To investigate the prognostic value of different deep learning models constructed using integrated multi-omics data for patients with cervical squamous cell carcinoma and to compare their predictive performance.

Using preprocessed multi-omics data from the The Cancer Genome Atlas (TCGA) database for cervical squamous cell carcinoma (CESC), three prognostic prediction models—AutoSurv, DeepSurv, and Cox-nnet—were constructed. The predictive performance of the models was evaluated using the C-index, time-dependent area under the receiver operating characteristic curve (AUC), and the P-value of the log-rank test. The AutoSurv model, which demonstrated superior predictive capability, was selected for interpretability analysis using DeepSHAP to identify mRNAs and miRNAs that help distinguish high-risk from low-risk patients.

The C-index, 1-year AUC, 3-year AUC, and 5-year AUC values for the different model test sets showed that the AutoSurv model had the best predictive performance (C-index: 0.78 (95%CI: 0.75-0.81), AUC (1-year): 0.84 (95%CI: 0.81-0.87), AUC (3-year): 0.78 (95%CI: 0.76-0.85), AUC (5-year): 0.77 (95%CI: 0.75-0.81). The Log-rank test of the Kaplan-Meier survival curves indicated that the survival difference between the high- and low-risk groups classified by the AutoSurv model was statistically significant (P=0.013). The AutoSurv model outperformed the DeepSurv and Cox-nnet models in predicting the prognosis of patients with cervical squamous cell carcinoma. After quantifying the mRNAs and miRNAs that significantly influence the prediction results using DeepSHAP, multiple variables associated with the prognosis of cervical squamous cell carcinoma were identified, including POLR2H, FN1, FYN, UBA52, and RPS27A, among others, which were enriched in molecular signaling pathways such as DNA damage repair and NF-κB.

The AutoSurv model can accurately predict survival risk in CESC patients (C-index: 0.78 (95%CI: 0.75-0.81)). Combined with the DeepSHAP method, it provides valuable interpretability analysis, identifies key predictive variables influencing the prognosis of cervical squamous cell carcinoma, and reveals underlying biological mechanisms. This provides a scientific basis for clinicians to develop personalized treatment strategies and offers a research direction for targeted cancer therapy.