Methodologies to evaluate, visualize, and characterize molecules and molecular populations designed with the INVENTON computer program

scINVENTON is a molecular design system that was initiated to demonstrate that, if properly programmed, computers can design molecules better than chemists. This thesis describes the chemical perception, property prediction, attribute evaluation, and molecular population analysis, visualization, and characterization components of scINVENTON. Thymidylate synthase inhibitors, designed de novo by the computer, and enzyme co-crystallized HIV protease inhibitors serve to exemplify these evaluatory aspects of the system.;scINVENTON computes numerous properties and attributes for each candidate structure it proposes. This allows the program to rank its proposals so that the most acceptable and/or interesting are presented to the chemist first. Chemical perception is a central component of scINVENTON, allowing the computer to predict chemical properties, evaluate binding interactions, and recognize other structural attributes. Other computed candidate attributes include molecular volume, ring and chiral center counts, volume-based and surface-based shape similarity, and heuristics for describing structural flexibility and conformational reasonableness.;The potentially thousands of unique candidates that scINVENTON proposes to satisfy a common set of design constraints comprise a molecular-population. Several methodologies to evaluate, visualize, and characterize molecular populations--in terms of structural diversity and quality--are key components of the system. Molecular population fogs are computer graphic visualizations that readily communicate spatial trends of atomic placement in entire candidate populations. In addition, color may be used to indicate patterns of chemical character, such as hydrogen bonding atoms, different size populations may be directly compared, and spatial diversity may be evaluated. Chemical functionality occurrence patterns allow populations to be chemically characterized, and chemical diversity to be assessed. Distributions of candidate attributes serve to describe population acceptability and diversity. They also may be used to compute quantitative diversity metrics, allowing direct comparison of population diversity. scINVENTON'S tools for molecular population analysis make it possible to judge the effectiveness of different design constraints and/or structure generation algorithms.