Drug discovery is a long, costly and high risk endeavour which is further convoluted by high attrition rates later in development stages. Poor toxicity profiles have been one of the main causes of failure during pre-clinical and clinical trials and it is estimated that success rates could quadruple if human toxicity was eliminated at preclinical and development stages. The accumulation of characterised small molecule toxicity data holds the key for the development of computational tools to facilitate optimisation of toxicity profiles of drug candidates, reducing costs, increasing success rates by identifying toxicity issues early on and assisting prioritization of compounds, also reducing use of animal models.
Here, we present toxCSM, the most comprehensive computational platform for the study and optimisation of toxicity profiles of small molecules. toxCSM leverages our well established graph-based signatures concept to develop scalable and accurate predictive models using supervised learning, outperforming alternative methods. Using these signatures we have developed 36 models for predicting a wide range of toxicity properties, from nuclear and stress responses to environmental toxicity, which can assist in the development of safer and less toxic drugs as well as herbicides and pesticides. Based on these features, we believe toxCSM web server will provide an easy, precise and reliable way to evaluate small molecule toxicity profiles.