19.02.2013 Views

Reviews in Computational Chemistry Volume 18

Reviews in Computational Chemistry Volume 18

Reviews in Computational Chemistry Volume 18

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Application of Scor<strong>in</strong>g Functions <strong>in</strong> Virtual Screen<strong>in</strong>g 65<br />

also teach how to recognize typical failure cases. Recent examples of library<br />

rank<strong>in</strong>g experiments <strong>in</strong>clude those by Charifson et al., 198 Bissantz, Folkers,<br />

and Rognan, 77 and Stahl and Rarey. 78 Charifson and co-workers compiled<br />

sets of several hundred active molecules for three different targets: p38<br />

MAP k<strong>in</strong>ase, <strong>in</strong>os<strong>in</strong>e monophosphate dehydrogenase, and HIV protease.<br />

The members of these sets were then docked <strong>in</strong>to the correspond<strong>in</strong>g active sites<br />

together with 10,000 randomly chosen, but drug-like, commercial compounds<br />

us<strong>in</strong>g DOCK 98 and the Vertex <strong>in</strong>-house dock<strong>in</strong>g tool Gambler. Three scor<strong>in</strong>g<br />

functions performed consistently well <strong>in</strong> enrich<strong>in</strong>g active compounds, namely,<br />

ChemScore, 71,199 the DOCK AMBER force field score, and PLP. 62 The f<strong>in</strong>d<strong>in</strong>g<br />

that these three scor<strong>in</strong>g functions performed so well was partially attributed<br />

to the fact that a rigid-body optimization could be carried out with these<br />

functions, because the functions <strong>in</strong>clude repulsive terms <strong>in</strong> contrast to many<br />

of the other tested functions. The study by Stahl and Rarey 78 compared the<br />

performance of DrugScore 92 and PMF 86 to that of PLP 62 and FlexX score<br />

us<strong>in</strong>g the dock<strong>in</strong>g program FlexX. 64–66 Interest<strong>in</strong>gly, the two knowledgebased<br />

scor<strong>in</strong>g functions showed significantly different behavior for extreme<br />

cases of active sites. DrugScore coped well with situations where ligands are<br />

tightly bound <strong>in</strong> narrow lipophilic cavities (e.g., COX-2 and the thromb<strong>in</strong><br />

S1 pocket), whereas PMF did not lead to good enrichment <strong>in</strong> such cases. Conversely,<br />

for the very polar b<strong>in</strong>d<strong>in</strong>g site of neuram<strong>in</strong>idase, PMF gave better<br />

enrichment than any other scor<strong>in</strong>g function, whereas DrugScore failed. The<br />

description of complexes <strong>in</strong> which many hydrogen bonds play a role seems<br />

to be a general strength of PMF. This has also been noted by Bissantz, Folkers<br />

and Rognan, 77 who found PMF to perform well for the polar target thymid<strong>in</strong>e<br />

k<strong>in</strong>ase and less well for the estrogen receptor.<br />

Hydrogen Bond<strong>in</strong>g versus Hydrophobic Interactions<br />

It is of central importance <strong>in</strong> virtual screen<strong>in</strong>g to achieve a balanced<br />

description of hydrogen bond<strong>in</strong>g and hydrophobic contributions to the score<br />

<strong>in</strong> order to avoid a bias toward either highly polar or completely hydrophobic<br />

molecules. Empirical scor<strong>in</strong>g functions have the advantage that they can be<br />

quickly reparameterized to achieve such a balance, whereas such an adjustment<br />

is impossible with knowledge-based functions. Because this is such an<br />

important topic, we will illum<strong>in</strong>ate it with a number of examples.<br />

Consider the follow<strong>in</strong>g database rank<strong>in</strong>g experiment. A database of<br />

about 7600 compounds was flexibly docked <strong>in</strong>to the ATP b<strong>in</strong>d<strong>in</strong>g site of<br />

p38 MAP k<strong>in</strong>ase. The database consisted of ca. 7500 random compounds<br />

from the World Drug Index (WDI) 200 and 72 <strong>in</strong>hibitors of p38 MAP k<strong>in</strong>ase,<br />

which <strong>in</strong> turn consisted of 30 <strong>in</strong>hibitors form<strong>in</strong>g two hydrogen bonds with<br />

the receptor and 20 <strong>in</strong>hibitors form<strong>in</strong>g only one. Both groups covered the<br />

same activity range from low micromolar (mM) to about 10 nM. For each<br />

of the docked compounds, up to 800 alternative dock<strong>in</strong>g solutions were

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!