An Interpretable Machine Learning Model for Predicting 5‐Year Survival in Breast Cancer Based on Integration of Proteomics and Clinical Data

ABSTRACT Breast cancer is a highly heterogeneous malignancy among women worldwide. Traditional prognostic models relying solely on clinicopathological features offer limited predictive accuracy and lack molecular‐level insights. Unlike such conventional approaches, this study integrates proteomic and clinical data within an interpretable deep learning framework to improve prognostic precision and biological interpretability. We aimed to develop a more reliable model to accurately predict the 5‐year survival status of patients with breast cancer using multi‐omics data. The model integrating proteomics and clinical features demonstrated superior performance (AUC = 0.8136) compared to other feature combination models. The optimized model with 13 key features (4 clinical features and 9 proteins) achieved an AUC of 0.864 with the precision of 0.970, the recall of 0.810, and F1‐score of 0.883. SHapley Additive exPlanations analysis identified MPHOSPH10, EGFR, ARL3, KRT18, lymph node status, and HER2 status as the most influential features, while Kolmogorov–Arnold Network analysis provided explicit mathematical relationships between key contributors and prediction outcomes. Collectively, our interpretable multi‐modal model demonstrates robust performance in predicting 5‐year survival in breast cancer patients and offers mechanistic insights, thereby enhancing its potential for clinical translation through the development of an accessible prediction tool.