Predicting gene compensation in disease with graph embedding techniques

Genetic compensation plays a critical role in mitigating the effects of deleterious mutations in genetic diseases. Identifying functionally compensatory gene relationships represents a promising strategy for discovering therapeutic targets in inherited genetic disorders and cancer. We present a novel approach that combines multiplexed network analysis with graph embedding techniques to predict compensatory genes. Our method constructs a multi-source gene similarity network, embedding each source with kernels and node2vec methods, to finally integrate all sources in a comprehensive functional similarity gene network. Our approach demonstrates high predictive performance, successfully prioritizing known compensatory genes in disorders such as Duchenne muscular dystrophy, spinal muscular atrophy, and β-thalassemia. Furthermore, we extend its application to cancer, where genetic compensation mechanisms contribute to treatment resistance. Notable examples include androgen receptor (AR) in prostate cancer and RBL1 suppression compensation by RBL2 in breast cancer. These results demonstrate that embedding network representation is useful for prioritizing compensatory genes.