The Monte Carlo technique as a tool to predict LOAEL

Quantitative structure activity relationships (QSARs) for the Lowest Observed Adverse Effect Level (LOAEL) for a large set of organic compounds (n = 341) are suggested. The molecular structures of these compounds are represented by Simplified Molecular Input-Line Entry Systems (SMILES). A criteria for the estimation quality of split into the visible training set (used for developing a model) and invisible external validation set is suggested. The correlation between the above criterion and the predictive potential of developed QSAR model (root-mean-square error for invisible validation set) has been detected. One-variable models are built up for several different splits into the visible training set and invisible validation set. The statistical quality of these models is quite good. Mechanistic interpretation and the domain of applicability for these models are defined according to probabilistic point of view. The methodology for defining applicability domain in QSAR modeling with SMILES notation based optimal descriptors is presented. (C) 2016 Elsevier Masson SAS. All rights reserved.