Using Chou’s Five-steps Rule to Classify and Predict Glutathione S-transferases with Different Machine Learning Algorithms and Pseudo Amino Acid Composition

The Glutathione S-Transferases (GSTs) are detoxification enzymes which exist in variety of living organisms such as bacteria, fungi, plants and animals. These multifunctional enzymes play important roles in the biosynthesis of steroids, prostaglandins, apoptosis regulation, and stress signaling. In this study, we designed a method to independently predict the structures of animal, fungal and plant GSTs using Chou’s pseudo-amino acid composition concept. Support vector machine (SVM), Random Forests (RF), Covariance Discrimination (CD) and Optimized Evidence-Theoretic K-nearest Neighbor (OET-KNN) were used as powerful machine learnings algorithms. Based on our results, Random Forests demonstrated the best prediction for animal GSTs with 0.9339 accuracy and SVM showed the best results for fungal and plant GSTs with 0.8982 and 0.9655 accuracy, respectively. Our study provided an effective prediction for GSTs based on the concept of PseAAC and four different machine learning algorithms.