摘要 : Initial combinatorial library designs were based on 2D substituent properties. Subsequently, two important extensions were introduced to improve the approach: use of pharmacophores to introduce 3D information, and performing calcu... 展开 Initial combinatorial library designs were based on 2D substituent properties. Subsequently, two important extensions were introduced to improve the approach: use of pharmacophores to introduce 3D information, and performing calculations on the enumerated library products rather than just on the substituents. Unfortunately, practical compromises due to the large number of possible products, the large number of conformations per product, and the explicit dependence on the scaffold limit the application of these extensions in five important ways: (1) to small virtual libraries, (2) to only 3- or 4-point pharmacophores, (3) to inadequate conformations sampling, (4) to simplistic diversity measures, and (5) to requiring a complete new calculation for every new library. The 3D oriented substituent pharmacophores have been developed to overcome these limitations. These add two additional points and corresponding distances to each substituent pharmacophore. This adds little additional computation beyond a normal 3D pharmacophore calculation on the substituents, but recaptures most of the orienting information lost in breaking up the enumerated products into fragments. Two main approximations are still implicitly required: the combinatorial conformer assumption and the template alignment assumption. In turn, however, they are designed to account not just for the 3- and 4- point pharmacophores, but for pharmacophores with up to 9 points in enumerated products with three sites of diversity. Perhaps more importantly, pharmacophore calculations are shown to be very sensitive to conformational sampling. The small number of substituents, plus the small number of rotatable bonds per substituent, permits very thorough conformational sampling. For a rigid scaffold with three diversity sites of 1, 000 candidate substituents each, the number of molecules to analyze is reduced by a factor of 10~6, and the number of conformations per molecule is reduced by another 10~4. In addition, the modest number of pairwise substituent similarities permits the creation of a Euclidean property space by MDS. This allows for sophisticated experimental design methods that require coordinates, rather than just the counting of the number of set bits in a library union fingerprint. Finally, oriented substituent calculations are scaffold independent and transferable. They can be stored in a database and need not be repeated for every new library. Thus, there are some approximations in the correspondence between oriented substituent pharmacophore similarities and enumerated product pharmacophore similarities. However, these errors are minor compared to the five advantages that the new method enables: large virtual library sizes, thorough conformational sampling, accounting for 1- to 9-point pharmacophores, creation of a Euclidean property space, and a reusable database of precomputed substituent values. 收起