The ultimate test of a model is the extent to which experimental data matches its predictions. The simplest biological and physical models frequently used in toxicology are those that predict a response which occurs in direct proportion to the amount chemical or physical agent applied (a linear exposure-response relationship). In some cases, the linear response is modified by assuming a threshold (level below which the stimulus is undetected or repaired) and a saturation level (a level beyond which no additional response is possible) which results in a sigmoid exposure-response curve. Often, however, important biological processes can't be adequately simulated using linear processes because these models assume systems that are at chemical/thermal equilibrium. When a biological system is at static chemical or thermal equilibrium, this usually implies it is a dormant or dead system.
Living, metabolizing systems, which are dynamically interactive, may be best represented by non-linear response models such as IPR. When such a model does correctly predict complex responses exhibited by the biological system, the model can be assumed to contain elements consistent with the affected biological process. As such, it can be very useful to help establish the underlying basis for the non-linear exposure-response relationships.