Prediction of Affinities of Molecular Complexes: A Hybrid Approach

Monday, April 7, 1997 - 9:00am - 10:00am
Keller 3-180
Garland Marshall (Washington University)
For efficient design of ligands, estimation of their affinities helps determine their relative synthetic priorities. For receptors of unknown structure, 3D QSAR can be derived for a set of ligands which are both self-consistent and predictive. For receptors of known 3D structure, two approaches have been developed. The first uses simulations and mutates a compound with known affinity to the compound of interest and estimates DDG. Alternative approaches use heuristics and include CoMFA, neural networks, etc. Hybrid approach, such as VALIDATE, estimate the affinities of novel receptor complexes based on properties calculated from a minimized complex. We have used HIV protease inhibitors as a test case to explore use of various parameters to represent the molecular complex and their impact on the predictability of resulting models.