Testing the Diversity Hypothesis

• Subset Selection
  • Dissimilarity-Based Compound Selection:
    • pick 1st compound that is most dissimilar from the rest
    • pick 2nd compound that is most dissimilar from the 1st
    • pick 3rd compound that is most dissimilar from 1st and 2nd
    • and so on

• Dissimilarity Measure
  • Descriptors: 1024 Daylight Fingerprints
  • Cosine coefficient (allows use of O(nN) centroid algorithm)
    • (Holliday et al. QSAR 1995, 14, 501-506)

• Diversity Measure
  • Normalised sum of pairwise dissimilarities O(N)

next / prev