ABSTRACT
Combinatorial chemistry, a major drug discovery tool, heavily relies on chemoinformatics ( 1 ,2) and molecular modeling to manage the huge flux of structural information related to potentially feasible combinatorial products, and to intelligently direct synthesis efforts toward products with a maximal chance of fulfilling the stringent conditions required of a drug molecule. Until recently, even the numberof combinatorial products that potentially could have been obtained on the basis of commercially available starting materials and relatively simple two-or three-step chemistries would have largely exceeded the available modeling capacities. In response to these novel constraints, molecular modeling tools dedicated to combinatorial chemistry (3-5) have been successfully developed. Soft-
ware packages aimed at processing large sets of molecules are nevertheless restricted to the fast bidimensional (topological) (6-8) description of combinatorial products, thus avoiding the computational effort due to geometry buildup and conformational sampling. Conformer generation may require seconds to minutes of CPU time per molecule, depending on the effort spent to score the relative relevance of the visited phase space region (using a simple bump check criterion toreject impossible geometries vs. performing a full-blown potential energy evaluation). Therefore, 3D descriptors may be routinely used to characterize libraries containing l<f-lOS compounds.
