Quantitative structure-activity relationships (QSARs) using the novel marine algal toxicity data of phenols

The present study reports for the first time in its entirety the toxicity of 30 phenolic compounds to marine alga Dunaliella tertiolecta. Toxicity of polar narcotics and respiratory uncouplers was strongly correlated to hydrophobicity as described by the logarithm of the octanol/water partition coefficient (Log P). Compounds expected to act by more reactive mechanisms, particularly hydroquinones, were shown to have toxicity in excess of that predicted by Log P. A quality quantitative structure-activity relationship (QSAR) was obtained with Log P and a 2D autocorrelation descriptor weighted by atomic polarizability (MATS3p) only after the removal of hydroquinones from the data set. In an attempt to model the whole data set including hydroquinones, 3D descriptors were included in the modeling process and three quality QSARs were developed using multiple linear regression (MLR). One of the most significant results of the present study was the superior performance of the consensus MLR model, obtained by averaging the predictions from each individual linear model, which provided excellent prediction accuracy for the test set (Q(test)²=0.94). The four-parameter Counter Propagation Artificial Neural Network (CP ANN) model, which was constructed using four out of six descriptors that appeared in the linear models, also provided an excellent external predictivity (Q(test)²=0.93). The proposed algal QSARs were further tested in their predictivity using an external set comprising toxicity data of 44 chemicals on freshwater alga Pseudokirchneriella subcapitata. The two-parameter global model employing a 3D descriptor (Mor24m) and a charge-related descriptor (C(ortho)) not only had high external predictivity (Q(ext)²=0.74), but it also had excellent external data set coverage (%97).