Statistical prediction schemes for the coiled-coil motif

Abstract

This thesis proposes a new multi-dimensional scoring approach for identifying and distinguishing trimeric and dimeric coiled coils. Practical issues in the implementation of the two-stranded coiled coil prediction algorithm PairCoil suggested by Berger are discussed. This algorithm is naturally extended to the domain of three-stranded coiled coils in the implementation of the MultiCoil program. The computations are probabilistically justified and based upon data gathered from a newly constructed three-stranded Coiled coils database comprising 6319 amino acid residues, as well as from the previously constructed two-stranded coiled-coil database. In addition to identifying coiled coils not predicted by previous two-stranded database programs, MultiCoil accurately classifies the oligomerization states of known dimeric and trimeric coiled coils. Analysis of the MultiCoil scores provides insight into structural features of coiled coils, including statistically justifiable estimates of the fraction of all protein residues that form three-stranded coiled coils and the fraction that form two-stranded coiled coils. Several methods for accounting for sampling errors in the databases are suggested and empirically analyzed with regard to the performance of the MultiCoil program. A second probabilistic algorithm for classifying a given coiled coil as dimeric or trimeric is also derived and implemented.