RE: QSAR - sensitivity to atomic charge models?

Dear Jim,
 
This is a very interesting and important question which becomes more
pertinent with the increased availability and diversity of modelling
software!
 
Regarding the second part of your post, I often use Principal Components
Analysis to detect sensitivity to different ways of calculating molecular
descriptors. This is also useful in cases where different tautomers exist
etc. Observing the change in position in scores space then allows you to
assess whether the difference is significant in the "grand scheme of
things". By looking for the 'latent variables' in this way models are more
stable than making decisions based on individual variables which may in
themselves be subject to large errors. The idea is to take the contribution
of the regularities in the data (from the correlation in many variables) and
leave behind the 'noise'.
 
The differently calculated variables themselves may be included in the same
model, or two different models and the effects observed in the loadings
plots. Any changes in correlation structure can then be assessed and
interpreted. If large changes are seen then it is cause for investigation,
if the models and variables are insensitive to the method of calculation
then there is less cause for alarm. Interpreting the models can then focus
attention on the influential and sensitive variables. This is akin to a 'QC
check' for modelling parameters.
 
I would be interested to hear how other QSAR members assess the effects of
variation in descriptors,
 
Best regards,
 
Mark
-- --------------------------------------------------------------------
Mark Earll CChem MRSC Umetrics
Senior Consultant (Scientific Data Analysis)
Umetrics UK Ltd
Woodside House, Woodside Road,
Winkfield, Windsor, SL4 2DX

Phone: 01344 885615 Mobile: 07765 402673
Email: mark.earll/./umetrics.co.uk
Fax: 01344 885410
Web: http://www.umetrics.com <http://www.umetrics.com/>
----------------------------------------------------------------------------
------------

-----Original Message-----
From: james.metz__abbott.com [mailto:james.metz++abbott.com]
Sent: 09 June 2004 01:32
To: qsar_society[#]accelrys.com
Cc: james.metz%abbott.com
Subject: QSAR - sensitivity to atomic charge models?

QSAR Society,

        I am looking for literature publications, white papers, or other
presentations which show that using
various charge models for small molecules can result in dramatically
different QSAR descriptor-based models.

        I was quite surprised recently to observe that two classification
trees built using the same descriptor, but
calculated using two different charge models had very different R2 values
for the same training set. If I had used only
the first set of charges, I would have concluded that the descriptor does
not fit the data well ! However, using the
second set of charges, I conclude that the descriptor does fit the data
reasonably well. Crossvalidation shows that the
model constructed using the second set of charges is OK. However, why
should I believe that my second set of charges is
"optimal" ?

        Hence, how much effort do you make testing the sensitivity of your
descriptors and resulting models to various
methods of computing atomic charges - Gasteiger-Huckel, MNDO, AM1, etc.

        I would appreciate any thoughts and comments on this issue.

        Regards,
        Jim Metz

James T. Metz, Ph.D.
Research Investigator Chemist

GPRD R46Y AP10-2
Abbott Laboratories
100 Abbott Park Road
Abbott Park, IL 60064-6100
U.S.A.

Office (847) 936 - 0441
FAX (847) 935 - 0548

james.metz_._abbott.com

This communication may contain information that is legally privileged,
confidential, or exempt from disclosure. If you are not the intended
recipient, please note that any dissemination, distribution, use, or copying
of this communication is strictly prohibited. Anyone who receives this
message in error should notify the sender immediately by telephone or return
email and delete it from his or her computer.
Received on 2004-06-09 - 12:24 GMT