Evolved neural networks for quantitative structure-activity relationships of anti-HIV compounds
Dana G Landavazo, Gary B. Fogel · 2003
This paper compares the utility of an evolved neural network to a linear model to describe the activity of a set of anti-HIV compounds. The results indicate that significant nonlinearity exists within the descriptors for these molecules. This nonlinearity can be captured in a neural network architecture for significantly increased predictive performance.