Drug design with improved MM-PBSA: tackling the entropy problem

Abstract

Small molecules intended as active ingredient in drugs usually bind proteins to assert their effect. Of course, side-effects occur for any drug, and lowering the dose used usually lowers side effects. Therefore, for designing drugs, it is crucial to find a compound that requires a low dose to assert its desired effect. The dose that can be used, depends on how strong the drug binds the target protein. Therefore, it is crucial to find strongly binding compounds. Since physically synthesizing all compounds considered would be resource and time intensive, methods to predict binding strength with computers can save lots of time and resources. Only the few compounds that bind strongest according to the predictions would have to be synthesized and tested. One computational method for estimating binding strength with intermediate speed and accuracy, is called molecular-mechanics-Poisson Boltzmann surface accessible (MM-PBSA). This method considers some enthalpic terms that increase binding strength, such as electric attraction between drug and protein, and it has an entropy term, which reflects the energy cost of lost drug- and protein flexibility once they bind to each other. However, the enthalpic terms are way faster to calculate than the entropy term. Additionally, when comparing two similarly shaped drug-candidate molecules, which bind to the protein in the same way, the entropy term is often similar. Therefore, reasonably predictions of relative binding strength have been obtained by using MM-PBSA on similar drug molecules binding to the same protein, while ignoring the entropy term. Because drug candidates are not always similar, and the entropy term in MM-PBSA is not always similar for these candidates, it remains important to find methods that can estimate the entropy without slowing MM-PBSA too much, while improving its accuracy. In this literature review, several historical and new methods for entropy estimation are explained and discussed. Benefits and limitations are listed for each method. Since each method makes different assumptions to speed up calculation, which may be applicable in some cases, but not all, it is listed what the most appropriate use-case would be for each method. The main conclusion is that there are many promising methods, but some require excessive amounts of simulations of protein and ligand to reach accurate entropy estimates. Only when computational resources would increase, would they be useable with MM-PBSA for investigating drug-protein binding. The most promising method is the interaction-entropy method. It is very fast to calculate, while it only relies on the same assumptions frequently used in MM-PBSA calculations. Furthermore, it has been proven to improve MM-PBSA predictions on average, even when including protein-drug couples that do not fall within its ideal use-case. Finally, although several separate methods exist that are compatible with MM-PBSA, only limited performance comparisons in terms of accuracy and speed are available. It is therefore too early to tell which method should become the new standard for use with MM-PBSA, or to draw definite conclusions regarding the most applicable method for a given use-case. Nevertheless, we hope that this review can guide fellow scientists in choosing an entropy method for use with MM-PBSA.