Abstract

A methodology designed to address the inverse globular protein-folding problem (the identification of which sequences are compatible with a given three-dimensional structure) is described. By using a library of protein finger-prints, defined by the side chain interaction pattern, it is possible to match each structure to its own sequence in an exhaustive data base search. It is shown that this is a permissive requirement for the validation of the methodology. To pass the more rigorous test of identifying proteins that are not close sequence homologs, but that have similar structure, the method has been extended to include insertions and deletions in the sequence, which is compared to the fingerprint. This allows for the identification of sequences having little or no sequence homology to the fingerprint. Examples include plastocyanin/azurin/pseudoazurin, the globin family, different families of proteases and cytochromes, including cytochromes c' and b-562, actinidin/papain, and lysozyme/alpha-lactalbumin. Turning to supersecondary structure prediction, we find that alpha/beta/alpha fragments possess sufficient specificity to identify their own and related sequences. By threading a beta-hairpin through a sequence, it is possible to predict the location of such hairpins and turns with remarkable fidelity. Thus, the method greatly extends existing techniques for the prediction of both global structural homology and local supersecondary structure.