| Abstract
| - We developed a group contribution method (GCM) to correlate acute toxicity (96 h LC50) forthe fathead minnow (Pimephales promelas) for 607 organic chemicals. Unlike most of theexisting methods, the new one makes no use of predefined groups as descriptors. A simplegeneral rule is proposed to break down any molecule into diatomic fragments. The entire dataset was partitioned three times. Each time, a training set and a test set were obtained with aratio of 2:1. For each partition quantitative structure−activity relationship, models weredeveloped using Powell's minimization method, multilinear regression, neural networks, andpartial least squares. The GCM method achieved a good correlation of the data for both trainingand test sets, regardless of the partition considered. The method is therefore robust and canbe generally applied. Further model improvements are described.
|
