Consider reaction rate equations (RRE) system of ODE's where the coefficients of the right hand side are considered unknown parameters. We are interested in the problem of recovering the reaction network from the RRE. As illustrated by GC recent paper, these networks are in general "unidentifiable" in the sense that different chemical reactions networks may give the the same RRE. The matters are further complicated as the coefficients of RRE are in practice estimated from experimental data and hence subject to measurement error. In the presentation we describe a statistical approach for identifying the "most likely network" from a given set of RRE coefficient estimates. The idea relies on mapping the estimated reaction constants into an appropriate convex region of a network and use the underlying geometry to identify reactions which are most likely to span that region. This approach reduces the problem to inference for the parameters of a multinomial distribution parametrized by a toric model which may be then solved using classical likelihood methods along with some ideas from the algebraic statistics. Joined work with Gheorge Craciun, University of Wisconsin.