Chemistry Reference
In-Depth Information
host of other scenarios where the atomic level resolution they provide can lend valu-
able insights into the systems of interest. Using computer simulation, one can even
study processes that are impossible or unphysical in real life (e.g., computational
alchemy). Thus, the potential of biomolecular computer simulation is undeniable.
Despite the many exciting uses of molecular simulation, there is no formal guar-
antee that the results will reflect reality. The quality of a molecular dynamics (MD)
or Monte Carlo (MC) simulation depends on the degree of sampling achieved during
the simulation, which we shall not consider here, together with a satisfactory descrip-
tion of the intra- and intermolecular interactions in the system, that is, an accurate
force field (FF). Decades have passed since the first MD simulations were performed,
but the effort to continuously improve the quality of the underlying FFs continues to
this day (Cheatham and Brooks 1998; Wang et al. 2001; Mu, Kosov, and Stock 2003;
Mackerell 2004; Allison et al. 2011; Schmid et al. 2011; Zhu, Lopes, and Mackerell
2012). It has been expressed that both the general philosophy and the parameters for
biomolecular FFs are probably converging (Wang et al. 2001). However, we argue
and demonstrate here that there is still significant room for improvement, and that a
markedly different parameterization philosophy is probably necessary.
5.2 CURRENTLY IDENTIFIED FORCE FIELD (FF) CHALLENGES
Two of the main biomolecular FF challenges are: (i) The generation of accurate
torsional potentials for peptides and proteins, and (ii) improved parameters for the
description of nonbonded interactions. A commonly cited example of the former is
Simmerling's (and others) exposure of a bias toward the α-helical secondary structure
in the AMBER ff94 and ff99 FFs (Okur et al. 2003). Simmerling was able to achieve
an improved balance between the possible secondary structure propensities through
modification of the torsional potentials, resulting in the AMBER99sb FF (Hornak et al.
2006). While incorrect torsional potentials may have contributed to other reported
artifacts (Mu, Kosov, and Stock 2003; Allison et al. 2011), they are not addressed here
because FST theory will not help with this part of the parameterization. Our focus will
only be on the use of FST/KB (Kirkwood-Buff) theory to help improve intermolecu-
lar nonbonded interactions for simple, effective charge, nonpolarizable FFs.
One strict test of the quality of the nonbonded parameters is the ability to repro-
duce appropriate protein-ligand association interactions and binding free energies.
In FF-based computational drug design, it is notoriously difficult to predict binding
free energies quantitatively, or to rank ligands based upon their binding affinities,
even in cases where the correct ligand-receptor pose has been predicted (Lazaridis
2002; Bonnet and Bryce 2004). Consequently, it has been suggested that the non-
bonded parameters may be deficient (Lazaridis, Masunov, and Gandolfo 2002). This
issue extends further to question the quantitative features concerning assembly and
aggregation equilibria.
Furthermore, because many proteins are only marginally stable (Dill, Ghosh, and
Schmit 2011), a correct description of nonbonded interactions is necessary for attain-
ing quantitative conformational equilibria, melting temperatures, and other thermo-
dynamic properties associated with protein folding. For example, it was reported that
a 28-residue miniprotein that adopts a ββα motif had a simulated melting temperature
