Computerized system PASS (Prediction of Activity Spectra for Substances) is described. PASS predcits more than 100 pharmacological effects, mechanisms of action and specific toxicities simultaneously. The average accuracy of prediction is about 80% both in leave-one-out cross-validation procedure for the total training set and by prediction for independent sample of 5000 biologically active compounds. The effectiveness of PASS used for heterogenic set of new drug-candidates possess different biological activities is 800% better than random guess-work and 300% better than estimation by skilled experts. PASS used in drug R&D might help to select compounds having structures dissimilar to well-known drugs and to estimate probable activity spectra for new compounds to optimize their synthesis abd testing.
Bioactive Compound Design: Possibilities for Industrial Use, BIOS Scientific Publishers, Oxford (UK), 1996, p.47-56.