In Chemosphere
Endocrine disruptors have been subjected to health risk assessments. Bioassays and chemoinformatics are very useful tools to characterize their chemical nature. By performing rat hyperactivity assays, we screened some endocrine disruptors, resulting in the classification of two groups: hyperactivity-associated and hyperactivity-negative chemicals. Moreover, many epidemiological studies have reported the correlation between most of the hyperactivity-associated chemicals identified in our bioassay and patients with attention deficit hyperactivity disorder (ADHD); thus, these chemicals are emerging as a subfamily of hyperactivity-associated chemicals among endocrine disruptors. Using RDKit, chemoinformatic analyses revealed no significant differences in the distribution of molecular weight between the two groups, but significant differences in "Fraction CSP3" (number of sp3-hybridized carbons/total carbon count) and the Tanimoto coefficient were observed. Additionally, hyperactivity-associated chemicals were distinguished from two known classes of dopaminergic toxins by the Tanimoto coefficient. Machine learning methods were also applied for classification, regression analyses, and prediction. A neural network model classified the two groups. Random forest methods also showed good prediction (R = 0.9, MAE (mean absolute error) = 0.06). Using a junction tree variational autoencoder, the core structure was interpolated between phthalate and phenol in the hyperactivity-associated group. Thus, I describe the chemical nature of a new chemical family that might promote the development of ADHD in humans.
Ishido Masami
2022-Dec-08
ADHD, Chemoinformatics, Environmental chemicals, Machine learning, Toxicoinformatics
