Character N-gram Model for Toxicity Prediction

Abstract

Prediction of molecular toxicity plays an important role in drug discovery. It has a direct relationship with human health and medical destiny. Accurately assessing a molecule’s toxicity can aid in the weeding out of low-quality com pounds early in the drug discovery phase, avoiding depletion later in the drug development process. Computational models have been used to automatically predict the molecular toxicity. In this paper, a machine learning-based model has been proposed. TF/IDF representation scheme has been used for N-gram and integrated with SMILE. Multiple machine learning classifiers have been im plemented such as Decision Tree, KNN, MLP, Random Forests, SGD and SVM. RF outperforms all the other classifiers. A wide range of N-gram models have been implemented and trigram reported the best results.