A fragment-free probabilistic graphical model for conformational sampling in continuous space. FUSION captures local relationships between protein sequence and structural features and allows for probabilistic sampling of conformational space of the protein backbone in full-atomic detail (i.e., at the same granularity as fragment assembly) from a continuous space different from the discrete space of fragment assembly. FUSION assesses its accuracy using 'blind' protein targets with a length up to 250 residues from the CASP11 structure prediction exercise. The method reduces sampling bottlenecks, exhibits strong convergence, and demonstrates better performance than the popular fragment assembly method, ROSETTA, on relatively larger proteins with a length of more than 150 residues in our benchmark set.
Department of Computer Science, University of Missouri, Columbia, MO, USA; Informatics Institute, University of Missouri, Columbia, MO, USA; Bond Life Science Center, University of Missouri, Columbia, MO, USA
FUSION funding source(s)
This work was supported by US National Institutes of Health grant (R01GM093123).