We discuss methods for the determination of the effective pairwise
interactions between amino acids in globular proteins in order to be
able to easily recognize the native state conformation of any protein
sequence among a set of decoy structures. The first method entails the
application of a numerical strategy to a training set of proteins that
maximizes the native fold stability with respect to alternative
structures. The extracted parameters are shown to be very reliable for
identifying the native states of proteins (unrelated to those in the
training set) among thousands of conformations. Folding transition
temperatures are estimated for a few proteins for which reliable
alternative structures have recently been generated. The only poor
performers are proteins with stabilizing heme groups whose complexity
cannot be captured by standard pairwise energy functionals. The key
ingredient of this technique is the knowledge of viable decoys for each
protein sequence in the training set. We then present a second strategy
which circumvents this difficulty. This method relies on the fact that
protein sequences are special compared to random heteropolymers and are
characterized by high thermodynamic stability in their native
conformations. We validate the technique on a lattice model of proteins,
we apply it to real proteins and carry out tests of the quality of the
extracted interaction parameters. We find that this novel technique
leads to good results that are comparable to those obtained with the
first method.
