Designing a Quantitative Structure-Activity Relationship for the Intrinsic Metabolic Clearance of Environmentally Relevant Chemicals
Citation:
Ingle, B., B. Veber, J. Nichols, J. Wambaugh, AND R. Tornero-Velez. Designing a Quantitative Structure-Activity Relationship for the Intrinsic Metabolic Clearance of Environmentally Relevant Chemicals. Presented at Society of Toxicology 56th Annual Meeting and ToxExpo, Baltimore, MD, March 12 - 16, 2017.
Impact/Purpose:
The highlighted research focuses on the development of a QSAR for intrinsic metabolic clearance, which it is highly relevant to the conference. The QSAR models will feed into high-throughput toxicokinetic models that bridge the gap between exposure estimates and potential in vitro toxicity. The abstract will allow for the presentation of ongoing research from NERL/CED/HEDMB at the Society of Toxicology 56th Annual Meeting and ToxExpo. Not only will the presentation increase exposure for the research, it will also foster discussions with experts in the field of toxiokinetics and computational toxicology regarding next steps and possible improvements.
Description:
Toxicokinetic models serve a vital role in risk assessment by bridging the gap between chemical exposure and potentially toxic endpoints. While intrinsic metabolic clearance rates have a strong impact on toxicokinetics, limited data is available for environmentally relevant chemicals including nearly 8000 chemicals tested for in vitro bioactivity in the Tox21 program. To address this gap, a quantitative structure-activity relationship (QSAR) for intrinsic metabolic clearance rate was developed to offer reliable in silico predictions for a diverse array of chemicals. Models were constructed with curated in vitro assay data for both pharmaceutical-like chemicals (ChEMBL database) and environmentally relevant chemicals (ToxCast screening) from human liver microsomes (2176 from ChEMBL) and human hepatocytes (757 from ChEMBL and 332 from ToxCast). Due to variability in the experimental data, a binned approach was utilized to classify metabolic rates. Machine learning algorithms, such as random forest and k-nearest neighbor, were coupled with open source molecular descriptors and fingerprints to provide reasonable estimates of intrinsic metabolic clearance rates. Applicability domains defined the optimal chemical space for predictions, which covered environmental chemicals well. A reduced set of informative descriptors (including relative charge and lipophilicity) and a mixed training set of pharmaceuticals and environmentally relevant chemicals provided the best intrinsic clearance rate predictions for both chemical classes. Ultimately, these QSARs will aid in the construction of toxicokinetic models for the high throughput identification of those chemicals with the highest probability of triggering an adverse outcome, and subsequent prioritization of those chemicals for further experimental testing.
