Abstract

CORAL software (http://www.insilico.eu/coral/) has been used for modeling of carcinogenicity (logTD50) of 401 compounds, and anticancer activity (-logIC50) of 100 compounds, on the basis of quantitative structure – activity relationships (QSAR). The simplified molecular input line entry system (SMILES) was used for the representation of the molecular structures. A new additional global invariant of the molecular structure was tested. This is an indicator of the presence of pairs of chemical elements (F, Cl, Br, N, O, S, and P). Three random splits into sub-training, calibration, and test set were examined. Molecular features (calculated with SMILES and statistically significant), which increase the anticancer activity have been identified: their presence in the molecular structure could be helpful criterion in the search for new anticancer agents.

Keywords: artificial neural networks, anticancer activity, threshold, Carcinogenicity, QSAR, SMILES, Validation, CORAL software