Boolean and Bayesian networks are widely used for modeling from gene expression data. Here I leave you two references, not very recent but quite useful, so you can have find other tools.
It depends on what kind of model you want to build. In fact you can use gene expression data to obtain a model's structure and/or to identify the parameter values. For instance, one possible approach is to use something like CLR, PLSnet or Aracne to derive a gene regulatory network. Then you can translate it into a gene regulatory model of different complexity levels, from boolean to fully mechanistic. In the latter, one has to identify parameter values, which is obtained again using gene expression data. Simplest case is using linear relationships such that you can use linear regression for estimating the parameters. One very used kind of function for modeling gene regulation is the Hill function, which is a sigmoid and therefore requires slightly more data than a linear function for parameter estimation. If you write the model in Copasi you can also identify the parameters with that piece of software.
Alternative, is that you have the structure of the model e.g. who regulates who and you only use expression data to do the parameter estimation step.
Gene expression data can also be used proficiently with metabolic models, as the abundance of a transcript can be taken as a proxy (better if in prokaryotes) of the protein abundance and therefore of the enzymatic activity.