QSAR-modeling of toxicity of organometallic compounds by means of the balance of correlations for InChI-based optimal descriptors

Abstract

Quantitative structure-activity relationships (QSAR) for toxicity toward rats (pLD50) have been built by means of optimal descriptors. Comparison of the optimal descriptors calculated using the International Chemical Identifier (InChI) with the optimal descriptors calculated using the simplified molecular input line entry system (SMILES) has shown that the InChI-based models give more accurate prediction for the above-mentioned toxicity of organometallic compounds. These models were obtained by means of the balance of correlation: one subset of the training set (subtraining set) plays role of the training; the second subset (calibration set) plays role of the preliminary check of the models. It has been shown that the balance of correlations is a more robust predictor for the toxicity than the classic scheme (training set-test set: without the calibration set). Three splits into the subtraining set, calibration set, and test set were examined.