Impact of massively parallel computation on protein structure determination.

For the past two decades, an important paradigm in protein chemistry has been the assertion that a biologically active protein is at thermodynamic equilibrium and therefore adopts its minimum free energy structure. Although some evidence now suggests that not all proteins conform to this notion, it is true often enough to remain an important guiding principle in structure determination, whether by direct computation or by the computationally assisted approaches of diffraction and resonance. Among the difficulties in predicting structure from sequence are the lack of a useful potential function incorporating the influence of solvent and the inability to sample the phase space efficiently or even to determine whether a free energy minimum is, in fact, the global minimum. These problems are general. Although they are greatly mitigated by experimental information, they become increasingly severe as empirical constraints are reduced. We review the difficulties involved in the general problem of protein structure prediction and discuss the impact of increased computer power in the context of new approaches to solvation and parallel algorithm design. A general focus of our discussion is the need to understand the theoretical basis for effective theories and to accommodate in numerical methods the interplay of different temporal and spatial scales.