Early attempts in designing lead-finding combinatorial libraries have focused on the size and diversity of the libraries [20-23]. More recently, the ''drug-likeness'' of compounds in combinatorial libraries has been addressed. The attribute of drug-likeness combines synthetic feasibility, protein-binding functionality, favorable ADME (absorption, distribution, metabolism, excretion), and toxicological properties of small molecules [24, 25]. In order to create drug-like libraries the phys-
icochemical parameters and functionality of the compounds must be optimized. Simple counting rules such as the ''rule of five''  from Lipinski and coworkers, related criteria such as the number of rotatable bonds or rings in a molecule , or related functional group filters  are usually employed to profile a combinatorial library. More sophisticated techniques such as neural nets [28, 29], genetic algorithms , or recursive partitioning  can in principle also be used to increase the drug-likeness of a library.
Teague and coworkers have recently pointed out that it may be more useful to design ''lead-like'' rather than drug-like combinatorial libraries . They argue that in a typical lead-optimization phase of a drug discovery program, key phys-icochemical parameters such as log P and molecular weight (MW) increase by 0.5-4 and 1-200, respectively, thereby placing the resulting optimized compounds outside of the drug-like space. Based on this assumption, Teague and coworkers suggested a design of lead-like libraries that have a log P profile of 1-3 and a MW of 100-350, well below the ''rule of five'' thresholds of clog P = 5 and MW = 500.
An important point, especially for smaller compounds as starting points for a lead optimization effort, is the appropriate functionality of the molecules. Teague and coworkers suggest using highly polar compounds or compounds bearing a single charge at physiological pH. Suitable functionality of small molecules accounts for an increased chance of sufficient binding affinity to a protein target but mainly addresses the specific binding of the compound to a particular target. It is therefore important to optimize the degree of functionality of a small molecule. Recently, we demonstrated how simple pharmacophore filter rules help in generating a drug-like profile of compounds in a library . Scheme 27.1 shows four simple functional groups that are sufficient to capture the most important functional properties of small drug-like molecules. These functional groups can be combined with what we refer to as pharmacophore points. These pharmacophore points include the following functional groups: amine, amide, alcohol, ketone, sul-fone, sulfonamide, carboxylic acid, carbamate, guanidine, amidine, urea, and ester. Using these simple pharmacophore points a simple set of rules has been derived from ''chemical wisdom'' to classify drug-like or lead-like compounds :
1 Pharmacophore points are fused and counted as one when their heteroatoms are not separated by more than one carbon atom.
2 Compounds with fewer than two or more than seven pharmacophore points are dismissed.
3 Primary, secondary, and tertiary amines are considered to be pharmacophore points but not pyrrole, indole, thiazole, isoxazole, other azoles or diazines.
4 Compounds with more than one carboxylic acid are dismissed.
5 Compounds without a ring structure are dismissed.
6 Intracyclic amines that occur in the same ring are fused (e.g. piperazine), i.e. they count as only one pharmacophore point.
Based on these simple rules it has been shown that about two-thirds of drug-like compounds as defined by their presence in drug-like databases such as Comprehensive Medicinal Chemistry (CMC) or MACCS-II Drug Data Report (MDDR) are classified as drug-like while only one-third of the compounds in the Available Chemicals Directory (ACD) pass the filter. Together with statistical tools that prune distributions of such important parameters as MW or clog P to be commensurate with those of drug-like molecules the above described pharmacophore filter provides a simple guiding tool in the design of virtual libraries that can easily be implemented.
Was this article helpful?