We present a detailed study of the performance and reliability of design
procedures based on energy minimization. The analysis is carried out for
model proteins where exact results can be obtained through exhaustive
enumeration. The efficiency of design techniques is assessed as a
function of protein length and the number of classes into which amino
acids are coarse grained. It turns out that, while energy minimization
strategies can identify correct solutions in most circumstances, it may
be impossible for numerical implementations of design algorithms to meet
the efficiency required to yield correct solutions in realistic
contexts. Alternative design strategies based on an approximate
treatment of the free energy are shown to be much more efficient than
energy-based methods while requiring nearly the same CPU time. Finally,
we present a novel iterative design strategy that incorporates negative
design with the use of selected decoy structures that compete
significantly with the target native state in housing the designed
sequences. This procedure allows one to identify systematically all
sequences that fold on a given target structure.
