Computational prediction of toxicity

With increasing number of chemicals produced each year, it still remains a daunting task to keep up with the toxicity profile of each chemical. In this paper, we attempt to predict toxicity of compounds using computational techniques, where results from certain in vitro assays applied on 309 chemicals, along with computed properties of chemicals are used to predict the toxicity caused by them at a particular endpoint. We show that both Random Forest (RF) and Naïve Bayes (NB) have a good performance. We also show that using small and related trees in RF helps to further improve the performance.