(316e) Prion Protein Conformational Statistics Via on-the-Fly Free-Energy Parameterization

On-the-fly free-energy parameterization is a collective-variable biasing approach akin to metadynamics and the adaptive biasing force method, with one important distinction: rather than attempting to build the bias pontential during the simulation using history-dependent accumulation of bias increments, it simply accelerates sampling of collective-variable space using temperature-acceleration and accumulates increments of thermodynamic mean forces.  These increments are tallied and integrated into potentials of mean force which ultimately converge.  Because it does not use a history-dependent bias, on-the-fly parameterization is not subject to the same pathologies arising from hidden variables that plague metadynamics and ABF.  We demonstrate this approach's advantages over metadynamics by revisiting the classical case of beta-hairpin unfolding in the prion protein PrPC.  We predict that the free-energy minimum conformation is that observed in the crystal structure, in contrast to previous predictions of metadynamics.  We confirm our prediction using several different collective variables.  We confirm that well-tempered metadynamics also converges, but we observe that the errors in its prediction of the potentials of mean force are larger than those obtained from on-the-fly parameterization for the same amount of computation, and the metadynamics convergence behavior is more strongly dependent on sampling of hidden variables.  Although hidden variables remain a central problem in free-energy methods, on-the-fly free-energy parameterization represents a new approach which may be less susceptible to errors arising from hidden variables than more popular and well-established methods.