Application of membrane protein topology prediction

Sammanfattning: Membrane proteins often have essential functions in the cell and many are important drug targets, yet only a small fraction of available protein structures are of membrane proteins. Experimental techniques for elucidating membrane protein structures have proven laborious and expensive, opening the field for comparatively inexpensive computational modeling. Topology prediction addresses a sub-problem of structure prediction for α-helical membrane proteins by modeling which parts of the peptide chain are in, and which parts are on either side, of the membrane.This work describes an algorithm for combining the results of several topology prediction methods to increase prediction accuracy and to quantify prediction reliability, and a faster implementation of the algorithm applicable to large-scale genome data.Further, topology prediction is applied, together with other sequence-based methods, to detect duplications in membrane proteins in whole genomes. We find more duplications in the genomes of yeast and E. coli than in human, possibly due to the abundance of nonduplicated GPCRs in human. A gene duplication and subsequent fusion event constitute a likely origin for duplicated proteins, yet only for one superfamily, the AcrB Multidrug Efflux Pump, do we find the duplicated unit in its nonduplicated form. This apparent scarcity of nonduplicated forms is confirmed when extending the study to the whole human genome.Finally, a benchmark study of topology prediction on several comparably large datasets is described. We confirm previous results showing that methods utilizing homology information top the ranking of topology prediction methods. We also see that the separation of membrane proteins from non-membrane proteins has a partially different set of requirements than topology prediction of membrane proteins, and we suggest a pipeline using different methods for these two tasks.