A machine learning-based QSAR model for predicting phenols cytotoxicity

Abstract

Cytotoxicity is a very important aspect that gains big research interest in toxicology and pharmacology as it is related to whether a compound may cause cell damage, necrosis or apoptosis and this contributes effectively to determine the toxicity potential of a compound and to cancer treatment studies as well. The complexity and the sensitivity of cytotoxicity assays and the involvement of animal tests in many instances increase the need for rapid and reliable alternative methods. Quantitative structure-activity relationships (QSAR) are relevant techniques that provide mathematical models and help in chemicals screening and in predicting biological activities and eventually cytotoxicity. Because of their capacity to handle complex problems, machine learning contributed substantially to the QSAR field's evolvement. In this study, we established a predictive QSAR model based on machine learning, namely deep neural network to predict the cytotoxicity of phenols. The model exhibited high performances in predicting new compounds. It was proved that hydrophobic, steric and electronic effects are relevant in determining the cytotoxicity variability of phenols against Tetrahymena pyriformis.