Machine Learning Algorithms for Prediction of Chemical Toxicity
D. P. Gaikwad, Shambhavi S. Singh · Apple Academic Press eBooks · 2025
Now-a-days, human being is exposing to an plenty of chemical compounds through the drugs, cosmetics, atmosphere, and nutrition. For protecting human being from potential harmful effects, drugs, and medicines must be passed steady tests for adverse effects and toxicity. Therefore, there is a need of developing more efficient and less time-consuming methods to predict toxicity of drugs and medicines. The computational models are more efficient approach to predict the toxicity because these models require less time to screen large numbers of compounds at low costs. The researchers have proposed many machine learning (ML) algorithms to implement of toxicity prediction systems. In this chapter, the significant concepts of machine learning are used in human safety and chemical health have summarized in detail. Many freely available tools for toxicity prediction have outlined with their training datasets. Different training datasets are available to build the ML based models. Due to poor annotation in toxicity training dataset, it is very difficult to retrieve and combine these datasets for research in toxicity. This chapter presents the performance of different supervised ML algorithms used to predict chemical toxicity. Specifically, Random Forest, Support Vector Machine (SVM), Classification and Regression Tree (CART), Gaussian Process, Linear Regression, K-Nearest Neighbors (K-NN), Linear Discriminant Analysis (LDA), and Naïve Bayes algorithms were assessed in terms of receiver operating characteristic (ROC) curves and classification accuracy. Initially, the models normalize the chemical representation of chemical compounds. Chemical descriptors have been computed and used as input to ML models. Considering best the performance of Random Forest, it is planned for prediction of toxicity of drugs. The proposed model has been trained and test using dataset provided by Tox21 Data Challenge. The model has trained and evaluated using training dataset. The proposed model predicts the toxicity of new compounds successfully with highest accuracy.