Discovery of gene-gene co-perturbation through big data

Abstract

AbstractUnderstanding the associations between genes is crucial to understanding the relationship between diseases. In order to learn these gene-gene associations and generate the huge gene network, more than 12,000 experiments have been analyzed. However, the correlation between genes is context-dependent, so two genes may not always be concordantly or discordantly co-dysregulated in all experiments, and the batch effect between experiments decreases the quality of the integrated data. We therefore developed a new co-perturbation model to identify reliable gene-gene correlations in the big integrated data, which significantly outperformed the widely used co-expression approach and can avoid Simpson’s paradox. Disease-related genes in our co-perturbation network are also more likely to be the hub genes, and the correlation between disease-related genes can be context dependent and non-linear. [We also validated biological findings between GeneA and B…]