Abstract Breast cancer heterogeneity is still a primary concern, with a large variation in the prognosis necessitating the development of personalised treatment plans. Single biomarkers can't fully encompass the breast cancer complexity and variations, and therefore, multi‐omics approaches offer a wide comprehension of the cancer‐specific biology. In this study, we developed a multi‐omics framework for the prediction of immune‐active features framework integrating four‐omics data, mainly genomic, proteomic and transcriptomic. By leveraging deep learning along with survival‐based feature selection, we constructed an autoencoder that generated compressed multi‐omics features to stratify the patients into two optimal immune subtypes with a significantly different overall survival ( p < 0.002) and a high C‐index of 0.74 (95% confidence interval: 0.62–0.83). An XGBoost classifier was trained to predict these subtypes by integrating all omics data as well as each omic individually, and was validated using both internal and external (the Cancer Genome Atlas and Gene Expression Omnibus) datasets. The integrated model achieved high predictive performance (ACC = 0.95). Omics‐unique classifiers showed consistently strong validation on independent datasets, particularly the immunotherapy‐treated cohort (GSE241876, p = 4.80×10 −2 ). We further investigated the biological mechanisms across the clusters and discovered that the C2, low‐risk cluster, exhibited an immune‐active landscape, characterised by a high infiltration of cells and more immune‐related pathways, making it a better candidate for a favourable immunotherapy response. On the other hand, the C1 cluster, the immune‐cold group, displayed an immunosuppressive microenvironment and poor prognosis. This methodology demonstrated the promising potential of deep learning‐driven multi‐omics integration to support precision oncology by enhancing prognostic prediction and tailoring treatment decisions.
Bendani et al. (Tue,) studied this question.