The need to reduce costs as well as environmental pollution has reinforced the trend to predict the possible biological effects of chemical compounds by computers. So a large number of studies have been carried out on the search for quantitative structure-activity relationships (QSAR), i.e. the search for empirical or theoretical parameters that are directly correlated to some biological response [1]- [7]. Since empirical data is not available for any thinkable structure [1,5] and experiments or quantum chemical calculations are expensive for larger sets of compounds, a lot of interest lies currently in the use of topological indices [3,4],[8]-[10] as discrimination criteria and prediction tools. A topological index is a numerical value computed only from the molecular graph. Among the different types of indices, the information content indices have been shown to be very useful in QSAR research [4],[11]-[13].
But even if close correlations have been found, the most active compound may have been overlooked, since the indices are usually only computed for a set of structures which the researcher considers relevant. Structure generators [14]-[18] , i.e. computer programs capable of calculating all possible isomers to given conditions, can help to overcome this lack. They can provide the whole variety of structures and ensure that all possiblities have been taken into account. In the present study, we will describe some principles of structure generation and apply this to predicting the activities of some barbiturates and monoketones.