Improving protein tertiary structure prediction by deep learning and distance prediction in CASP14

Substantial progresses in protein structure prediction have been made by utilizing deep-learning and residue-residue distance prediction since CASP13. Inspired by the advances, we improve our CASP14 MULTICOM protein structure prediction system by incorporating three new components: (a) a new deep learning-based protein inter-residue distance predictor to improve template-free (ab initio) tertiary structure prediction, (b) an enhanced template-based tertiary structure prediction method, and (c) distance-based model quality assessment methods empowered by deep learning. In the 2020 CASP14 experiment, MULTICOM predictor was ranked seventh out of 146 predictors in tertiary structure prediction and ranked third out of 136 predictors in inter-domain structure prediction. The results demonstrate that the template-free modeling based on deep learning and residue-residue distance prediction can predict the correct topology for almost all template-based modeling targets and a majority of hard targets (template-free targets or targets whose templates cannot be recognized), which is a significant improvement over the CASP13 MULTICOM predictor. Moreover, the template-free modeling performs better than the template-based modeling on not only hard targets but also the targets that have homologous templates. The performance of the template-free modeling largely depends on the accuracy of distance prediction closely related to the quality of multiple sequence alignments. The structural model quality assessment works well on targets for which enough good models can be predicted, but it may perform poorly when only a few good models are predicted for a hard target and the distribution of model quality scores is highly skewed. MULTICOM is available at https://github.com/jianlin-cheng/MULTICOM_Human_CASP14/tree/CASP14_DeepRank3 and https://github.com/multicom-toolbox/multicom/tree/multicom_v2.0 .     

Contact order and ab initio protein structure prediction

Although much of the motivation for experimental studies of protein folding is to obtain insights for improving protein structure prediction, there has been relatively little connection between experimental protein folding studies and computational structural prediction work in recent years. In the present study, we show that the relationship between protein folding rates and the contact order (CO) of the native structure has implications for ab initio protein structure prediction. Rosetta ab initio folding simulations produce a dearth of high CO structures and an excess of low CO structures, as expected if the computer simulations mimic to some extent the actual folding process. Consistent with this, the majority of failures in ab initio prediction in the CASP4 (critical assessment of structure prediction) experiment involved high CO structures likely to fold much more slowly than the lower CO structures for which reasonable predictions were made. This bias against high CO structures can be partially alleviated by performing large numbers of additional simulations, selecting out the higher CO structures, and eliminating the very low CO structures; this leads to a modest improvement in prediction quality. More significant improvements in predictions for proteins with complex topologies may be possible following significant increases in high-performance computing power, which will be required for thoroughly sampling high CO conformations (high CO proteins can take six orders of magnitude longer to fold than low CO proteins). Importantly for such a strategy, simulations performed for high CO structures converge much less strongly than those for low CO structures, and hence, lack of simulation convergence can indicate the need for improved sampling of high CO conformations. The parallels between Rosetta simulations and folding in vivo may extend to misfolding: The very low CO structures that accumulate in Rosetta simulations consist primarily of local up-down beta-sheets that may resemble precursors to amyloid formation.     

Functional inferences from blind ab initio protein structure predictions

Ab initio protein structure prediction methods have improved dramatically in the past several years. Because these methods require only the sequence of the protein of interest, they are potentially applicable to the open reading frames in the many organisms whose sequences have been and will be determined. Ab initio methods cannot currently produce models of high enough resolution for use in rational drug design, but there is an exciting potential for using the methods for functional annotation of protein sequences on a genomic scale. Here we illustrate how functional insights can be obtained from low-resolution predicted structures using examples from blind ab initio structure predictions from the third and fourth critical assessment of structure prediction (CASP3, CASP4) experiments.     

Ab initio and template-based prediction of multi-class distance maps by two-dimensional recursive neural networks

Background

Prediction of protein structures from their sequences is still one of the open grand challenges of computational biology. Some approaches to protein structure prediction, especially ab initio ones, rely to some extent on the prediction of residue contact maps. Residue contact map predictions have been assessed at the CASP competition for several years now. Although it has been shown that exact contact maps generally yield correct three-dimensional structures, this is true only at a relatively low resolution (3–4 Å from the native structure). Another known weakness of contact maps is that they are generally predicted ab initio, that is not exploiting information about potential homologues of known structure.

Results

We introduce a new class of distance restraints for protein structures: multi-class distance maps. We show that C _α trace reconstructions based on 4-class native maps are significantly better than those from residue contact maps. We then build two predictors of 4-class maps based on recursive neural networks: one ab initio, or relying on the sequence and on evolutionary information; one template-based, or in which homology information to known structures is provided as a further input. We show that virtually any level of sequence similarity to structural templates (down to less than 10%) yields more accurate 4-class maps than the ab initio predictor. We show that template-based predictions by recursive neural networks are consistently better than the best template and than a number of combinations of the best available templates. We also extract binary residue contact maps at an 8 Å threshold (as per CASP assessment) from the 4-class predictors and show that the template-based version is also more accurate than the best template and consistently better than the ab initio one, down to very low levels of sequence identity to structural templates. Furthermore, we test both ab-initio and template-based 8 Å predictions on the CASP7 targets using a pre-CASP7 PDB, and find that both predictors are state-of-the-art, with the template-based one far outperforming the best CASP7 systems if templates with sequence identity to the query of 10% or better are available. Although this is not the main focus of this paper we also report on reconstructions of C _α traces based on both ab initio and template-based 4-class map predictions, showing that the latter are generally more accurate even when homology is dubious.

Conclusion

Accurate predictions of multi-class maps may provide valuable constraints for improved ab initio and template-based prediction of protein structures, naturally incorporate multiple templates, and yield state-of-the-art binary maps. Predictions of protein structures and 8 Å contact maps based on the multi-class distance map predictors described in this paper are freely available to academic users at the url http://distill.ucd.ie/.     

Cover Image,Volume 87, Issue 12

CASP (critical assessment of structure prediction) assesses the state of the art in modeling protein structure from amino acid sequence. The most recent experiment (CASP13 held in 2018) saw dramatic progress in structure modeling without use of structural templates (historically “ab initio” modeling). Progress was driven by the successful application of deep learning techniques to predict inter-residue distances. In turn, these results drove dramatic improvements in three-dimensional structure accuracy: With the proviso that there are an adequate number of sequences known for the protein family, the new methods essentially solve the long-standing problem of predicting the fold topology of monomeric proteins. Further, the number of sequences required in the alignment has fallen substantially. There is also substantial improvement in the accuracy of template-based models. Other areas—model refinement, accuracy estimation, and the structure of protein assemblies—have again yielded interesting results. CASP13 placed increased emphasis on the use of sparse data together with modeling and chemical crosslinking, SAXS, and NMR all yielded more mature results. This paper summarizes the key outcomes of CASP13. The special issue of PROTEINS contains papers describing the CASP13 assessments in each modeling category and contributions from the participants.     

Protein tertiary structure modeling driven by deep learning and contact distance prediction in CASP13

Predicting residue-residue distance relationships (eg, contacts) has become the key direction to advance protein structure prediction since 2014 CASP11 experiment, while deep learning has revolutionized the technology for contact and distance distribution prediction since its debut in 2012 CASP10 experiment. During 2018 CASP13 experiment, we enhanced our MULTICOM protein structure prediction system with three major components: contact distance prediction based on deep convolutional neural networks, distance-driven template-free (ab initio) modeling, and protein model ranking empowered by deep learning and contact prediction. Our experiment demonstrates that contact distance prediction and deep learning methods are the key reasons that MULTICOM was ranked 3rd out of all 98 predictors in both template-free and template-based structure modeling in CASP13. Deep convolutional neural network can utilize global information in pairwise residue-residue features such as coevolution scores to substantially improve contact distance prediction, which played a decisive role in correctly folding some free modeling and hard template-based modeling targets. Deep learning also successfully integrated one-dimensional structural features, two-dimensional contact information, and three-dimensional structural quality scores to improve protein model quality assessment, where the contact prediction was demonstrated to consistently enhance ranking of protein models for the first time. The success of MULTICOM system clearly shows that protein contact distance prediction and model selection driven by deep learning holds the key of solving protein structure prediction problem. However, there are still challenges in accurately predicting protein contact distance when there are few homologous sequences, folding proteins from noisy contact distances, and ranking models of hard targets.     

Fully automated ab initio protein structure prediction using I-SITES,HMMSTR and ROSETTA

MOTIVATION: The Monte Carlo fragment insertion method for protein tertiary structure prediction (ROSETTA) of Baker and others, has been merged with the I-SITES library of sequence structure motifs and the HMMSTR model for local structure in proteins, to form a new public server for the ab initio prediction of protein structure. The server performs several tasks in addition to tertiary structure prediction, including a database search, amino acid profile generation, fragment structure prediction, and backbone angle and secondary structure prediction. Meeting reasonable service goals required improvements in the efficiency, in particular for the ROSETTA algorithm. RESULTS: The new server was used for blind predictions of 40 protein sequences as part of the CASP4 blind structure prediction experiment. The results for 31 of those predictions are presented here. 61% of the residues overall were found in topologically correct predictions, which are defined as fragments of 30 residues or more with a root-mean-square deviation in superimposed alpha carbons of less than 6A. HMMSTR 3-state secondary structure predictions were 73% correct overall. Tertiary structure predictions did not improve the accuracy of secondary structure prediction.     

Improving protein structure prediction using multiple sequence-based contact predictions

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.     

HCPM--program for hierarchical clustering of protein models

HCPM is a tool for clustering protein structures from comparative modeling, ab initio structure prediction, etc. A hierarchical clustering algorithm is designed and tested, and a heuristic is provided for an optimal cluster selection. The method has been successfully tested during the CASP6 experiment.     

Prediction of protein structure by simulating coarse-grained folding pathways: a preliminary report

A set of software tools designed to study protein structure and kinetics has been developed. The core of these tools is a program called Folding Machine (FM) which is able to generate low resolution folding pathways using modest computational resources. The FM is based on a coarse-grained kinetic ab initio Monte-Carlo sampler that can optionally use information extracted from secondary structure prediction servers or from fragment libraries of local structure. The model underpinning this algorithm contains two novel elements: (a) the conformational space is discretized using the Ramachandran basins defined in the local phi-psi energy maps; and (b) the solvent is treated implicitly by rescaling the pairwise terms of the non-bonded energy function according to the local solvent environments. The purpose of this hybrid ab initio/knowledge-based approach is threefold: to cover the long time scales of folding, to generate useful 3-dimensional models of protein structures, and to gain insight on the protein folding kinetics. Even though the algorithm is not yet fully developed, it has been used in a recent blind test of protein structure prediction (CASP5). The FM generated models within 6 A backbone rmsd for fragments of about 60-70 residues of alpha-helical proteins. For a CASP5 target that turned out to be natively unfolded, the trajectory obtained for this sequence uniquely failed to converge. Also, a new measure to evaluate structure predictions is presented and used along the standard CASP assessment methods. Finally, recent improvements in the prediction of beta-sheet structures are briefly described.     

Ab initio construction of protein tertiary structures using a hierarchical approach

We present a hierarchical method to predict protein tertiary structure models from sequence. We start with complete enumeration of conformations using a simple tetrahedral lattice model. We then build conformations with increasing detail, and at each step select a subset of conformations using empirical energy functions with increasing complexity. After enumeration on lattice, we select a subset of low energy conformations using a statistical residue-residue contact energy function, and generate all-atom models using predicted secondary structure. A combined knowledge-based atomic level energy function is then used to select subsets of the all-atom models. The final predictions are generated using a consensus distance geometry procedure. We test the feasibility of the procedure on a set of 12 small proteins covering a wide range of protein topologies. A rigorous double-blind test of our method was made under the auspices of the CASP3 experiment, where we did ab initio structure predictions for 12 proteins using this approach. The performance of our methodology at CASP3 is reasonably good and completely consistent with our initial tests.     

Progress in protein structure prediction: assessment of CASP3.

The third comparative assessment of techniques of protein structure prediction (CASP3) was held during 1998. This is a blind trial in which structures are predicted prior to having knowledge of the coordinates, which are then revealed to enable the assessment. Three sections at the meeting evaluated different methodologies - comparative modelling, fold recognition and ab initio methods. For some, but not all of the target coordinates, high quality models were submitted in each of these sections. There have been improvements in prediction techniques since CASP2 in 1996, most notably for ab initio methods.     

GalaxyTBM: template-based modeling by building a reliable core and refining unreliable local regions

ABSTRACT: BACKGROUND: Protein structures can be reliably predicted by template-based modeling (TBM) when experimental structures of homologous proteins are available. However, it is challenging to obtain structures more accurate than the single best templates by either combining information from multiple templates or by modeling regions that vary among templates or are not covered by any templates. RESULTS: We introduce GalaxyTBM, a new TBM method in which the more reliable core region is modeled first from multiple templates and less reliable, variable local regions, such as loops or termini, are then detected and re-modeled by an ab initio method. This TBM method is based on "Seok-server," which was tested in CASP9 and assessed to be amongst the top TBM servers. The accuracy of the initial core modeling is enhanced by focusing on more conserved regions in the multiple-template selection and multiple sequence alignment stages. Additional improvement is achieved by ab initio modeling of up to 3 unreliable local regions in the fixed framework of the core structure. Overall, GalaxyTBM reproduced the performance of Seok-server, with GalaxyTBM and Seok-server resulting in average GDT-TS of 68.1 and 68.4, respectively, when tested on 68 single-domain CASP9 TBM targets. For application to multi-domain proteins, GalaxyTBM must be combined with domain-splitting methods. CONCLUSION: Application of GalaxyTBM to CASP9 targets demonstrates that accurate protein structure prediction is possible by use of a multiple-template-based approach, and ab initio modeling of variable regions can further enhance the model quality.     

Long-range information and physicality constraints improve predicted protein contact maps

Protein topology representations such as residue contact maps are an important intermediate step towards ab initio prediction of protein structure, but the problem of predicting reliable contact maps is far from solved. One of the main pitfalls of existing contact map predictors is that they generally predict unphysical maps, i.e. maps that cannot be embedded into three-dimensional structures or, at best, violate a number of basic constraints observed in real protein structures, such as the maximum number of contacts for a residue. Here, we focus on the problem of learning to predict more "physical" contact maps. We do so by first predicting contact maps through a traditional system (XXStout), and then filtering these maps by an ensemble of artificial neural networks. The filter is provided as input not only the bare predicted map, but also a number of global or long-range features extracted from it. In a rigorous cross-validation test, we show that the filter greatly improves the predicted maps it is input. CASP7 results, on which we report here, corroborate this finding. Importantly, since the approach we present here is fully modular, it may be beneficial to any other ab initio contact map predictor.     

Ab initio prediction of the three-dimensional structure of a de novo designed protein: a double-blind case study

Ab initio structure prediction and de novo protein design are two problems at the forefront of research in the fields of structural biology and chemistry. The goal of ab initio structure prediction of proteins is to correctly characterize the 3D structure of a protein using only the amino acid sequence as input. De novo protein design involves the production of novel protein sequences that adopt a desired fold. In this work, the results of a double-blind study are presented in which a new ab initio method was successfully used to predict the 3D structure of a protein designed through an experimental approach using binary patterned combinatorial libraries of de novo sequences. The predicted structure, which was produced before the experimental structure was known and without consideration of the design goals, and the final NMR analysis both characterize this protein as a 4-helix bundle. The similarity of these structures is evidenced by both small RMSD values between the coordinates of the two structures and a detailed analysis of the helical packing.     

Assessment of protein assembly prediction in CASP13

We present the assembly category assessment in the 13th edition of the CASP community-wide experiment. For the second time, protein assemblies constitute an independent assessment category. Compared to the last edition we see a clear uptake in participation, more oligomeric targets released, and consistent, albeit modest, improvement of the predictions quality. Looking at the tertiary structure predictions, we observe that ignoring the oligomeric state of the targets hinders modeling success. We also note that some contact prediction groups successfully predicted homomeric interfacial contacts, though it appears that these predictions were not used for assembly modeling. Homology modeling with sizeable human intervention appears to form the basis of the assembly prediction techniques in this round of CASP. Future developments should see more integrated approaches where subunits are modeled in the context of the assemblies they form.     

Prediction of protein tertiary structure using PROFESY, a novel method based on fragment assembly and conformational space annealing

A novel method for ab initio prediction of protein tertiary structures, PROFESY (PROFile Enumerating SYstem), is proposed. This method utilizes the secondary structure prediction information of a query sequence and the fragment assembly procedure based on global optimization. Fifteen-residue-long fragment libraries are constructed using the secondary structure prediction method PREDICT, and fragments in these libraries are assembled to generate full-length chains of a query protein. Tertiary structures of 50 to 100 conformations are obtained by minimizing an energy function for proteins, using the conformational space annealing method that enables one to sample diverse low-lying local minima of the energy. We apply PROFESY for benchmark tests to proteins with known structures to demonstrate its feasibility. In addition, we participated in CASP5 and applied PROFESY to four new-fold targets for blind prediction. The results are quite promising, despite the fact that PROFESY was in its early stages of development. In particular, PROFESY successfully provided us the best model-one structure for the target T0161.     

Protein contact prediction using patterns of correlation

We describe a new method for using neural networks to predict residue contact pairs in a protein. The main inputs to the neural network are a set of 25 measures of correlated mutation between all pairs of residues in two "windows" of size 5 centered on the residues of interest. While the individual pair-wise correlations are a relatively weak predictor of contact, by training the network on windows of correlation the accuracy of prediction is significantly improved. The neural network is trained on a set of 100 proteins and then tested on a disjoint set of 1033 proteins of known structure. An average predictive accuracy of 21.7% is obtained taking the best L/2 predictions for each protein, where L is the sequence length. Taking the best L/10 predictions gives an average accuracy of 30.7%. The predictor is also tested on a set of 59 proteins from the CASP5 experiment. The accuracy is found to be relatively consistent across different sequence lengths, but to vary widely according to the secondary structure. Predictive accuracy is also found to improve by using multiple sequence alignments containing many sequences to calculate the correlations.     

Fold recognition and ab initio structure predictions using hidden markov models and β-strand pair potentials

Protein structure predictions were submitted for 9 of the target sequences in the competition that ran during 1994. Targets sequences were selected that had no known homology with any sequence of known structure and were members of a reasonably sized family of related but divergent sequences. The objective was either to recognize a compatible fold for the target sequence in the database of known structures or to predict ab initio its rough 3D topology. The main tools used were Hidden Markov models (HMM) for fold recognition, a β- strand pair potential to predict β-sheet topology, and the PHD server for secondary structure prediction. Compatible folds were correctly identified in a number of cases and the β-strand pair potential was shown to be a useful tool for ab initio topology prediction. © 1995 Wiley-Liss, Inc.     

Ab initio protein structure prediction using pathway models

Ab initio prediction is the challenging attempt to predict protein structures based only on sequence information and without using templates. It is often divided into two distinct sub-problems: (a) the scoring function that can distinguish native, or native-like structures, from non-native ones; and (b) the method of searching the conformational space. Currently, there is no reliable scoring function that can always drive a search to the native fold, and there is no general search method that can guarantee a significant sampling of near-natives. Pathway models combine the scoring function and the search. In this short review, we explore some of the ways pathway models are used in folding, in published works since 2001, and present a new pathway model, HMMSTR-CM, that uses a fragment library and a set of nucleation/propagation-based rules. The new method was used for ab initio predictions as part of CASP5. This work was presented at the Winter School in Bioinformatics, Bologna, Italy, 10-14 February 2003.