Experimentally Biased Molecular Dynamics Simulations Using Neutron Reflection Data

Structural modeling of proteins and characterization of their equilibrium properties is an essential part of understanding the complexities of organic matter. In addition, a wide array of biological functions occur at or are regulated by a lipid membrane. For this reason, characterization of proteins at

bilayer interfaces is a crucial research pursuit. Neutron reflectometry (NR) is a powerful experimental technique for characterizing some of the interaction of proteins with lipid bilayers. Molecular dynamics (MD) simulations provide

a computational framework to model details beyond what is accessible in experiment. Traditionally, these two methods are performed independently and their results are used in complement to one another. In this work, a new method is developed and explored in order to integrate NR data into MD simulations. The algorithm takes the real{space

model of the protein derived from NR scattering data and compares it to the corresponding data calculated from simulation, updating this information as the simulation progresses. This comparison is used to construct a potential

that biases the protein's dynamics in order to garner better agreement between the MD and NR data. First, a review is conducted of MD simulations, NR scattering, and the

specific experimental systems under investigation. This review gives a general sense of how other experimental techniques and MD simulations are integrated and how NR data may be used in simulation. Next, an analysis

of the effect biasing has on the dynamics of a system is explored. This motivates the use of a linear bias potential.

After supplying motivation for the method, details of the implementation are presented. The linear difference of the NR and MD data is used in the potential, and the MD data is calculated using two methods. In one method, the data is calculated only taking into account the protein conformation

locally in time (memoryless bias). In the other, the running average of the data is tracked and used in the comparison to NR (histories bias). The effectiveness of these two methods is explored on two model systems. The first system is a small helical peptide that has degrees of freedom

equivalent to a rigid rod, under the energy scales explored. This system is not studied using neutron data, but rather a highly artificial potential is generated to represent only a confined subset of available conformations. Since the data is a reduction of the structure from three dimensions to one, the effects of the bias are not known before the simulations are performed.