ABSTRACT

Calculating free energy changes associated with a process is a major application of computer simulations. For example, knowing the free energy required to move small molecules across a biological membrane is a prerequisite to estimate the translocation rate of drugs that enter the cytoplasm by passive membrane permeation. These processes often include significant free energy barriers, which makes the calculation impractical using straightforward simulations. A large number of methods have been proposed to overcome this problem by biasing the system toward states that are unlikely to be observed in an unbiased simulation. We compare three widely used methods, well-tempered metadynamics, umbrella sampling, and replica exchange umbrella sampling, in their ability to accurately and efficiently calculate free energy profiles for the translocation of various solutes across a phospholipid bilayer. We show that for charge-neutral permeants and those with small dipole moments, all three methods perform equally well. For charged and strongly polar solutes, however, we find significant differences between these methods, including the possibility of convergence failure over typical simulation time scales. We discuss how such failure can be detected and overcome. Furthermore, we demonstrate how we can infer useful information about the physical behavior of a system from the differences in performance among the three methods.