Physics-based models do not have this deciency. Fully atomistic simulations of coupled folding-binding with explicit solvent in principle provide the highest accuracy feasible, albeit requiring large computational resources in addition to enhanced sampling methods in order to sample binding. However, there are a number of reasons why atomistic simulations may not be the best choice for studying folding-binding transitions at the time of writing. e rst is that the energy functions, or force elds, themselves have some limitations, which may be particularly detrimental to an accurate treatment of unfolded or disordered proteins. For example, unfolded states using current protein force elds and water models are too collapsed and structured in comparison to any experimental measure , and nonspecic protein-protein association appears to be too favorable . Since this deciency could clearly bias any binding mechanism obtained in simulation, use of such force elds for studying coupled folding and binding may be questionable. e second issue relates to the sort of questions that are being asked. If a set of binding events at atomistic detail is needed, certainly all atom simulations are the only feasible approach. However, for a great number of interesting questions, this level of detail is unnecessary and may even make it harder to identify the e ects that are really important. Lastly, the computational cost for studying binding events, even with enhanced sampling, is still very high. is limits the possibility of varying the protein sequence or other conditions in order to test hypotheses. Coarsegrained simulation models, in which the number of congurational degrees of freedom as well as the complexity of the energy function are reduced, can help to overcome some of these deciencies. ey make it easier to tune both sequence and interactions, and, in some cases, to make accurate predictions of binding mechanism. eir major drawback is their limited predictive value for specic cases, particularly if a structure of the bound complex is not known. Nonetheless, there are many generic and specic questions that can be more straightforwardly investigated with the aid of coarse-grained models.