EVALUATION OF QUANTITATIVE STRUCTURE-ACTIVITY PREDICTIONS - COMPARISON OF THE PREDICTIVE POWER OF AN ARTIFICIAL-INTELLIGENCE SYSTEM WITH HUMAN EXPERTS

In this paper, we present three new mathematical techniques for evaluating the predictive skills of structure-activity experts. The question addressed in this paper is how to evaluate the predictive ability of structure-activity experts in identifying the most active compounds of a set of drug candidates. The three proposed mathematical techniques are based on the Phi-square Distance, the Rank Comparison, and the Shuffle method, respectively. They have been used to evaluate the performance of a new computer system and three human experts in predicting the antibacterial potencies of a series of chemical compounds in five different biological tests. The expert system, an artificial intelligence structure-activity program called MULTICASE, performed significantly better than one of the human experts and somewhat better than the other two. © 1990 J.C. Baltzer AG, Scientific Publishing Company.