[Product Releases]
|
|
|
|
|
[Blog]
|
|
Most recent post
|
[News]
|
|
Can we trust docking results?
Sept 2010 IBM Systems and Technology Group releases a white paper with eHiTS and Cell
Oct 2008
EPA's ToxCastTM project will use SimBioSys' eHiTS as docking engine
Nov, 2007
|
[Events]
|
| 243rd ACS
Mar 25-29, 2012
San Diego, CA
see >> more
|
|
|
|
|
eHiTS LASSO:
Ligand Activity in Surface Similarity Order
|
 |
Spot all the actives
quickly and easily with eHiTS LASSO !
|
|
The
activity surface point types (shown as triangles) of a Faxtor Xa
inhibitor ZK-807834 (CI-1031).
|
What is LASSO? A short overview 
Download
eHiTS LASSO datasheet with the latest results.
A new 3D ligand activity surfaced-based
similarity tool
from SImBioSys gives users the power to quickly screen large datasets
for structurally diverse active
molecules. This proven scaffold-hopping capability is possible because
eHiTS LASSO uses chemical features, not just 2D or 3D structural
similarity, of active ligands to rapidly
identify molecules with potentially similar activity.
-
eHiTS LASSO provides a very easy-to-use training
utility that will capture the chemical features of your active
molecules and uses this information to look for molecule with similar
features. In addition eHiTS LASSO ships with a large set of
pre-trained knowledge files that can have you screening databases
within
minutes of installing the software.
-
Trails have shown that
eHiTS LASSO is able to retrieve high percentages of actives in a
seeded dataset when trained on a very small number of actives. When
tested on very diverse ligands, eHiTS LASSO has shown a very strong
ability to retrieve structurally diverse actives, showing its ability
to scaffold hop and identify actives of varying chemotypes.
-
The
eHiTS LASSO has been seamlessly integrated with the eHiTS docking
program, giving you the ability to screen very large databases
quickly while still generating docking poses of the most promising
molecules.
Evaluating eHiTS
LASSO Performance
|
Following the ligand diversity classification of Hert
et. al J. Chem. Inf Model, 46, 462-470, 2006, three groups of ligands
with low, medium and high Mean Pairwise Similarity (MPS) were extracted
from the MDDR database. For each of these 3 groups 5 families, i.e. 15
sets in total were chosen. Please see detailed results in a table here or a quick view on the chart.
To test eHiTS LASSO, 2% of the actives from each
dataset were
used to train the neural net used
|
|
|
by the LASSO. The eHiTS LASSO was
then run on a cleaned version of MDDR database and the ability of the
LASSO utility to rank actives was captured. The enrichment chart at the
top of the page is showing the % of actives recovered in the top 2%, 5%
and 10% of the screened and ranked database. It can be seen that the
eHiTS LASSO, trained on only 2% of
the actives is capable of
retrieving over 50% of the actives
in the top 10% of the screened
database in 11 out of the 15 cases tested. Even for the most
structurally diverse data set eHiTS LASSO was able to place at least
20% of the actives in the top 2% of the screened results.
These
results
show that the eHiTS LASSO is capable of identifying
actives from large datasets using a very small number of actives for
training. Training with just 2% of a know set of actives gives very
strong enrichment results over a very diverse range of activity
classes. It is clear, however, that if the actives have low
diversity (i.e. are structurally very similar) then this ligand based
similarity search is far more effective than if the actives are
highly diverse. In the test shown, eHiTS LASSO retrieved over 90%
of the actives in the top 10% of the screened database for all 5 test
sets. However it is clear that there are situations where the eHiTS
LASSO does not perform as well. This is likely due to the small
number of actives used to train the LASSO and the large diversity of
actives being tested. Therefore it is important to use as many
actives as you have available to train the eHiTS LASSO for optimal
results in real life screening situations.
For more detailed supporting data and information
on
data preparation, please click here.
|
|
|
|
