Machine learning predictive classification models for the carcinogenic activity of activated metabolites derived from aromatic amines and nitroaromatics. (June 2022)

• Record Type: Journal Article
• Authors: Halabi, Andrés; Rincón, Elizabeth; Chamorro, Eduardo
• Is Part Of: Toxicology in vitro. Volume 81(2022)
• Publication Date: 2022-06
• Subjects: Machine learning -- QSAR -- DFT -- Aromatic amines -- Nitroaromatics -- Carcinogenic activity -- Carcinogenic potency -- J48Consolidated -- RandomTree -- JCHAIDStar -- SPAARC -- WEKA -- Activated Metabolites -- Solvent Effects; Toxicity testing -- In vitro -- Periodicals; Toxicology -- Periodicals; 615.9
• DOI: 10.1016/j.tiv.2022.105347
• Languages: English
• ISSNs: 0887-2333

Abstract

Abstract: A 3D-QSAR study based on DFT descriptors and machine learning calculations is presented in this work. Our goal has been to build predictive models for classifying the carcinogenic activity of a set of aromatic amines (AA) and nitroaromatic (NA) compounds. As the main result, we stress that calculations must consider both the activated metabolites (derived from AA and NA species) and the water solvent to obtain reliable predictive classification models. We have obtained eight decision tree models that presented an accuracy of over 90% by using either Gázquez-Vela chemical potential (μ + ) or the chemical hardness (η) of the activated metabolites in aqueous solvent. Highlights: Activated metabolites and solvent information is needed for predicting carcinogenic activity of nitro and aromatic amines. Our models predicts carcinogenic activity with over 90% accuracy accordingly with high Cohen's kappa statistic values For activated metabolites in the aqueous solvent phase, the electron-accepting chemical potential and hardness are essential.