Virtual screening and similarity searching using binary kernel discrimination


Peter Willett, Department of Information Studies, Department of Information Studies, University of Sheffield, Western Bank, S10 2TN Sheffield, United Kingdom
Binary kernel discrimination (BKD) is a machine learning technique that has recently been suggested for use in virtual screening. A molecule is scored by calculating its similarities with sets of known active and known inactive molecules, the number of these similarities contributing to the overall score for that molecule being determined by an optimisable parameter. This paper reports the use of BKD in simulated virtual screening experiments with public and corporate datasets in which the molecules are characterised by 2D fragment bit-strings. Our results suggest that BKD is fully competitive with existing approaches to 2D virtual screening in terms of its ability to prioritise compounds for biological testing. We also demonstrate that a simple modification of the method provides an effective way of carrying out similarity searches when multiple reference structures are available.

Herman Skolnik Award Symposium
8:15 AM-11:45 AM, Tuesday, August 24, 2004 Pennsylvania Convention Center -- 110A&B, Oral

Division of Chemical Information

The 228th ACS National Meeting, in Philadelphia, PA, August 22-26, 2004