Multiple Model Approach: Exploring the Limits of Comparative Modeling

One of the biggest problems in modeling distantly related proteins is the quality of the target-template alignment. This problem often results in low quality models that do not utilize all the information available in the template structure. The divergence of alignments at a low sequence identity level, which is a hindrance in most modeling attempts, is used here as a basis for a new technique of Multiple Model Approach (MMA). Alternative alignments prepared here using different mutation matrices and gap penalties, combined with automated model building, are used to create a set of models that explore a range of possible conformations for the target protein. Models are evaluated using different techniques to identify the best model. In the set of examples studied here, the correct target structure is known, which allows the evaluation of various alignment and evaluation strategies. For a randomly selected group of distantly homologous protein pairs representing all structural classes and various fold types, it is shown that a threading score based on simplified statistical potentials of mean force can identify the best models and, consequently, the most reliable alignment. In cases where the difference between target and template structures is significant, the threading score shows clearly that all models are wrong, therefore disqualifying the template.     

Systematic analysis of the effect of multiple templates on the accuracy of comparative models of protein structure

Background  

Although multiple templates are frequently used in comparative modeling, the effect of inclusion of additional template(s) on model accuracy (when compared to that of corresponding single-template based models) is not clear. To address this, we systematically analyze two-template models, the simplest case of multiple-template modeling. For an existing target-template pair (single-template modeling), a two-template based model of the target sequence is constructed by including an additional template without changing the original alignment to measure the effect of the second template on model accuracy.     

Homology-Modelling Protein-Ligand Interactions: Allowing for Ligand-Induced Conformational Change

Current homology-modelling methods do not consider small molecules in their automated processes. Therefore, the development of a reliable tool for protein-ligand homology modelling is an important next step in generating plausible models for molecular interactions. Two automated protein-ligand homology-modelling strategies, requiring no expert knowledge from the user, are investigated here. Both employ the “induced fit” concept with flexibility in side chains and ligand. The most successful strategy superimposes the new ligand over the original ligand before homology modelling, allowing the new ligand to be taken into consideration during protein modelling (rather than after), facilitating conformational change in the local backbone if necessary. We show that this approach results in successful modelling of the ligand and key binding-site residues of angiotensin-converting enzyme 2 (ACE2) from its homologue ACE, which is not possible via conventional homology modelling or by homology modelling followed by docking. Several other difficult target complexes are also successfully modelled, reproducing native protein-ligand contacts with significantly different biological substrates and different binding-site conformations. These include the modelling of Cdk5 (cyclin-dependent kinase 5) from Cdk2, thymidine phosphorylase from a bacterial homologue, and dihydrofolate reductase from a recombinant variant with a markedly different inhibitor. In terms of average modelling quality across 82 targets, the ligand RMSD with respect to the experimental structure is 1.4 Å (and 2.0 Å for the protein binding site) for “easy” cases and 2.9 Å for the ligand (and 2.7 Å for the protein binding site) in “hard” cases. This demonstrates the importance of selecting an optimal template. Ligand-modelling accuracy is strongly dependent on target-template ligand structural similarity, rather than target-template sequence identity. However, protein-modelling accuracy is dependent on both. Our automated protein-ligand homology-modelling strategy generates a higher degree of accuracy than homology modelling followed by docking, generating an average ligand RMSD that is 1-2 Å better than docking with homology models.     

Refinement of protein structures by iterative comparative modeling and CryoEM density fitting

We developed a method for structure characterization of assembly components by iterative comparative protein structure modeling and fitting into cryo-electron microscopy (cryoEM) density maps. Specifically, we calculate a comparative model of a given component by considering many alternative alignments between the target sequence and a related template structure while optimizing the fit of a model into the corresponding density map. The method relies on the previously developed Moulder protocol that iterates over alignment, model building, and model assessment. The protocol was benchmarked using 20 varied target-template pairs of known structures with less than 30% sequence identity and corresponding simulated density maps at resolutions from 5A to 25A. Relative to the models based on the best existing sequence profile alignment methods, the percentage of C(alpha) atoms that are within 5A of the corresponding C(alpha) atoms in the superposed native structure increases on average from 52% to 66%, which is half-way between the starting models and the models from the best possible alignments (82%). The test also reveals that despite the improvements in the accuracy of the fitness function, this function is still the bottleneck in reducing the remaining errors. To demonstrate the usefulness of the protocol, we applied it to the upper domain of the P8 capsid protein of rice dwarf virus that has been studied by cryoEM at 6.8A. The C(alpha) root-mean-square deviation of the model based on the remotely related template, bluetongue virus VP7, improved from 8.7A to 6.0A, while the best possible model has a C(alpha) RMSD value of 5.3A. Moreover, the resulting model fits better into the cryoEM density map than the initial template structure. The method is being implemented in our program MODELLER for protein structure modeling by satisfaction of spatial restraints and will be applicable to the rapidly increasing number of cryoEM density maps of macromolecular assemblies.     

Systematic analysis of added-value in simple comparative models of protein structure

Added-value is the additional information that a model carries with respect to the template structure used for model building. Thousands of single-template models, corresponding to proteins of known structure, were analyzed. The accuracy of structure-derived properties, such as residue accessibility, surface area, electrostatic potential, and others, was determined as a function of template:target sequence identity by comparing the models with their corresponding experimental structures. Added-value was determined by comparing the accuracy in models with that from templates. Geometry-dependent properties such as neighborhood of buried residues and accessible surface area showed low added-value. Properties that also depend on the protein sequence, such as presence of polar areas and electrostatic potential, showed high added-value. In general added-value increases when template:target sequence identity decreases, but it is also affected by alignment errors. This study justifies the use of models instead of the use of templates to estimate structure-derived properties of a target protein.     

Accuracy of Protein-Protein Binding Sites in High-Throughput Template-Based Modeling

The accuracy of protein structures, particularly their binding sites, is essential for the success of modeling protein complexes. Computationally inexpensive methodology is required for genome-wide modeling of such structures. For systematic evaluation of potential accuracy in high-throughput modeling of binding sites, a statistical analysis of target-template sequence alignments was performed for a representative set of protein complexes. For most of the complexes, alignments containing all residues of the interface were found. The full interface alignments were obtained even in the case of poor alignments where a relatively small part of the target sequence (as low as 40%) aligned to the template sequence, with a low overall alignment identity (<30%). Although such poor overall alignments might be considered inadequate for modeling of whole proteins, the alignment of the interfaces was strong enough for docking. In the set of homology models built on these alignments, one third of those ranked 1 by a simple sequence identity criteria had RMSD<5 Å, the accuracy suitable for low-resolution template free docking. Such models corresponded to multi-domain target proteins, whereas for single-domain proteins the best models had 5 Å<RMSD<10 Å, the accuracy suitable for less sensitive structure-alignment methods. Overall, ∼50% of complexes with the interfaces modeled by high-throughput techniques had accuracy suitable for meaningful docking experiments. This percentage will grow with the increasing availability of co-crystallized protein-protein complexes.     

Improved scoring function for comparative modeling using the M4T method

Improvements in comparative protein structure modeling for the remote target-template sequence similarity cases are possible through the optimal combination of multiple template structures and by improving the quality of target-template alignment. Recently developed MMM and M4T methods were designed to address these problems. Here we describe new developments in both the alignment generation and the template selection parts of the modeling algorithms. We set up a new scoring function in MMM to deliver more accurate target-template alignments. This was achieved by developing and incorporating into the composite scoring function a novel statistical pairwise potential that combines local and non-local terms. The non-local term of the statistical potential utilizes a shuffled reference state definition that helped to eliminate most of the false positive signal from the background distribution of pairwise contacts. The accuracy of the scoring function was further increased by using BLOSUM mutation table scores.     

Refinement of comparative models of protein structure by using multicanonical molecular dynamics simulations

Comparative modelling is a powerful method that easily predicts a considerably accurate structure of a protein by using a template structure having a similar amino-acid sequence to the target protein. However, in the region where the amino-acid sequence is different between the target and the template, the predicted structure remains unreliable. In such a case, the model has to be refined. In the present study, we explored the possibility of a molecular dynamics-based method, using the human SAP Src Homology 2 (SH2) domain as the modelling target. The multicanonical method was used to alleviate the multiple-minima problem and the generalised Born/surface area model was used to reduce the computational cost. In addition, position restraints were imposed on the atoms in the reliable regions to avoid unnecessary conformational sampling. We analyzed the conformational distribution of the ligand-recognition loop of the domain and found that the most populated conformational clusters in the ensemble of the model agreed well with one of the two major clusters in the ensemble of the reference simulation starting from the crystal structure. This demonstrates that the current refinement method can significantly improve the accuracy of an unreliable region in a comparative model.     

A novel protocol of energy optimisation for predicted protein structures built by homology modelling

Homology modelling was applied to predict the three-dimensional (3D) structures of six sets of lipase proteins. Sequence identities between the target and template were 34.6, 44.9, 57.4, 69.9, 79.0 and 86.2%, respectively. Then, eight different protocols including three optimising factors [periodically bounded cell (PBC) water, molecular dynamics (MD) simulation, ‘grade-unpacking’ strategy or ‘combinatorial’ strategy] were used to refine the initial model of each system. By comparing the energy-optimised models with the true 3D structure of the target protein in terms of all backbone atoms' root mean square deviation, we determined a novel but all-purpose protocol for model refinement. The protocol refined a homology model by adopting the ‘grade-unpacking’ strategy for energy minimisation while the model was solvated in PBC water. Furthermore, by comparing the influence of each single optimising factor on the accuracy of the refined structure, we found that introducing the MD simulation into the model refinement method would decrease the accuracy of the final protein structure while methods with either PBC water or the ‘grade-unpacking’ strategy would increase the accuracy of the final model.     

Homology modelling of the human eukaryotic initiation factor 5A (eIF-5A).

Homology modelling of the human eIF-5A protein has been performed by using a multiple predictions strategy. As the sequence identity between the target and the template proteins is nearly 30%, which is lower than the commonly used threshold to apply with confidence the homology modelling method, we developed a specific predictive scheme by combining different sequence analyses and predictions, as well as model validation by comparison to structural experimental information. The target sequence has been used to find homologues within sequence databases and a multiple alignment has been created. Secondary structure for each single protein has been predicted and compared on the basis of the multiple sequence alignment, in order to evaluate and adjust carefully any gap. Therefore, comparative modelling has been applied to create the model of the protein on the basis of the optimized sequence alignment. The quality of the model has been checked by computational methods and the structural features have been compared to experimental information, giving us a good validation of the reliability of the model and its correspondence to the protein structure in solution. Last, the model was deposited in the Protein Data Bank to be accessible for studies on the structure-function relationships of the human eIF-5A.     

Benchmarking of dimeric threading and structure refinement

The understanding of protein-protein interactions is a major goal in the postgenomic era. The prediction of interaction from sequence and the subsequent generation of full-length dimeric models is therefore of great interest especially because the number of structurally characterized protein-protein complexes is sparse. A quality assessment of a benchmark comprised of 170 weakly homologous dimeric target-template pairs is presented. They are predicted in a two-step method, similar to the previously described MULTIPROSPECTOR algorithm: each target sequence is assigned to a monomeric template structure by threading; then, those templates that belong to the same physically interacting dimer template are selected. Additionally we use structural alignments as the "gold standard" to assess the percentage of correctly assigned monomer and dimer templates and to evaluate the threading results with a focus on the quality of the alignments in the interfacial region. This work aims to give a quantitative picture of the quality of dimeric threading. Except for one, all monomer templates are identified correctly, but approximately 40% of the dimer templates are still problematic or incorrect. Preliminary results for three full-length dimeric models generated with the TASSER method show on average a significant improvement of the final model over the initial template.     

An evaluation of automated homology modelling methods at low target template sequence similarity

MOTIVATION: There are two main areas of difficulty in homology modelling that are particularly important when sequence identity between target and template falls below 50%: sequence alignment and loop building. These problems become magnified with automatic modelling processes, as there is no human input to correct mistakes. As such we have benchmarked several stand-alone strategies that could be implemented in a workflow for automated high-throughput homology modelling. These include three new sequence-structure alignment programs: 3D-Coffee, Staccato and SAlign, plus five homology modelling programs and their respective loop building methods: Builder, Nest, Modeller, SegMod/ENCAD and Swiss-Model. The SABmark database provided 123 targets with at least five templates from the same SCOP family and sequence identities 相似文献   

Issues in high-throughput comparative modelling: a case study using the ubiquitin E2 conjugating enzymes

Sequences of the ubiquitin-conjugating enzyme (UBC or E2) family were used as a test set to investigate issues associated with the high-throughput comparative modelling of protein structures. A semi-automatic method was initially developed with particular emphasis on producing models of a quality suitable for structural comparison. Structural and sequence features of the E2 family were used to improve the sequence alignment and the quality of the structural templates. Initially, failure to correct for subtle structural inconsistencies between templates lead to problems in the comparative analysis of the UBC electrostatic potentials. Modelling of known UBC structures using Modeller 4.0 showed that multiple templates produced, on average, no better models than the use of just one template, as judged by the root-mean-squared deviation between the comparative model and crystal structure backbones. Using four different quality-checking methods, for a given target sequence, it was not possible to distinguish the model most similar to the experimental structure. The UBC models were thus finally modelled using only the crystal structure template with the highest sequence identity to the target to be modelled, and producing only one model solution. Quality checking was used to reject models with obvious structural anomalies (e.g., bad side-chain packing). The resulting models have been used for a comparison of UBC structural features and of their electrostatic potentials. The work was extended through the development of a fully automated pipeline that identifies E2 sequences in the sequence databases, aligns and models them, and calculates the associated electrostatic potential.     

A composite score for predicting errors in protein structure models

Reliable prediction of model accuracy is an important unsolved problem in protein structure modeling. To address this problem, we studied 24 individual assessment scores, including physics-based energy functions, statistical potentials, and machine learning-based scoring functions. Individual scores were also used to construct approximately 85,000 composite scoring functions using support vector machine (SVM) regression. The scores were tested for their abilities to identify the most native-like models from a set of 6000 comparative models of 20 representative protein structures. Each of the 20 targets was modeled using a template of <30% sequence identity, corresponding to challenging comparative modeling cases. The best SVM score outperformed all individual scores by decreasing the average RMSD difference between the model identified as the best of the set and the model with the lowest RMSD (DeltaRMSD) from 0.63 A to 0.45 A, while having a higher Pearson correlation coefficient to RMSD (r=0.87) than any other tested score. The most accurate score is based on a combination of the DOPE non-hydrogen atom statistical potential; surface, contact, and combined statistical potentials from MODPIPE; and two PSIPRED/DSSP scores. It was implemented in the SVMod program, which can now be applied to select the final model in various modeling problems, including fold assignment, target-template alignment, and loop modeling.     

Benchmarking template selection and model quality assessment for high-resolution comparative modeling

Comparative modeling is presently the most accurate method of protein structure prediction. Previous experiments have shown the selection of the correct template to be of paramount importance to the quality of the final model. We have derived a set of 732 targets for which a choice of ten or more templates exist with 30-80% sequence identity and used this set to compare a number of possible methods for template selection: BLAST, PSI-BLAST, profile-profile alignment, HHpred HMM-HMM comparison, global sequence alignment, and the use of a model quality assessment program (MQAP). In addition, we have investigated the question of whether any structurally defined subset of the sequence could be used to predict template quality better than overall sequence similarity. We find that template selection by BLAST is sufficient in 75% of cases but that there are examples in which improvement (global RMSD 0.5 A or more) could be made. No significant improvement is found for any of the more sophisticated sequence-based methods of template selection at high sequence identities. A subset of 118 targets extending to the lowest levels of sequence similarity was examined and the HHpred and MQAP methods were found to improve ranking when available templates had 35-40% maximum sequence identity. Structurally defined subsets in general are found to be less discriminative than overall sequence similarity, with the coil residue subset performing equivalently to sequence similarity. Finally, we demonstrate that if models are built and model quality is assessed in combination with the sequence-template sequence similarity that a extra 7% of "best" models can be found.     

SCWRL and MolIDE: computer programs for side-chain conformation prediction and homology modeling

SCWRL and MolIDE are software applications for prediction of protein structures. SCWRL is designed specifically for the task of prediction of side-chain conformations given a fixed backbone usually obtained from an experimental structure determined by X-ray crystallography or NMR. SCWRL is a command-line program that typically runs in a few seconds. MolIDE provides a graphical interface for basic comparative (homology) modeling using SCWRL and other programs. MolIDE takes an input target sequence and uses PSI-BLAST to identify and align templates for comparative modeling of the target. The sequence alignment to any template can be manually modified within a graphical window of the target-template alignment and visualization of the alignment on the template structure. MolIDE builds the model of the target structure on the basis of the template backbone, predicted side-chain conformations with SCWRL and a loop-modeling program for insertion-deletion regions with user-selected sequence segments. SCWRL and MolIDE can be obtained at (http://dunbrack.fccc.edu/Software.php).     

Improving the quality of protein structure models by selecting from alignment alternatives

Background  

In the area of protein structure prediction, recently a lot of effort has gone into the development of Model Quality Assessment Programs (MQAPs). MQAPs distinguish high quality protein structure models from inferior models. Here, we propose a new method to use an MQAP to improve the quality of models. With a given target sequence and template structure, we construct a number of different alignments and corresponding models for the sequence. The quality of these models is scored with an MQAP and used to choose the most promising model. An SVM-based selection scheme is suggested for combining MQAP partial potentials, in order to optimize for improved model selection.     

Estimating quality of template-based protein models by alignment stability

The error in protein tertiary structure prediction is unavoidable, but it is not explicitly shown in most of the current prediction algorithms. Estimated error of a predicted structure is crucial information for experimental biologists to use the prediction model for design and interpretation of experiments. Here, we propose a method to estimate errors in predicted structures based on the stability of the optimal target-template alignment when compared with a set of suboptimal alignments. The stability of the optimal alignment is quantified by an index named the SuboPtimal Alignment Diversity (SPAD). We implemented SPAD in a profile-based threading algorithm and investigated how well SPAD can indicate errors in threading models using a large benchmark dataset of 5232 alignments. SPAD shows a very good correlation not only to alignment shift errors but also structure-level errors, the root mean square deviation (RMSD) of predicted structure models to the native structures (i.e. global errors), and local errors at each residue position. We have further compared SPAD with seven other quality measures, six from sequence alignment-based measures and one atomic statistical potential, discrete optimized protein energy (DOPE), in terms of the correlation coefficient to the global and local structure-level errors. In terms of the correlation to the RMSD of structure models, when a target and a template are in the same SCOP family, the sequence identity showed a best correlation to the RMSD; in the superfamily level, SPAD was the best; and in the fold level, DOPE was best. However, in a head-to-head comparison, SPAD wins over the other measures. Next, SPAD is compared with three other measures of local errors. In this comparison, SPAD was best in all of the family, the superfamily and the fold levels. Using the discovered correlation, we have also predicted the global and local error of our predicted structures of CASP7 targets by the SPAD. Finally, we proposed a sausage representation of predicted tertiary structures which intuitively indicate the predicted structure and the estimated error range of the structure simultaneously.     

Accuracy of structure-derived properties in simple comparative models of protein structures

The accuracy of comparative models of proteins is addressed here. A set of 12732 single-template models of sequences of known high-resolution structures was built by an automated procedure. Accuracy of several structure-derived properties, such as surface area, residue accessibility, presence of pockets, electrostatic potential and others, was determined as a function of template:target sequence identity by comparing models with their corresponding experimental structures. As expected, the average accuracy of structure-derived properties always increases with higher template:target sequence identity, but the exact shape of this relationship can differ from one property to another. A comparison of structure-derived properties measured from NMR and X-ray structures of the same protein shows that for most properties, the NMR/X-ray difference is of the same order as the error in models based on ~40% template:target sequence identity. The exact sequence identity at which properties reach that accuracy varies between 25 and 50%, depending on the property being analyzed. A general characteristic of simple comparative models is that their surface has increased area as a consequence of being more rugged than that of experimental structures. This suggests that including solvent effects during model building or refinement could significantly improve the accuracy of surface properties in comparative models.