Pharmaceutical drug design using dynamic connectionist ensemble networks

This article presents a dynamic ensemble neural network model for a pharmaceutical drug design problem. Designing drugs is a current problem in the pharmaceutical research domain. By designing a drug, we mean to choose some variables of drug formulation (inputs), for obtaining optimal characteristics of drug (outputs). To solve such a problem, we propose a dynamic ensemble neural network model and the performance is compared with several neural network architectures and learning approaches. The idea is to build a dynamic ensemble neural network depicting the dependence between inputs and outputs for the drug design problem. Bootstrap techniques were used to generate more samples of data since the number of experimental data is reduced due to the costs and time durations of experimentations. We obtain in this way a better estimation of some drug parameters. Experiment results indicate that the proposed method is efficient.