•We modify the BPSO for wrapper feature selection with two mechanisms.•The memory renewal mechanism has an effect on local and global optimum, helps the particle overstep the local extremum.•The mutation-enhanced mechanism increases the particle mutation probability to avoid premature convergence.•We examine our modified algorithm, compared with previous versions of BPSO and other algorithms.•The novel algorithm results show the better accuracy and fewer feature number.Feature selection (FS) is an essential component of data mining and machine learning. Most researchers devoted to get more effective method with high accuracy and fewer features, it has become one of the most challenging problems in FS. Certainly, some algorithms have been proven to be effectively, such as binary particle swarm optimization (BPSO), genetic algorithm (GA) and support vector machine (SVM). BPSO is a metaheuristic algorithm having been widely applied to various fields and applications successfully, including FS. As a wrapper method of FS, BPSO-SVM tends to be trapped into premature easily. In this paper, we present a novel mutation enhanced BPSO-SVM algorithm by adjusting the memory of local and global optimum (LGO) and increasing the particles’ mutation probability for feature selection to overcome convergence premature problem and achieve high quality features. Typical simulated experimental results carried out on Sonar, LSVT and DLBCL datasets indicated that the proposed algorithm improved the accuracy and decreased the number of feature subsets, comparing with existing modified BPSO algorithms and GA.Download high-res image (130KB)Download full-size image

Support vector machine (SVM) was used to develop a nonlinear quantitative structure–activity relationship (QSAR) model for the prediction of the activities of the adenosine A2A receptor antagonists. Six molecular descriptors selected by the heuristic method (HM) in CODESSA were used as inputs for SVM. The results obtained by SVM were compared with those obtained by HM. The mean squared errors (MSEs) for the training set given by HM and SVM are 0.08 and 0.05, respectively, which shows the performance of SVM model is better than that of the HM model.

A quantitative structure–activity relationship study of a series of HIV-1 reverse transcriptase inhibitors (2-amino-6-arylsulfonylbenzonitriles and their thio and sulfinyl congeners) was performed. Topological and geometrical, as well as quantum mechanical energy-related and charge distribution-related descriptors generated from CODESSA, were selected to describe the molecules. Principal component analysis (PCA) was used to select the training set. Six techniques: multiple linear regression (MLR), multivariate adaptive regression splines (MARS), radial basis function neural networks (RBFNN), general regression neural networks (GRNN), projection pursuit regression (PPR) and support vector machine (SVM) were used to establish QSAR models for two data sets: anti-HIV-1 activity and HIV-1 reverse transcriptase binding affinity. Results showed that PPR and SVM models provided powerful capacity of prediction.A QSAR study of a series of HIV-1 NNRTIs was performed based on six different methods: MLR, MARS, RBFNN, GRNN, PPR and SVM. PPR and SVM yielded the best models.

Quantitative structure–activity relationship (QSAR) models were developed to predict for CCR5 binding affinity of substituted 1-(3,3-diphenylpropyl)-piperidinyl amides and ureas using linear free energy relationship (LFER). Eight molecular descriptors selected by the heuristic method (HM) in CODESSA were used as inputs to perform multiple linear regression (MLR), support vector machine (SVM) and projection pursuit regression (PPR) studies. Compared with MLR model, the SVM and PPR models give better results with the predicted correlation coefficient (R2) of 0.867 and 0.834 and the squared standard error (s2) of 0.095 and 0.119 for the training set and R2R2 of 0.732 and 0.726 and s2s2 of 0.210 and 0.207 for the test set, respectively. It indicates that the SVM and PPR approaches are more adapted to the set of molecules we studied. In addition, methods used in this paper are simple, practical and effective for chemists to predict the human CCR5 chemokine receptor.Definition of substituted 1-(3,3-diphenylpropyl)-piperidinyl amides and ureas. R1-position (showing limited substitution pattern), X position (showing limited structural variations), and the R2, R3, R4 positions of the phenyl rings (showing diverse substitution pattern).

Molecular modeling of a series of HIV reverse transcriptase (RT) non-nucleoside inhibitors (2-amino-6-arylsulfonylbenzonitriles and their thio and sulfinyl congeners) was carried out by comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) approaches. Docking simulations were employed to position the inhibitors into RT active site to determine the most probable binding mode and most reliable conformations. The study was conducted using a complex receptor-based and ligand-based alignment procedure and different alignment modes were studied to obtain highly reliable and predictive CoMFA and CoMSIA models with cross-validated q2 value of 0.723 and 0.760, respectively. Furthermore, the CoMFA and CoMSIA contour maps with the 3D structure of the target (the binding site of RT) inlaid were obtained to better understand the interaction between the RT protein and the inhibitors and the structural requirements for inhibitory activity against HIV-1. We show that for 2-amino-6-arylsulfonylbenzonitriles inhibitors to have appreciable inhibitory activity, bulky and hydrophobic groups in 3- and 5-position of the B ring are required. Moreover, H-bond donor groups in 2-position of the A ring to build up H-bonding with the Lys101 residue of the RT protein are also favorable to activity.A CoMSIA model with high predictive ability is developed. Favorable positions for bulky, hydrophobic and H-bond donor or acceptor substituents are discussed.

The least-squares support vector machine (LS-SVM), as an effective machine learning algorithm, was used to develop a nonlinear binary classification model of novel piperazines-bis- piperazines as antagonists for the melanocortin-4 (MC4) receptor based on their activity. Each compound was represented by calculated structural descriptors that encode constitutional, topological, geometrical, electrostatic, quantum-chemical features. Five descriptors selected by forward stepwise linear discriminant analysis (LDA) were used as inputs of the LS-SVM model. The nonlinear model developed from LS-SVM algorithm (with prediction accuracy of 95% on the test set) outperformed LDA (test accuracy of 90%). The proposed method is very useful for chemists to screen antagonists for the MC4 receptor.