Feature Selection in Predicting the Activity of Cyclooxygenase-2 Inhibitors

S. Zhang, Y. Zhao, J. Wang, and Z. Yang (PRC)


feature selection, hybrid model (filter and wrapper), Principal Component Analysis, Multiple Linear Regression, BP Net, Support Vector Machine


Feature selection is the prerequisite condition of predicting the activity of drugs accurately and quickly. In this paper, we use a method similar to Hybrid model, which combines Filter and Wrapper model, to select the best feature subset and use the subset to build the model of predicting activity of cyclooxygenase-2. First of all we extract the ten features that might influence the activity of cox-2 inhibitors. We get the general trend of data by Principal Component Analysis and get the contribution value of each feature by Multiple Linear Regression Analysis, and then we can remove those irrelevant or redundant features from the initial feature set and get three possible best feature subsets. In order to decide which subset should be selected, we build two classification models, BP Neural Network and Support Vector Machine, to predict the activity of cox-2 inhibitors, and then we can evaluate the feature subsets and find the best feature subset to predict the activity of cox-2 inhibitors.

Important Links:

Go Back