Machine learning-based multimodal radiomics and transcriptomics models for predicting radiotherapy sensitivity and prognosis in esophageal cancer

Abstract

Radiotherapy plays a critical role in treating esophageal cancer, but individual responses vary significantly, impacting patient outcomes. This study integrates machine learning-driven multimodal radiomics and transcriptomics to develop predictive models for radiotherapy sensitivity and prognosis in esophageal cancer. We applied the SEResNet101 deep learning model to imaging and transcriptomic data from the UCSC Xena and TCGA databases, identifying prognosis-associated genes such as STUB1, PEX12, and HEXIM2. Using Lasso regression and Cox analysis, we constructed a prognostic risk model that accurately stratifies patients based on survival probability. Notably, STUB1, an E3 ubiquitin ligase, enhances radiotherapy sensitivity by promoting the ubiquitination and degradation of SRC, a key oncogenic protein. In vitro and in vivo experiments confirmed that STUB1 overexpression or SRC silencing significantly improves radiotherapy response in esophageal cancer models. These findings highlight the predictive power of multimodal data integration for individualized radiotherapy planning and underscore STUB1 as a promising therapeutic target for enhancing radiotherapy efficacy in esophageal cancer.

Keywords: SEResNet101; SRC ubiquitination; STUB1; esophageal cancer; prognostic risk model; radiotherapy sensitivity.

Figure 1

Application of radiomics and machine learning models in prognosis prediction for esophageal cancer.A, schematic diagram of the SEResNet101 model construction. B, confusion matrix displaying the classification results of the SEResNet101 model for esophageal cancer patients, with the horizontal axis representing the predicted results and the vertical axis representing the actual outcomes. C, accuracy curve demonstrating a rapid increase in accuracy during the initial training epochs, stabilizing above 98%. D, F1 score curve showing the model's stability in balancing precision and recall across training epochs. E, loss curve depicting the progressive reduction of the loss value during training, ultimately reaching approximately 0.38. F, precision curve indicating the model's precision consistently ranging between 97% and 99%, reflecting its high accuracy in predicting positive samples. G, recall curve illustrating the model's sensitivity to positive samples. H, survival curves comparing radiotherapy (blue) and nonradiotherapy (orange) groups, demonstrating the improved survival probability of patients receiving radiotherapy.

Figure 2

Transcriptomic analysis of esophageal cancer.A, schematic diagram showing the distribution of esophageal cancer patients in the TCGA dataset. B, volcano plot of differentially expressed genes between untreated tumor samples (62 cases) and normal control samples (11 cases) in the TCGA dataset. Red dots represent upregulated genes in tumor tissues, while blue dots represent downregulated genes. C, volcano plot of differentially expressed genes between untreated tumor samples (62 cases) and radiotherapy-treated tumor samples (31 cases). Red dots indicate genes upregulated in the radiotherapy group, and blue dots indicate downregulated genes. D, Venn diagram showing the intersection of genes downregulated in untreated tumor samples and upregulated in radiotherapy-treated samples. E, prognosis analysis identifying genes positively correlated with prognosis. The x-axis represents survival time, and the y-axis represents survival probability, with 46 cases each in the high-expression and low-expression groups for each gene.

Figure 3

Construction of prognosis model for esophageal cancer patients.A, lasso coefficient distribution of eight genes in esophageal cancer. B, selection of the optimal parameter (lambda) in the Lasso analysis for esophageal cancer. C, forest plot of multivariate Cox analysis, showing gene names on the left, p-values in the center, and hazard ratios (HR) representing risk rates. D and E, risk distribution of esophageal cancer patients divided into high-risk and low-risk groups based on multivariate Cox analysis. Green represents the low-risk group (46 cases), and red represents the high-risk group (46 cases). The x-axis shows patients sorted by increasing risk scores, and the y-axis represents the risk scores. F, survival status of high-risk and low-risk esophageal cancer patients. The x-axis shows patients sorted by increasing risk scores, and the y-axis represents survival time, with green dots indicating surviving patients and red dots indicating deceased patients. G, survival curves for high-risk and low-risk esophageal cancer patients, with the x-axis representing survival time and the y-axis representing survival probability. The red line represents the high-risk group, and the blue line represents the low-risk group. H, ROC curve analysis of the prognosis risk model's accuracy in predicting esophageal cancer prognosis, with green, blue, and red lines representing 1-, 3-, and 5-years ROC curve analyses, respectively.

Figure 4

Clinical correlation analysis of prognosis risk model in esophageal cancer patients.A, univariate Cox analysis evaluating the prognostic ability of the risk model. B, multivariate Cox analysis assessing the independent prognostic capability of the risk model. The left panel lists clinical traits and risk scores from the prognosis model, the center shows p-values (∗p < 0.05 indicates independent prognostic factors), and HR represents risk rates. C, heatmap displaying the correlation between the prognosis risk model and clinical traits in esophageal cancer patients. ∗p < 0.05, ∗∗p < 0.01.

Figure 5

Potential molecular mechanisms affecting radiotherapy sensitivity in esophageal cancer patients.A–C, heatmaps showing the clinical correlations of high and low expression of HEXIM2, PEX12, and STUB1 genes with clinical traits. ∗p < 0.05, ∗∗p < 0.01. D, volcano plot of differentially expressed proteins between untreated (59 cases) and radiotherapy-treated (28 cases) esophageal cancer tissues. Red dots represent upregulated proteins in the radiotherapy group, while blue dots indicate downregulated proteins. E, PPI network generated using the STRING database, illustrating the interaction between SRC and STUB1.

Figure 6

Effects of STUB1 overexpression or SRC silencing on radiotherapy sensitivity in KYSE520 esophageal cancer cells.A, schematic diagram of the construction of the radioresistant cell model. B, clonogenic survival assays evaluating the radioresistance of esophageal cancer cells. C, Western blot analysis of STUB1 and SRC protein expression in esophageal cancer cells. D, Western blot analysis of STUB1 and SRC protein expression following STUB1 overexpression or SRC silencing. E, clonogenic survival assays assessing the radioresistance of esophageal cancer cells across different groups. F, transwell assays evaluating the migration and invasion abilities of esophageal cancer cells in each group. G, flow cytometry analysis of apoptosis rates in esophageal cancer cells across different groups. H, Western blot analysis of apoptosis-related proteins Bax and Bcl-2 in esophageal cancer cells across different groups. ∗p < 0.05, all cell experiments were repeated three times.

Figure 7

Effects of STUB1 on SRC ubiquitination and degradation in regulating radiotherapy sensitivity in KYSE520 esophageal cancer cells.A, Western blot analysis of SRC protein expression following STUB1 overexpression in esophageal cancer cells. B, RT-qPCR analysis of SRC transcription levels after STUB1 overexpression. C, Co-IP assay to assess STUB1-mediated SRC ubiquitination. D, analysis of SRC protein stability following CHX treatment. E, Western blot analysis of SRC protein expression after E1 inhibitor (PYR-41) treatment. F, Western blot analysis of STUB1 and SRC protein expression in each experimental group. G, clonogenic survival assays evaluating radioresistance in esophageal cancer cells across different groups. H, transwell assays assessing migration and invasion abilities of esophageal cancer cells in each group. I, flow cytometry analysis of apoptosis rates in esophageal cancer cells across different groups. J, Western blot analysis of apoptosis-related proteins Bax and Bcl-2 in esophageal cancer cells across different groups. K, schematic representation of the mechanism by which STUB1 promotes SRC ubiquitination and degradation to enhance radiotherapy sensitivity in esophageal cancer cells. ∗p < 0.05, all cell experiments were performed in triplicate.

Figure 8

Effects of STUB1 on SRC in enhancing radiotherapy sensitivity in esophageal cancer mouse models.A, representative images of subcutaneous xenograft tumors in nude mice from different groups. B, tumor growth curve statistics for different groups. C, tumor weight statistics for different groups. D, Western blot analysis of STUB1 and SRC protein expression in tumor tissues from different group. E and F, TUNEL staining showing apoptotic cells in tumor tissues from each group (Scale bar represents 50 μm). G, immunohistochemistry analysis of Ki67-positive expression in tumor tissues from each group (Scale bar represents 50 μm). H, schematic illustration of the role of STUB1 in modulating SRC to enhance radiotherapy sensitivity in esophageal cancer mouse models. ∗p < 0.05, five mice per group.