Welcome to the Optibrium Community





Forgot login?
Register

Forum

Search Forum

Welcome, Guest
Please Login or Register.    Lost Password?

CNS scoring - surprising result
(1 viewing) (1) Guest
General discussions about using StarDrop. Please use one of the other forums for more specific issues.
#137
CNS scoring - surprising result
stevesmedchem
Fresh Boarder
Posts: 1
graphgraph
User Offline Click here to see the profile of this user
Karma: 0
I’ve had a curious result with the standard property calculations when inorganics are included. So, for example, I’ve downloaded the ‘Approved’ list of drugs from DrugBank, removed Nutraceuticals and then calculated properties. I then scored on the standard ‘Oral CNS scoring profile’ Using the snake plot it appeared that around 240 compounds of the 1353 originals looked statistically better than the rest. I then took this subset and used the selection tool to choose 25 compounds based on 50:50 structure and CNS score. This gave the attached list which includes 7 inorganics including three metals, three salts and S=Se The calculated columns all seem to have reasonable statistics for these 7 even though there are instances of ‘inf’ for some of the more regular molecules. Clearly this is a rather artificial process but I wouldn’t have expected these results.
 
The administrator has disabled public write access.
#138
Re:CNS scoring - surprising result
matt
Moderator
Posts: 72
graph
User Offline Click here to see the profile of this user
Karma: 1
Thanks for the interesting question Steve! I think a couple of factors are coming into play here.

The inorganics are clearly very different from the compounds that were used to build the models, so the results a unlikely to be reliable. As you point out, this should ideally show up as an 'inf' in the uncertainty in the prediction. However, the inorganics are so different from compounds we typically work with that we haven't include descriptors that capture these inorganics. This, in turn, means that the model isn't sensitive to the difference. We'll look at ways of including detection and flagging of such inorganics in the future.

This effect is then amplified by a selection based on 50:50 diversity to score. This means that the algorithm will look for high scoring compounds that are as diverse as possible... The inorganics are very diverse with respect both to drugs and to each other, so the selection algorithm will preferentially pick these over more drug-like molecules.

Of course, as you also point out, this is a very artificial process. StarDrop users are very unlikely to be modelling and selecting from data sets including S=Se and the like in practice. (All that being said, ions such as K+ do penetrate the BBB, albeit slowly, are soluble and well absorbed, so would meet the scoring criteria in this case! )
 
The administrator has disabled public write access.
Matt Segall
CEO, Optibrium
matt.segall@optibrium.com
Go to topPage: 1
Latest Forums

Read more >