|
|
PASS
predicts with high accuracy (70-80%)
up to 2000 biological activities for your compounds. As input you only need the
2D structure. This is one of the most successful programs giving you early
indications if your compounds might be useful. This is a "one click" tool for
in silico
screening.
A typical example:
PASS was trained on a
special set of compounds. There were 5000 compounds containing 200 hits. With
PASS 50 structures were selected, 25 were hits. Assume one test is 1 dollar. If
the distribution of the hits would be equal in the database of 5000, you need to
spend 5000/200*25 = 625 dollars to find 25 hits. With PASS you would need 50
dollars. This is a saving of 625/50 = 1250%!
Why should you use PASS?
- Find out unwanted side effects, quickly.
|
- Analyze your
dataset:
- drug versus non-drug like (see
reference Anzali et.al.)
- cluster the dataset using the PASS parameters or the
MNA descriptor keys
- select compounds that belong to an activity class
- Buy compounds that are promising candidates (click here to
see an application).
- Select
the most likely candidates for screening.
- Adjust
your screens according to the predicted activities.
- Find
new therapeutic areas for compounds with known activity.
All
you require is the structure in the form of a MOL
or SDFile. The prediction goes very fast. Calculation of biological activity
spectra for 10,000 compounds on an ordinary IBM PC takes about 5 min. So you can
do it for large data sets.
The
training set includes
at the moment ca.
60'000 substances with more than
2'000 biological
effects. PASS
predicts simultaneously the probabilities of presence
(Pa)/absence
(Pi). PASS Pro
is open; therefore the user can add to the training set some new
biologically active compounds and new activities. Mean accuracy of prediction in LOO cross-validation is ~87%.
|
Go an read about
applications of PASS in R&D. |
