Our models and algorithms relie on the alpha-shapes theory wich is briefly discribed in another section. As a summary, we consider atoms as spheres with their Van der Waals radii plus a constant probe radius. A molecule is thus depicted as a union of such spheres. We then inrtoduce a polyhedral model of the molecule where the vertices of the polyhedron are the centers of thoose atoms that have a part of their surface not completely engulfed by other spheres. The faces (triangles and edges) of this polyedron also depict intersections of atoms that occur freely outside all other atoms. This polyhedron is also often reffered as the dual-complex of the molecule and is desrcibed more formaly in another section.
Important remark : the dual-complex depicts exactly the molecular surface, but formaly it is not a representation of the molecule's surface. For instance it's vertices are not placed on the molecular surface, they are the centers of the molecule's atoms.
Remind that except for mouths, a facet in the polyhedral model indicates a blockage of the probe by the three atoms centered at the facet vertices. In our software we discriminate between five types of facets in the dual complex :
Irregular facets reffer to the case where a probe sphere is blocked on both sides by the same three atoms.
Important remark : it is worth to insist on the fact that Pck seeks for pockets solely on geometric criteria. Thus, when we talk of pockets we should precise geometric pockets, as opposed to "ligand binding pockets" or "biological pockets". Nevertheless, as expressed in [REFS], biological pockets most often dwell in geometric pockets, thus a geometric definition of pockets might be a preliminary step for the study of biological pockets.
Pockets are usualy discriminated according to their shape : one talks of cavities for thoose voids that are totaly buried in the protein core, isolated from solvant space, of (closed) pockets for thoose that shrink in a narrow mouth before opening to the solvant; and of clefts for thoose that open wide to the solvant.
![]() |
In our software the detection of pockets is achieved in the polyhedral model. We implemented two algorithms meant to detect different kind of pockets :
We define a convexity index for a pocket by comparing volumetric values of it's polyhedral model and of it's convex hull.
IndexV = 100 * V(ConvexHull) / V(Pocket)
This provides information on the "globularity" of the pocket. See (fig 3.a) for a 2D exemple.
As a supplementary information we also provide the same information with regard to the surface instead of the volume. This information is to handle with care : The volume of a polyhedron is always smaller than the one of it's convex hull. As for the surface this is no longer the case. In 2D, the surface of an object is always greater or equal to the one of it's convex hull, which is not necessarily the case in 3D. Consider for instance a triangular flat box (Fig 3.b) with a long pin comming out from one of it's corners. The surface of the object will bee approximatively the surface of the box, whereas the surface of the convex hull will be approximatively the one of the tetrahedron composed of the corners of the box plus the one at the top of the pin.