The rapid development of biological and medical examination methods has vastly expanded personal biomedical information, including molecular, cellular, image, and electronic health record datasets. Integrating this wealth of information enables precise disease diagnosis, biomarker identification, and treatment design in clinical settings. Artificial intelligence (AI) techniques, particularly deep learning models, have been extensively employed in biomedical applications, demonstrating increased precision, efficiency, and generalization. The success of the large language and vision models further significantly extends their biomedical applications. However, challenges remain in learning these multimodal biomedical datasets, such as data privacy, fusion, and model interpretation. In this review, we provide a comprehensive overview of various biomedical data modalities, multimodal representation learning methods, and the applications of AI in biomedical data integrative analysis. Additionally, we discuss the challenges in applying these deep learning methods and how to better integrate them into biomedical scenarios. We then propose future directions for adapting deep learning methods with model pretraining and knowledge integration to advance biomedical research and benefit their clinical applications.

Understanding the prognosis of cancer patients is crucial for enabling precise diagnosis and treatment by clinical practitioners. Multimodal fusion models based on artificial intelligence (AI) offer a comprehensive depiction of the tumor heterogeneity landscape, facilitating more accurate predictions of cancer patient prognosis. However, in the real‐world, the lack of complete multimodal data from patients often hinders the practical clinical utility of such models. To address this limitation, an interpretable bridged multimodal fusion model is developed that combines histopathology, genomics, and transcriptomics. This model assists clinical practitioners in achieving more precise prognosis predictions, particularly when patients lack corresponding molecular features. The predictive capabilities of the model are validated across 12 cancer types, achieving optimal performance in both complete and missing modalities. The work highlights the promise of developing a clinically applicable medical multimodal fusion model. This not only aids in reducing the healthcare burden on cancer patients but also provides improved assistance for clinical practitioners in precise diagnosis and treatment.

Background: Tertiary lymphoid structures (TLSs) exert a crucial role in the tumor microenvironment (TME), impacting tumor development, immune escape, and drug resistance. Nonetheless, the heterogeneity of TLSs in colorectal cancer (CRC) and their impact on prognosis and treatment response remain unclear. Methods: The authors collected genome, transcriptome, clinicopathological information, and digital pathology images from multiple sources. An unsupervised clustering algorithm was implemented to determine diverse TLS patterns in CRC based on the expression levels of 39 TLS signature genes (TSGs). Comprehensive explorations of heterogeneity encompassing mutation landscape, TME, biological characteristics, response to immunotherapy, and drug resistance were conducted using multiomics data. TLSscore was then developed to quantitatively assess TLS patterns of individuals for further clinical applicability. Results: Three distinct TLS patterns were identified in CRC. Cluster 1 exhibited upregulation of proliferation-related pathways, high metabolic activity, and intermediate prognosis, while Cluster 2 displayed activation of stromal and carcinogenic pathways and a worse prognosis. Both Cluster 1 and Cluster 2 may potentially benefit from adjuvant chemotherapy. Cluster 3, characterized by the activation of immune regulation and activation pathways, demonstrated a favorable prognosis and enhanced responsiveness to immunotherapy. The authors subsequently employed a regularization algorithm to construct the TLSscore based on nine core genes. Patients with lower TLSscore trended to prolonged prognosis and a more prominent presence of TLSs, which may benefit from immunotherapy. Conversely, those with higher TLSscore exhibited increased benefits from adjuvant chemotherapy. Conclusions: The authors identified distinct TLS patterns in CRC and characterized their heterogeneity through multiomics analyses. The TLSscore held promise for guiding clinical decision-making and further advancing the field of personalized medicine in CRC.

Background: Colorectal cancer (CRC) is currently the third most common cancer worldwide, with high heterogeneity and poor prognosis. Gene expression-based molecular subtypes can effectively dissect tumor heterogeneity, but their clinical translation remains challenging. This study aims to conduct radiogenomic analysis regarding molecular subtypes and establish prognostic signatures for survival prediction of colorectal cancer. Methods: In this retrospective study involving 2948 CRC patients from 8 cohorts, we utilized a supervised deep learning framework to extract quantitative feature representations of molecular subtypes. Through correlation analysis, we selected key gene expression features related to these subtypes to establish a prognostic signature. A similar pipeline was applied to derive a non-invasive radiomic prognostic signature. Finally, we validated the prognostic value of both signatures in multiple cohorts and explored their biological interpretation. Results: We successfully established a molecular subtype-associated gene signature and a non-invasive radiogenomic signature. The gene signature classified patients into high-risk and low-risk groups with significantly different prognoses. The low-risk group had a better prognosis and showed a greater potential benefit from immunotherapy. Similarly, the radiogenomic signature exhibited characteristics related to molecular subtypes and comparable performance in prognostic prediction. Multivariate analysis confirmed the independent prognostic value of both signatures. In summary, this retrospective study demonstrates that our framework translates molecular subtypes into cost-effective biomarkers for risk stratification and treatment guidance.