Save the rats! Test less - predict more!

The program PASS predicts ca. 5000 biological effects using as input 2D structures in the form of an SDFile. The accuracy of prediction is high and proven in many publications.  Convince your-self and submit a molfile using the online version for free: http://ibmc.msk.ru/PASS.

 PASS was developed to solve the problem that it is practically impossible to test experimentally a compound for all possible biological activities.

We use PASS to find selectively compounds that show a desired biological activity. For this we need clustering techniques and other programs that develop descriptors, such as fingerprints to present the most interesting structures to the researcher.

We present with CWM Lead Finder an application that allows the researcher to find new active compounds that are similar in biological activity to a set of compounds with known biological activity, but not necessarily similar in structure.

For this we need an SDFile with compounds of known activity (KNOWNS), let's chose Alzheimer Treatment, and a SDFile with compounds of untested effects (UNKNOWS). We predict all the PASS coefficients Pa-PI for the KNOWNS. Most of the activities have values around zero, and are noise. We select those activities that show for the largest number of compounds highest coefficients. This subset of activities we consider to be the "biological profile".

Using this subset of activities we predict the PASS coefficients for the UNKNOWNS. We cluster the profiles, and select those compounds that appear in clusters of KNOWNS and UNKNOWNS. We calculate fingerprints; this allows us to sort by structure similarity - value "T" in the picture to the right.  We sum the PASS coefficients for each compound "Passcoefficient"; this is a measure how well the compound fits the biological profile. We display the number of possible activities of Pa >Pi, "Selectivity"; this is measure how selective the compound is. Small compound show often effects for a lot of activities, but are not very interesting compounds for drug research. We calculate the PASS parameters for about 40 toxicities and color code compounds by yellow if the are above a selected threshold.

As result we show the compounds that are predicted in a grid, and the researcher can sort by structure - the most dissimilar structures of the UNKNOWNS compared to the KNOWNS might be most interesting. He can sort additionally by profile similarity and "selectivity".

We created a completely diverse database of 329 compounds from 2.4 million. 303 compounds could be computed by PASS. We seeded this set of compounds with 9 compounds from Symyx Drug Data Report (MDDR) database with known Alzheimer treatment activity. As KNOWNS we used 21 drugs on the market from the database Comprehensive Medicinal Chemistry (CMC). CWM Lead Finder found as only hits 5 compounds of the 9 from MDDR. This proved that CWM Lead Finder can find with high selectivity compounds that have a desired biological activity.

We tested CWM Lead Finder presently mainly with databases in the size of 5000 compounds. If you are interested to participate in the CWM Lead Finder-beta testing, please let us know and send an email to akos@akosgmbh.de.

On more comment for people that have experience with PASS. If your desired effect is not yet incorporated in PASS, you had to train PASS with your own data, provided you had enough data. The CWM Lead Finder overcomes this problem. However, even we cannot do wonders and the knowledge base of PASS has to be trained with some related effect. However, this related effect needs not to be known by you.

This screenshots shows the clusters, cluster that contain "Known" and "Unknown" are color coded.

This screenshot shows ths members of one cluster, teh blue ones are the "Knowns", the red ones the "Unknowns", and structures with potential toxicity are colored yeallow.

Finding new potential acetylcholine esterase Inhibitors in SDFiles using CWM Lead Finder and PASS (Prediction of Activity Spectra for Substances)

corresponding author email

from 3rd German Conference on Chemoinformatics
Goslar, Germany. 11-13 November 2007

 

Chemistry Central Journal 2008, 2(Suppl 1):P42doi:10.1186/1752-153X-2-S1-P42

The electronic version of this abstract is the complete one and can be found online at: http://www.journal.chemistrycentral.com/content/2/S1/P42