Abstract:
Methods for the assessment of the environmental hazard of chemicals include the detection of acute aquatic
toxicity of compounds for various small organisms, these are test objects. Experimental determination of toxicity
presents significant difficulties; therefore, computational methods are developed. The article is devoted
to predicting the toxicity parameter of organic substances of different classes in relation to the test organism
from the class of crustaceans Daphnia magna. The possibility of using QSAR / QSTR (Quantitative Structure
Activity Relationships / Quantitative Structure Toxicity Relationships) method for estimating the toxicity index
of pLC50 using a large set of descriptors was investigated. To describe the structure of molecules, a set of
2644 descriptors generated by the Dragon 7 program was used. The calculations were carried out using the
computer program Program Robustness Calculation (PROGROC). This program allows the number of descriptors
to be used without preliminary selection exceeding the number of substances in the set. 546 Substances
were used to construct the models, 170 compounds of which made up a control set, and 376 entered
the training set. The obtained models are characterized by a high correlation between calculation and experiment.
For a control sample of 170 compounds, the correlation coefficient is R = 0.952 and the standard deviation
is s = 0.45, for a training sample of 376 compounds, R = 0.971 and s = 0.41. The model for all samples is
characterized by the following statistical parameters: correlation coefficient is R = 0.966 and standard deviation
is s = 0.42.
