A longstanding and fundamental open problem is to predict the three-dimensional structure of a protein from its amino acid sequence. Because function derives not just from the molecular sequence, but from the spatial arrangements of atoms, structure prediction has the potential to add tremendous value to genomic data. The combinatorial complexity of the problem still presents a major hurdle, although recent advances in pattern recognition, simplified representations of protein chains, efficient scoring functions and efficient search algorithms, combined with escalating computing power, are making some head way. In addition to the three-dimensional structures of proteins, biological function also depends on dynamic processes at the molecular level. New simulation techniques, combined with advances in computer hardware and software, allow classical simulations of important reactions such as protein folding. Stanford researchers recently implemented a clever parallel use of tens of thousands of otherwise idle processors over the internet to enable modeling of reactions that take over a second. The combined use of classical and quantum mechanical representations is another frontier area that is providing access for the first time to the details of enzyme reactions. In the coming decades, increasingly complex simulations of molecular motions have the potential to provide crucial insights into the most complex biochemical processes.