Department of Structural Biology, Stanford University, Stanford, CA.
Simplified models have played an important role in attempts to model protein folding by computer simulation: By eliminating much of the detail, such models are computationally more tractable. While such models are useful, a central question is whether there is sufficient detail in a particular model to make the native fold stand out from the vast number of alternate non-native folds. More specifically, one hopes that the native fold has a very low free-energy value.
In the first part of this talk we use a highly simplified representation of proteins that allows all possible conformations to be generated. We search this conformational space in order to find those folds that have low energies with a simple pair-wise residues contact potential. Results on a wide variety of small proteins show that, even at this low resolution, there is significant selectivity. Using the correct sequence of the protein, selects for those folds that are more similar to the native conformation of the protein. Using a shuffled sequence does not show such selectivity.
In the second part we use a more detailed model specially to chosen to represent protein native structure accurately while still having a small number of possible conformations. By keeping segments of secondary structure fixed at the native positions, it is possible to exhaustively search this model. We show that different commonly accepted energy functions differ very much in their ability to distinguish the correct native folded structure from a large number of well- constructed decoy structures. Certain combinations of energy functions are much more successful at such discrimination, suggesting they will be useful for both folding and inverse folding studies.