There is no 'best' way to model pathways. The research question comes first. How you 'best' model the pathways depends on what you want to investigate and find out.

Several different optimization techniques have been used. You can find review papers on the use of global optimization methods like the one in Genome Res. 2003 Nov;13(11):2467-74. "Parameter estimation in biochemical pathways: a comparison of global optimization methods" by Moles C.G., Mendes P., and Banga JR.

The answer will differ depending on what you are looking for. In the abvovementioned paper they look for values of model parameters, that is an identification problem is solved. You may also like to investigate the pathway itslef that is looking for the mathematical form (structure, equations, order, ...). This is a different problem and in turn, the most appropriate methods will not necessarily be the same.

I sympathize thus with Oliver's comment. It is impossible to answer such a question. This is indeed the case everytime that a question calls for the "best method" or the "best way". There is generally no answer since this is highly case dependent. 

Thanks, Gregory, for trying to be a bit more constructive and my apologies, Himanshu, for not providing anything too useful.

In my opinion, a model is a 'good' model not necessarily because it is very detailed and reproduces experiments perfectly, but what makes a model 'good' is the fact that you learn something from it.

If, for example, you are interested in understanding fundamental principles how a response is achieved, a highly abstracted model might be the best because it helps you to understand how things work. If, on the other hand, you want to make precise predictions on the effects of a drug, then you need a quantitative model which has been experimentally validated under as many conditions as possible. Therefore, two models of the same pathway may look extremely different, but both may be good for what they have been designed for.

To start, you may want to look at 'classical' models of the wnt pathway (Lee, E.; Salic, A.; Krüger, R.; Heinrich, R. & Kirschner, M. W., PLoS Biol, 2003, 1, E10) and the noth-delta pathway (Sprinzak, D.; Lakhanpal, A.; Lebon, L.; Santat, L. A.; Fontes, M. E.; Anderson, G. A.; Garcia-Ojalvo, J. & Elowitz, M. B, Nature, 2010, 465, 86-90)

Hope that helps

I am also very much of this opinion:

"but what makes a model 'good' is the fact that you learn something from it."

This is a message that is hard to convey to many engineers or physicists since many of them have a kind of binary view on modeling. I mean a lot of people are convinced that if your model does not include everything, it cannot be informative or useful. This is wrong in my opinion.

Modeling is always performed with "a goal" and models should only be qualified w.r.t. the achievement of this goal. If you need a model for designing a PID controller, then an open-loop step response followed by the identification of a low-order linear model will be enough. If your goal is to do optimization, then it needs to be adequate for optimization in that it should be able to let you predict the conditions of optimality of your system (see e.g. the adequacy conditions in Forbes and Marlin papers). You may also like to validate/infirm assumptions. In such a case your model should elaborate around identifiable parameters whose identified values will help you decide. Finally if you want to predict accurately the behavior of several measured quantities you need to develop your model with this in mind. This is in my opinion what detailed models are intended for (although often being to detailed and complex lead to good fitting but poor predictions), the best being always to focus on the outputs (i.e. the measured quantities) that you are really interested in.

Thank You so much for informative discussions...