Derivation of class II force fields. VIII. Derivation of a general quantum mechanical force field for organic compounds.

A class II valence force field covering a broad range of organic molecules has been derived employing ab initio quantum mechanical "observables." The procedure includes selecting representative molecules and molecular structures, and systematically sampling their energy surfaces as described by energies and energy first and second derivatives with respect to molecular deformations. In this article the procedure for fitting the force field parameters to these energies and energy derivatives is briefly reviewed. The application of the methodology to the derivation of a class II quantum mechanical force field (QMFF) for 32 organic functional groups is then described. A training set of 400 molecules spanning the 32 functional groups was used to parameterize the force field. The molecular families comprising the functional groups and, within each family, the torsional angles used to sample different conformers, are described. The number of stationary points (equilibria and transition states) for these molecules is given for each functional group. This set contains 1324 stationary structures, with 718 minimum energy structures and 606 transition states. The quality of the fit to the quantum data is gauged based on the deviations between the ab initio and force field energies and energy derivatives. The accuracy with which the QMFF reproduces the ab initio molecular bond lengths, bond angles, torsional angles, vibrational frequencies, and conformational energies is then given for each functional group. Consistently good accuracy is found for these computed properties for the various types of molecules. This demonstrates that the methodology is broadly applicable for the derivation of force field parameters across widely differing types of molecular structures. Copyright 2001 John Wiley & Sons, Inc. J Comput Chem 22: 1782-1800, 2001

Partially unified multiple property recursive partitioning (PUMP-RP): a new method for predicting and understanding drug selectivity.

The decision tree method for classification problems has been extended to accommodate multiple dependent properties. When applied to drug discovery efforts this means a separate activity class can be predicted for each of several targets with a single tree model. A new tree representation and growth procedure, PUMP-RP, has been developed. The final architecture of the tree allows for easy interpretation as to which independent variables and split values are important for all targets and which are specific to a given target. It should thus be usefully applied to studies of drug specificity. A side benefit of the new method is that it can make use of data with missing (or even sparse) dependent property values. This has the potential to leverage copious data from an older, well-studied target while beginning to study a newer target for which only a small amount of data are available.

Partially unified multiple property recursive partitioning (PUMP-RP) analyses of cyclooxygenase (COX) inhibitors.

We have carried out partially unified multiple property recursive partitioning (PUMP-RP) analyses on a database of cyclooxygenase (COX) inhibitors, using CART methods implemented in Cerius(2). Three sets of physicochemical descriptors (ISIS public keys, DAYLIGHT Fingerprints, and Cerius(2)) were computed for the database molecules which were divided into two groups, assigned as training (89%) and test (11%-selected using diversity analyses tools in Cerius(2)) sets. The descriptors which led to the discrimination of active and selective COX-2 inhibitors included ISIS Key #59 (Snot%A%A), Balaban electrotopological index JY, partition coefficient AlogP, and Jurs surface area descriptors (FNSA, FPSA, and PPSA). A strong correlation is obtained between the predicted and experimental COX-2 inhibitory activity and a moderate correlation for selectivity of the COX-2 inhibitors, both in the training and test sets. Application of the RP trees to a validation set of Merck cyclooxygenase inhibitors shows good consistency with the COX-1 and COX-2 activity data, albeit moderate consistency with the selectivity data. Compared to the independent RP models (obtained by considering each activity separately), the PUMP-RP decision trees provide easier identification and interpretation of those descriptors that are common to both COX-1 and COX-2 activities. Similarly, they are easier to distinguish the descriptors that discriminate the two activities. The study represents a preliminary validation of the PUMP-RP method described in the previous article of this issue.