Predicting adverse effects of toxic compounds [email protected] can contribute to overcoming some shortcomings of existing safety assessments . In this study, we generated and evaluated the performance of about 19,000 predictive models using combinations of many [email protected] algorithms to classify compounds to 17 in vivo toxicity endpoints . A unique aspect of this study is dealing with incompletely labeled datasets (i.e., some toxicity endpoints are unknown for some compounds). This limitation can inhibit algorithms’ ability to identify correlations between toxicity endpoints and may reduce their performance. Recently, some predictive algorithms were applied to such toxicity datasets. However, in these studies different datasets, pre-‐processing steps or performance evaluation metrics were used. Therefore, the goal of this study is to provide a systematic evaluation of many predictive models to identify causes of variability in predictive performance across endpoints and compounds.
Future direction would involve: • aggregating predictions from all models; • Modeling continuous endpoints; and • Modeling more specific endpoints with respect to strain, and route and duration of exposure.
1. Arwa Bin Raies, and Vladimir B. Bajic. In silico toxicology: computational methods for the prediction of chemical toxicity. WIREs Comput Mol Sci, 6, pp. 147172, (2016).
2. Arwa Bin Raies, and Vladimir B. Bajic. In silico toxicology: comprehensive benchmarking of multi-‐label [email protected] applied to toxicity data. WIREs Comput Mol Sci, (2017)