The automatic search for ligand binding sites in proteins of known three-dimensional structure using only geometric criteria

The biological function of a protein typically depends on the structure of specific binding sites. These sites are located at the surface of the protein molecule and are determined by geometrical arrangements and physico-chemical properties of tens of non-hydrogen atoms. In this paper we describe a new algorithm called APROPOS, based purely on geometric criteria for identifying such binding sites using atomic co-ordinates. For the description of the protein shape we use an alpha-shape algorithm which generates a whole family of shapes with different levels of detail. Comparing shapes of different resolution we find cavities on the surface of the protein responsible for ligand binding. The algorithm correctly locates more than 95% of all binding sites for ligands and prosthetic groups of molecular mass between about 100 and 2000 Da in a representative set of proteins. Only in very few proteins does the method find binding sites of single ions outside the active site of enzymes. With one exception, we observe that interfaces between subunits show different geometric features compared to binding sites of ligands. Our results clearly support the view that protein-protein interactions occur between flat areas of protein surface whereas specific interactions of smaller ligands take place in pockets in the surface. (C) 1996 Academic Press Limited.