Deep-learning contact-map guided protein structure prediction in CASP13

We report the results of two fully automated structure prediction pipelines, “Zhang-Server” and “QUARK”, in CASP13. The pipelines were built upon the C-I-TASSER and C-QUARK programs, which in turn are based on I-TASSER and QUARK but with three new modules: (a) a novel multiple sequence alignment (MSA) generation protocol to construct deep sequence-profiles for contact prediction; (b) an improved meta-method, NeBcon, which combines multiple contact predictors, including ResPRE that predicts contact-maps by coupling precision-matrices with deep residual convolutional neural-networks; and (c) an optimized contact potential to guide structure assembly simulations. For 50 CASP13 FM domains that lacked homologous templates, average TM-scores of the first models produced by C-I-TASSER and C-QUARK were 28% and 56% higher than those constructed by I-TASSER and QUARK, respectively. For the first time, contact-map predictions demonstrated usefulness on TBM domains with close homologous templates, where TM-scores of C-I-TASSER models were significantly higher than those of I-TASSER models with a P-value <.05. Detailed data analyses showed that the success of C-I-TASSER and C-QUARK was mainly due to the increased accuracy of deep-learning-based contact-maps, as well as the careful balance between sequence-based contact restraints, threading templates, and generic knowledge-based potentials. Nevertheless, challenges still remain for predicting quaternary structure of multi-domain proteins, due to the difficulties in domain partitioning and domain reassembly. In addition, contact prediction in terminal regions was often unsatisfactory due to the sparsity of MSAs. Development of new contact-based domain partitioning and assembly methods and training contact models on sparse MSAs may help address these issues.     

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.     

Evaluation measures of multiple sequence alignments.

Multiple sequence alignments (MSAs) are frequently used in the study of families of protein sequences or DNA/RNA sequences. They are a fundamental tool for the understanding of the structure, functionality and, ultimately, the evolution of proteins. A new algorithm, the Circular Sum (CS) method, is presented for formally evaluating the quality of an MSA. It is based on the use of a solution to the Traveling Salesman Problem, which identifies a circular tour through an evolutionary tree connecting the sequences in a protein family. With this approach, the calculation of an evolutionary tree and the errors that it would introduce can be avoided altogether. The algorithm gives an upper bound, the best score that can possibly be achieved by any MSA for a given set of protein sequences. Alternatively, if presented with a specific MSA, the algorithm provides a formal score for the MSA, which serves as an absolute measure of the quality of the MSA. The CS measure yields a direct connection between an MSA and the associated evolutionary tree. The measure can be used as a tool for evaluating different methods for producing MSAs. A brief example of the last application is provided. Because it weights all evolutionary events on a tree identically, but does not require the reconstruction of a tree, the CS algorithm has advantages over the frequently used sum-of-pairs measures for scoring MSAs, which weight some evolutionary events more strongly than others. Compared to other weighted sum-of-pairs measures, it has the advantage that no evolutionary tree must be constructed, because we can find a circular tour without knowing the tree.     

Ensembling multiple raw coevolutionary features with deep residual neural networks for contact-map prediction in CASP13

We report the results of residue-residue contact prediction of a new pipeline built purely on the learning of coevolutionary features in the CASP13 experiment. For a query sequence, the pipeline starts with the collection of multiple sequence alignments (MSAs) from multiple genome and metagenome sequence databases using two complementary Hidden Markov Model (HMM)-based searching tools. Three profile matrices, built on covariance, precision, and pseudolikelihood maximization respectively, are then created from the MSAs, which are used as the input features of a deep residual convolutional neural network architecture for contact-map training and prediction. Two ensembling strategies have been proposed to integrate the matrix features through end-to-end training and stacking, resulting in two complementary programs called TripletRes and ResTriplet, respectively. For the 31 free-modeling domains that do not have homologous templates in the PDB, TripletRes and ResTriplet generated comparable results with an average accuracy of 0.640 and 0.646, respectively, for the top L/5 long-range predictions, where 71% and 74% of the cases have an accuracy above 0.5. Detailed data analyses showed that the strength of the pipeline is due to the sensitive MSA construction and the advanced strategies for coevolutionary feature ensembling. Domain splitting was also found to help enhance the contact prediction performance. Nevertheless, contact models for tail regions, which often involve a high number of alignment gaps, and for targets with few homologous sequences are still suboptimal. Development of new approaches where the model is specifically trained on these regions and targets might help address these problems.     

Predicting residue-residue contacts and helix-helix interactions in transmembrane proteins using an integrative feature-based random forest approach

Integral membrane proteins constitute 25-30% of genomes and play crucial roles in many biological processes. However, less than 1% of membrane protein structures are in the Protein Data Bank. In this context, it is important to develop reliable computational methods for predicting the structures of membrane proteins. Here, we present the first application of random forest (RF) for residue-residue contact prediction in transmembrane proteins, which we term as TMhhcp. Rigorous cross-validation tests indicate that the built RF models provide a more favorable prediction performance compared with two state-of-the-art methods, i.e., TMHcon and MEMPACK. Using a strict leave-one-protein-out jackknifing procedure, they were capable of reaching the top L/5 prediction accuracies of 49.5% and 48.8% for two different residue contact definitions, respectively. The predicted residue contacts were further employed to predict interacting helical pairs and achieved the Matthew's correlation coefficients of 0.430 and 0.424, according to two different residue contact definitions, respectively. To facilitate the academic community, the TMhhcp server has been made freely accessible at http://protein.cau.edu.cn/tmhhcp.     

PROCAIN: protein profile comparison with assisting information

Detection of remote sequence homology is essential for the accurate inference of protein structure, function and evolution. The most sensitive detection methods involve the comparison of evolutionary patterns reflected in multiple sequence alignments (MSAs) of protein families. We present PROCAIN, a new method for MSA comparison based on the combination of ‘vertical’ MSA context (substitution constraints at individual sequence positions) and ‘horizontal’ context (patterns of residue content at multiple positions). Based on a simple and tractable profile methodology and primitive measures for the similarity of horizontal MSA patterns, the method achieves the quality of homology detection comparable to a more complex advanced method employing hidden Markov models (HMMs) and secondary structure (SS) prediction. Adding SS information further improves PROCAIN performance beyond the capabilities of current state-of-the-art tools. The potential value of the method for structure/function predictions is illustrated by the detection of subtle homology between evolutionary distant yet structurally similar protein domains. ProCAIn, relevant databases and tools can be downloaded from: http://prodata.swmed.edu/procain/download. The web server can be accessed at http://prodata.swmed.edu/procain/procain.php.     

M-Coffee: combining multiple sequence alignment methods with T-Coffee

We introduce M-Coffee, a meta-method for assembling multiple sequence alignments (MSA) by combining the output of several individual methods into one single MSA. M-Coffee is an extension of T-Coffee and uses consistency to estimate a consensus alignment. We show that the procedure is robust to variations in the choice of constituent methods and reasonably tolerant to duplicate MSAs. We also show that performances can be improved by carefully selecting the constituent methods. M-Coffee outperforms all the individual methods on three major reference datasets: HOMSTRAD, Prefab and Balibase. We also show that on a case-by-case basis, M-Coffee is twice as likely to deliver the best alignment than any individual method. Given a collection of pre-computed MSAs, M-Coffee has similar CPU requirements to the original T-Coffee. M-Coffee is a freeware open-source package available from http://www.tcoffee.org/.     

Use of secondary structural information and C<Emphasis Type="Italic">α</Emphasis>-C<Emphasis Type="Italic">α</Emphasis> distance restraints to model protein structures with MODELLER

Protein secondary structure predictions and amino acid long range contact map predictions from primary sequence of proteins have been explored to aid in modelling protein tertiary structures. In order to evaluate the usefulness of secondary structure and 3D-residue contact prediction methods to model protein structures we have used the known Q3 (alpha-helix, beta-strands and irregular turns/loops) secondary structure information, along with residue-residue contact information as restraints for MODELLER. We present here results of our modelling studies on 30 best resolved single domain protein structures of varied lengths. The results shows that it is very difficult to obtain useful models even with 100% accurate secondary structure predictions and accurate residue contact predictions for up to 30% of residues in a sequence. The best models that we obtained for proteins of lengths 37, 70, 118, 136 and 193 amino acid residues are of RMSDs 4.17, 5.27, 9.12, 7.89 and 9.69, respectively. The results show that one can obtain better models for the proteins which have high percent of alpha-helix content. This analysis further shows that MODELLER restrain optimization program can be useful only if we have truly homologous structure(s) as a template where it derives numerous restraints, almost identical to the templates used. This analysis also clearly indicates that even if we satisfy several true residue-residue contact distances, up to 30% of their sequence length with fully known secondary structural information, we end up predicting model structures much distant from their corresponding native structures.     

Use of secondary structural information and C alpha-C alpha distance restraints to model protein structures with MODELLER

Protein secondary structure predictions and amino acid long range contact map predictions from primary sequence of proteins have been explored to aid in modelling protein tertiary structures. In order to evaluate the usefulness of secondary structure and 3D-residue contact prediction methods to model protein structures we have used the known Q3 (alpha-helix,beta-strands and irregular turns/loops) secondary structure information, along with residue-residue contact information as restraints for MODELLER. We present here results of our modelling studies on 30 best resolved single domain protein structures of varied lengths. The results shows that it is very difficult to obtain useful models even with 100% accurate secondary structure predictions and accurate residue contact predictions for up to 30% of residues in a sequence. The best models that we obtained for proteins of lengths 37, 70, 118, 136 and 193 amino acid residues are of RMSDs 4.17, 5.27, 9.12, 7.89 and 9.69,respectively. The results show that one can obtain better models for the proteins which have high percent of alpha-helix content. This analysis further shows that MODELLER restrain optimization program can be useful only if we have truly homologous structure(s) as a template where it derives numerous restraints, almost identical to the templates used. This analysis also clearly indicates that even if we satisfy several true residue-residue contact distances, up to 30%of their sequence length with fully known secondary structural information, we end up predicting model structures much distant from their corresponding native structures.     

The partitioned LASSO-patternsearch algorithm with application to gene expression data

Background

The generation of multiple sequence alignments (MSAs) is a crucial step for many bioinformatic analyses. Thus improving MSA accuracy and identifying potential errors in MSAs is important for a wide range of post-genomic research. We present a novel method called MergeAlign which constructs consensus MSAs from multiple independent MSAs and assigns an alignment precision score to each column.

Results

Using conventional benchmark tests we demonstrate that on average MergeAlign MSAs are more accurate than MSAs generated using any single matrix of sequence substitution. We show that MergeAlign column scores are related to alignment precision and hence provide an ab initio method of estimating alignment precision in the absence of curated reference MSAs. Using two novel and independent alignment performance tests that utilise a large set of orthologous gene families we demonstrate that increasing MSA performance leads to an increase in the performance of downstream phylogenetic analyses.

Conclusion

Using multiple tests of alignment performance we demonstrate that this novel method has broad general application in biological research.     

Improved Contact Predictions Using the Recognition of Protein Like Contact Patterns

Given sufficient large protein families, and using a global statistical inference approach, it is possible to obtain sufficient accuracy in protein residue contact predictions to predict the structure of many proteins. However, these approaches do not consider the fact that the contacts in a protein are neither randomly, nor independently distributed, but actually follow precise rules governed by the structure of the protein and thus are interdependent. Here, we present PconsC2, a novel method that uses a deep learning approach to identify protein-like contact patterns to improve contact predictions. A substantial enhancement can be seen for all contacts independently on the number of aligned sequences, residue separation or secondary structure type, but is largest for β-sheet containing proteins. In addition to being superior to earlier methods based on statistical inferences, in comparison to state of the art methods using machine learning, PconsC2 is superior for families with more than 100 effective sequence homologs. The improved contact prediction enables improved structure prediction.     

Accurate Simulation and Detection of Coevolution Signals in Multiple Sequence Alignments

Background

While the conserved positions of a multiple sequence alignment (MSA) are clearly of interest, non-conserved positions can also be important because, for example, destabilizing effects at one position can be compensated by stabilizing effects at another position. Different methods have been developed to recognize the evolutionary relationship between amino acid sites, and to disentangle functional/structural dependencies from historical/phylogenetic ones.

Methodology/Principal Findings

We have used two complementary approaches to test the efficacy of these methods. In the first approach, we have used a new program, MSAvolve, for the in silico evolution of MSAs, which records a detailed history of all covarying positions, and builds a global coevolution matrix as the accumulated sum of individual matrices for the positions forced to co-vary, the recombinant coevolution, and the stochastic coevolution. We have simulated over 1600 MSAs for 8 protein families, which reflect sequences of different sizes and proteins with widely different functions. The calculated coevolution matrices were compared with the coevolution matrices obtained for the same evolved MSAs with different coevolution detection methods. In a second approach we have evaluated the capacity of the different methods to predict close contacts in the representative X-ray structures of an additional 150 protein families using only experimental MSAs.

Conclusions/Significance

Methods based on the identification of global correlations between pairs were found to be generally superior to methods based only on local correlations in their capacity to identify coevolving residues using either simulated or experimental MSAs. However, the significant variability in the performance of different methods with different proteins suggests that the simulation of MSAs that replicate the statistical properties of the experimental MSA can be a valuable tool to identify the coevolution detection method that is most effective in each case.     

miREE: miRNA recognition elements ensemble

Background

The accurate prediction of ligand binding residues from amino acid sequences is important for the automated functional annotation of novel proteins. In the previous two CASP experiments, the most successful methods in the function prediction category were those which used structural superpositions of 3D models and related templates with bound ligands in order to identify putative contacting residues. However, whilst most of this prediction process can be automated, visual inspection and manual adjustments of parameters, such as the distance thresholds used for each target, have often been required to prevent over prediction. Here we describe a novel method FunFOLD, which uses an automatic approach for cluster identification and residue selection. The software provided can easily be integrated into existing fold recognition servers, requiring only a 3D model and list of templates as inputs. A simple web interface is also provided allowing access to non-expert users. The method has been benchmarked against the top servers and manual prediction groups tested at both CASP8 and CASP9.

Results

The FunFOLD method shows a significant improvement over the best available servers and is shown to be competitive with the top manual prediction groups that were tested at CASP8. The FunFOLD method is also competitive with both the top server and manual methods tested at CASP9. When tested using common subsets of targets, the predictions from FunFOLD are shown to achieve a significantly higher mean Matthews Correlation Coefficient (MCC) scores and Binding-site Distance Test (BDT) scores than all server methods that were tested at CASP8. Testing on the CASP9 set showed no statistically significant separation in performance between FunFOLD and the other top server groups tested.

Conclusions

The FunFOLD software is freely available as both a standalone package and a prediction server, providing competitive ligand binding site residue predictions for expert and non-expert users alike. The software provides a new fully automated approach for structure based function prediction using 3D models of proteins.     

PROMALS3D: a tool for multiple protein sequence and structure alignments

Although multiple sequence alignments (MSAs) are essential for a wide range of applications from structure modeling to prediction of functional sites, construction of accurate MSAs for distantly related proteins remains a largely unsolved problem. The rapidly increasing database of spatial structures is a valuable source to improve alignment quality. We explore the use of 3D structural information to guide sequence alignments constructed by our MSA program PROMALS. The resulting tool, PROMALS3D, automatically identifies homologs with known 3D structures for the input sequences, derives structural constraints through structure-based alignments and combines them with sequence constraints to construct consistency-based multiple sequence alignments. The output is a consensus alignment that brings together sequence and structural information about input proteins and their homologs. PROMALS3D can also align sequences of multiple input structures, with the output representing a multiple structure-based alignment refined in combination with sequence constraints. The advantage of PROMALS3D is that it gives researchers an easy way to produce high-quality alignments consistent with both sequences and structures of proteins. PROMALS3D outperforms a number of existing methods for constructing multiple sequence or structural alignments using both reference-dependent and reference-independent evaluation methods.     

Modeling ensembles of transmembrane beta-barrel proteins

Transmembrane beta-barrel (TMB) proteins are embedded in the outer membrane of gram-negative bacteria, mitochondria, and chloroplasts. Despite their importance, very few nonhomologous TMB structures have been determined by X-ray diffraction because of the experimental difficulty encountered in crystallizing transmembrane proteins. We introduce the program partiFold to investigate the folding landscape of TMBs. By computing the Boltzmann partition function, partiFold estimates inter-beta-strand residue interaction probabilities, predicts contacts and per-residue X-ray crystal structure B-values, and samples conformations from the Boltzmann low energy ensemble. This broad range of predictive capabilities is achieved using a single, parameterizable grammatical model to describe potential beta-barrel supersecondary structures, combined with a novel energy function of stacked amino acid pair statistical potentials. PartiFold outperforms existing programs for inter-beta-strand residue contact prediction on TMB proteins, offering both higher average predictive accuracy as well as more consistent results. Moreover, the integration of these contact probabilities inside a stochastic contact map can be used to infer a more meaningful picture of the TMB folding landscape, which cannot be achieved with other methods. Partifold's predictions of B-values are competitive with recent methods specifically designed for this problem. Finally, we show that sampling TMBs from the Boltzmann ensemble matches the X-ray crystal structure better than single structure prediction methods. A webserver running partiFold is available at http://partiFold.csail.mit.edu/.     

Improving accuracy of protein contact prediction using balanced network deconvolution

Residue contact map is essential for protein three‐dimensional structure determination. But most of the current contact prediction methods based on residue co‐evolution suffer from high false‐positives as introduced by indirect and transitive contacts (i.e., residues A–B and B–C are in contact, but A–C are not). Built on the work by Feizi et al. (Nat Biotechnol 2013; 31:726–733), which demonstrated a general network model to distinguish direct dependencies by network deconvolution, this study presents a new balanced network deconvolution (BND) algorithm to identify optimized dependency matrix without limit on the eigenvalue range in the applied network systems. The algorithm was used to filter contact predictions of five widely used co‐evolution methods. On the test of proteins from three benchmark datasets of the 9th critical assessment of protein structure prediction (CASP9), CASP10, and PSICOV (precise structural contact prediction using sparse inverse covariance estimation) database experiments, the BND can improve the medium‐ and long‐range contact predictions at the L/5 cutoff by 55.59% and 47.68%, respectively, without additional central processing unit cost. The improvement is statistically significant, with a P‐value < 5.93 × 10^?3 in the Student's t‐test. A further comparison with the ab initio structure predictions in CASPs showed that the usefulness of the current co‐evolution‐based contact prediction to the three‐dimensional structure modeling relies on the number of homologous sequences existing in the sequence databases. BND can be used as a general contact refinement method, which is freely available at: http://www.csbio.sjtu.edu.cn/bioinf/BND/ . Proteins 2015; 83:485–496. © 2014 Wiley Periodicals, Inc.     

Recent advances in features generation for membrane protein sequences: From multiple sequence alignment to pre-trained language models

Membrane proteins play a crucial role in various cellular processes and are essential components of cell membranes. Computational methods have emerged as a powerful tool for studying membrane proteins due to their complex structures and properties that make them difficult to analyze experimentally. Traditional features for protein sequence analysis based on amino acid types, composition, and pair composition have limitations in capturing higher-order sequence patterns. Recently, multiple sequence alignment (MSA) and pre-trained language models (PLMs) have been used to generate features from protein sequences. However, the significant computational resources required for MSA-based features generation can be a major bottleneck for many applications. Several methods and tools have been developed to accelerate the generation of MSAs and reduce their computational cost, including heuristics and approximate algorithms. Additionally, the use of PLMs such as BERT has shown great potential in generating informative embeddings for protein sequence analysis. In this review, we provide an overview of traditional and more recent methods for generating features from protein sequences, with a particular focus on MSAs and PLMs. We highlight the advantages and limitations of these approaches and discuss the methods and tools developed to address the computational challenges associated with features generation. Overall, the advancements in computational methods and tools provide a promising avenue for gaining deeper insights into the function and properties of membrane proteins, which can have significant implications in drug discovery and personalized medicine.     

Predicting the accuracy of multiple sequence alignment algorithms by using computational intelligent techniques

Multiple sequence alignments (MSAs) have become one of the most studied approaches in bioinformatics to perform other outstanding tasks such as structure prediction, biological function analysis or next-generation sequencing. However, current MSA algorithms do not always provide consistent solutions, since alignments become increasingly difficult when dealing with low similarity sequences. As widely known, these algorithms directly depend on specific features of the sequences, causing relevant influence on the alignment accuracy. Many MSA tools have been recently designed but it is not possible to know in advance which one is the most suitable for a particular set of sequences. In this work, we analyze some of the most used algorithms presented in the bibliography and their dependences on several features. A novel intelligent algorithm based on least square support vector machine is then developed to predict how accurate each alignment could be, depending on its analyzed features. This algorithm is performed with a dataset of 2180 MSAs. The proposed system first estimates the accuracy of possible alignments. The most promising methodologies are then selected in order to align each set of sequences. Since only one selected algorithm is run, the computational time is not excessively increased.     

A guideline to proteome-wide α-helical membrane protein topology predictions

For current state-of-the-art methods, the prediction of correct topology of membrane proteins has been reported to be above 80%. However, this performance has only been observed in small and possibly biased data sets obtained from protein structures or biochemical assays. Here, we test a number of topology predictors on an "unseen" set of proteins of known structure and also on four "genome-scale" data sets, including one recent large set of experimentally validated human membrane proteins with glycosylated sites. The set of glycosylated proteins is also used to examine the ability of prediction methods to separate membrane from nonmembrane proteins. The results show that methods utilizing multiple sequence alignments are overall superior to methods that do not. The best performance is obtained by TOPCONS, a consensus method that combines several of the other prediction methods. The best methods to distinguish membrane from nonmembrane proteins belong to the "Phobius" group of predictors. We further observe that the reported high accuracies in the smaller benchmark sets are not quite maintained in larger scale benchmarks. Instead, we estimate the performance of the best prediction methods for eukaryotic membrane proteins to be between 60% and 70%. The low agreement between predictions from different methods questions earlier estimates about the global properties of the membrane proteome. Finally, we suggest a pipeline to estimate these properties using a combination of the best predictors that could be applied in large-scale proteomics studies of membrane proteins.     

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.