Neural networks as data mining tools in drug design

Neural networks are powerful data mining tools with a wide range of applications in drug design. This paper largely concentrates on self‐organizing neural networks that can be used for investigating datasets both by unsupervised and by supervised learning. The representation of chemical structures is the key to success in establishing useful relationships. Applications are shown for exploring different structure representations, for establishing quantitative structure–activity relationships and for handling compounds having multicategory activities. The applications comprise the separation of compounds according to different biological activities, the location of biologically active compounds in large chemical spaces, the analysis of high‐throughput screening data and the classification of compounds according to mode of toxic action.