Molecular diversity in chemical databases: comparison of medicinal chemistry knowledge bases and databases of commercially available compounds.

A molecular descriptor space has been developed which describes structural diversity. Large databases of molecules have been mapped into it and compared. This analysis used five chemical databases, CMC and MDDR, which represent knowledge bases containing active medicinal agents, ACD and SPECS, two databases of commercially available compounds, and finally the Wellcome Registry. Together these databases contained more than 300,000 structures. Topological indices and the free energy of solvation were computed for each compound in the databases. Factor analysis was used to reduce the dimensionality of the descriptor space. Low density observations were deleted as a way of removing outliers, which allowed a further reduction in the descriptor space of interest. The five databases could then be compared on an efficient basis using a metric developed for this purpose. A Riemann gridding scheme was used to subdivide the factor space into subhypercubes to obtain accurate comparisons. Most of the 300,000 structures were highly clustered, but unique structures were found. An analysis of overlap between the biological and commercial databases was carried out. The metric provides a useful algorithm for choosing screening sets of diverse compounds from large databases.