Taking a Bath by Computer

Computer simulations of proteins or DNA bathed in water rely on models that describe the interactions among water molecules and between water and the biomolecule under study. These models are often quite simplistic, says Jordan. "Water-water interaction models are usually tested to see how well they describe the properties of bulk water because experimentally we know a lot about bulk water. But in biological systems, water is in a wide range of environments. Far away from a protein, water should have a structure similar to the bulk. But near the protein, the structure can be very different. The water molecules close to the protein, on average, will be involved in less hydrogen bonding than those in the bulk. In this sense, one has a situation somewhat analogous to a cluster."

Scientists have long realized that water clusters could test the usefulness of water-interaction models for environments other than bulk water. The lack of experimental data and the difficulty of doing accurate quantum chemical computations, however, have hampered progress. Most modeling, points out Jordan, treats the interaction between two water molecules as independent of interactions with other water molecules in the system, even though it is well known that these other interactions have an effect. In a cluster of three water molecules, for example, the interaction between two of them distorts the electron distribution on the third, which in turn modifies the first interaction. How to account for this distortion -- known as "polarization" -- is the main challenge, says Jordan, to developing reliable models, and this is where quantum mechanical calculations come into play.

Jordan is one among a handful of scientists using quantum calculations to study water clusters. It is only recently, he says, that the computer firepower needed to do these computations on small clusters became available. Such calculations give the geometries and binding energies of the clusters and provide other data important for testing water-interaction models. The CRAY C90 at Pittsburgh made it feasible for researchers C. J. Tsai and K. Kim in Jordan's group to carry out these calculations using the GAUSSIAN 92 and Molpro software packages.

go back to the main screen