The experiments are normally performed in vitro or in vivo (with rodents , preferably mice). You have to run an experiment to obtain a log dose-response curve for both therapeutic or efficacious effect and another for toxic or lethal effect. Hence, draw the two sigmoid curves and establish the corresponding the NO Observed Adverse and Lowest Observed Adverse Effect Levels , consecutive and respectivally, for a type of Adverse Response you are looking for. If it is a non therapeutic agent just the later (Toxic).
There are many references regarding these methods. They are quite simple.