CSSEC: An adaptive approach integrating consensus and specific self-expressive coefficients for multi-omics cancer subtyping.

Cancer is a complex and heterogeneous disease, and accurate cancer subtyping can significantly improve patient survival rates. The complexity of cancer spans multiple omics levels, and analyzing multi-omics data for cancer subtyping is a focus of considerable attention. However, extracting complementary information from different omics data sources and adaptively integrating them remains a major challenge. To address this, we proposed an adaptive approach integrating consensus and specific self-expressive coefficients for multi-omics cancer subtyping (CSSEC). First, independent self-expressive networks are applied to each omics to calculate coefficient matrices to measure patient similarity. Then, two feature graph convolutional network modules capture consensus and specific similarity features by the topK relevant features. Finally, the multi-omics self-expression coefficient matrix is constructed by consensus and specific similarity features. Furthermore, joint consistency and disparity constraints are applied to regularize the fusion process of the self-expressive coefficients. Experimental results demonstrate that CSSEC outperforms existing state-of-the-art methods in survival analysis. Moreover, case studies on Kidney cancer confirm that the cancer subtypes identified by CSSEC have significant biological relevance. The complete code can be available at https://github.com/ykxhs/CSSEC.