An integrative network inference approach to predict mechanisms of cancer chemoresistance

We present an integrative general network inference methodology to infer genetic and metabolic pathways associated with oncological drug chemoresistance. This methodology is general because it can infer different kinds of networks from different kinds of data. It is integrative because it integrates model simulation in its framework and it assembles into a larger integrated network all the inferred networks. The inference model is a variational approximation of Bayesian inference for stochastic processes. We used the Bayesian framework due to its ability to incorporate prior knowledge and constraints in the inference procedure and to treat both partial data and a large amount of data whose dynamics laws are mostly unknown. We show the performance of this approach using a case study of the gemcitabine chemoresistance in pancreatic cancer cells. Our method, inferred from time series data of gene expressions and metabolites, concentrates first the network of interactions of genes responsible for the sensitivity and resistance to gemcitabine, then the metabolic network, and finally it merges the two networks into a larger network predicting the correlations between genes and metabolizing enzymes.