Chemistry Department Seminar
"Physicochemical Properties in Functional Genomics; or What Do Polyprotic Acid Equilibria Have To Do with Protein Function Annotation?"
Presented by Mary Jo Ondrechen, Northeastern University
Friday, October 13, 2006, 10:30 am — Hamilton Seminar Room, Bldg. 555
Some simple concepts from physical chemistry form the basis for a computational methodology with valuable predictive capabilities for proteomics and genomics. Our method, called THEMATICS (for Theoretical Microscopic Titration Curves), enables the identification of local interaction sites in protein structures and opens the door to the design of specific ligands for protein structures in advance of biochemical characterization. The ionizable residues in catalytic sites and recognition sites in proteins tend to have theoretical titration curves that deviate significantly from standard Henderson-Hasselbalch (H-H) behavior. It is argued that such non-H-H titration behavior enables both local protonation states to be populated over a wide pH range and that this facilitates catalysis and reversible recognition. THEMATICS uses a Poisson-Boltzmann calculation to obtain the approximate electrical potential function from the 3D structure of a protein, then computes the titration curves (mean charge as a function of pH) for all of the ionizable residues in the structure, then identifies those residues that deviate most from typical H-H form. Clusters of two or more such deviant residues in physical proximity are reliable predictors of catalytic and binding sites in proteins. With optimized statistical criteria, our method predicts sites correctly for 93% of the 170 enzymes in the Catalytic Site Atlas (CSA) (http://www.ebi.ac.uk/thornton-srv/databases/CSA/) with a low average filtration ratio of just 3%. Our optimized statistical selection returns an average sensitivity to CSA-annotated catalytic residues of 50% for the CSA proteins; this sensitivity increases to 76% using Support Vector Machines (SVMs). For the electrostatic method with either statistical or SVM selection, average precision rates on a representative sample set of enzymes are two to three times higher than for any other accessible method, with significantly better filtration ratios. Some examples illustrate th
Hosted by: John Miller
3245 | INT/EXT | Events Calendar