Institute for Protein Design

High-Resolution Comparative Modeling with RosettaCM

October 8, 2013    Song, Y., DiMaio F., et al. (2013).  High-resolution comparative modeling with RosettaCM. Structure. 21(10), 1735-42. 

Researchers in the Baker group describe an improved method for comparative modeling, RosettaCM, which optimizes a physically realistic all-atom energy function over the conformational space defined by homologous structures. Given a set of sequence alignments, RosettaCM assembles topologies by recombining aligned segments in Cartesian space and building unaligned regions de novo in torsion space. The junctions between segments are regularized using a loop closure method combining fragment superposition with gradient-based minimization. The energies of the resulting models are optimized by all-atom refinement, and the most representative low-energy model is selected. The CASP10 experiment suggests that RosettaCM yields models with more accurate side-chain and backbone conformations than other methods when the sequence identity to the templates is greater than ∼15%.

RosettaCM

Examples of improvements over starting templates in CASP10. Top row, difference between per-residue deviations of best template to native structure and deviations of server model to native structure for T0667 (left), and T0702 (middle) and T0685 (right). Values less than zero indicate regions in which the submitted model is closer to the true structure than the best template. Results are shown for first submitted models: green, RosettaCM; blue, HHpredA; magenta, Zhang-server. The structural comparisons in rows 2-4 are over the region with the largest improvements over the templates indicated by red arrow in the first row. (D-L). The native structures are in black; the best template is in orange (D-F); and models from RosettaServer are in green (G-I). HHpredA and ZhangServer models are in blue and magenta for comparison (J-L). Orange labels indicate aligned template residue identities; black labels, the target residue identities.