Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2011 Aug 10;19(8):1182-91.
doi: 10.1016/j.str.2011.05.004.

Improving Protein Structure Prediction Using Multiple Sequence-Based Contact Predictions

Affiliations
Free PMC article

Improving Protein Structure Prediction Using Multiple Sequence-Based Contact Predictions

Sitao Wu et al. Structure. .
Free PMC article

Abstract

Although residue-residue contact maps dictate the topology of proteins, sequence-based ab initio contact predictions have been found little use in actual structure prediction due to the low accuracy. We developed a composite set of nine SVM-based contact predictors that are used in I-TASSER simulation in combination with sparse template contact restraints. When testing the strategy on 273 nonhomologous targets, remarkable improvements of I-TASSER models were observed for both easy and hard targets, with p value by Student's t test<0.00001 and 0.001, respectively. In several cases, template modeling score increases by >30%, which essentially converts "nonfoldable" targets into "foldable" ones. In CASP9, I-TASSER employed ab initio contact predictions, and generated models for 26 FM targets with a GDT-score 16% and 44% higher than the second and third best servers from other groups, respectively. These findings demonstrate a new avenue to improve the accuracy of protein structure prediction especially for free-modeling targets.

Figures

Figure 1
Figure 1
TM-score of the first-ranking models generated by the normal implementation of I-TASSER on 164 non-homologous Hard targets (stars) and 109 non-homologous Easy/Medium targets (circles) versus that of I-TASSER when using ab initio contact predictions.
Figure 2
Figure 2
An illustrative example of I-TASSER modeling for the target protein “1kafA”. (A) Experimental structure. (B) Model generated by I-TASSER without using ab initio contact prediction. (C) Model generated by I-TASSER with ab initio contact prediction. (D) Map of native Cα contacts (green “*”), template-based predicted Cα contacts (red “*”), and consensus sequence- and template-based Cα contact predictions (blue “+”). (E) Map of native side-chain center contacts (green “*”), template-based predicted side-chain center contacts (red “*”), and consensus sequence- and template-based side-chain center contacts (blue “+”).
Figure 3
Figure 3
An illustrative example of I-TASSER modeling for the target protein “1zkeA”. (A) Experimental structure. (B) Model generated by I-TASSER without ab initio contact prediction. (C) Model generated by I-TASSER with ab initio contact prediction. (D) Map of native Cα contacts (green “*”), template-based predicted Cα contacts (red “*”), and consensus sequence- and template-based Cα contact predictions (blue “+”). (E) Map of native side-chain center contacts (green “*”), template-based predicted side-chain center contacts (red “*”), and consensus sequence- and template-based side-chain center contacts (blue “+”).
Figure 4
Figure 4
An illustrative example of I-TASSER modeling for the target protein “1zc1A”. (A) Experimental structure. (B) Model generated by I-TASSER without ab initio contact prediction. (C) Model generated by I-TASSER with ab initio contact prediction. (D) Map of native Cα contacts (green “*”), template-based predicted Cα contacts (red “*”), and consensus sequence- and template-based Cα contact predictions (blue “+”). (E) Map of native side-chain center contacts (green “*”), template-based predicted side-chain center contacts (red “*”), and consensus sequence- and template-based side-chain center contacts (blue “+”).
Figure 5
Figure 5
Analysis of I-TASSER modeling for the CASP8 “easy” target “T0437”. (A) The best template protein “2jz5A” (top panel) and its optimal structural alignment by TM-align (Zhang and Skolnick, 2005) on the experimental structure (RMSD=1.34 Å, bottom panel); (B) Template-based (top panel) and sequence-based (lower panel) contact predictions represented by thin sticks (red color: true positive predictions; green color: false positive predictions); (C) The I-TASSER model and its superimposition to the native structure (RMSD=1.13 Å). In the bottom subfigures of A and C, the native structure is displayed in thin lines while the template (or model) is in thick lines. Blue to red color runs from N- to C-terminus. The region encircled in white is where most of the improvement occurs.
Figure 6
Figure 6
I-TASSER modeling on two CASP9 hard targets: “T0553” (top panel) and “T0604_1” (bottom panel). The left panels are backbones of the native structures with cross line representing SVMSEQ Cα contact predictions at distance <8Å (red solid lines: true positive predictions; blue dashed lines: false positive predictions). The middle and right panels show the experimental structures and the I-TASSER models, respectively. Blue to red runs from N- to C-terminus. For T0553, TM-score=0.493, RMSD=7.3Å; for T0604_1, TM-score=0.691, RMSD=2.7Å. See also Table S1.

Similar articles

See all similar articles

Cited by 30 articles

See all "Cited by" articles

Publication types

LinkOut - more resources

Feedback