IntNet: A novel framework using reconstructed mutation samples and a biologically informed neural network for pathway analysis

Most pathway analysis methods based on simple nucleotide variations overlook mutations occurring in genes outside the pathway, fail to address mutation heterogeneity with targeted measures, and do not consider the functional relationships among gene components within biomolecular networks when modeling the contributions of biomarkers to pathways. So, we proposed a novel framework, named IntNet, for pathway analysis based on simple nucleotide variations. The core components of IntNet are sample reconstruction (SC) and a new biologically informed neural network model (BINN). To address the difficulty of extracting effective features caused by the mutation heterogeneity, IntNet reconstructs mutation samples based on the hypothesis that mutations in patients with the same phenotype may cluster in similar network regions. BINN uses the human gene network and expanded biological pathways to construct its neural network architecture. BINN implements mutation signaling propagation along the human gene network and integration within the pathways. IntNet's performance was validated using multiple cancer datasets. Experimental results demonstrated that IntNet could identify more accurate and reliable pathways.