Analysis of the internal representations developed by neural networks for structures applied to quantitative structure-activity relationship studies of benzodiazepines

An application of recursive cascade correlation (CC) neural networks to qua ntitative structure-activity relationship (QSAR) studies is presented, with emphasis on the study of the internal representations developed by the neu ral networks. Recursive CC is a neural network model recently proposed for the processing of structured data. It allows the direct handling of chemica l compounds as labeled ordered directed graphs, and constitutes a novel app roach to QSAR. The adopted representation of molecular structure captures, in a quite general and flexible way, significant topological aspects and ch emical functionalities for each specific class of molecules showing a parti cular chemical reactivity or biological activity. A class of 1,4-benzodiaze pin-2-ones is analyzed by the proposed approach. It compares favorably vers us the traditional QSAR treatment based on equations. To show the ability o f the model in capturing most of the structural features that account for t he biological activity, the internal representations developed by the netwo rks are analyzed by principal component analysis. This analysis shows that the networks are able to discover relevant structural features just on the basis of the association between the molecular morphology and the target pr operty (affinity).