Machine Learning models for prediction of acute dermal toxicity

The acute dermal toxicity is the collection of adverse effects that a substance may cause by contact with the skin within 24 hours. The lethal dose (LD50) in rats is the preferred method to evaluate this endpoint.
However, due to recent bans and backlash on animal testing, it is necessary to develop methods to circumvent the use of animals in laboratory. Here we developed and validated machine learning models for prediction of chemical acute dermal toxicity. The data obtained through public databases were integrated, curated, and standardized according to a well-accepted protocol. Models were developed using multiple molecular descriptors and machine learning algorithms with 5-fold cross validation according to the best practices proposed by OECD. The best model presented CCR of 79%, sensitivity of 80%, specificity of 78%,PPV of 78%, and NPV of 79%. The models developed in this work can reliably identify putative dermal toxicants without animal testing.