Deep learning in multi-omics integration for gastrointestinal cancer biomarker discovery

Gastrointestinal (GI) cancers remain leading causes of cancer-related morbidity and mortality worldwide, driven by complex molecular mechanisms involving genomic, transcriptomic, proteomic, metabolomic, and radiomic alterations. The discovery of reliable biomarkers is essential for improving early detection, risk stratification, prognosis prediction, and personalized treatment strategies. Recent advances in deep learning (DL) have enabled the integration of multi-omics and imaging data, offering unprecedented opportunities to identify clinically relevant biomarkers in these malignancies. This review synthesizes representative DL architectures specifically optimized for GI cancer biomarker research, highlighting methodological advances in feature extraction, multi-modal data fusion, and predictive modeling. Beyond detailing technical architectures, we critically examine current challenges, including the reliance on public datasets, the impact of intratumoral heterogeneity, and the need for rigorous external validation. We further discuss emerging strategies such as biologically-informed deep learning and multimodal foundation models for enhancing model interpretability, addressing data heterogeneity, and supporting translational adoption within regulatory frameworks (e.g., FDA/NMPA). By bridging computational innovations with clinical needs, DL-enabled multi-omics integration holds promise for advancing predictive, preventive, and personalized medicine in gastrointestinal oncology.