The prediction of polypharmocol. effects from existing networks requires a method of establishing the relationship between mols. with unknown biol. properties and those which have well established pharmacol.Towards this end, several techniques have been developed to estimate the degree of similarity between the query mol. and the compounds in the polypharmacol. network.In our own work, we have developed a method that attempts to define the probability of iso-activity of a compound to others in the data set by calibrating the similarity coefficients against a large dataset of measured activities.We have used the resultant "Belief Theory" estimates of similarity in the generation and querying of the pharmacol. networks.This presentation will highlight the advantages of using this approach when defining networks, as well as efforts to extend the work from two-dimensional fingerprints to three-dimensional shape and electrostatic descriptors.