19th-banner-rev.gif

Book of Abstracts: Albany 2009

category image Albany 2009
Conversation 16
June 16-20 2009
© Adenine Press (2008)

Pharmacophore Modeling Using a Reduced Protein Representation as a Tool for Srtucture-Based Protein Function Prediction

Using the double centroid reduced representation (DCRR) of proteins, we have modeled the pharamacophores for ATP and GTP in ser/thr protein kinases (stPK) and small Ras-type G-proteins (RtGP) . In DCRR, each amino acid in the protein is represented by two points, namely, the centroids of its backbone and sidechain atoms. The pharmacophore model, which we call the ?3D search motif? (3D SM), is a tetrahedron with a unique root node, R, and three branch nodes, n1, n2 and n3; it also has three root-branch edges, Rn1, Rn2 and Rn3, and three branch-branch edges, n1n2, n1n3 and n2n3, all of specific lengths. These four nodes correspond to the four amino acids with the most dominant interactions (hydrogen bonds and van der Waals interactions) with the ligand atoms. We next developed an analytical algorithm (written in Fortran 90) for screening protein 3D structures for the 3D SM. The ATP and GTP 3D SMs were determined from sets of experimentally solved training structures, all of which contain the bound ligand. Validation tests performed on ?unseen? positive and negative structures reveal that the specificity of the method is nearly 100% for both protein families, and a sensitivity of 60% for the stPK family and approximately 93% for the RtGP family. Further tests reveal that our algorithm can distinguish effectively between GTP and GTP-like ligands, and between ATP- and ATP-like ligands. It is also shown that the method, which is local structure-based, works successfully in cases where global structure-based methods fail. These results show that the combined modeling and screening methods might be effective for the prediction of proteins belonging to the RtGP and stPK families. Finally, as a benchmark experiment, the method was applied to a set of protein 3D structures predicted by 123D threading and partially refined by Modeller v6.2 from the proteome of Dictyostelium discoideum, with promising results.

Vicente M. Reyes

Dept. of Biological Sciences
Rochester Institute of Technology
Rochester, NY 14623

Ph: 585-475-4115
vmrsbi@rit.edu