EXPLORE PUBLICATIONS BY COUNTRIES

Genetic Evolutionary Algorithms, HIV, Data Predictive Data Mining.

In this work, we utilized Quantitative Structure-Activity Relationship (QSAR) techniques to develop predictive models for inhibitors of the HIV-1 enzymes Integrase, HIV-Protease, and Reverse Transcriptase. Each predictive model was composed of quantitative drug characteristics that were selected by genetic evolutionary algorithms, such as Genetic Algorithm (GE), Differential Evolutionary Algorithm (DE), Binary Particle Swarm Optimization (BPSO), and Differential Evolution with Binary Particle Swarm Optimization (DE-BPSO). After characteristic selection, each model was tested with machine-learning algorithms such as Multiple Linear Regression (MLR), Support Vector Machine (SVM), and Multi-Layer Perceptron neural networks (MLP/ANN). We found that a combination of DE-BPSO combined with Multi-Layer Perceptron produced the most accurate predictive models as measured by R2, the statistical measure of proportion of variance in prediction values, and root-mean-square-error (RMSE) of prediction values compared to observed values. As for the models themselves: the best predictors for Integrase inhibitor included mass-weighted centred Broto-Moreau autocorrelation values, Moran autocorrelations, and eigenvalues of Burden matrices weighted by I-states; the best predictors for HIV-Protease inhibitors included the second Zagreb index value, the normalized spectral positive sum from Laplace matrix, and the connectivity-like index of order 0 from edge adjacency mat; and the best predictors for Reverse Transcriptase inhibitors included the number of hydrogen atoms, the molecular path count of order 7, the centred Broto-Moreau autocorrelation of lag 2 weighted by Sanderson electronegativity, the P_VSA-like on ionization potential, and the frequency of C – N bonds at topological distance 3.