Methods for estimating the structure and energetics of water around biomolecules are presented with the objective of improving the treatment of the biomolecular simulation environment as well as facilitating the use of complex, rigorous water models in the context of structure prediction problems that demand cheap solutions. Salt solutions around biomolecules are studied using an implicit solvent model with explicitly represented ions, revealing that the structure of the ion atmosphere is much more sensitive to the assumption of a particular biomolecular dielectric constant than the dynamical properties of ions in solution. Extreme accumulation of ions was observed around the phosphate groups of a highly charged B-DNA dodecamer, suggesting that the structures of salt ions in solution around nucleic acids are extremely sensitive to the details of the biomolecular model. Next, a method is developed for converting a computationally expensive solvent model into a fast solvent energy estimator that takes as input only the identities and distances between atoms in the biomolecular solute. A robust method for fitting the solvent energy estimator based on any choice of benchmark model is developed based on setting up and solving a simple linear least squares problem. In this work, Poisson electrostatics and molecular surface area approximations make up the benchmark model used to parameterize the estimator, but when applied to biomolecular docking problems the estimator reveals that its benchmark model wrongly identifies numerous theoretical complexes between two proteins with greater stability than the experimentally observed native complex. One possible source of the inaccuracies in Poisson electrostatics, the lack of a solvent structure, is then studied. The striking result that water molecules create a strong, positive electrostatic potential within solvated cavities with no electrostatic properties of their own is rediscovered and explored for the case of a complex cavity created by a protein. This electrostatic potential arises from the polarity and finite size of individual water molecules and is not captured by Poisson electrostatics. It is also found that consideration of this effect greatly increases the correspondence between the energies obtained by Poisson electrostatics and a model that represents the water atomistically. These result suggests that, a least for structure prediction problems where energetics are most important, it may be possible to augment implicit solvent models to capture the effects of explicit solvent models, and that it is possible to parameterize a computationally cheap pairwise-additive model to closely mimic much more complicated solvation energy functions. As problems in biomolecular structure prediction begin to involve more challenging molecules such as RNAs, it may be fruitful to revisit the studies of ionic structure around biomolecules with better treatment of solvation effects pertaining to individual ions. Future studies in biomolecular structure prediction will be able to incorporate highly sophisticated solvent models at little additional cost
