Generating and testing protein folds

Abstract

Prediction of three-dimensional protein structure has become once more the stage of intense activity. New approaches combine methods from computational chemistry and statistical mechanics with sequence alignment procedures and analyses of known structures. A key role is played by effective potentials derived from the database of protein structures, which provide a simpler description of the protein conformation. The novelty in the prediction methods is the use of sequence and structure libraries, which are screened against each other in the search for compatible sequence-structure matches. The ultimate goal is to predict which of the known folds is compatible with a given sequence in the absence of detectable sequence homology. Although this cannot yet be achieved reliably, results obtained so far are very promising, a sign for some that a practical solution to the protein folding problem may be in sight. In ab initio tertiary structure predictions, which do not rely on an existing fold, progress has been slower. Computer experiments with very simple model systems are however providing valuable insight, and developments of smarter strategies for searching conformational space hold promise for the future.