Data Mining Analysis of HIV-1 Protease Crystal Structures

A data mining study has been done using HIV-1 protease crystal structures complexed with FDA approved HIV-1 protease inhibitor drugs. Chemical descriptors have been computed for the binding pockets of each crystal structure, yielding approximately 600 constitutional, topological, geometric, elecrotostatic, and quantum mechanical descriptors for each structure. Several supervised (hybrid binary particle swarm optimizationartificial neural network and random forest) and unsupervised learning (locally linear embedding) techniques have been explored for feature selection to determine a quantitative structureactivity relationship (QSAR) model containing the most relevant descriptors needed to cluster each crystal structure according to their bound ligand. This method of computational modeling and screening process would aid in the understanding of the effect HIV mutations have on the binding affinity of various present and future HIV-1 protease inhibitors due to structural changes arising from the mutations.