Reactivity Prediction Models Applied to the Selection of Novel Candidate Building Blocks for High-Throughput Organic Synthesis of Combinatorial Libraries

Quantitative structure−property relationships (QSPRs) expressing the reactivity of compounds on the basis of molecular descriptors have been developed and applied to the computer-aided selection of synthons of appropriate reactivity for the high-throughput synthesis of combinatorial libraries. Our approach explicitly models the influence of substituents on the activity of the reactive center (RC), introducing specific molecular descriptors for their electronic, steric, and field effects (including the solvent effects) as a function of the 2D and 3D substituent−RC distances. Therefore, the approach requires a much smaller number of empirical "substituent constants" than the classical Hammett approach. These constants only depend on the chemical nature of the substituents and not on their relative position with respect to the RC. A general pKa prediction model was obtained by calibrating the weighting factors that express the relative influences of the electronic and field effect descriptors on the acidity of functional groups, using a learning set of about 500 organic amines and acids. A QSPR model expressing the degrees of conversion of a reference amine in the amide synthesis reaction, in terms of the descriptors of the carboxylic acids, was then derived. The used learning set included 100 out of the 150 acids for which the conversions were experimentally determined at the first stage of a typical selection process of building blocks for combinatorial synthesis. The predicted percentages of conversion of the acids not included in the learning set showed (absolute) errors not exceeding ±20%. As a consequence, this model is a useful computational tool in discriminating between reactive and inappropriate compounds from molecular databases, retrieving the building blocks that are most likely to comply with the reactivity criteria.