Computer prediction of protein surface features and antigenic determinants.

For many years it has been appreciated that the arrangement of amino acids in the linear sequence of a protein is responsible for the three dimensional structure of the folded protein. However, until recently very little practical information could be obtained from amino acid sequences, because of our imperfect understanding of the way that the individual amino acids influence the conformation of a peptide chain. At present there are several useful ways of predicting conformation from sequence, including methods based on energy minimization, frequency of occurrence of amino acids in particular secondary structures, and consideration of the solubility of the amino acids in aqueous and organic solvents. These methods seldom generate information that is useful on a practical level, partly because they attempt to predict too much detailed information from a given sequence. In the development of my method, I asked a simpler question, namely, is it possible to predict the locations of antibody binding sites on a protein sequence, regardless of any consideration of the precise conformation of the peptide chain? Using this criterion and a data set of twelve well-known protein antigens, I developed a simple hydrophilicity analysis that reliably predicts the locations of antigenic residues in protein sequences. A listing of experiments where the outcome was predictable by my method is presented in Table I.