Statistical methods for matching protein-ligand binding sites
Year: 2009
Type: dissertation
Abstract: Faculty of Maths and Physical Sciences School of Mathematics Doctor of Philosophy by John Richard Davies SitesBase is a binding site database that contains over 33000 binding sites extracted from protein structures taken from the PDB that have been cocrystallised with ligands. These sites have been matched in a pairwise fashion using a geometric hashing algorithm to generate a raw matching score which is the number of commonly superimposed atoms. However it is difficult to interpret the quality of a matching pair solely from this raw score, particularly because the size of the binding sites in question show large variation in size. We develop a statistical model for matching protein-ligand binding sites called the Poisson Index (PI) and show that this model is better suited to finding significant binding site matches than other similarity measures such as the Tanimoto Index (TI). We use PI to investigate statistically significant cross fold similarities in a representative set of sites taken from SitesBase. We investigate the nature of these cross fold matches using several different structural and functional classifiers. We find that overall, cross fold matches are dominated by binding sites whose function is critical to maintaining fundamental biological processes that would have been important in the earliest forms of life. We also generate networks of significant fold-fold interactions and find that four of the five most ancestral folds [1] are important members of these networks. These results support a model of protein evolutionary history that is dominated by divergent evolution, where functions such as nucleotide binding and redox chemistry have diversified into other novel functions over time. Finally we investigated the use of PI and binding site similarity in predicting “off-target liabilities” in drug design studies by calculating enrichment of protein superfamilies predicted to contain off-target liabilities to protein kinases using the Poisson Index and SitesBase. We found that the superfamilies predicted using SitesBase and PI were enriched in these datasets, suggesting that as a proof of concept the principle of identifying off-target liabilities using a similarity score such as PI is plausible provided that it is used in concert with other more comprehensive methods.
Author: John R. Davies
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W2588507604