Demonstration and framework of MMPDENB-RBM model. (A) The biological explanation of multimodal PDENBs in the MMPDENB-RBM model. Firstly, we obtained gene expression data of paired samples (i.e. normal and tumor samples) of individual patients and reference samples as well as gene mutation data. Then, the paired-SSN method was used to construct a PGIN for each patient by integrating the aforementioned gene expression data. Subsequently, we used edge-network analysis to transform the node-network (PGIN) into an edge-network, which comprised rich dynamic and high-dimensional information of omics data, to characterize the development of cancer. Finally, we defined PDENB identification as LMOP and designed an MMEA based on latent space search to identify multimodal PDENBs. The multimodal PDENBs can be used to detect early-warning signals in the development of cancer as well as provide potential drug targets and synthetic lethality edge-biomarkers for cancer treatment, realizing multi-purpose early disease prediction. (B) The framework of MMEA with latent space search in MMPDENB-RBM model. Firstly, the PEN was divided into several subnetworks that formed the initial population via the population initialization strategy. Then, we calculated the objective function of the initial solution. Subsequently, these solutions were non-dominated sorted according to the value of the objective function, and the solutions with better quality were selected as parents. Then, each offspring was determined whether produced in original space or latent space according to decision conditions. The offspring and their parents were merged, and better solutions were retained by environmental selection operators. Finally, if iteration conditions were met, Pareto optimal solutions were outputted; otherwise, the search for better PDENBs was continued.