Research on Drug Molecular Computing Based on Support Vector Machine and Neural Network

Abstract

With the continuous development of science, a large number of drugs, molecules and genes have been stored in various databases. The relationship between drug like molecules and molecular properties can be identified by computer science methods according to the diversity of compound library, drug molecular characteristics and the differences between different molecules, so as to achieve the purpose of drug screening. In this paper, 300 FBPase inhibitors were selected from previous literatures to convert lC50 into plC50. The whole data set is divided into 236 molecules of training set and 64 molecules of test set. Two machine learning algorithms, namely Neural Network and Support Vector Machine, are selected to construct the prediction model with the training set as the learning object. And the reliability of the model is verified by the test set, and it is applied to the prediction of FBPase inhibitors. The prediction results show that the model is stable and reliable. In addition, the randomization test shows that the current model is not caused by chance correlation. The explanation of the selected molecular descriptors proves that the polarity of the molecule plays an important role in the inhibitory activity of FBPase. The model can provide some useful guidance for drug researchers and can screen ideal drug molecules before drug development.