Topological indices based on 2- or 3-eccentricity to predict anti-HIV activity

Abstract

As molecular descriptors, topological indices based on k-eccentricity of a graph are introduced. Firstly we devise an algorithm for these indices based on 3-eccentricity and analyze the computing complexity. As their applications, we employ the topological indcies based on 2- and 3-eccentricity as the feature vectors to predict anti-HIV activity by devising machine learning predicting models with the help of Support Vector Machine (SVM), K Nearest Neighbor (KNN) and Decision Tree (DecTree), respectively. Experiment results show that the highest accuracy is 99.7%, while the lowest is 97.7% except the special cases. The special cases are that the experiments are with the single 2-CEI or 3-CEI as the feature vector, respectively. Through these experiments, we find that these topological indices based on the k-eccentricity (k=2,3) have good applications in predicting anti-HIV activity. But not every feature vector is generally applicable. Different feature vectors may be used to different models. Furthermore, there is no clear relationship between the dimension of feature vectors and the accuracy of prediction.