Sunghwan Kim - PubChem3D: Biologically relevant 3-D similarity

Version 4

      Publication Details (including relevant citation   information):

      S. Kim, E.E.   Bolton, and S.H. Bryant;

      Journal of Cheminformatics, 2011, 26,   2011.




      The use of 3-D similarity techniques in the analysis of   biological data and virtual screening is pervasive, but what is a   biologically meaningful 3-D similarity value? Can one find   statistically significant separation between "active/active" and   "active/inactive" spaces? These questions are explored using   734,486 biologically tested chemical structures, 1,389 biological   assay data sets, and six different 3-D similarity types utilized   by PubChem analysis tools.


      The similarity value distributions of 269.7 billion unique   conformer pairs from 734,486 biologically tested compounds   (all-against-all) from PubChem were utilized to help work towards   an answer to the question: what is a biologically meaningful 3-D   similarity score? The average and standard deviation for the six   similarity measures STST-opt, CTST-opt,   ComboTST-opt, STCT-opt,   CTCT-opt, and ComboTCT-opt were 0.54 ±   0.10, 0.07 ± 0.05, 0.62 ± 0.13, 0.41 ± 0.11, 0.18 ± 0.06, and   0.59 ± 0.14, respectively. Considering that this random   distribution of biologically tested compounds was constructed   using a single theoretical conformer per compound (the "default"   conformer provided by PubChem), further study may be necessary   using multiple diverse conformers per compound; however, given   the breadth of the compound set, the single conformer per   compound results may still apply to the case of multi-conformer   per compound 3-D similarity value distributions. As such, this   work is a critical step, covering a very wide corpus of chemical   structures and biological assays, creating a statistical   framework to build upon. The second part of this study explored   the question of whether it was possible to realize a   statistically meaningful 3-D similarity value separation between   reputed biological assay "inactives" and "actives". Using the   terminology of noninactive-noninactive (NN) pairs and the   noninactive-inactive (NI) pairs to represent comparison of the   "active/active" and "active/inactive" spaces, respectively, each   of the 1,389 biological assays was examined by their 3-D   similarity score differences between the NN and NI pairs and   analyzed across all assays and by assay category types. While a   consistent trend of separation was observed, this result was not   statistically unambiguous after considering the respective   standard deviations. While not all "actives" in a biological   assay are amenable to this type of analysis, e.g., due to   different mechanisms of action or binding configurations, the   ambiguous separation may also be due to employing a single   conformer per compound in this study. With that said, there were   a subset of biological assays where a clear separation between   the NN and NI pairs found. In addition, use of combo Tanimoto   (ComboT) alone, independent of superposition optimization type,   appears to be the most efficient 3-D score type in identifying   these cases.


      This study provides a statistical guideline for analyzing   biological assay data in terms of 3-D similarity and PubChem   structure-activity analysis tools. When using a single conformer   per compound, a relatively small number of assays appear to be   able to separate "active/active" space from "active/inactive"   space.


      Address (URL):