Neural-network applications for analysis of adverse drug reactions

A neural-network analysis has been applied to predict adverse or side effects of drugs using a database of antidepressant agents with known effects as well as input from a database with both patient parameters and drug information. A NeuralWork software package was implemented on a Macintosh Quadra 700 and trained on a database of ten drugs with known adverse effects. Another agent (not in the database) was used to test the ability of the network to predict the relative incidence of its side effects. Despite the small number of drugs used for training, the adverse effects of some drugs, such as doxepin, were predicted with 90-100% accuracy. These results indicate that neural-network analysis can be used to predict adverse drug effects for drugs within a given class and ultimately can be extended to include patient parameters to predict the mechanisms of action of drugs from relatively large databases.