Cervical cancer survival analysis by integrating multi-omics data through an adaptive weighted deep neural network

Accurately analyzing cervical cancer prognosis holds significant importance for enhancing clinical outcomes within the framework of precision medicine. Since single-omics data only capture a limited perspective of cervical cancer patients, integrating multiple types of omics data is essential to provide more comprehensive insights. However, capturing the key information in multi-omics data fusion remains challenging owing to the presence of a large number of redundant variables and a relatively small sample size. In this study, we propose a deep neural network called AWMP to better integrate multi-omics data for cervical cancer survival analysis. The experiment demonstrated that AWMP enhances cancer prognosis prediction by 4.74% relative to the mean performance of the four comparator methods, benefiting from its adaptive weighted mechanism that effectively addresses the greedy nature of learning in multi-omics integration. In this study, based on the risk subgroups stratified by AWMP, four key genes (CDC42BPB, KLRG1, MAP7, LIMD2) are identified as been highly associated with cervical cancer prognosis and immune cell infiltration, highlighting their role in tumor heterogeneity. Additionally, single-cell tumor heterogeneity analysis revealed that the B-cell C2 subset correlates with both prognosis and immune response in cervical cancer, suggesting its potential as a therapeutic target. These findings underscore the accuracy and biological relevance of our deep learning framework for multi-omics integration in cervical cancer prognosis. Collectively, the identification of B-cell C2 subset advances the understanding of the precise role of B cells in cancer immune evasion and paves the way for more effective immunotherapy for cervical cancer.