Estimating environmental pollution by xenobiotic chemicals using QSAR (QSBR) models based on artificial intelligence
Water Science & Technology · 1998
An attempt was made to construct QSAR (Quantitative Structure-Activity Relationships) or QSBR (Quantitative Structure-Biodegradation Relationships) formulae and models for predicting biodegradability of chemicals in aqueous aerobic environment with machine learning (ML) tools of artificial intelligence (AI). Inverse of biodegradability is environmental persistence, from which possible dynamics of soil, groundwater and water pollution can be inferred. We tried to predict the biodegradability with several programs that can learn from examples and construct decision or regression trees and/or can construct equations. Besides the given basic topological properties, the main contribution was inclusion of connectivity indices. Above all, normalization of properties to molecular weight or the number of carbon atoms significantly improved prediction. The obtained results are comparable (or better) to the best achieved results with other approaches. Contrary to the statistical methods, ML tools present the information (learned knowledge) in a compact, easily understandable manner which can help identify and understand the key properties of chemicals and mechanisms important for assessing biodegradation (and thus possible environmental contamination) from chemical structure only.