H. Matter et T. Potter, Comparing 3D pharmacophore triplets and 2D fingerprints for selecting diverse compound subsets, J CHEM INF, 39(6), 1999, pp. 1211-1225
Citations number
64
Categorie Soggetti
Chemistry
Journal title
JOURNAL OF CHEMICAL INFORMATION AND COMPUTER SCIENCES
The performance of two important 2D and 3D molecular descriptors for ration
al design to maximize the structural diversity of databases is investigated
in this publication. Those methods are based either on a 2D description us
ing a binary fingerprint, which accounts for the absence or presence of mol
ecular fragments, or a 3D description based on the geometry of pharmacophor
ic features encoded in a fingerprint (pharmacophoric definition triplets, P
DTs). Both descriptors in combination with maximum dissimilarity selections
, complete linkage hierarchical cluster analysis, or sequential dissimilari
ty selections were compared to random subsets as reference. This comparison
is based on their ability to cover representative biological classes from
parent databases (coverage analysis) and the degree of separation between a
ctive and inactive compounds for a biological target from hierarchical clus
tering (cluster separation analysis). While the similarity coefficients (Ta
nimoto, cosine) show only a minor influence, the number of conformations to
generate the 3D PDT fingerprint lead to remarkably different results. PDT
fingerprints derived from a lower number of conformers perform significantl
y better, but they are not comparable to a 2D fingerprint-based design. Whe
n 2D and 3D descriptors are combined with weighting factors > 0.5 for 2D fi
ngerprints, a significant improvement of coverage and cluster separation re
sults is observed for a small number of PDT conformers and medium sized sub
sets. Some combined descriptors outperform 2D fingerprints, but not for all
subset populations. Applying sequential dissimilarity selection to PDT des
criptors reveals that its performance is dependent on the initial ordering
of compounds, while presorting according to 2D fingerprint diversity does n
ot improve results. Finally the relationship between biological activity an
d similarity was investigated, showing that PDTs quantify smaller structura
l differences due to the large number of bits in the fingerprint.