Our protein folding program, Chorus, works by using a knowledge base of known structures in a Monte Carlo search procedure. The knowledge base is represented in the form of phi-psi probability tables. We found in the last year that this table can predict the regions of the sequence of the protein that act as the folding initiation sites. This is the first time that such sites could be predicted from the sequence alone. The fact that the probability tables can predict such sites demonstrates the power of the information contained in these tables. We found two major problems with Chorus when tried on a number of real systems. The main problem was the potential energy function. a useful potential energy function should be able to discriminate true structure from a myriad of wrong structures. However, no potential function that we have tried so far could do this. The potential that has the best theoretical foundation and therefore the best chance for success is that which combines the hydrogen bond potential and hydrophobic free energy. The problem here is the enormous computer time required to compute the latter. We are currently working a simplified scheme for computing this free energy and also on using the Intel parallel machine for this purpose.
