Search

Selection of optimal descriptors in quantitative structure–activity–property relationship (QSAR/QSPR) studies has been a perennial problem. Artificial Neural Networks (ANNs) have been used widely in QSAR/QSPR studies but less widely in descriptor selection. The current study used ANNs to select an optimal set of descriptors using large numbers of input variables. The effects of clean, noisy, and random input descriptors with linear, nonlinear, and periodic data on synthetic and real data QSAR/QSPR sets were examined. The optimal set of descriptors could be determined using a signal‐to‐noise ratio method. The optimal values for the rho parameter, which relates sample size to network architecture, were found to vary with the type of data. ANNs were able to detect meaningful descriptors in the presence of large numbers of random false descriptors. © 2003 Wiley Periodicals, Inc. J Comput Chem 24: 891–897, 2003

The goal of quantitative structure–pharmacokinetic relationship analyses is to develop useful models that can predict one or more pharmacokinetic properties of a particular compound. In the present study, a multiple-output artificial neural network model was constructed to predict human half-life, renal and total body clearance, fraction excreted in urine, volume of distribution, and fraction bound to plasma proteins for a series of cephalosporins. Descriptors generated solely from drug structure were used as inputs for the model, and the six pharmacokinetic parameters were simultaneously predicted as outputs. The final 10 descriptor model contained sufficient information for successful predictions using both internal and external test compounds. Descriptors were found to contribute to individual pharmacokinetic parameters to differing extents, such that descriptor importance was independent of the relationships between pharmacokinetic parameters. This technique provides the advantage of simultaneous prediction of multiple parameters using information obtained by nonexperimental means, with the potential for use during the early stages of drug development. © 2003 Wiley-Liss, Inc. and the American Pharmaceutical Association J Pharm Sci 92:552–559, 2003

The goal of quantitative structure–pharmacokinetic relationship analyses is to develop useful models that can predict one or more pharmacokinetic properties of a particular compound. In the present study, a multiple‐output artificial neural network model was constructed to predict human half‐life, renal and total body clearance, fraction excreted in urine, volume of distribution, and fraction bound to plasma proteins for a series of cephalosporins. Descriptors generated solely from drug structure were used as inputs for the model, and the six pharmacokinetic parameters were simultaneously predicted as outputs. The final 10 descriptor model contained sufficient information for successful predictions using both internal and external test compounds. Descriptors were found to contribute to individual pharmacokinetic parameters to differing extents, such that descriptor importance was independent of the relationships between pharmacokinetic parameters. This technique provides the advantage of simultaneous prediction of multiple parameters using information obtained by nonexperimental means, with the potential for use during the early stages of drug development.

This review examines the role of soft computing methods such as artificial neural networks (ANNs), genetic algorithms (GAs), fuzzy logic (FL), chaos, fractals and cellular automata (CA) and their hybrids in the field of drug design. They have been found to be useful in a wide variety of areas including quantitative structure-activity relationship (QSAR), quantitative structure-property relationship (QSPR), variable selection, conformation searching, receptor docking, pharmacophore development, molecular design, combinatorial libraries, surface phenomena, kinetics and complex system studies. Based upon the studies examined, the use of soft computing techniques is likely to grow significantly in the future.