deletions | additions       diff --git a/subsection_Shape_Representation_A_molecular__.tex b/subsection_Shape_Representation_A_molecular__.tex  index 3b3e54a..6dd1fcd 100644  --- a/subsection_Shape_Representation_A_molecular__.tex  +++ b/subsection_Shape_Representation_A_molecular__.tex 
...
We grow/shrink molecular shapes by adding/removing the appropriate number of surface voxels.  \subsection*{ Shape Constraints}  Since we assume all shapes are registered to a common coordinate system defined by the anchor fragment, it is possible to exactly specify regions of space that a molecule should and should not occupy. We refer to these constraints as included and excluded shapes. The \textit{included shape} is a set of voxels that must be part of the target shape. An included shape can be used to require that the desired shape makes key contacts with the receptor or has sufficient bulk to fill a binding pocket.  An \textit{excluded shape} is a set voxels that must not be present in the desired shape. An excluded shape can be used to require that the desired shape does not clash with the receptor or exceed some maximum volume. Included shape and excluded shape constraints permit a user to `sculpt' a precise specification of the desired shape. An expert user may be able to create highly specific custom shape constraints. However, for purposes of our evaluation we consider only shape constraints derived directly and automatically from existing receptor-ligand complexes. These shape constraints would serve as starting points for modifications by an expert user, for example, to better incorporate tolerances to receptor flexibility in specific areas.  The structure of receptor-ligand complexes provide a natural starting point for the development of shape constraints. The receptor shape provides an excluded constraint, which can be shrunk to increase tolerance of minor clashes and to compensate for the potential plasticity of the binding site. The ligand shape can be used to develop an included constraint. However, constructing an included constraint directly from the ligand shape, even if the shape is shrunk or skeletonized, results in a highly specific shape query. Although this may be appropriate for `soft' similarity comparisons, which can tolerate small deviations in shape, a highly specific included shape constraint will only be able to identify molecules with nearly identical chemical scaffolds.  As an alternative to directly using the shape of a bound ligand, we derive \textit{interaction points} from the ligand-receptor complex. Interaction points are points at the center of clusters of ligand atoms that are potentially interacting with the receptor. Interacting ligand atoms are defined to be those no more than 6{\AA} away from a receptor atom, as measured between atom centers. These atoms are then clustered into groups of three or more that are now more than 4{\AA} across. The center of the cluster is an interaction point. An example of interaction points identified from a ligand-receptor complex is shown with the corresponding included and excluded shape constraints in Figure~\ref{iptsshape}. Interaction points provide a more general specification of the binding mode of a ligand that is less dependent on the ligand chemical scaffold and ignores non-interaction components of the ligand.