Nonnegative Network Component Analysis by Linear Programming for Gene Regulatory Network Reconstruction

We consider a systems biology problem of reconstructing gene regulatory network from time-course gene expression microarray data, a special blind source separation problem for which conventional methods cannot be applied. Network component analysis (NCA), which makes use of the structural information of the mixing matrix, is a tailored method for this specific blind source separation problem. In this paper, a new NCA method called nonnegative NCA (nnNCA) is proposed to take into account of the non-negativity constraint on the mixing matrix that is based on a reasonable biological assumption. The nnNCA problem is formulated as a linear programming problem which can be solved effectively. Simulation results on spectroscopy data and experimental results on time-course microarray data of yeast cell cycle demonstrate the effectiveness and anti-noise robustness of the proposed nnNCA method.