My research program currently has two principal foci. The major focus is structure-based prediction of protein function, with emphasis on the prediction of ligand binding sites and protein-protein interaction partners, two of the most generally accepted methods of predicting protein function based on 3D structure. As for the former, our approach consists of modeling the ligand binding site (LBS) from experimentally solved training structures using a reduced representation of the protein we term as 'double-centroid' (DC); this is essentially pharmacophore modeling of the ligand in question, using a tetrahedral pharmacophore model. We then developed and implemented a screening algorithm to detect such pharmacophore models in protein 3D structures. In poster I and abstract (see below), we establish this method using the ATP and GTP LBSs from ser/thr protein kinases (stPK) and small Ras-type G-proteins (srtGP), respectively. In poster II and abstract (see below), we apply the method to the prediction of ATP-binding proteins of the stPK type, GTP-binding proteins of the srtGP type, and sialic acid-, retinoic acid- and heme-bound and –unbound nitric oxide (NO) binding proteins, using as application set some 800 unannotated protein structures from the PDB. As for the latter, we extended the pharmacophore modeling approach above to the modeling of the interactions (hydrogen-bonding and van der Waals) at the protein-protein interface, with one pharmacophore on each molecule of the complex, which we term as 'interfacial pharmacophore' (IP). In poster III and abstract (see below), we describe the modeling of IPs from nine different protein binary complexes (9 x 2 = 18 IPs in all) and the screening for these IPs in the 800 unannotated protein PDB structures.
A minor focus of my research program is the formulation of novel methods for representing and analyzing protein 3D structures. For example, since protein 3D structures have always been represented in rectangular Cartesian coordinates, it would be interesting to find out what additional information, if any, can be gleaned by representing globular ones in spherical coordinates, elongated ones in cylindrical coordinates, etc. In poster IV and abstract (see below), we transform protein 3D structures from Cartesian to spherical coordinates and demonstrate two applications of this novel representation, namely, the investigation of protein surface topography as a way of finding potential LBSs, and the separation of the protein hydrophilic outer layer (HOL) from the hydrophobic inner core (HIC). In poster V and abstract (see below), we introduce the 'cutting plane' and 'tangent sphere' methods - two complementary methods for quantifying the depth of LBS burial in the receptor protein, a parameter that has implications on protein flexibility.
In addition, we have a number of projects that are either in the planning stages or whose foundations have been laid out and ready to be taken off the ground. For example, we are trying to transform all proteins in the PDB to the DC reduced representation and building a graphical web interface to visualize them together with the protein intramolecular interactions (H-bonds and van der Waals interactions) as a tool and resource for the proteomics community and at the same as a launching pad for our future projects. As another example, being able to separate the HIC from the HOL, we are investigating whether it would allow the prediction of catalytic sites and protein-protein interaction interfaces by searching for clusters of hydrophilic residues in the HIC and patches of hydrophobic residues in the HOL, respectively. As a third example, we are trying to investigate a couple of possible metrics for measuring the 'distance' between two protein structures: one involves projecting the protein atoms (or centroids, as in DC reduced representation) in two dimensions and scoring the coincidences after a search for maximum overlap, and the other by RMSD minimization that takes into account not just the position of the backbone Cα's of the proteins involved but also the sidechain positions (as centroids), volume, pKa, and hydrophobicity index, thus capturing the chemical properties of the protein residues. Still another example, we are trying to build α-helix, β-sheet and loop libraries in order to mine the resulting databases for possible 'rules of association' among these secondary structures to form the intact protein 3D structure. Finally, we are trying a method to predict disease-related SNPs by assessing the proximity of these sites to protein functional sites (in the case of expressed SNPs). |
