Computational models for pan-cancer classification based on multi-omics data.

Tumor heterogeneity presents a significant challenge in cancer treatment, limiting the ability of clinicians to achieve accurate early-stage diagnoses and develop customized therapeutic strategies. Early diagnosis is crucial for effective intervention, yet current methods lack robust solutions to overcome this challenge. The Pan-Cancer Atlas has emerged as a pivotal framework to investigate cancer heterogeneity by integrating multi-omics data (genomics, transcriptomics, proteomics) across tumor types. This initiative systematically maps inter- and intratumor variations, providing insight for clinical decision making. However, such frameworks often struggle to integrate dynamic temporal changes and spatial heterogeneity within tumors, limiting their real-time clinical applicability. In this review, we first summarize the available multi-omics data and public biomedical databases used in pan-cancer research. Then, we examine current pan-cancer classification approaches based on the computational models they employed, including machine learning and deep learning. We also provide a comparison of these classification methods to explore their advantages and limitations. Finally, we conclude by discussing the key challenges in pan-cancer research and suggesting potential directions for future studies.