MetaPSICOV: combining coevolution methods for accurate prediction of contacts and long range hydrogen bonding in proteins

Co-Evolution of Intrinsically Disordered Proteins with Folded Partners Witnessed by Evolutionary Couplings

Although improved strategies for the detection and analysis of evolutionary couplings (ECs) between protein residues already enable the prediction of protein structures and interactions, they are mostly restricted to conserved and well-folded proteins. Whereas intrinsically disordered proteins (IDPs) are central to cellular interaction networks, due to the lack of strict structural constraints, they undergo faster evolutionary changes than folded domains. This makes the reliable identification and alignment of IDP homologs difficult, which led to IDPs being omitted in most large-scale residue co-variation analyses. By preforming a dedicated analysis of phylogenetically widespread bacterial IDP–partner interactions, here we demonstrate that partner binding imposes constraints on IDP sequences that manifest in detectable interprotein ECs. These ECs were not detected for interactions mediated by short motifs, rather for those with larger IDP–partner interfaces. Most identified coupled residue pairs reside close (<10 Å) to each other on the interface, with a third of them forming multiple direct atomic contacts. EC-carrying interfaces of IDPs are enriched in negatively charged residues, and the EC residues of both IDPs and partners preferentially reside in helices. Our analysis brings hope that IDP–partner interactions difficult to study could soon be successfully dissected through residue co-variation analysis.

High precision in protein contact prediction using fully convolutional neural networks and minimal sequence features

Motivation

In addition to substitution frequency data from protein sequence alignments, many state-of-the-art methods for contact prediction rely on additional sources of information, or features, of protein sequences in order to predict residue-residue contacts, such as solvent accessibility, predicted secondary structure, and scores from other contact prediction methods. It is unclear how much of this information is needed to achieve state-of-the-art results. Here, we show that using deep neural network models, simple alignment statistics contain sufficient information to achieve state-of-the-art precision. Our prediction method, DeepCov, uses fully convolutional neural networks operating on amino-acid pair frequency or covariance data derived directly from sequence alignments, without using global statistical methods such as sparse inverse covariance or pseudolikelihood estimation.

Results

Comparisons against CCMpred and MetaPSICOV2 show that using pairwise covariance data calculated from raw alignments as input allows us to match or exceed the performance of both of these methods. Almost all of the achieved precision is obtained when considering relatively local windows (around 15 residues) around any member of a given residue pairing; larger window sizes have comparable performance. Assessment on a set of shallow sequence alignments (fewer than 160 effective sequences) indicates that the new method is substantially more precise than CCMpred and MetaPSICOV2 in this regime, suggesting that improved precision is attainable on smaller sequence families. Overall, the performance of DeepCov is competitive with the state of the art, and our results demonstrate that global models, which employ features from all parts of the input alignment when predicting individual contacts, are not strictly needed in order to attain precise contact predictions.

Availability and implementation

DeepCov is freely available at https://github.com/psipred/DeepCov.

Supplementary information

Supplementary data are available at Bioinformatics online.

Structural biology meets data science: does anything change?

Data science has emerged from the proliferation of digital data, coupled with advances in algorithms, software and hardware (e.g., GPU computing). Innovations in structural biology have been driven by similar factors, spurring us to ask: can these two fields impact one another in deep and hitherto unforeseen ways? We posit that the answer is yes. New biological knowledge lies in the relationships between sequence, structure, function and disease, all of which play out on the stage of evolution, and data science enables us to elucidate these relationships at scale. Here, we consider the above question from the five key pillars of data science: acquisition, engineering, analytics, visualization and policy, with an emphasis on machine learning as the premier analytics approach.

Deep generative models of genetic variation capture the effects of mutations

The functions of proteins and RNAs are defined by the collective interactions of many residues, and yet most statistical models of biological sequences consider sites nearly independently. Recent approaches have demonstrated benefits of including interactions to capture pairwise covariation, but leave higher-order dependencies out of reach. Here we show how it is possible to capture higher-order, context-dependent constraints in biological sequences via latent variable models with nonlinear dependencies. We found that DeepSequence ( https://github.com/debbiemarkslab/DeepSequence ), a probabilistic model for sequence families, predicted the effects of mutations across a variety of deep mutational scanning experiments substantially better than existing methods based on the same evolutionary data. The model, learned in an unsupervised manner solely on the basis of sequence information, is grounded with biologically motivated priors, reveals the latent organization of sequence families, and can be used to explore new parts of sequence space.

Predicting improved protein conformations with a temporal deep recurrent neural network

Accurate protein structure prediction from amino acid sequence is still an unsolved problem. The most reliable methods centre on template based modelling. However, the accuracy of these models entirely depends on the availability of experimentally resolved homologous template structures. In order to generate more accurate models, extensive physics based molecular dynamics (MD) refinement simulations are performed to sample many different conformations to find improved conformational states. In this study, we propose a deep recurrent network model, called DeepTrajectory, that is able to identify these improved conformational states, with high precision, from a variety of different MD based sampling protocols. The proposed model learns the temporal patterns of features computed from MD trajectory data in order to classify whether each recorded simulation snapshot is an improved quality conformational state, decreased quality conformational state or whether there is no perceivable change in state with respect to the starting conformation. The model was trained and tested on 904 trajectories from 42 different protein systems with a cumulative number of more than 1.7 million snapshots. We show that our model outperforms other state of the art machine-learning algorithms that do not consider temporal dependencies. To our knowledge, DeepTrajectory is the first implementation of a time-dependent deep-learning protocol that is re-trainable and able to adapt to any new MD based sampling procedure, thereby demonstrating how a neural network can be used to learn the latter part of the protein folding funnel.

Enhancing coevolution-based contact prediction by imposing structural self-consistency of the contacts

Based on the development of new algorithms and growth of sequence databases, it has recently become possible to build robust higher-order sequence models based on sets of aligned protein sequences. Such models have proven useful in de novo structure prediction, where the sequence models are used to find pairs of residues that co-vary during evolution, and hence are likely to be in spatial proximity in the native protein. The accuracy of these algorithms, however, drop dramatically when the number of sequences in the alignment is small. We have developed a method that we termed CE-YAPP (CoEvolution-YAPP), that is based on YAPP (Yet Another Peak Processor), which has been shown to solve a similar problem in NMR spectroscopy. By simultaneously performing structure prediction and contact assignment, CE-YAPP uses structural self-consistency as a filter to remove false positive contacts. Furthermore, CE-YAPP solves another problem, namely how many contacts to choose from the ordered list of covarying amino acid pairs. We show that CE-YAPP consistently improves contact prediction from multiple sequence alignments, in particular for proteins that are difficult targets. We further show that the structures determined from CE-YAPP are also in better agreement with those determined using traditional methods in structural biology.

Combined approaches from physics, statistics, and computer science for ab initio protein structure prediction: ex unitate vires (unity is strength)?

Connecting the dots among the amino acid sequence of a protein, its structure, and its function remains a central theme in molecular biology, as it would have many applications in the treatment of illnesses related to misfolding or protein instability. As a result of high-throughput sequencing methods, biologists currently live in a protein sequence-rich world. However, our knowledge of protein structure based on experimental data remains comparatively limited. As a consequence, protein structure prediction has established itself as a very active field of research to fill in this gap. This field, once thought to be reserved for theoretical biophysicists, is constantly reinventing itself, borrowing ideas informed by an ever-increasing assembly of scientific domains, from biology, chemistry, (statistical) physics, mathematics, computer science, statistics, bioinformatics, and more recently data sciences. We review the recent progress arising from this integration of knowledge, from the development of specific computer architecture to allow for longer timescales in physics-based simulations of protein folding to the recent advances in predicting contacts in proteins based on detection of coevolution using very large data sets of aligned protein sequences.

Comparing co-evolution methods and their application to template-free protein structure prediction

Motivation

Co-evolution methods have been used as contact predictors to identify pairs of residues that share spatial proximity. Such contact predictors have been compared in terms of the precision of their predictions, but there is no study that compares their usefulness to model generation.

Results

We compared eight different co-evolution methods for a set of ∼3500 proteins and found that metaPSICOV stage 2 produces, on average, the most precise predictions. Precision of all the methods is dependent on SCOP class, with most methods predicting contacts in all α and membrane proteins poorly. The contact predictions were then used to assist in de novo model generation. We found that it was not the method with the highest average precision, but rather metaPSICOV stage 1 predictions that consistently led to the best models being produced. Our modelling results show a correlation between the proportion of predicted long range contacts that are satisfied on a model and its quality. We used this proportion to effectively classify models as correct/incorrect; discarding decoys classified as incorrect led to an enrichment in the proportion of good decoys in our final ensemble by a factor of seven. For 17 out of the 18 cases where correct answers were generated, the best models were not discarded by this approach. We were also able to identify eight cases where no correct decoy had been generated.

Availability and Implementation

Data is available for download from: http://opig.stats.ox.ac.uk/resources.

Contact

saulo.deoliveira@dtc.ox.ac.uk

Supplimentary Information

Supplementary data are available at Bioinformatics online.

Contact prediction is hardest for the most informative contacts, but improves with the incorporation of contact potentials

Co-evolution between pairs of residues in a multiple sequence alignment (MSA) of homologous proteins has long been proposed as an indicator of structural contacts. Recently, several methods, such as direct-coupling analysis (DCA) and MetaPSICOV, have been shown to achieve impressive rates of contact prediction by taking advantage of considerable sequence data. In this paper, we show that prediction success rates are highly sensitive to the structural definition of a contact, with more permissive definitions (i.e., those classifying more pairs as true contacts) naturally leading to higher positive predictive rates, but at the expense of the amount of structural information contributed by each contact. Thus, the remaining limitations of contact prediction algorithms are most noticeable in conjunction with geometrically restrictive contacts—precisely those that contribute more information in structure prediction. We suggest that to improve prediction rates for such “informative” contacts one could combine co-evolution scores with additional indicators of contact likelihood. Specifically, we find that when a pair of co-varying positions in an MSA is occupied by residue pairs with favorable statistical contact energies, that pair is more likely to represent a true contact. We show that combining a contact potential metric with DCA or MetaPSICOV performs considerably better than DCA or MetaPSICOV alone, respectively. This is true regardless of contact definition, but especially true for stricter and more informative contact definitions. In summary, this work outlines some remaining challenges to be addressed in contact prediction and proposes and validates a promising direction towards improvement.

Inter-residue, inter-protein and inter-family coevolution: bridging the scales

Interacting proteins coevolve at multiple but interconnected scales, from the residue-residue over the protein-protein up to the family-family level. The recent accumulation of enormous amounts of sequence data allows for the development of novel, data-driven computational approaches. Notably, these approaches can bridge scales within a single statistical framework. Although being currently applied mostly to isolated problems on single scales, their immense potential for an evolutionary informed, structural systems biology is steadily emerging.

ConKit: a python interface to contact predictions

Summary

Recent advances in protein residue contact prediction algorithms have led to the emergence of many new methods and a variety of file formats. We present ConKit, an open source, modular and extensible Python interface which allows facile conversion between formats and provides an interface to analyses of sequence alignments and sets of contact predictions.

Availability and Implementation

ConKit is available via the Python Package Index. The documentation can be found at http://www.conkit.org. ConKit is licensed under the BSD 3-Clause.

Supplementary information

Supplementary data are available at Bioinformatics online.

Identification of residue pairing in interacting β-strands from a predicted residue contact map

Background

Despite the rapid progress of protein residue contact prediction, predicted residue contact maps frequently contain many errors. However, information of residue pairing in β strands could be extracted from a noisy contact map, due to the presence of characteristic contact patterns in β-β interactions. This information may benefit the tertiary structure prediction of mainly β proteins. In this work, we propose a novel ridge-detection-based β-β contact predictor to identify residue pairing in β strands from any predicted residue contact map.

Results

Our algorithm RDb₂C adopts ridge detection, a well-developed technique in computer image processing, to capture consecutive residue contacts, and then utilizes a novel multi-stage random forest framework to integrate the ridge information and additional features for prediction. Starting from the predicted contact map of CCMpred, RDb₂C remarkably outperforms all state-of-the-art methods on two conventional test sets of β proteins (BetaSheet916 and BetaSheet1452), and achieves F1-scores of ~ 62% and ~ 76% at the residue level and strand level, respectively. Taking the prediction of the more advanced RaptorX-Contact as input, RDb₂C achieves impressively higher performance, with F1-scores reaching ~ 76% and ~ 86% at the residue level and strand level, respectively. In a test of structural modeling using the top 1 L predicted contacts as constraints, for 61 mainly β proteins, the average TM-score achieves 0.442 when using the raw RaptorX-Contact prediction, but increases to 0.506 when using the improved prediction by RDb₂C.

Conclusion

Our method can significantly improve the prediction of β-β contacts from any predicted residue contact maps. Prediction results of our algorithm could be directly applied to effectively facilitate the practical structure prediction of mainly β proteins.

Availability

All source data and codes are available at http://166.111.152.91/Downloads.html or the GitHub address of https://github.com/wzmao/RDb2C.

Electronic supplementary material

The online version of this article (10.1186/s12859-018-2150-1) contains supplementary material, which is available to authorized users.

Identifying functionally informative evolutionary sequence profiles

Motivation

Multiple sequence alignments (MSAs) can provide essential input to many bioinformatics applications, including protein structure prediction and functional annotation. However, the optimal selection of sequences to obtain biologically informative MSAs for such purposes is poorly explored, and has traditionally been performed manually.

Results

We present Selection of Alignment by Maximal Mutual Information (SAMMI), an automated, sequence-based approach to objectively select an optimal MSA from a large set of alternatives sampled from a general sequence database search. The hypothesis of this approach is that the mutual information among MSA columns will be maximal for those MSAs that contain the most diverse set possible of the most structurally and functionally homogeneous protein sequences. SAMMI was tested to select MSAs for functional site residue prediction by analysis of conservation patterns on a set of 435 proteins obtained from protein-ligand (peptides, nucleic acids and small substrates) and protein-protein interaction databases. Availability and implementation: A freely accessible program, including source code, implementing SAMMI is available at https://github.com/nelsongil92/SAMMI.git.

Contact

andras.fiser@einstein.yu.edu.

Supplementary information

Supplementary data are available at Bioinformatics online.

Large-scale structure prediction by improved contact predictions and model quality assessment

Motivation

Accurate contact predictions can be used for predicting the structure of proteins. Until recently these methods were limited to very big protein families, decreasing their utility. However, recent progress by combining direct coupling analysis with machine learning methods has made it possible to predict accurate contact maps for smaller families. To what extent these predictions can be used to produce accurate models of the families is not known.

Results

We present the PconsFold2 pipeline that uses contact predictions from PconsC3, the CONFOLD folding algorithm and model quality estimations to predict the structure of a protein. We show that the model quality estimation significantly increases the number of models that reliably can be identified. Finally, we apply PconsFold2 to 6379 Pfam families of unknown structure and find that PconsFold2 can, with an estimated 90% specificity, predict the structure of up to 558 Pfam families of unknown structure. Out of these, 415 have not been reported before.

Availability and Implementation

Datasets as well as models of all the 558 Pfam families are available at http://c3.pcons.net/. All programs used here are freely available.

Sequential search leads to faster, more efficient fragment-based de novo protein structure prediction

Motivation

Most current de novo structure prediction methods randomly sample protein conformations and thus require large amounts of computational resource. Here, we consider a sequential sampling strategy, building on ideas from recent experimental work which shows that many proteins fold cotranslationally.

Results

We have investigated whether a pseudo-greedy search approach, which begins sequentially from one of the termini, can improve the performance and accuracy of de novo protein structure prediction. We observed that our sequential approach converges when fewer than 20 000 decoys have been produced, fewer than commonly expected. Using our software, SAINT2, we also compared the run time and quality of models produced in a sequential fashion against a standard, non-sequential approach. Sequential prediction produces an individual decoy 1.5-2.5 times faster than non-sequential prediction. When considering the quality of the best model, sequential prediction led to a better model being produced for 31 out of 41 soluble protein validation cases and for 18 out of 24 transmembrane protein cases. Correct models (TM-Score > 0.5) were produced for 29 of these cases by the sequential mode and for only 22 by the non-sequential mode. Our comparison reveals that a sequential search strategy can be used to drastically reduce computational time of de novo protein structure prediction and improve accuracy.

Availability and implementation

Data are available for download from: http://opig.stats.ox.ac.uk/resources. SAINT2 is available for download from: https://github.com/sauloho/SAINT2.

Contact

saulo.deoliveira@dtc.ox.ac.uk.

Supplementary information

Supplementary data are available at Bioinformatics online.

Opportunities and obstacles for deep learning in biology and medicine

Deep learning describes a class of machine learning algorithms that are capable of combining raw inputs into layers of intermediate features. These algorithms have recently shown impressive results across a variety of domains. Biology and medicine are data-rich disciplines, but the data are complex and often ill-understood. Hence, deep learning techniques may be particularly well suited to solve problems of these fields. We examine applications of deep learning to a variety of biomedical problems-patient classification, fundamental biological processes and treatment of patients-and discuss whether deep learning will be able to transform these tasks or if the biomedical sphere poses unique challenges. Following from an extensive literature review, we find that deep learning has yet to revolutionize biomedicine or definitively resolve any of the most pressing challenges in the field, but promising advances have been made on the prior state of the art. Even though improvements over previous baselines have been modest in general, the recent progress indicates that deep learning methods will provide valuable means for speeding up or aiding human investigation. Though progress has been made linking a specific neural network's prediction to input features, understanding how users should interpret these models to make testable hypotheses about the system under study remains an open challenge. Furthermore, the limited amount of labelled data for training presents problems in some domains, as do legal and privacy constraints on work with sensitive health records. Nonetheless, we foresee deep learning enabling changes at both bench and bedside with the potential to transform several areas of biology and medicine.

Applications of sequence coevolution in membrane protein biochemistry

Recently, protein sequence coevolution analysis has matured into a predictive powerhouse for protein structure and function. Direct methods, which use global statistical models of sequence coevolution, have enabled the prediction of membrane and disordered protein structures, protein complex architectures, and the functional effects of mutations in proteins. The field of membrane protein biochemistry and structural biology has embraced these computational techniques, which provide functional and structural information in an otherwise experimentally-challenging field. Here we review recent applications of protein sequence coevolution analysis to membrane protein structure and function and highlight the promising directions and future obstacles in these fields. We provide insights and guidelines for membrane protein biochemists who wish to apply sequence coevolution analysis to a given experimental system.

NeBcon: protein contact map prediction using neural network training coupled with naïve Bayes classifiers

Motivation

Recent CASP experiments have witnessed exciting progress on folding large-size non-humongous proteins with the assistance of co-evolution based contact predictions. The success is however anecdotal due to the requirement of the contact prediction methods for the high volume of sequence homologs that are not available to most of the non-humongous protein targets. Development of efficient methods that can generate balanced and reliable contact maps for different type of protein targets is essential to enhance the success rate of the ab initio protein structure prediction.

Results

We developed a new pipeline, NeBcon, which uses the naïve Bayes classifier (NBC) theorem to combine eight state of the art contact methods that are built from co-evolution and machine learning approaches. The posterior probabilities of the NBC model are then trained with intrinsic structural features through neural network learning for the final contact map prediction. NeBcon was tested on 98 non-redundant proteins, which improves the accuracy of the best co-evolution based meta-server predictor by 22%; the magnitude of the improvement increases to 45% for the hard targets that lack sequence and structural homologs in the databases. Detailed data analysis showed that the major contribution to the improvement is due to the optimized NBC combination of the complementary information from both co-evolution and machine learning predictions. The neural network training also helps to improve the coupling of the NBC posterior probability and the intrinsic structural features, which were found particularly important for the proteins that do not have sufficient number of homologous sequences to derive reliable co-evolution profiles.

Availiablity and Implementation

On-line server and standalone package of the program are available at http://zhanglab.ccmb.med.umich.edu/NeBcon/ .

Contact

zhng@umich.edu.

Supplementary information

Supplementary data are available at Bioinformatics online.

Inverse statistical physics of protein sequences: a key issues review

In the course of evolution, proteins undergo important changes in their amino acid sequences, while their three-dimensional folded structure and their biological function remain remarkably conserved. Thanks to modern sequencing techniques, sequence data accumulate at unprecedented pace. This provides large sets of so-called homologous, i.e. evolutionarily related protein sequences, to which methods of inverse statistical physics can be applied. Using sequence data as the basis for the inference of Boltzmann distributions from samples of microscopic configurations or observables, it is possible to extract information about evolutionary constraints and thus protein function and structure. Here we give an overview over some biologically important questions, and how statistical-mechanics inspired modeling approaches can help to answer them. Finally, we discuss some open questions, which we expect to be addressed over the next years.

Assessment of contact predictions in CASP12: Co-evolution and deep learning coming of age

Following up on the encouraging results of residue-residue contact prediction in the CASP11 experiment, we present the analysis of predictions submitted for CASP12. The submissions include predictions of 34 groups for 38 domains classified as free modeling targets which are not accessible to homology-based modeling due to a lack of structural templates. CASP11 saw a rise of coevolution-based methods outperforming other approaches. The improvement of these methods coupled to machine learning and sequence database growth are most likely the main driver for a significant improvement in average precision from 27% in CASP11 to 47% in CASP12. In more than half of the targets, especially those with many homologous sequences accessible, precisions above 90% were achieved with the best predictors reaching a precision of 100% in some cases. We furthermore tested the impact of using these contacts as restraints in ab initio modeling of 14 single-domain free modeling targets using Rosetta. Adding contacts to the Rosetta calculations resulted in improvements of up to 26% in GDT_TS within the top five structures.

[From sequence variability to structural and functional prediction: modeling of homologous protein families]

Thanks to next-generation sequencing, the number of sequenced genomes grows rapidly, providing in particular ample examples for the sequence variability between homologous proteins. This article discusses data-driven probabilistic sequence models, which are able to extract a multitude of information from sequence data alone, including (i) structural features like residue-residue contacts, which are formed in the folded protein, (ii) protein-protein interaction interfaces and (iii) phenotypic effects of amino-acid substitutions in proteins.

Finding the needle in the haystack: towards solving the protein-folding problem computationally

Prediction of protein tertiary structures from amino acid sequence and understanding the mechanisms of how proteins fold, collectively known as "the protein folding problem," has been a grand challenge in molecular biology for over half a century. Theories have been developed that provide us with an unprecedented understanding of protein folding mechanisms. However, computational simulation of protein folding is still difficult, and prediction of protein tertiary structure from amino acid sequence is an unsolved problem. Progress toward a satisfying solution has been slow due to challenges in sampling the vast conformational space and deriving sufficiently accurate energy functions. Nevertheless, several techniques and algorithms have been adopted to overcome these challenges, and the last two decades have seen exciting advances in enhanced sampling algorithms, computational power and tertiary structure prediction methodologies. This review aims at summarizing these computational techniques, specifically conformational sampling algorithms and energy approximations that have been frequently used to study protein-folding mechanisms or to de novo predict protein tertiary structures. We hope that this review can serve as an overview on how the protein-folding problem can be studied computationally and, in cases where experimental approaches are prohibitive, help the researcher choose the most relevant computational approach for the problem at hand. We conclude with a summary of current challenges faced and an outlook on potential future directions.

Data-assisted protein structure modeling by global optimization in CASP12

In CASP12, 2 types of data-assisted protein structure modeling were experimented. Either SAXS experimental data or cross-linking experimental data was provided for a selected number of CASP12 targets that the CASP12 predictor could utilize for better protein structure modeling. We devised 2 separate energy terms for SAXS data and cross-linking data to drive the model structures into more native-like structures that satisfied the given experimental data as much as possible. In CASP11, we successfully performed protein structure modeling using simulated sparse and ambiguously assigned NOE data and/or correct residue-residue contact information, where the only energy term that folded the protein into its native structure was the term which was originated from the given experimental data. However, the 2 types of experimental data provided in CASP12 were far from being sufficient enough to fold the target protein into its native structure because SAXS data provides only the overall shape of the molecule and the cross-linking contact information provides only very low-resolution distance information. For this reason, we combined the SAXS or cross-linking energy term with our regular modeling energy function that includes both the template energy term and the de novo energy terms. By optimizing the newly formulated energy function, we obtained protein models that fit better with provided SAXS data than the X-ray structure of the target. However, the improvement of the model relative to the 1 modeled without the SAXS data, was not significant. Consistent structural improvement was achieved by incorporating cross-linking data into the protein structure modeling.

CONFOLD2: improved contact-driven ab initio protein structure modeling

Background

Contact-guided protein structure prediction methods are becoming more and more successful because of the latest advances in residue-residue contact prediction. To support contact-driven structure prediction, effective tools that can quickly build tertiary structural models of good quality from predicted contacts need to be developed.

Results

We develop an improved contact-driven protein modelling method, CONFOLD2, and study how it may be effectively used for ab initio protein structure prediction with predicted contacts as input. It builds models using various subsets of input contacts to explore the fold space under the guidance of a soft square energy function, and then clusters the models to obtain the top five models. CONFOLD2 obtains an average reconstruction accuracy of 0.57 TM-score for the 150 proteins in the PSICOV contact prediction dataset. When benchmarked on the CASP11 contacts predicted using CONSIP2 and CASP12 contacts predicted using Raptor-X, CONFOLD2 achieves a mean TM-score of 0.41 on both datasets.

Conclusion

CONFOLD2 allows to quickly generate top five structural models for a protein sequence when its secondary structures and contacts predictions at hand. The source code of CONFOLD2 is publicly available at https://github.com/multicom-toolbox/CONFOLD2/.

Electronic supplementary material

The online version of this article (10.1186/s12859-018-2032-6) contains supplementary material, which is available to authorized users.

Enhancing Evolutionary Couplings with Deep Convolutional Neural Networks

While genes are defined by sequence, in biological systems a protein's function is largely determined by its three-dimensional structure. Evolutionary information embedded within multiple sequence alignments provides a rich source of data for inferring structural constraints on macromolecules. Still, many proteins of interest lack sufficient numbers of related sequences, leading to noisy, error-prone residue-residue contact predictions. Here we introduce DeepContact, a convolutional neural network (CNN)-based approach that discovers co-evolutionary motifs and leverages these patterns to enable accurate inference of contact probabilities, particularly when few related sequences are available. DeepContact significantly improves performance over previous methods, including in the CASP12 blind contact prediction task where we achieved top performance with another CNN-based approach. Moreover, our tool converts hard-to-interpret coupling scores into probabilities, moving the field toward a consistent metric to assess contact prediction across diverse proteins. Through substantially improving the precision-recall behavior of contact prediction, DeepContact suggests we are near a paradigm shift in template-free modeling for protein structure prediction.

Protein structure modeling and refinement by global optimization in CASP12

For protein structure modeling in the CASP12 experiment, we have developed a new protocol based on our previous CASP11 approach. The global optimization method of conformational space annealing (CSA) was applied to 3 stages of modeling: multiple sequence-structure alignment, three-dimensional (3D) chain building, and side-chain re-modeling. For better template selection and model selection, we updated our model quality assessment (QA) method with the newly developed SVMQA (support vector machine for quality assessment). For 3D chain building, we updated our energy function by including restraints generated from predicted residue-residue contacts. New energy terms for the predicted secondary structure and predicted solvent accessible surface area were also introduced. For difficult targets, we proposed a new method, LEEab, where the template term played a less significant role than it did in LEE, complemented by increased contributions from other terms such as the predicted contact term. For TBM (template-based modeling) targets, LEE performed better than LEEab, but for FM targets, LEEab was better. For model refinement, we modified our CASP11 molecular dynamics (MD) based protocol by using explicit solvents and tuning down restraint weights. Refinement results from MD simulations that used a new augmented statistical energy term in the force field were quite promising. Finally, when using inaccurate information (such as the predicted contacts), it was important to use the Lorentzian function for which the maximal penalty arising from wrong information is always bounded.

Approaches to ab initio molecular replacement of α-helical transmembrane proteins

α-Helical transmembrane proteins are a ubiquitous and important class of proteins, but present difficulties for crystallographic structure solution. Here, the effectiveness of the AMPLE molecular replacement pipeline in solving α-helical transmembrane-protein structures is assessed using a small library of eight ideal helices, as well as search models derived from ab initio models generated both with and without evolutionary contact information. The ideal helices prove to be surprisingly effective at solving higher resolution structures, but ab initio-derived search models are able to solve structures that could not be solved with the ideal helices. The addition of evolutionary contact information results in a marked improvement in the modelling and makes additional solutions possible.

The challenge of modeling protein assemblies: the CASP12-CAPRI experiment

We present the quality assessment of 5613 models submitted by predictor groups from both CAPRI and CASP for the total of 15 most tractable targets from the second joint CASP-CAPRI protein assembly prediction experiment. These targets comprised 12 homo-oligomers and 3 hetero-complexes. The bulk of the analysis focuses on 10 targets (of CAPRI Round 37), which included all 3 hetero-complexes, and whose protein chains or the full assembly could be readily modeled from structural templates in the PDB. On average, 28 CAPRI groups and 10 CASP groups (including automatic servers), submitted models for each of these 10 targets. Additionally, about 16 groups participated in the CAPRI scoring experiments. A range of acceptable to high quality models were obtained for 6 of the 10 Round 37 targets, for which templates were available for the full assembly. Poorer results were achieved for the remaining targets due to the lower quality of the templates available for the full complex or the individual protein chains, highlighting the unmet challenge of modeling the structural adjustments of the protein components that occur upon binding or which must be accounted for in template-based modeling. On the other hand, our analysis indicated that residues in binding interfaces were correctly predicted in a sizable fraction of otherwise poorly modeled assemblies and this with higher accuracy than published methods that do not use information on the binding partner. Lastly, the strengths and weaknesses of the assessment methods are evaluated and improvements suggested.

Template-based and free modeling of I-TASSER and QUARK pipelines using predicted contact maps in CASP12

We develop two complementary pipelines, "Zhang-Server" and "QUARK", based on I-TASSER and QUARK pipelines for template-based modeling (TBM) and free modeling (FM), and test them in the CASP12 experiment. The combination of I-TASSER and QUARK successfully folds three medium-size FM targets that have more than 150 residues, even though the interplay between the two pipelines still awaits further optimization. Newly developed sequence-based contact prediction by NeBcon plays a critical role to enhance the quality of models, particularly for FM targets, by the new pipelines. The inclusion of NeBcon predicted contacts as restraints in the QUARK simulations results in an average TM-score of 0.41 for the best in top five predicted models, which is 37% higher than that by the QUARK simulations without contacts. In particular, there are seven targets that are converted from non-foldable to foldable (TM-score >0.5) due to the use of contact restraints in the simulations. Another additional feature in the current pipelines is the local structure quality prediction by ResQ, which provides a robust residue-level modeling error estimation. Despite the success, significant challenges still remain in ab initio modeling of multi-domain proteins and folding of β-proteins with complicated topologies bound by long-range strand-strand interactions. Improvements on domain boundary and long-range contact prediction, as well as optimal use of the predicted contacts and multiple threading alignments, are critical to address these issues seen in the CASP12 experiment.

Three-body interactions improve contact prediction within direct-coupling analysis

The prediction of residue contacts in a protein solely from sequence information is a promising approach to computational structure prediction. Recent developments use statistical or information theoretic methods to extract contact information from a multiple sequence alignment. Despite good results, accuracy is limited due to usage of two-body interactions within a Potts model. In this paper we generalize this approach and propose a Hamiltonian with an additional three-body interaction term. We derive a mean-field approximation for inference of three-body couplings within a Potts model which is fast enough on modern computers. Finally, we show that our model has a higher accuracy in predicting residue contacts in comparison with the plain two-body-interaction model.

Forecasting residue-residue contact prediction accuracy

Motivation

Apart from meta-predictors, most of today's methods for residue-residue contact prediction are based entirely on Direct Coupling Analysis (DCA) of correlated mutations in multiple sequence alignments (MSAs). These methods are on average ∼40% correct for the 100 strongest predicted contacts in each protein. The end-user who works on a single protein of interest will not know if predictions are either much more or much less correct than 40%, which is especially a problem if contacts are predicted to steer experimental research on that protein.

Results

We designed a regression model that forecasts the accuracy of residue-residue contact prediction for individual proteins with an average error of 7 percentage points. Contacts were predicted with two DCA methods (gplmDCA and PSICOV). The models were built on parameters that describe the MSA, the predicted secondary structure, the predicted solvent accessibility and the contact prediction scores for the target protein. Results show that our models can be also applied to the meta-methods, which was tested on RaptorX.

Availability and implementation

All data and scripts are available from http://comprec-lin.iiar.pwr.edu.pl/dcaQ/.

Contact

malgorzata.kotulska@pwr.edu.pl.

Supplementary information

Supplementary data are available at Bioinformatics online.

Protein contact prediction by integrating deep multiple sequence alignments, coevolution and machine learning

In this study, we report the evaluation of the residue-residue contacts predicted by our three different methods in the CASP12 experiment, focusing on studying the impact of multiple sequence alignment, residue coevolution, and machine learning on contact prediction. The first method (MULTICOM-NOVEL) uses only traditional features (sequence profile, secondary structure, and solvent accessibility) with deep learning to predict contacts and serves as a baseline. The second method (MULTICOM-CONSTRUCT) uses our new alignment algorithm to generate deep multiple sequence alignment to derive coevolution-based features, which are integrated by a neural network method to predict contacts. The third method (MULTICOM-CLUSTER) is a consensus combination of the predictions of the first two methods. We evaluated our methods on 94 CASP12 domains. On a subset of 38 free-modeling domains, our methods achieved an average precision of up to 41.7% for top L/5 long-range contact predictions. The comparison of the three methods shows that the quality and effective depth of multiple sequence alignments, coevolution-based features, and machine learning integration of coevolution-based features and traditional features drive the quality of predicted protein contacts. On the full CASP12 dataset, the coevolution-based features alone can improve the average precision from 28.4% to 41.6%, and the machine learning integration of all the features further raises the precision to 56.3%, when top L/5 predicted long-range contacts are evaluated. And the correlation between the precision of contact prediction and the logarithm of the number of effective sequences in alignments is 0.66.

Simultaneous refinement of inaccurate local regions and overall structure in the CASP12 protein model refinement experiment

Advances in protein model refinement techniques are required as diverse sources of protein structure information are available from low-resolution experiments or informatics-based computations such as cryo-EM, NMR, homology models, or predicted residue contacts. Given semi-reliable or incomplete structural information, structure quality of a protein model has to be improved by ab initio methods such as energy-based simulation. In this study, we describe a new automatic refinement server method designed to improve locally inaccurate regions and overall structure simultaneously. Locally inaccurate regions may occur in protein structures due to non-convergent or missing information in template structures used in homology modeling or due to intrinsic structural flexibilities not resolved by experimental techniques. However, such variable or dynamic regions often play important functional roles by participating in interactions with other biomolecules or in transitions between different functional states. The new refinement method introduced here utilizes diverse types of geometric operators which drive both local and global changes, and the effect of structure changes and relaxations are accumulated. This resulted in consistent refinement of both local and global structural features. Performance of this method in CASP12 is discussed.

Protein-protein interactions leave evolutionary footprints: High molecular coevolution at the core of interfaces

Protein-protein interactions are essential to all aspects of life. Specific interactions result from evolutionary pressure at the interacting interfaces of partner proteins. However, evolutionary pressure is not homogeneous within the interface: for instance, each residue does not contribute equally to the binding energy of the complex. To understand functional differences between residues within the interface, we analyzed their properties in the core and rim regions. Here, we characterized protein interfaces with two evolutionary measures, conservation and coevolution, using a comprehensive dataset of 896 protein complexes. These scores can detect different selection pressures at a given position in a multiple sequence alignment. We also analyzed how the number of interactions in which a residue is involved influences those evolutionary signals. We found that the coevolutionary signal is higher in the interface core than in the interface rim region. Additionally, the difference in coevolution between core and rim regions is comparable to the known difference in conservation between those regions. Considering proteins with multiple interactions, we found that conservation and coevolution increase with the number of different interfaces in which a residue is involved, suggesting that more constraints (i.e., a residue that must satisfy a greater number of interactions) allow fewer sequence changes at those positions, resulting in higher conservation and coevolution values. These findings shed light on the evolution of protein interfaces and provide information useful for identifying protein interfaces and predicting protein-protein interactions.

Biomolecular coevolution and its applications: Going from structure prediction toward signaling, epistasis, and function

Evolution leads to considerable changes in the sequence of biomolecules, while their overall structure and function remain quite conserved. The wealth of genomic sequences, the ‘Biological Big Data’, modern sequencing techniques provide allows us to investigate biomolecular evolution with unprecedented detail. Sophisticated statistical models can infer residue pair mutations resulting from spatial proximity. The introduction of predicted spatial adjacencies as constraints in biomolecular structure prediction workflows has transformed the field of protein and RNA structure prediction toward accuracies approaching the experimental resolution limit. Going beyond structure prediction, the same mathematical framework allows mimicking evolutionary fitness landscapes to infer signaling interactions, epistasis, or mutational landscapes.

Improved protein contact predictions with the MetaPSICOV2 server in CASP12

In this paper, we present the results for the MetaPSICOV2 contact prediction server in the CASP12 community experiment (http://predictioncenter.org). Over the 35 assessed Free Modelling target domains the MetaPSICOV2 server achieved a mean precision of 43.27%, a substantial increase relative to the server's performance in the CASP11 experiment. In the following paper, we discuss improvements to the MetaPSICOV2 server, covering both changes to the neural network and attempts to integrate contact predictions on a domain basis into the prediction pipeline. We also discuss some limitations in the CASP12 assessment which may have overestimated the performance of our method.

Folding Membrane Proteins by Deep Transfer Learning

Computational elucidation of membrane protein (MP) structures is challenging partially due to lack of sufficient solved structures for homology modeling. Here, we describe a high-throughput deep transfer learning method that first predicts MP contacts by learning from non-MPs and then predicts 3D structure models using the predicted contacts as distance restraints. Tested on 510 non-redundant MPs, our method has contact prediction accuracy at least 0.18 better than existing methods, predicts correct folds for 218 MPs, and generates 3D models with root-mean-square deviation (RMSD) less than 4 and 5 Å for 57 and 108 MPs, respectively. A rigorous blind test in the continuous automated model evaluation project shows that our method predicted high-resolution 3D models for two recent test MPs of 210 residues with RMSD ∼2 Å. We estimated that our method could predict correct folds for 1,345-1,871 reviewed human multi-pass MPs including a few hundred new folds, which shall facilitate the discovery of drugs targeting at MPs.

Analysis of deep learning methods for blind protein contact prediction in CASP12

Here we present the results of protein contact prediction achieved in CASP12 by our RaptorX-Contact server, which is an early implementation of our deep learning method for contact prediction. On a set of 38 free-modeling target domains with a median family size of around 58 effective sequences, our server obtained an average top L/5 long- and medium-range contact accuracy of 47% and 44%, respectively (L = length). A complete implementation has an average accuracy of 59% and 57%, respectively. Our deep learning method formulates contact prediction as a pixel-level image labeling problem and simultaneously predicts all residue pairs of a protein using a combination of two deep residual neural networks, taking as input the residue conservation information, predicted secondary structure and solvent accessibility, contact potential, and coevolution information. Our approach differs from existing methods mainly in (1) formulating contact prediction as a pixel-level image labeling problem instead of an image-level classification problem; (2) simultaneously predicting all contacts of an individual protein to make effective use of contact occurrence patterns; and (3) integrating both one-dimensional and two-dimensional deep convolutional neural networks to effectively learn complex sequence-structure relationship including high-order residue correlation. This paper discusses the RaptorX-Contact pipeline, both contact prediction and contact-based folding results, and finally the strength and weakness of our method.

Modeling protein quaternary structure of homo- and hetero-oligomers beyond binary interactions by homology

Cellular processes often depend on interactions between proteins and the formation of macromolecular complexes. The impairment of such interactions can lead to deregulation of pathways resulting in disease states, and it is hence crucial to gain insights into the nature of macromolecular assemblies. Detailed structural knowledge about complexes and protein-protein interactions is growing, but experimentally determined three-dimensional multimeric assemblies are outnumbered by complexes supported by non-structural experimental evidence. Here, we aim to fill this gap by modeling multimeric structures by homology, only using amino acid sequences to infer the stoichiometry and the overall structure of the assembly. We ask which properties of proteins within a family can assist in the prediction of correct quaternary structure. Specifically, we introduce a description of protein-protein interface conservation as a function of evolutionary distance to reduce the noise in deep multiple sequence alignments. We also define a distance measure to structurally compare homologous multimeric protein complexes. This allows us to hierarchically cluster protein structures and quantify the diversity of alternative biological assemblies known today. We find that a combination of conservation scores, structural clustering, and classical interface descriptors, can improve the selection of homologous protein templates leading to reliable models of protein complexes.

RRCRank: a fusion method using rank strategy for residue-residue contact prediction

Background

In structural biology area, protein residue-residue contacts play a crucial role in protein structure prediction. Some researchers have found that the predicted residue-residue contacts could effectively constrain the conformational search space, which is significant for de novo protein structure prediction. In the last few decades, related researchers have developed various methods to predict residue-residue contacts, especially, significant performance has been achieved by using fusion methods in recent years. In this work, a novel fusion method based on rank strategy has been proposed to predict contacts. Unlike the traditional regression or classification strategies, the contact prediction task is regarded as a ranking task. First, two kinds of features are extracted from correlated mutations methods and ensemble machine-learning classifiers, and then the proposed method uses the learning-to-rank algorithm to predict contact probability of each residue pair.

Results

First, we perform two benchmark tests for the proposed fusion method (RRCRank) on CASP11 dataset and CASP12 dataset respectively. The test results show that the RRCRank method outperforms other well-developed methods, especially for medium and short range contacts. Second, in order to verify the superiority of ranking strategy, we predict contacts by using the traditional regression and classification strategies based on the same features as ranking strategy. Compared with these two traditional strategies, the proposed ranking strategy shows better performance for three contact types, in particular for long range contacts. Third, the proposed RRCRank has been compared with several state-of-the-art methods in CASP11 and CASP12. The results show that the RRCRank could achieve comparable prediction precisions and is better than three methods in most assessment metrics.

Conclusions

The learning-to-rank algorithm is introduced to develop a novel rank-based method for the residue-residue contact prediction of proteins, which achieves state-of-the-art performance based on the extensive assessment.

Electronic supplementary material

The online version of this article (10.1186/s12859-017-1811-9) contains supplementary material, which is available to authorized users.

EigenTHREADER: analogous protein fold recognition by efficient contact map threading

MOTIVATION

Protein fold recognition when appropriate, evolutionarily-related, structural templates can be identified is often trivial and may even be viewed as a solved problem. However in cases where no homologous structural templates can be detected, fold recognition is a notoriously difficult problem ( Moult et al., 2014 ). Here we present EigenTHREADER, a novel fold recognition method capable of identifying folds where no homologous structures can be identified. EigenTHREADER takes a query amino acid sequence, generates a map of intra-residue contacts, and then searches a library of contact maps of known structures. To allow the contact maps to be compared, we use eigenvector decomposition to resolve the principal eigenvectors these can then be aligned using standard dynamic programming algorithms. The approach is similar to the Al-Eigen approach of Di Lena et al. (2010) , but with improvements made both to speed and accuracy. With this search strategy, EigenTHREADER does not depend directly on sequence homology between the target protein and entries in the fold library to generate models. This in turn enables EigenTHREADER to correctly identify analogous folds where little or no sequence homology information is.

RESULTS

EigenTHREADER outperforms well-established fold recognition methods such as pGenTHREADER and HHSearch in terms of True Positive Rate in the difficult task of analogous fold recognition. This should allow template-based modelling to be extended to many new protein families that were previously intractable to homology based fold recognition methods.

AVAILABILITY AND IMPLEMENTATION

All code used to generate these results and the computational protocol can be downloaded from https://github.com/DanBuchan/eigen_scripts . EigenTHREADER, the benchmark code and the data this paper is based on can be downloaded from: http://bioinfadmin.cs.ucl.ac.uk/downloads/eigenTHREADER/ .

CONTACT

d.t.jones@ucl.ac.uk.

Improved protein structure reconstruction using secondary structures, contacts at higher distance thresholds, and non-contacts

Background

Residue-residue contacts are key features for accurate de novo protein structure prediction. For the optimal utilization of these predicted contacts in folding proteins accurately, it is important to study the challenges of reconstructing protein structures using true contacts. Because contact-guided protein modeling approach is valuable for predicting the folds of proteins that do not have structural templates, it is necessary for reconstruction studies to focus on hard-to-predict protein structures.

Results

Using a data set consisting of 496 structural domains released in recent CASP experiments and a dataset of 150 representative protein structures, in this work, we discuss three techniques to improve the reconstruction accuracy using true contacts – adding secondary structures, increasing contact distance thresholds, and adding non-contacts. We find that reconstruction using secondary structures and contacts can deliver accuracy higher than using full contact maps. Similarly, we demonstrate that non-contacts can improve reconstruction accuracy not only when the used non-contacts are true but also when they are predicted. On the dataset consisting of 150 proteins, we find that by simply using low ranked predicted contacts as non-contacts and adding them as additional restraints, can increase the reconstruction accuracy by 5% when the reconstructed models are evaluated using TM-score.

Conclusions

Our findings suggest that secondary structures are invaluable companions of contacts for accurate reconstruction. Confirming some earlier findings, we also find that larger distance thresholds are useful for folding many protein structures which cannot be folded using the standard definition of contacts. Our findings also suggest that for more accurate reconstruction using predicted contacts it is useful to predict contacts at higher distance thresholds (beyond 8 Å) and predict non-contacts.

Electronic supplementary material

The online version of this article (10.1186/s12859-017-1807-5) contains supplementary material, which is available to authorized users.

An overview of comparative modelling and resources dedicated to large-scale modelling of genome sequences

Computational modelling of proteins has been a major catalyst in structural biology. Bioinformatics groups have exploited the repositories of known structures to predict high-quality structural models with high efficiency at low cost. This article provides an overview of comparative modelling, reviews recent developments and describes resources dedicated to large-scale comparative modelling of genome sequences. The value of subclustering protein domain superfamilies to guide the template-selection process is investigated. Some recent cases in which structural modelling has aided experimental work to determine very large macromolecular complexes are also cited.

Co-evolution techniques are reshaping the way we do structural bioinformatics

Co-evolution techniques were originally conceived to assist in protein structure prediction by inferring pairs of residues that share spatial proximity. However, the functional relationships that can be extrapolated from co-evolution have also proven to be useful in a wide array of structural bioinformatics applications. These techniques are a powerful way to extract structural and functional information in a sequence-rich world.

XSuLT: a web server for structural annotation and representation of sequence-structure alignments

The web server XSuLT, an enhanced version of the protein alignment annotation program JoY, formats a submitted multiple-sequence alignment using three-dimensional (3D) structural information in order to assist in the comparative analysis of protein evolution and in the optimization of alignments for comparative modelling and construct design. In addition to the features analysed by JoY, which include secondary structure, solvent accessibility and sidechain hydrogen bonds, XSuLT annotates each amino acid residue with residue depth, chain and ligand interactions, inter-residue contacts, sequence entropy, root mean square deviation and secondary structure and disorder prediction. It is also now integrated with built-in 3D visualization which interacts with the formatted alignment to facilitate inspection and understanding. Results can be downloaded as stand-alone HTML for the formatted alignment and as XML with the underlying annotation data. XSuLT is freely available at http://structure.bioc.cam.ac.uk/xsult/.

EPSILON-CP: using deep learning to combine information from multiple sources for protein contact prediction

BACKGROUND

Accurately predicted contacts allow to compute the 3D structure of a protein. Since the solution space of native residue-residue contact pairs is very large, it is necessary to leverage information to identify relevant regions of the solution space, i.e. correct contacts. Every additional source of information can contribute to narrowing down candidate regions. Therefore, recent methods combined evolutionary and sequence-based information as well as evolutionary and physicochemical information. We develop a new contact predictor (EPSILON-CP) that goes beyond current methods by combining evolutionary, physicochemical, and sequence-based information. The problems resulting from the increased dimensionality and complexity of the learning problem are combated with a careful feature analysis, which results in a drastically reduced feature set. The different information sources are combined using deep neural networks.

RESULTS

On 21 hard CASP11 FM targets, EPSILON-CP achieves a mean precision of 35.7% for top- L/10 predicted long-range contacts, which is 11% better than the CASP11 winning version of MetaPSICOV. The improvement on 1.5L is 17%. Furthermore, in this study we find that the amino acid composition, a commonly used feature, is rendered ineffective in the context of meta approaches. The size of the refined feature set decreased by 75%, enabling a significant increase in training data for machine learning, contributing significantly to the observed improvements.

CONCLUSIONS

Exploiting as much and diverse information as possible is key to accurate contact prediction. Simply merging the information introduces new challenges. Our study suggests that critical feature analysis can improve the performance of contact prediction methods that combine multiple information sources. EPSILON-CP is available as a webservice: http://compbio.robotics.tu-berlin.de/epsilon/.

Membrane protein contact and structure prediction using co-evolution in conjunction with machine learning

De novo membrane protein structure prediction is limited to small proteins due to the conformational search space quickly expanding with length. Long-range contacts (24+ amino acid separation)-residue positions distant in sequence, but in close proximity in the structure, are arguably the most effective way to restrict this conformational space. Inverse methods for co-evolutionary analysis predict a global set of position-pair couplings that best explain the observed amino acid co-occurrences, thus distinguishing between evolutionarily explained co-variances and these arising from spurious transitive effects. Here, we show that applying machine learning approaches and custom descriptors improves evolutionary contact prediction accuracy, resulting in improvement of average precision by 6 percentage points for the top 1L non-local contacts. Further, we demonstrate that predicted contacts improve protein folding with BCL::Fold. The mean RMSD100 metric for the top 10 models folded was reduced by an average of 2 Å for a benchmark of 25 membrane proteins.