Background Technology
[0002] Prostate cancer is one of the most common new-onset malignant tumors in men worldwide and the second leading cause of cancer-related deaths in men, with its incidence and mortality rates continuing to rise. Radical prostatectomy (RP) and radical radiotherapy (RT) are currently the main treatments for patients with localized prostate cancer. Most patients with primary prostate cancer can be cured with radical prostatectomy, with a 5-year survival rate approaching 100%. However, approximately 20%–40% of patients undergoing radical prostatectomy and 30%–50% of patients undergoing radiotherapy will experience biochemical recurrence (BCR) within 10 years. BCR is defined as two consecutive increases in PSA ≥0.2 ng / mL after RP, or an increase in PSA greater than 2 ng / mL from its lowest value after RT. Biochemical recurrence in prostate cancer usually indicates disease progression, which can further develop into castration-resistant prostate cancer (CRPC) and may be accompanied by distant metastasis, ultimately leading to the patient's death.
[0003] Stem cell nature refers to the potential of cells to differentiate and develop into various cell types. Tumor cells acquire stem cell characteristics while losing their differentiation phenotype, a crucial mechanism driving tumor progression. Stem cell nature is regulated by multiple factors, including metabolism. Most tumors primarily utilize glycolysis for energy, while prostate cancer cells tend to obtain energy through fatty acid metabolism reprogramming. Castration-resistant prostate cancer (CRPC) also exhibits significantly enhanced fatty acid synthesis and metabolic activity. Current research suggests that biochemical recurrence of prostate cancer is closely related to stem cell nature, and fatty acid metabolism can participate in regulating tumor cell nature.
[0004] Machine learning can extract patterns, regularities, and correlations from massive amounts of data, handle complex relationships between data, accelerate the analysis process, and provide a scientific basis for analytical decisions. It is currently widely used in cancer-related research. Although there are clinical models for predicting prostate cancer recurrence based on cancer cell stem genes, these models, which only focus on cancer cell stem genes, neglect the crucial role of tumor metabolic reprogramming. Summary of the Invention
[0005] The purpose of this application is to address the technical problem in existing predictive models that focus only on stem genes in cancer cells and neglect tumor metabolic reprogramming.
[0006] To achieve the above objectives, this application provides the following technical solution.
[0007] A method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells includes the following steps:
[0008] S1: Data collection: Obtain transcriptome data of prostate cancer samples with biochemical recurrence information from the database, and divide the data into training set and test set for the model;
[0009] S2: Dryness score analysis;
[0010] S3: Fatty acid metabolism score analysis;
[0011] S4: Based on the analysis results of S2 and S3, co-expressed gene modules related to fatty acid metabolism and stemness characteristics in prostate cancer are identified by co-expression similarity algorithm and hierarchical clustering algorithm, and a gene set related to fatty acid metabolism and stemness characteristics is obtained.
[0012] S5: Constructing a prognostic risk prediction model for prostate cancer BCR: Based on the gene set in S4, the data is trained and tested using algorithms from the R package.
[0013] S6: Constructing a nomogram model: Construct a nomogram model of prostate cancer-specific antigen, Gleason score, and fat_stemness_BCR, and construct calibration curves based on the stdca algorithm. Then construct decision curves and ROC curves, and evaluate the predictive advantage of the fat_stemness_BCR model based on the above results.
[0014] S7: Extract RNA from the sample to be tested and construct a cDNA library. Quantify the expression of the above genes and calculate the prognostic risk level of prostate cancer based on the expression level.
[0015] Preferably, the database in S1 includes the databases GEO, CIT, ICGC, TCGA, PCTA, and CPGEA, and the data in GEO is used as the training set, while the data in the other databases are used as the external test set.
[0016] Preferably, the data inclusion and processing in S1 follow the following standards and steps:
[0017] (1) The tissue samples were obtained from prostate cancer patients;
[0018] (2) Clinical information on biochemical relapse can be obtained;
[0019] (3) For the GEO database, obtain prostate cancer expression profile data, merge multiple GEO datasets and perform batch effect correction;
[0020] (4) For the CIT database, the original microarray data is standardized and the expression values are transformed.
[0021] Preferably, the R package in S5 is an R package for machine learning algorithms, including gbm, glmnet, plsRcox, randomForestSRC, superpc, survcomp, survival, CoxBoost, survivorsvm, and BART.
[0022] The machine learning algorithms in the R package used for machine learning algorithms include Lasso, Ridge, Enet, StepCox, survivalSVM, CoxBoost, SuperPC, plsRcox, RSF, and GBM.
[0023] Preferably, in step S2, a cell stemness model is constructed using a single-class logistic regression function of the R package Gelnet, and based on this model and prostate cancer expression profile data, the stemness score of the prostate cancer sample is calculated.
[0024] Preferably, in step S3, the fatty acid metabolism fraction of the prostate cancer sample is calculated using the single-sample gene set enrichment analysis (ssGSEA) algorithm of the R package GSVA.
[0025] Preferably, in step S5, during model training, 10-fold cross-validation is used for feature selection and model tuning. The model tuning includes obtaining the optimal λ value for Lasso, Ridge, and Elastic Network (Enet); the optimal penalty parameter for CoxBoost; the optimal threshold for SuperPC; the optimal nt parameter for plsRcox; and the optimal n.trees for GBM.
[0026] Preferably, in step S5, a grid search method is also used to select the optimal direction and α parameter for the StepCox model and the Enet model respectively, so as to ensure that the model achieves the best performance.
[0027] Preferably, in step S6, a nomogram model of prostate cancer-specific antigen (PSA), Gleason score, and fat_stemness_BCR is constructed using the R package (nomogramEx, survival). A calibration curve is constructed based on the stdca algorithm, and then a decision curve and ROC curve are constructed using the coxph and cph functions of the R package (survival). The predictive advantage of the fat_stemness_BCR model is evaluated based on these results.
[0028] This application also provides a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stem cell genes of cancer cells, which is constructed using the construction method described above.
[0029] Compared with the prior art, this application has the following beneficial effects:
[0030] This invention successfully constructed a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and cancer cell stemness gene sets, using bioinformatics and machine learning methods. The Cox proportional hazards regression model constructed based on this model score and clinical information showed good accuracy (AUC=0.779). These findings explore the potential role of fatty acid metabolism and cancer cell stemness in predicting the prognostic risk of biochemical recurrence in prostate cancer, providing important clues for the clinical prediction of prostate cancer biochemical recurrence. Attached Figure Description
[0031] Figure 1 Stem cell characteristic scores for normal and prostate cancer samples predicted by the stem cell characteristic model. (A) Stem cell characteristic scores in normal and prostate cancer samples. (B) Stem cell characteristic scores in prostate cancer samples are significantly higher than those in normal samples.
[0032] Figure 2 Fatty acid metabolism scores for normal and prostate cancer (PCa) samples. (A) Fatty acid metabolism scores in normal and prostate cancer samples. (B) Fatty acid metabolism scores in prostate cancer samples were significantly higher than those in normal samples.
[0033] Figure 3 To investigate the correlation between stem cell characteristics and fatty acid metabolism in prostate cancer and to identify co-expressed modules. (A) In prostate cancer, fatty acid metabolism is positively correlated with the stemness characteristics of cancer cells. (B) Co-expressed modules related to fatty acid metabolism and stemness characteristics of cancer cells in prostate cancer.
[0034] Figure 4 This is a machine learning model based on prostate cancer cell stemness and fatty acid metabolism. The C-index heatmap shows 101 combinations of 10 machine learning algorithms across 6 datasets (GEO, TCGA, ICGC, CIT, PCTA, and CPGEA).
[0035] Figure 5To predict the risk of biochemical recurrence (BCR) in prostate cancer based on clinical characteristics and the fat_stemness_BCR model score. (A) Nonograph constructed using a multivariate Cox proportional hazards regression model that integrates the fat_stemness_BCR model score, Gleason score, and prostate-specific antigen (PSA) to predict BCR risk. (B, C) Calibration curves for predicting 1-year and 3-year BCR probabilities. (D, E) Decision curve analysis for 1-year and 3-year BCR predictions. (F) Receiver operating characteristic (ROC) curves comparing the predictive performance of different models. Wherein, Model 1: PSA-based prediction model; Model 2: Gleason score-based prediction model; Model 3: fat_stemness_BCR-based prediction model; Model 4: Integrated prediction model combining PSA, Gleason score, and fat_stemness_BCR; AUC: Area under the curve. Detailed Implementation
[0036] This application provides a method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and cancer cell stemness genes, comprising the following steps:
[0037] S1: Data Acquisition: 1224 prostate cancer expression profiles and biochemical recurrence data were obtained from the GEO, CIT, ICGC, TCGA, PCTA and CPGEA databases. 786 of these data were used as the training set for the model, and 438 data were used as the test set for the model.
[0038] Specifically, in one embodiment, step S1 is as follows:
[0039] Transcriptome data from 1224 prostate cancer (PCa) samples with biochemical recurrence (BCR) information were obtained from the GEO, CIT, ICGC, TCGA, PCTA, and CPGEA databases. Of these, 786 PCa samples with BCR information from the GEO database were used as the training set, while 438 PCa samples with BCR information from the TCGA, ICGC, CIT, PCTA, and CPGEA cohorts served as the external test set. In these datasets, GSE21032 and CIT were derived from raw microarray data; GSE40272, GSE70770, and GSE116918 were derived from SOFT format data; and GSE54460, TCGA, ICGC, PCTA, and CPGEA were derived from processed transcriptome matrix files. Data inclusion and processing followed the following standards and procedures: (1) tissue samples were derived from prostate cancer patients; (2) clinical information on biochemical recurrence was available; (3) for the GEO database, prostate cancer expression profile data were obtained, multiple GEO datasets were merged, and batch effect correction was performed; (4) for the CIT database, raw microarray data were standardized and expression values were converted. For the CIT, ICGC, TCGA, PCTA, and CPGEA datasets, expression data were processed as follows: the average expression level of genes with multiple probes or repetitive gene names was calculated, and RNA-seq data (ICGC, TCGA, and CPGEA) were converted from FPKM or raw counts to TPM matrices. The ComBat function (default parameter) in the R package (sva) was used to perform cross-batch effect correction on all datasets.
[0040] S2: Stemness Score Analysis: Based on transcriptome data of pluripotent stem cells and embryonic stem cells from the Progenitor Cell Biology Consortium, a cell stemness model was constructed using a single-class logistic regression function of the R package Gelnet. Based on this model and prostate cancer expression profile data, the stemness score of prostate cancer samples was calculated.
[0041] The results are as follows Figure 1 As shown, the dryness score of prostate cancer samples was significantly higher than that of normal samples.
[0042] S3: Fatty acid metabolism score analysis: Based on the fatty acid metabolism gene sets in the HALLMARK, KEGG, and REACTOME databases, the fatty acid metabolism scores of prostate cancer samples were calculated using the single-sample gene set enrichment analysis (ssGSEA) algorithm of the R package GSVA.
[0043] The results are as follows Figure 2 As shown, the fatty acid metabolism score of prostate cancer samples was significantly higher than that of normal samples.
[0044] S4: Obtain candidate genes: Based on fatty acid metabolism and stemness score, the co-expression gene modules related to fatty acid metabolism and stemness characteristics in prostate cancer are identified by the co-expression similarity algorithm and hierarchical clustering algorithm of the R package WGCNA, and the gene set related to fatty acid metabolism and stemness is obtained.
[0045] In one embodiment, Pearson correlation analysis was used to calculate the dryness score and fatty acid metabolism score of normal prostate samples and prostate cancer samples, and a positive correlation was found between the two. Figure 3 A). Then, through weighted co-expression network analysis (WGCNA), combined with co-expression similarity algorithms and hierarchical clustering methods, four co-expression gene modules (MEsteelblue, MEturquoise, Mered, MEdarkorange) related to fatty acid metabolism and cancer cell stemness characteristics in prostate cancer were identified, and 1402 candidate genes were obtained. Figure 3 B).
[0046] S5: Constructing a prostate cancer BCR prognostic risk prediction model: Based on fatty acid and stemness-related gene sets, the model (fat_stemness_BCR) is constructed by training and testing data using 101 combinations of 10 machine learning algorithms (Lasso, Ridge, Enet, StepCox, survivalSVM, CoxBoost, SuperPC, plsRcox, RSF, and GBM) from the R package (gbm, glmnet, plsRcox, randomForestSRC, superpc, survcomp, survival, CoxBoost, survivalsvm, BART).
[0047] Specifically, in one implementation, transcriptomic data from prostate cancer (PCa) samples with clinical information on biochemical recurrence (BCR) in the GEO database were used as the training set, while data from the CIT, ICGC, TCGA, PCTA, and CPGEA databases were used as the test set. A prostate cancer BCR prognostic model was constructed using 10 different combinations of machine learning algorithms (e.g., Lasso + StepCox).
[0048] The R packages used for machine learning algorithms in this study include gbm, glmnet, plsRcox, randomForestSRC, superpc, survcomp, survival, CoxBoost, survivoralsvm, and BART. During model training, 10-fold cross-validation was used for feature selection and model tuning.
[0049] In one implementation, parameter optimization includes obtaining the optimal λ values for Lasso, Ridge, and Elastic Network (Enet); the optimal penalty parameter for CoxBoost; the optimal threshold for SuperPC; the optimal nt parameter for plsRcox; and the optimal n.trees for GBM. Furthermore, this application employs a grid search method to select the optimal orientation and α parameters for the StepCox and Enet models respectively, ensuring optimal model performance. In each tradeoff, the data is automatically divided into training and validation subsets, and this process is performed only on the training set. The test set does not participate in any steps of model training or optimization. The C-index values calculated on the training and test sets for each model combination are sorted and displayed to obtain the optimal combined model (…). Figure 4 The combined Lasso+StepCox model (fat_stemness_BCR) showed better prediction results, with a C-index of 0.628.
[0050] In one embodiment, the reagent for detecting the expression level of the biomarker includes primers for amplifying the genes JAGN1, EBPL, POLR2H, PRPF19, CYTH2, ZNF532, GHDC, ZNF696, RABGAP1, SCAP, BCAR1, ESRRA, PKP3, MAEA, CFLAR, VAMP2, MTIF3, SHC1, CDK7, ZCRB1, MESP1, OSBPL10, ATPAF2, KPTN, DNPEP, CHD2, AGFG2, APEH, PSENEN, PPTC7, LAS1L, TRIM35, and DOCK1.
[0051] S6: Constructing Nonograph Models: Nonograph models for prostate cancer-specific antigen (PSA), Gleason score, and fat_stemness_BCR are constructed using R packages (nomogramEx, survival). Calibration curves are built based on the stdca algorithm, and decision curves and ROC curves are then constructed using the coxph and cph functions of the survival package. The predictive advantage of the fat_stemness_BCR model is evaluated based on these results.
[0052] In one implementation, a nomogram model integrating prostate-specific antigen (PSA), Gleason score, and fat_stemness_BCR score was constructed to predict biochemical recurrence (BCR) of prostate cancer. Calibration curve, decision curve analysis, and ROC curve were used to evaluate the predictive accuracy of the model. Figure 5The calibration curve results showed a high degree of agreement between the predicted and observed probabilities of biochemical recurrence at the assessment time points, confirming the model's predictive accuracy. Decision curve analysis showed that the fat_stemness_BCR score provided a higher net benefit compared to prostate-specific antigen (PSA) and Gleason scores in predicting biochemical recurrence. The ROC curve showed that the fat_stemness_BCR model score outperformed PSA and Gleason scores in predicting biochemical recurrence, indicating its potential clinical application value.
[0053] S7: Extract RNA from the sample to be tested and construct a cDNA library to quantify the expression of the above-mentioned genes. Calculate the prognostic risk level of prostate cancer based on the expression level.
[0054] In one embodiment, the reagent for detecting the expression level of the biomarker includes primers for amplifying the genes JAGN1, EBPL, POLR2H, PRPF19, CYTH2, ZNF532, GHDC, ZNF696, RABGAP1, SCAP, BCAR1, ESRRA, PKP3, MAEA, CFLAR, VAMP2, MTIF3, SHC1, CDK7, ZCRB1, MESP1, OSBPL10, ATPAF2, KPTN, DNPEP, CHD2, AGFG2, APEH, PSENEN, PPTC7, LAS1L, TRIM35, and DOCK1.
[0055] In addition, this application also provides a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stem cell genes of cancer cells, which is constructed using the above-mentioned method.
[0056] Based on the aforementioned validation experiments, this application successfully constructed a prognostic risk prediction model for biochemical recurrence of prostate cancer using bioinformatics and machine learning methods, based on fatty acid metabolism and cancer cell stemness gene sets. The Cox proportional hazards regression model constructed based on this model score and clinical information showed good accuracy (AUC=0.779). These findings explore the potential role of fatty acid metabolism and cancer cell stemness in predicting the prognostic risk of biochemical recurrence of prostate cancer, providing important clues for the clinical prediction of biochemical recurrence of prostate cancer.
Claims
1. A method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells, characterized in that: Includes the following steps: S1: Data collection: Obtain transcriptome data of prostate cancer samples with biochemical recurrence information from the database, and divide the data into training set and test set for the model; S2: Dryness score analysis; S3: Fatty acid metabolism score analysis; S4: Based on the analysis results of S2 and S3, co-expressed gene modules related to fatty acid metabolism and stemness characteristics in prostate cancer are identified by co-expression similarity algorithm and hierarchical clustering algorithm, and a gene set related to fatty acid metabolism and stemness characteristics is obtained. S5: Constructing a prognostic risk prediction model for prostate cancer BCR: Based on the gene set in S4, the data is trained and tested using algorithms from the R package. S6: Constructing a nomogram model: Construct a nomogram model of prostate cancer-specific antigen, Gleason score, and fat_stemness_BCR, and construct calibration curves based on the stdca algorithm. Then construct decision curves and ROC curves, and evaluate the predictive advantage of the fat_stemness_BCR model based on the above results. S7: Extract RNA from the sample to be tested and construct a cDNA library. Quantify the expression of the above genes and calculate the prognostic risk level of prostate cancer based on the expression level.
2. The method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells according to claim 1, characterized in that: The database in S1 includes GEO, CIT, ICGC, TCGA, PCTA and CPGEA databases, with the data in GEO serving as the training set and the data in the other databases serving as the external test set.
3. The method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stem cell genes according to claim 1, characterized in that: The data inclusion and processing in S1 follow the following standards and procedures: (1) The tissue samples were obtained from prostate cancer patients; (2) Clinical information on biochemical relapse can be obtained; (3) For the GEO database, obtain prostate cancer expression profile data, merge multiple GEO datasets and perform batch effect correction; (4) For the CIT database, the original microarray data is standardized and the expression values are transformed.
4. The method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stem cell genes according to claim 1, characterized in that: The R packages in S5 are R packages used for machine learning algorithms. These R packages include gbm, glmnet, plsRcox, randomForestSRC, superpc, survcomp, survival, CoxBoost, survivorsvm, and BART. The machine learning algorithms in the R package used for machine learning algorithms include Lasso, Ridge, Enet, StepCox, survivalSVM, CoxBoost, SuperPC, plsRcox, RSF, and GBM.
5. The method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells according to claim 1, characterized in that: In S2, a cell stemness model is constructed using a single-class logistic regression function of the R package Gelnet. Based on this model and prostate cancer expression profile data, the stemness score of the prostate cancer sample is calculated.
6. The method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stem cell genes according to claim 1, characterized in that: In S3, the fatty acid metabolism fraction of prostate cancer samples is calculated using the single-sample gene set enrichment analysis (ssGSEA) algorithm of the R package GSVA.
7. The method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells according to claim 1, characterized in that: In S5, during model training, 10-fold cross-validation is used for feature selection and model tuning. The model tuning includes obtaining the optimal λ value for Lasso, Ridge, and Elastic Network (Enet); the optimal penalty parameter for CoxBoost; the optimal threshold for SuperPC; the optimal nt parameter for plsRcox; and the optimal n.trees for GBM.
8. The method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells according to claim 7, characterized in that: In S5, a grid search method is also used to select the optimal direction and α parameter for the StepCox model and the Enet model respectively, so as to ensure that the model achieves the best performance.
9. The method for constructing a prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells according to claim 1, characterized in that: In step S6, nomogram models of prostate cancer-specific antigen (PSA), Gleason score, and fat_stemness_BCR are constructed using the R package (nomogramEx, survival). Calibration curves are constructed based on the stdca algorithm, and decision curves and ROC curves are then constructed using the coxph and cph functions of the survival package. The predictive advantage of the fat_stemness_BCR model is evaluated based on the above results.
10. A prognostic risk prediction model for biochemical recurrence of prostate cancer based on fatty acid metabolism and stemness genes of cancer cells, characterized in that: It is constructed using any one of the construction methods of claims 1-9.