Swarm Intelligence Optimization for Feature Selection of Biomolecules

Abstract

The biological activity of molecules is usually measured in assays to establish the level of inhibition of signal transduction or metabolic pathways. Drug discovery involves the use of Quantitative Structure Activity Relationship (QSAR) to identify chemical structures that could have good inhibitory effects on specific targets and have low toxicity. QSAR has very complicated 3D structure. Therefore, the flower-based optimization model (FBOM) for molecules is proposed to solve the curse of dimensionality problems. Four performance measures: accuracy, precision, sensitivity and specificity are used to evaluate the proposed model. Molecules activity is predicted using support vector machine (SVM), Naive Bayesian (NB), K-Nearest Neighbor (KNN), Decision Tree (DT) and Neural Network (NN) Classifiers. The results of the proposed model are promising. The proposed model reduces the number of features to 8 features out of 1666 features. Moreover, the average classification accuracy reaches to 95%.