Improved de novo structure prediction in CASP11 by incorporating coevolution information into Rosetta

Abstract

We describe CASP11 de novo blind structure predictions made using the Rosetta structure prediction methodology with both automatic and human assisted protocols. Model accuracy was generally improved using coevolution derived residue-residue contact information as restraints during Rosetta conformational sampling and refinement, particularly when the number of sequences in the family was more than three times the length of the protein. The highlight was the human assisted prediction of T0806, a large and topologically complex target with no homologs of known structure, which had unprecedented accuracy-<3.0 Å root-mean-square deviation (RMSD) from the crystal structure over 223 residues. For this target, we increased the amount of conformational sampling over our fully automated method by employing an iterative hybridization protocol. Our results clearly demonstrate, in a blind prediction scenario, that coevolution derived contacts can considerably increase the accuracy of template-free structure modeling. Proteins 2016; 84(Suppl 1):67-75. © 2015 Wiley Periodicals, Inc.

Keywords: ab initio prediction; coevolution; contact prediction; protein structure prediction; rosetta.

Figure 1

Fully automated Robetta structure prediction protocol.

Figure 2

Choice of restraint functional form. A) Alternative functional form for restraint penalty functions. B) Use of sigmoidal restraints with no gradient beyond 8 Å requires a large amount of sampling, but reduces the impact of incorrect contact predictions on the resulting models, which maximize satisfaction of internally consistent sets of contacts.

Figure 3

Human assisted structure prediction protocol.

Figure 4

Comparison of Robetta results with (during CASP11) and without (post CASP control) GREMLIN predicted contacts.

Figure 5

Robetta highlights. A) T0790-D2 (residues 136–265); B) T0789-D1 (residues 6–81); C) T0761-D2 (residues 202–285); D) T0767-D2 (residues 133– 312). All the GREMLIN contact predictions used in modeling are shown in the scatter plot; most of the high scoring residue pairs were indeed in contact in the native structure (x-axis). Lines connect residue pairs with GREMLIN score ≥1.5 (for cases with more sequences, more contacts are shown, with cutoff as indicated in figure); green, closest heavy atom distance <5 Å; yellow, distance closest <10 Å and red, >10 Å.

Figure 6

Human assisted prediction highlights. A) T0806; B) T0824; C) T0836.

Figure 7

Re-ranking the five submitted models using a single-model MQA method, ProQ2, improves model 1 accuracy for FM domains.