The rapid advancement of disease diagnosis and biomarker identification has been facilitated by the integration of multi-omics data and graph-based deep learning techniques. This convergence of technologies has improved the accuracy and interpretability of disease classification and biomarker identification. Notably, the development of novel models such as MOGKAN and MSNGO has achieved state-of-the-art performance in multi-cancer classification and protein function prediction tasks. Furthermore, the application of these techniques to specific diseases, such as non-small cell lung cancer and Alzheimer's disease, has led to the identification of potential biomarkers and therapeutic targets. Additionally, advancements in computer vision, image processing, graph representation learning, person re-identification, multimodal processing, human activity recognition, and out-of-distribution detection have also been observed. The use of heterogeneous graph learning, parameter-efficient fine-tuning, and blind matching techniques has improved cross-modal correspondence and video-text retrieval. The integration of boundary enhancement features, multi-scale aggregation attention, and frequency-guided fusion modules has enhanced the accuracy and reliability of semantic segmentation models. Novel frameworks for image fusion, such as methods utilizing degradation and semantic dual-prior guidance or controllable image fusion with language-vision prompts, have also been proposed. The field of graph representation learning is rapidly evolving, with a focus on developing innovative methods to capture complex relationships and structural properties of graphs. The use of hypergraphs, multilevel graphs, and topology-aware vision transformers has improved graph representation learning. Overall, these advances have the potential to transform our understanding of disease mechanisms and improve the diagnosis and treatment of complex diseases.