Model Identification Using Correlation-Based Inference and Transfer Entropy Estimation

Biological network inference makes use of mathematical methods to deduce the topology of networks of biochemical interactions from observational data. Recently, many efforts have been directed towards the achievement of this goal, and an increasing literature is proposing new mathematical models of inference. However, this still remains a challenging task, requiring a combination of different methods in order to overcome the limitations of each single procedure. In this work, we propose three methods to infer the structure of a biochemical network from the abundance of reactants time series. The first method combines the evaluation of the time-lagged correlation between species with a probabilistic method of model calibration. The second method estimates the transfer entropy to detect the causal relationships between time series. The third method is a combination of the transfer entropy-based method with the probabilistic model of parameter estimation. We argue the motivations, the advantages and the limitations of the three methods, and we present their performances on data generated from models of experimentally validated metabolic networks.