Abstract
The REACH (Registration, Evaluation, Authorization and restriction of Chemicals) and BPR (Biocide Product Regulation) regulations strongly promote the use of non-animal testing techniques to evaluate chemical risk. This has renewed the interest towards alternative methods such as QSAR in the regulatory context. The assessment of Bioconcentration Factor (BCF) required by these regulations is expensive, in terms of costs, time, and laboratory animal sacrifices. Herein, we present QSAR models based on the ANTARES dataset, which is a large collection of known and verified experimental BCF data. Among the models developed, the best results were obtained from a nine-descriptor highly predictive model. This model was derived from a training set of 608 chemicals and challenged against a validation and blind set containing 152 and 76 chemicals. The model's robustness was further controlled through several validation strategies and the implementation of a multi-step approach for the applicability domain. Suitable safety margins were used to increase sensitivity. The easy interpretability of the model is ensured by the use of meaningful biokinetics descriptors. The satisfactory predictive power for external compounds suggests that the new models could represent a reliable alternative to the in vivo assay, helping the registrants to fulfill regulatory requirements in compliance with the ethical and economic necessity to reduce animal testing.
Highlights
Quantitative Structure-Activity Relationships (QSARs) are in silico approaches developed to quantitatively predict a certain property/activity for a substance of interest
10 of these are characterized by multiple bioconcentration factor (BCF) experimental values; – Footprint PPDB2 (2013): contains unique experimental values for 159 pesticides; – Arnot and Gobas (2006): comprises only experimental data for fish species and aquatic organisms indicated by OECD 305 guidelines (OECD, 2012) (Danio rerio, Pimephalespromelas, Cyprinus carpio, Oryziaslatipes, Poeciliareticulata, Lepomismacrochirus, Oncorhynchusmykiss, and Gasterosteusaculeatus) with an overall reliability score of 1; contains unique or multiple experimental BCF data for 759 compounds; – EURAS3: contains 511 reliable data points for fish species suggested by OECD 305
The comparison between models based on QikProp and Dragon descriptors disclosed that the former are characterized by a far better performance on training set (TS) and even higher on VS and blind set (BS)
Summary
Quantitative Structure-Activity Relationships (QSARs) are in silico approaches developed to quantitatively predict a certain property/activity (e.g., pharmacological effect or toxicity, defined as endpoint) for a substance of interest. There are several techniques to challenge models In this respect, the external validation was the most appropriate approach for model control. The external validation was the most appropriate approach for model control Such an approach allows simulation of real life uses of the model. This allows prediction of the response for those chemicals included in a so-called external set, i.e., excluded from the model derivation. This procedure represents the proof of the capability of a given model to predict the properties of unknown compounds (Golbraikh and Tropsha, 2002). REACH Article 1 encourages the use of alternative methods (in silico among others) for assessing the presence or absence of hazardous properties of chemical substances, which, at the same time, minimize the costs of experiments and the controversial use of vertebrate animals (EC, 2006)
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call