Reviews in Computational Chemistry Volume 18
Reviews in Computational Chemistry Volume 18
Reviews in Computational Chemistry Volume 18
- 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 71<br />
FlexX score<br />
top 5% = 382 cpds.<br />
FlexX score<br />
top 149<br />
14<br />
23<br />
consensus<br />
list, 149 cpds.<br />
PLP score<br />
top 5% = 382 cpds.<br />
PLP score<br />
top 149<br />
Figure 9 Analysis of the consensus scor<strong>in</strong>g concept with COX-2 as an example.<br />
Numbers <strong>in</strong> the shaded areas are numbers of active compounds. The larger pie charts<br />
at the top show the numbers of <strong>in</strong>hibitors <strong>in</strong> the top 5% of the database <strong>in</strong> terms of<br />
scores. The smaller pie charts refer to fewer top rank<strong>in</strong>g compounds for better<br />
comparison with the smaller size of the consensus list.<br />
scor<strong>in</strong>g for the top 100 compounds by means of PLP and FlexX scores would<br />
elim<strong>in</strong>ate all but one of the s<strong>in</strong>gly hydrogen-bonded <strong>in</strong>hibitors.<br />
Figure 9 shows a worked consensus scor<strong>in</strong>g example for a virtual screen<strong>in</strong>g<br />
experiment on COX-2. (Figure 5 shows the correspond<strong>in</strong>g FlexX and PLP<br />
enrichment curves.) There are 23 <strong>in</strong>hibitors <strong>in</strong> the top 5% of the FlexX score<br />
rank list and roughly twice as many <strong>in</strong> the PLP rank list. Consensus scor<strong>in</strong>g<br />
reta<strong>in</strong>s 22 of the actives. Because many <strong>in</strong>active compounds are filtered out,<br />
the ratio of actives to false positives <strong>in</strong>creases relative to either of the orig<strong>in</strong>al<br />
lists. A different picture is obta<strong>in</strong>ed when one regards only the top 149 compounds<br />
from the <strong>in</strong>dividual FlexX and PLP rank lists—the same number of<br />
compounds that are <strong>in</strong> the consensus list. It becomes clear that the PLP function<br />
alone performs significantly better than does consensus scor<strong>in</strong>g.<br />
Thus, if one does not know <strong>in</strong> advance which scor<strong>in</strong>g function will work<br />
better, more robust results can be obta<strong>in</strong>ed with consensus scor<strong>in</strong>g. If one has<br />
a rough idea which function works better, one can decrease the number of<br />
false positives more effectively by test<strong>in</strong>g fewer compounds from the top of<br />
a s<strong>in</strong>gle rank list. Experience from seed<strong>in</strong>g experiments with known <strong>in</strong>hibitors<br />
or an analysis of the type of b<strong>in</strong>d<strong>in</strong>g site of the target can help to identify a<br />
suitable scor<strong>in</strong>g function.<br />
22<br />
33<br />
44