Scalable reverse-engineering of gene regulatory networks from time-course measurements

Topological inference of biological interaction networks from experimental data is a fundamental research topic in the broad area of Systems Biology. Several algorithms presented in the literature have been devised in order to infer the topology of a network from time-course data. The present work introduces a novel method for reverse-engineering gene regulatory networks from time-course experiments, which combines the instrumental variables technique for the identification of dynamical systems with a regularization strategy for dealing with over-parametrized systems. Differently from least squares methods, the proposed approach can explicitly address the bias and nonconsistency issues that arise when dealing with time-course measurements, thus yielding improved performance with respect to methods designed for steady-state data. Moreover, the devised approach, which has been named RIVA (Reverse-engineering of biological networks via Instrumental VAriables), can simultaneously exploit multiple time-series, thus enabling one to get improved results by collecting and exploiting data from multiple experiments, and is computationally efficient, thus it can be also applied to large-scale (in the order of thousands of nodes) networks. To analyze the applicability and effectiveness of RIVA, we performed several tests, both with simulated data and with experimental data, and compared the results against other state-of-the-art inference methods designed for time-series data.