Classification analysis using support vector machine, decision tree, and neural network with principal component analysis to determine molecular structure relationship from its biological activity on dipeptidyl peptidase IV inhibitors
Haris Hamzah, Alhadi Bustamam, Arry Yanuar, Devvi Sarwinda · AIP conference proceedings · 2020
A chronic metabolic disease that often affects adults is type 2 diabetes. Dipeptidyl peptidase-IV (DPP-IV) inhibitors are drug targets for diabetes mellitus type 2 (T2DM) that can block the enzyme dipeptidyl peptidase-IV. At this time, there are adverse effects from these inhibitors. Therefore, novel DPP-IV inhibitors are still expected with minimal adverse effects. In this paper, a machine learning approach is used to predict the molecular structure of DPP-IV inhibitors. There are 3363 inhibitors consisting of 1849 inhibitors with active labels and 1514 inhibitors with inactive labels that are optimized using fingerprint topology as descriptors. However, fingerprint topology always produces high-dimensional data. So, the principal component analysis method is proposed to reduce the dimension of the data set. Then, support vector machine, decision tree, and neural network are used for classifying DPP-IV inhibitors. The overall classification using the support vector machine method produces specificity, sensitivity, accuracy, and Matthews coefficient correlation C, respectively 0.774, 0.826, 0.803, and 0.604. These results indicate that the support vector machine method has a good ability in the classification of active and inactive DPP-IV inhibitors based on topological fingerprint as descriptors.