The set of spheres for the whole protein is then processed to find spheres which appropriately describe the binding site.
First, score spheres by their LIGSITE solvent-accessibility score and remove the most exposed.
Merge remaining closely overlapped spheres and remove redundant spheres.
Cluster remaining spheres and sort clusters by their LIGSITE score to output 50-100 spheres per site.
Optionally specify macromolecule residues and atoms to define proximity to included or excluded sites. Sphere generation is fast enough to apply iteratively.
In the absence of information about where the ligand binds, sphinx/sublime automatically generate a sphere cluster which contains the binding site in most cases:
Binding site is in cluster 1 in 94/123 cases
Binding site is in cluster 2 in 15/123 cases
Binding site is in clusters 3-5 in remaining cases.
It is possible to use more than one cluster in a single run. DockIt will sample between the clusters.