Particle swarms for drug design

The design of new drugs to prevent diseases and improve the quality life for people around the world is a challenge faced by the pharmaceutical industry on a daily basis. This has motivated scientists to find new chemoinformatics tools that can significantly reduce the time and cost to bring a new drug to the market. Quantitative structure-activity relationship (QSAR) models are often used by scientists to evaluate the potential of new compounds. QSAR models provide medicinal chemists with mechanisms for predicting the biological activity of compounds using their chemical structure or properties. This work describes various particle swarms techniques for the development of QSAR models based on artificial neural networks and k-nearest neighbor and kernel regression. The particle swarm techniques are compared against models developed by simulated annealing and artificial ant systems. Particle swarm techniques are shown to compare favorably to the other techniques using three classical data sets from the QSAR literature.