Breaking the Complexity of Cancer Using Computational Transcriptomic Network Biology

Abstract

While the landscapes of cancer mutations have been mostly clarified, in this study, we focused on the connective aggregates between mutations and phenotypes, named here as “gene transcriptomic networks,” aiming to survey computational network biology processes that have achieved significant results in cancer biology. Two methodologies are considered as necessary to unravel hidden cancer biology: (a) gene network estimation and (b) interpretation of the huge and complex network. For (a), key statistical methods, including Bayesian network with nonparametric regression and the Gaussian state space model, are mathematically reviewed in the context of RNA expression data analysis. For (b), methods based on regularized regression model are reviewed to explain how important markers for anti-cancer drug resistance are derived from interpretable gene networks. Challenges of explainable AI (Xprediction, Tensor Reconstruction-based Interpretable Prediction) for such networks are impressive and send a message to the future for breaking the complexity of cancer.