Endocrine-disrupting chemicals (EDCs) are defined as exogenous compounds that interfere with the normal physiological process of natural blood-derived hormones during the growth, development, and reproduction of the body, resulting in dysfunction of the endocrine system. EDC exposure leads to adverse events and cauesd enormous harm to the environment. EDC-Predictor is an a web server for prediction of potential EDCs based computational target profile via machine learning methods. Through the prediction model, statistical test and user-friendly web interface, EDC-Predictor facilitate biologists to observe novel EDCs and accelerate ED-related researches.
The mechanism of EDCs is complex, and they mainly affect hormone receptors and hormone-responsive cells to interfere with the synthesis, transport, action, metabolism, and clearance of hormones. Therefore, the interaction of EDCs with hormone receptors becomes a crucial issue in this field of research.
The computational target profile consists of the network-based and machine learning-based target profiles for drug safety assessment and ED-related studies in drug discovery and development. The network-based target profiles contains the prediction scores of these compounds for all 1844 protein targets in the DTI network. The Machine learning-based target profile is represents as the predicted label of compounds for 11 ED-related target categories.
Six machine learning methods were explored in this study, including support vector machines (SVM), decision tree (DT), random forest (RF), k-nearest neighbors (kNN), , linear regression (LR), extreme gradient boosting (XGB). RF can be viewed as bagging many decision trees that utilize a random subset of features and combine them via a voting system. XGB are homogeneous individual learners, whose difference lies in whether there is a strong dependency relationship between individual learners. The other four methods are frequently used algorithms in the toxicology research.