Note: The workshop is full.
Biomolecular recognition can be very complex, involving not only van der Waals and electrostatic interactions between the partners, but also solvent contributions such as those arising from dielectric shielding or the hydrophobic effect. Binding involves solvent release from the binding site, and it can be accompanied by conformational rearrangements in one or both partners and the uptake or release of protons or ions. To understand and engineer biomolecular complexes, one would like to obtain a molecular view integrating dynamics and thermodynamics. For example, one should be able to understand the specific binding of a family of ligands, preferably with chemical accuracy for the binding affinities; at the same time, one should be able to elucidate the pathway for the binding/unbinding reaction and the nature of the conformational rearrangements involved. The essential tools in this area are (1) molecular dynamics simulations, to reveal time-dependent conformational dynamics [Frenkel96]; (2) free energy simulations, to compute binding affinities (both relative and absolute) and activation free energies [Chipot07].
The bottlenecks in this area have shifted [Aleksandrov10]. Accurate long-range electrostatic treatments are readily available, for example. On the other hand, to study broad families of ligands, a major problem is force field availability and accuracy, including the treatment of electronic polarizability. Similarly, while the conformational sampling, or `multiple minimum' problem has been alleviated by today's computers, it is often replaced by a `multiple state' problem [Aleksandrov10]: the need to establish protonation or oxidation states of sidechains or cofactors, the presence or absence of bound ions or waters, and sidechain and surface loop orientations. We see several main challenges and frontiers. The first is to integrate emerging tools for enhanced sampling, and bring them to bear on practical applications. For example, specialized efforts are made today to sample protonation/deprotonation of individual groups; constant-pH MD and coupled MD/MC tools come to mind [Khandogin06]. Related but different tools are used to sample the addition/subtraction of individual waters from a cavity or binding interface [Collins05]; mainly (but not exclusively) grand canonical MC simulations. Yet another set of sophisticated methods are needed to sample conformational relaxation of the protein on its rugged energy surface: replica exchange and multicanonical methods are important examples [Rosta09]. The first challenge is to improve these tools and their integration, both at the methodological and software levels.
A second major challenge is to connect the high level simulations to practical problems such as ligand (or drug) design and the inference of protein-protein interaction networks. For this, it is important to compare results from the most accurate simulation methods with results from much simpler models, where solvent is treated implicitly and some of the biomolecular degrees of freedom are coarse-grained [Simonson07]. Popular continuum dielectric methods are in this class. The physical basis and performance of these methods is far from being completely understood, as shown for example by the fast evolution of generalized Born solvent models in recent years [Feig04].
A third challenge is to expand the length scales and timescales that are typically simulated, to realistically explore the most exciting and biologically important systems. These include the ribosome, ATP synthase, and other cellular machines. For each of these systems, an integrated view is our ultimate goal, spanning a wide range of length scales. For example, we are beginning to have the ability to accurately predict and understand the interactions of small antibiotics with the ribosome [Aleksandrov08]; at the same time, methods are being developed to simulate its entire functional cycle, including translocation of tRNA molecules between several binding sites [Sanbonmatsu05]. Pushing the limits of system size and simulation length will be achieved with the help of new computational engines, such as volunteer computing and GPU processors. Volunteer computing has already led to exceptionally accurate solvation free energies [Shirts03].