First page Back Continue Last page Overview Graphics

Do Different Coefficients Give Similar Ranking?

  • UNITY 2D fingerprints generated for NCI Aids database of 37,124 compounds
  • 60 compounds randomly-selected to act as targets
  • 22 different association, correlation and distance coefficients
  • Experiments repeated using IDAlert database of 11,607 compound


    Given 22 similarity coefficients, we may wander if they all give similar ranking. An earlier work compared the rankings of 22 coefficients. They did the similarity searches for the 22 coefficients and counted the number of compound in common in the top structures for every pair of coefficients. And they output a dissimilarity value by using each of these number. Then used these dissimilarity values to produce a dissimilarity matrix for every target structure. Three clustering methods including single linkage, complete linkage and group average were applied to the matrix to cluster the coefficients. There were 60 target structures with three clustering methods. This makes 180 sets of groupings. And they identified the coefficients that are always clustered together. These were done with Daylight, BCI and Unity fingerprints using AIDS, IDAlert and MDDR databases.