Accounting for 3-D descriptors of conformers in QSAR modeling

CINF 99

Shaillay K Dogra, shaillay@strandls.com, Achintya Das, achintya@strandls.com, and Kalyanasundaram Subramanian, kas@strandls.com. Cheminformatics, Strand Life Sciences Pvt. Ltd, 237, C. V. Raman Avenue, Raj Mahal Vilas, Bangalore, India
3D descriptors computed from a single conformer are used in QSAR modeling. However, the molecule may exist in various conformers. How do we incorporate this variation in 3D descriptors in the modeling process? Does this affect modeling significantly? Averaging of descriptors based on probability-weights derived from a Boltzmann-distribution has been used as a method to address this issue. In an extensive study on a Solubility dataset (1305 compounds, ~45k conformers, 634 3D descriptors), we found that this approach did not make a significant difference. Comparable models were obtained for the two sets – a Boltzmann-averaged descriptors set and a minimum-conformer descriptors set. Though the values for a given descriptor differed across the two sets, these changed in a strongly correlated manner eventually giving us similar statistical models. Is there a better measure to account for the differences in conformers? We investigate different means of sampling and averaging and check their performance.
 

General Papers
1:00 PM-2:30 PM, Thursday, March 29, 2007 McCormick Place North -- Room N134, Level 1, Oral

Sci-Mix
8:00 PM-10:00 PM, Monday, March 26, 2007 Hyatt Regency Chicago -- Riverside Center, Sci-Mix

Division of Chemical Information

The 233rd ACS National Meeting, Chicago, IL, March 25-29, 2007