Peter Carlin Zaback

Bioinformatics & Computational Biology

My current research is focused on engineering zinc finger proteins (ZFPs) to target specific genomic sequences. Zinc finger proteins are the most frequently occurring DNA-binding domains in human transcription factors, and their modularity makes them attractive prospects for targeting DNA sequences of varying length. We are currently using modules selected for binding to a specific nucleotide triplet at the middle position of a three-finger ZFP in the context of the Zif268 target site for fingers 1 & 3 (F1 & F3, respectively) [Barbas ref]. From these F2-specific modules we have constructed several novel three-finger ZFPs to specifically target several 9 bp target sites, and have demonstrated in vitro binding to 3 of 4 targets tested thus far. In the immediate future we will expand our experiments to more thoroughly test the specificity of binding to these targets, with the goal of evaluating the principal of modular design. I am in the early stages of designing an experiment to test specificity of ZFP-DNA interactions in silico by using molecular dynamics simulations to study the change in free energy of binding that occurs upon mutation in either the DNA or protein sequence. I am also considering using a somewhat more simple modeling approach by using a rotamer approach, and allowing small variations in backbone angle, rather than performing a complete molecular dynamics simulation. While ZFPs are currently the primary focus of my research, I am interested in the more general problem of protein-DNA recognition and - yet more generally - in macromolecular recognition, particularly protein-protein and protein-nucleic acid interaction. I attempted to improve protein-protein interface prediction by modifying the kernel function of a support vector machine (SVM) to implement a substitution matrix-based kernel, but obtained nearly zero improvement. I hope to return to this problem in the future, by testing how this approach performs when separate classifiers are used for each target residue, an approach that has showed modest success in other prediction tasks. In exploring macromolecular interface prediction, I am also interested in incorporating features not yet widely utilized, such as intrinsic disorder in proteins and water-mediated contacts. Dunker and others have been gradually building the case for the significance of protein disorder in protein-DNA interaction, based on the prevalence of DNA-binding proteins in the set of proteins known to contain significant amounts of disorder. Water mediated bonds have been shown to be important in protein-DNA and protein-protein binding, and have been demonstrated to be particularly important in ZFP-DNA interaction. Indeed, Baker et. al noted the absence of water-mediated contacts in their model as a potential explanation for their relatively poorer performance in predicting the free energy of binding for ZFPs.