The development of new drugs today has become increasingly complex. It is not enough that a new compound be effective, it must have significant advantages over existing compounds--greater activity, fewer side effects, and greater purity. Further, some of the most serious human diseases, such as AIDS present such a difficult moving target for drugs, that the traditional random screening of compounds for the desired activity is simply not winning the battle. Our ability to invent new compounds to meet very specific objectives is important to saving lives. Although major advances have occurred in elucidating the three- dimensional structure of key enzymes, in computer graphic display and analysis of proteins and nucleic acids, and in correlating chemical structure with activity, chemists are still designing new compounds manually by as hoc guessing, and by minor perturbations of an original lead compound. A major aim of this project is to develop a new system, INVENTION, for inventing chemical structures to fit specific criteria. In contrast to computer-assisted design systems of today which help the chemist invent, the focus of INVENTION is invention by the computer. Our hypothesis, is that when properly programmed, the computer can be a better inventor than the chemist. The benefit of such a system is that it could improve the yield of marketable compounds and reduce the cost of developing new drugs. An appropriate computer-based invention system would be free from the bias that a chemist develops over the years of education and training. Today's chemical design problems are heavily rooted in the three-dimensional spatial arrangement of atoms, an area in which chemists have great difficulty mentally visualizing new arrangements. INVENTION will be applied to the design of inhibitors of HIV protease and the enzyme thymidylate synthase (TS). A second major aim in this research is to develop computer algorithms for reasoning by analogy from the growing database of protein-substrate crystal structures to help predict the potential site and strength of binding of a proposed enzyme inhibitor. We hypothesize that the analogical reasoning methodology we developed previously in AIMB can be useful in extracting knowledge that is inherent in the protein-substrate database, even though the complexity is so great as to virtually preclude manual human analysis. The knowledge extracted will be helpful to INVENTION in designing new lead compounds that will selectively bind to the HIV protease to inhibit virus replication. Additionally we expect this study to provide general information about the conformational changes in the protein caused by substrate binding, and about substrate conformational changes as a result of binding to the protein.
