Prediction of contact maps with neural networks and correlated mutations.

Contact maps of proteins are predicted with neural network-based methods, using as input codings of increasing complexity including evolutionary information, sequence conservation, correlated mutations and predicted secondary structures. Neural networks are trained on a data set comprising the contact maps of 173 non-homologous proteins as computed from their well resolved three-dimensional structures. Proteins are selected from the Protein Data Bank database provided that they align with at least 15 similar sequences in the corresponding families. The predictors are trained to learn the association rules between the covalent structure of each protein and its contact map with a standard back propagation algorithm and tested on the same protein set with a cross-validation procedure. Our results indicate that the method can assign protein contacts with an average accuracy of 0.21 and with an improvement over a random predictor of a factor >6, which is higher than that previously obtained with methods only based either on neural networks or on correlated mutations. Furthermore, filtering the network outputs with a procedure based on the residue coordination numbers, the accuracy of predictions increases up to 0.25 for all the proteins, with an 8-fold deviation from a random predictor. These scores are the highest reported so far for predicting protein contact maps.     

Prediction of contact maps by GIOHMMs and recurrent neural networks using lateral propagation from all four cardinal corners

MOTIVATION: Accurate prediction of protein contact maps is an important step in computational structural proteomics. Because contact maps provide a translation and rotation invariant topological representation of a protein, they can be used as a fundamental intermediary step in protein structure prediction. RESULTS: We develop a new set of flexible machine learning architectures for the prediction of contact maps, as well as other information processing and pattern recognition tasks. The architectures can be viewed as recurrent neural network implemantations of a class of Bayesian networks we call generalized input-output HMMs (GIOHMMs). For the specific case of contact maps, contextual information is propagated laterally through four hidden planes, one for each cardinal corner. We show that these architectures can be trained from examples and yield contact map predictors that outperform previously reported methods. While several extensions and improvements are in progress, the current version can accurately predict 60.5% of contacts at a distance cutoff of 8 A and 45% of distant contacts at 10 A, for proteins of length up to 300.     

Prediction of the transmembrane regions of beta-barrel membrane proteins with a neural network-based predictor

A method based on neural networks is trained and tested on a nonredundant set of beta-barrel membrane proteins known at atomic resolution with a jackknife procedure. The method predicts the topography of transmembrane beta strands with residue accuracy as high as 78% when evolutionary information is used as input to the network. Of the transmembrane beta-strands included in the training set, 93% are correctly assigned. The predictor includes an algorithm of model optimization, based on dynamic programming, that correctly models eight out of the 11 proteins present in the training/testing set. In addition, protein topology is assigned on the basis of the location of the longest loops in the models. We propose this as a general method to fill the gap of the prediction of beta-barrel membrane proteins.     

Exploring the effects of sparse restraints on protein structure prediction

One of the main barriers to accurate computational protein structure prediction is searching the vast space of protein conformations. Distance restraints or inter‐residue contacts have been used to reduce this search space, easing the discovery of the correct folded state. It has been suggested that about 1 contact for every 12 residues may be sufficient to predict structure at fold level accuracy. Here, we use coarse‐grained structure‐based models in conjunction with molecular dynamics simulations to examine this empirical prediction. We generate sparse contact maps for 15 proteins of varying sequence lengths and topologies and find that given perfect secondary‐structural information, a small fraction of the native contact map (5%‐10%) suffices to fold proteins to their correct native states. We also find that different sparse maps are not equivalent and we make several observations about the type of maps that are successful at such structure prediction. Long range contacts are found to encode more information than shorter range ones, especially for α and αβ‐proteins. However, this distinction reduces for β‐proteins. Choosing contacts that are a consensus from successful maps gives predictive sparse maps as does choosing contacts that are well spread out over the protein structure. Additionally, the folding of proteins can also be used to choose predictive sparse maps. Overall, we conclude that structure‐based models can be used to understand the efficacy of structure‐prediction restraints and could, in future, be tuned to include specific force‐field interactions, secondary structure errors and noise in the sparse maps.     

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.     

A contact map matching approach to protein structure similarity analysis

We modeled the problem of identifying how close two proteins are structurally by measuring the dissimilarity of their contact maps. These contact maps are colored images, in which the chromatic information encodes the chemical nature of the contacts. We studied two conceptually distinct image-processing algorithms to measure the dissimilarity between these contact maps; one was a content-based image retrieval method, and the other was based on image registration. In experiments with contact maps constructed from the protein data bank, our approach was able to identify, with greater than 80% precision, instances of monomers of apolipoproteins, globins, plastocyanins, retinol binding proteins and thioredoxins, among the monomers of Protein Data Bank Select. The image registration approach was only slightly more accurate than the content-based image retrieval approach.     

Modular DAG-RNN architectures for assembling coarse protein structures.

We develop and test machine learning methods for the prediction of coarse 3D protein structures, where a protein is represented by a set of rigid rods associated with its secondary structure elements (alpha-helices and beta-strands). First, we employ cascades of recursive neural networks derived from graphical models to predict the relative placements of segments. These are represented as discretized distance and angle maps, and the discretization levels are statistically inferred from a large and curated dataset. Coarse 3D folds of proteins are then assembled starting from topological information predicted in the first stage. Reconstruction is carried out by minimizing a cost function taking the form of a purely geometrical potential. We show that the proposed architecture outperforms simpler alternatives and can accurately predict binary and multiclass coarse maps. The reconstruction procedure proves to be fast and often leads to topologically correct coarse structures that could be exploited as a starting point for various protein modeling strategies. The fully integrated rod-shaped protein builder (predictor of contact maps + reconstruction algorithm) can be accessed at http://distill.ucd.ie/.     

A Study on Protein Residue Contacts Prediction by Recurrent Neural Network

1IntroductionThe three-dimensional(3D)structure of a proteinis perhaps the most important of all its features,since itdetermines completely how the protein functions andinteracts with other molecules.Most biological mech-anisms at the protein level are based on shape-complementarity,so that proteins present particularconcavities and convexities that allow them to bind toeach other and formcomplexstructures,and tendon.Forthis reason,for instance,the drug design problem con-sists primarily in th…     

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/.     

Beyond the Twilight Zone: Automated prediction of structural properties of proteins by recursive neural networks and remote homology information

The prediction of 1D structural properties of proteins is an important step toward the prediction of protein structure and function, not only in the ab initio case but also when homology information to known structures is available. Despite this the vast majority of 1D predictors do not incorporate homology information into the prediction process. We develop a novel structural alignment method, SAMD, which we use to build alignments of putative remote homologues that we compress into templates of structural frequency profiles. We use these templates as additional input to ensembles of recursive neural networks, which we specialise for the prediction of query sequences that show only remote homology to any Protein Data Bank structure. We predict four 1D structural properties – secondary structure, relative solvent accessibility, backbone structural motifs, and contact density. Secondary structure prediction accuracy, tested by five‐fold cross‐validation on a large set of proteins allowing less than 25% sequence identity between training and test set and query sequences and templates, exceeds 82%, outperforming its ab initio counterpart, other state‐of‐the‐art secondary structure predictors (Jpred 3 and PSIPRED) and two other systems based on PSI‐BLAST and COMPASS templates. We show that structural information from homologues improves prediction accuracy well beyond the Twilight Zone of sequence similarity, even below 5% sequence identity, for all four structural properties. Significant improvement over the extraction of structural information directly from PDB templates suggests that the combination of sequence and template information is more informative than templates alone. Proteins 2009. © 2009 Wiley‐Liss, Inc.     

CoCoNet—boosting RNA contact prediction by convolutional neural networks

Co-evolutionary models such as direct coupling analysis (DCA) in combination with machine learning (ML) techniques based on deep neural networks are able to predict accurate protein contact or distance maps. Such information can be used as constraints in structure prediction and massively increase prediction accuracy. Unfortunately, the same ML methods cannot readily be applied to RNA as they rely on large structural datasets only available for proteins. Here, we demonstrate how the available smaller data for RNA can be used to improve prediction of RNA contact maps. We introduce an algorithm called CoCoNet that is based on a combination of a Coevolutionary model and a shallow Convolutional Neural Network. Despite its simplicity and the small number of trained parameters, the method boosts the positive predictive value (PPV) of predicted contacts by about 70% with respect to DCA as tested by cross-validation of about eighty RNA structures. However, the direct inclusion of the CoCoNet contacts in 3D modeling tools does not result in a proportional increase of the 3D RNA structure prediction accuracy. Therefore, we suggest that the field develops, in addition to contact PPV, metrics which estimate the expected impact for 3D structure modeling tools better. CoCoNet is freely available and can be found at https://github.com/KIT-MBS/coconet.     

Generation and evaluation of dimension-reduced amino acid parameter representations by artificial neural networks

In order to process data of proteins, a numerical representation for an amino acid is often necessary. Many suitable parameters can be derived from experiments or statistical analysis of databases. To ensure a fast and efficient use of these sources of information, a reduction and extraction of relevant information out of these parameters is a basic need. In this approach established methods like principal component analysis (PCA) are supplemented by a method based on symmetric neural networks. Two different parameter representations of amino acids are reduced from five and seven dimensions, respectively, to one, two, three, or four dimensions by using a symmetric neural network approach alternatively with one or three hidden layers. It is possible to create general reduced parameter representations for amino acids. To demonstrate the ability of this approach, these reduced sets of parameters are applied for the ab initio prediction of protein secondary structure from primary structure only. Artificial neural networks are implemented and trained with a diverse representation of 430 proteins out of the PDB. An essentially faster training and also prediction without a decrease in accuracy is obtained for the reduced parameter representations in comparison with the complete set of parameters. The method is transferable to other amino acids or even other molecular building blocks, like nucleic acids, and therefore represents a general approach.Electronic Supplementary Material available.     

Deducing high-accuracy protein contact-maps from a triplet of coevolutionary matrices through deep residual convolutional networks

The topology of protein folds can be specified by the inter-residue contact-maps and accurate contact-map prediction can help ab initio structure folding. We developed TripletRes to deduce protein contact-maps from discretized distance profiles by end-to-end training of deep residual neural-networks. Compared to previous approaches, the major advantage of TripletRes is in its ability to learn and directly fuse a triplet of coevolutionary matrices extracted from the whole-genome and metagenome databases and therefore minimize the information loss during the course of contact model training. TripletRes was tested on a large set of 245 non-homologous proteins from CASP 11&12 and CAMEO experiments and outperformed other top methods from CASP12 by at least 58.4% for the CASP 11&12 targets and 44.4% for the CAMEO targets in the top-L long-range contact precision. On the 31 FM targets from the latest CASP13 challenge, TripletRes achieved the highest precision (71.6%) for the top-L/5 long-range contact predictions. It was also shown that a simple re-training of the TripletRes model with more proteins can lead to further improvement with precisions comparable to state-of-the-art methods developed after CASP13. These results demonstrate a novel efficient approach to extend the power of deep convolutional networks for high-accuracy medium- and long-range protein contact-map predictions starting from primary sequences, which are critical for constructing 3D structure of proteins that lack homologous templates in the PDB library.     

Predicting the secondary structure of globular proteins using neural network models

We present a new method for predicting the secondary structure of globular proteins based on non-linear neural network models. Network models learn from existing protein structures how to predict the secondary structure of local sequences of amino acids. The average success rate of our method on a testing set of proteins non-homologous with the corresponding training set was 64.3% on three types of secondary structure (alpha-helix, beta-sheet, and coil), with correlation coefficients of C alpha = 0.41, C beta = 0.31 and Ccoil = 0.41. These quality indices are all higher than those of previous methods. The prediction accuracy for the first 25 residues of the N-terminal sequence was significantly better. We conclude from computational experiments on real and artificial structures that no method based solely on local information in the protein sequence is likely to produce significantly better results for non-homologous proteins. The performance of our method of homologous proteins is much better than for non-homologous proteins, but is not as good as simply assuming that homologous sequences have identical structures.     

Building Disease-Specific Drug-Protein Connectivity Maps from Molecular Interaction Networks and PubMed Abstracts

The recently proposed concept of molecular connectivity maps enables researchers to integrate experimental measurements of genes, proteins, metabolites, and drug compounds under similar biological conditions. The study of these maps provides opportunities for future toxicogenomics and drug discovery applications. We developed a computational framework to build disease-specific drug-protein connectivity maps. We integrated gene/protein and drug connectivity information based on protein interaction networks and literature mining, without requiring gene expression profile information derived from drug perturbation experiments on disease samples. We described the development and application of this computational framework using Alzheimer''s Disease (AD) as a primary example in three steps. First, molecular interaction networks were incorporated to reduce bias and improve relevance of AD seed proteins. Second, PubMed abstracts were used to retrieve enriched drug terms that are indirectly associated with AD through molecular mechanistic studies. Third and lastly, a comprehensive AD connectivity map was created by relating enriched drugs and related proteins in literature. We showed that this molecular connectivity map development approach outperformed both curated drug target databases and conventional information retrieval systems. Our initial explorations of the AD connectivity map yielded a new hypothesis that diltiazem and quinidine may be investigated as candidate drugs for AD treatment. Molecular connectivity maps derived computationally can help study molecular signature differences between different classes of drugs in specific disease contexts. To achieve overall good data coverage and quality, a series of statistical methods have been developed to overcome high levels of data noise in biological networks and literature mining results. Further development of computational molecular connectivity maps to cover major disease areas will likely set up a new model for drug development, in which therapeutic/toxicological profiles of candidate drugs can be checked computationally before costly clinical trials begin.     

Self-organizing hierarchic networks for pattern recognition in protein sequence.

We present a method based on hierarchical self-organizing maps (SOMs) for recognizing patterns in protein sequences. The method is fully automatic, does not require prealigned sequences, is insensitive to redundancy in the training set, and works surprisingly well even with small learning sets. Because it uses unsupervised neural networks, it is able to extract patterns that are not present in all of the unaligned sequences of the learning set. The identification of these patterns in sequence databases is sensitive and efficient. The procedure comprises three main training stages. In the first stage, one SOM is trained to extract common features from the set of unaligned learning sequences. A feature is a number of ungapped sequence segments (usually 4-16 residues long) that are similar to segments in most of the sequences of the learning set according to an initial similarity matrix. In the second training stage, the recognition of each individual feature is refined by selecting an optimal weighting matrix out of a variety of existing amino acid similarity matrices. In a third stage of the SOM procedure, the position of the features in the individual sequences is learned. This allows for variants with feature repeats and feature shuffling. The procedure has been successfully applied to a number of notoriously difficult cases with distinct recognition problems: helix-turn-helix motifs in DNA-binding proteins, the CUB domain of developmentally regulated proteins, and the superfamily of ribokinases. A comparison with the established database search procedure PROFILE (and with several others) led to the conclusion that the new automatic method performs satisfactorily.     

Reconstruction of 3D structures from protein contact maps

The prediction of the protein tertiary structure from solely its residue sequence (the so called Protein Folding Problem) is one of the most challenging problems in Structural Bioinformatics. We focus on the protein residue contact map. When this map is assigned it is possible to reconstruct the 3D structure of the protein backbone. The general problem of recovering a set of 3D coordinates consistent with some given contact map is known as a unit-disk-graph realization problem and it has been recently proven to be NP-Hard. In this paper we describe a heuristic method (COMAR) that is able to reconstruct with an unprecedented rate (3-15 seconds) a 3D model that exactly matches the target contact map of a protein. Working with a non-redundant set of 1760 proteins, we find that the scoring efficiency of finding a 3D model very close to the protein native structure depends on the threshold value adopted to compute the protein residue contact map. Contact maps whose threshold values range from 10 to 18 Ångstroms allow reconstructing 3D models that are very similar to the proteins native structure.     

Hybridized distance- and contact-based hierarchical structure modeling for folding soluble and membrane proteins

Crystallography and NMR system (CNS) is currently a widely used method for fragment-free ab initio protein folding from inter-residue distance or contact maps. Despite its widespread use in protein structure prediction, CNS is a decade-old macromolecular structure determination system that was originally developed for solving macromolecular geometry from experimental restraints as opposed to predictive modeling driven by interaction map data. As such, the adaptation of the CNS experimental structure determination protocol for ab initio protein folding is intrinsically anomalous that may undermine the folding accuracy of computational protein structure prediction. In this paper, we propose a new CNS-free hierarchical structure modeling method called DConStruct for folding both soluble and membrane proteins driven by distance and contact information. Rigorous experimental validation shows that DConStruct attains much better reconstruction accuracy than CNS when tested with the same input contact map at varying contact thresholds. The hierarchical modeling with iterative self-correction employed in DConStruct scales at a much higher degree of folding accuracy than CNS with the increase in contact thresholds, ultimately approaching near-optimal reconstruction accuracy at higher-thresholded contact maps. The folding accuracy of DConStruct can be further improved by exploiting distance-based hybrid interaction maps at tri-level thresholding, as demonstrated by the better performance of our method in folding free modeling targets from the 12th and 13th rounds of the Critical Assessment of techniques for protein Structure Prediction (CASP) experiments compared to popular CNS- and fragment-based approaches and energy-minimization protocols, some of which even using much finer-grained distance maps than ours. Additional large-scale benchmarking shows that DConStruct can significantly improve the folding accuracy of membrane proteins compared to a CNS-based approach. These results collectively demonstrate the feasibility of greatly improving the accuracy of ab initio protein folding by optimally exploiting the information encoded in inter-residue interaction maps beyond what is possible by CNS.     

Cover Image,Volume 87, Issue 7

Template-based modeling is considered as one of the most successful approaches for protein structure prediction. However, reliably and accurately selecting optimal template proteins from a library of known protein structures having similar folds as the target protein and making correct alignments between the target sequence and the template structures, a template-based modeling technique known as threading, remains challenging, particularly for non- or distantly-homologous protein targets. With the recent advancement in protein residue-residue contact map prediction powered by sequence co-evolution and machine learning, here we systematically analyze the effect of inclusion of residue-residue contact information in improving the accuracy and reliability of protein threading. We develop a new threading algorithm by incorporating various sequential and structural features, and subsequently integrate residue-residue contact information as an additional scoring term for threading template selection. We show that the inclusion of contact information attains statistically significantly better threading performance compared to a baseline threading algorithm that does not utilize contact information when everything else remains the same. Experimental results demonstrate that our contact based threading approach outperforms popular threading method MUSTER, contact-assisted ab initio folding method CONFOLD2, and recent state-of-the-art contact-assisted protein threading methods EigenTHREADER and map_align on several benchmarks. Our study illustrates that the inclusion of contact maps is a promising avenue in protein threading to ultimately help to improve the accuracy of protein structure prediction.     

Distill: a suite of web servers for the prediction of one-, two- and three-dimensional structural features of proteins

Background  

We describe Distill, a suite of servers for the prediction of protein structural features: secondary structure; relative solvent accessibility; contact density; backbone structural motifs; residue contact maps at 6, 8 and 12 Angstrom; coarse protein topology. The servers are based on large-scale ensembles of recursive neural networks and trained on large, up-to-date, non-redundant subsets of the Protein Data Bank. Together with structural feature predictions, Distill includes a server for prediction of C _α traces for short proteins (up to 200 amino acids).