Sequence-structure Relationships In Proteins Research Articles

Exploring the sequence-structure map: joint embedding of protein sequence & structure

While the success of Google's AlphaFold2 (AF2) is widely recognized as a watershed moment in generative protein modeling, AF2 does not inherently have the ability to generate synthetic proteins with desired properties, nor does AF2's static 3D structure prediction give a characterization of protein dynamics. We seek to leverage the ability of AF2 to accurately predict protein structures to develop novel protein engineering artificial intelligence (AI). By embedding joint representations of protein sequences with their corresponding structures using autoencoder neural networks, we can encode protein sequence-structure relationships into a continuous latent variable space.

Improved protein structure prediction by deep learning irrespective of co-evolution information.

Predicting the tertiary structure of a protein from its primary sequence has been greatly improved by integrating deep learning and co-evolutionary analysis, as shown in CASP13 and CASP14. We describe our latest study of this idea, analyzing the efficacy of network size and co-evolution data and its performance on both natural and designed proteins. We show that a large ResNet (convolutional residual neural networks) can predict structures of correct folds for 26 out of 32 CASP13 free-modeling (FM) targets and L/5 long-range contacts with precision over 80%. When co-evolution is not used ResNet still can predict structures of correct folds for 18 CASP13 FM targets, greatly exceeding previous methods that do not use co-evolution either. Even with only primary sequence ResNet can predict structures of correct folds for all tested human-designed proteins. In addition, ResNet may fare better for the designed proteins when trained without co-evolution than with co-evolution. These results suggest that ResNet does not simply denoise co-evolution signals, but instead may learn important protein sequence-structure relationship. This has important implications on protein design and engineering especially when co-evolutionary data is unavailable.

Local sequence-structure relationships in proteins.

We seek to understand the interplay between amino acid sequence and local structure in proteins. Are some amino acids unique in their ability to fit harmoniously into certain local structures? What is the role of sequence in sculpting the putative native state folds from myriad possible conformations? In order to address these questions, we represent the local structure of each Cα atom of a protein by just two angles, θ and μ, and we analyze a set of more than 4,000 protein structures from the PDB. We use a hierarchical clustering scheme to divide the 20 amino acids into six distinct groups based on their similarity to each other in fitting local structural space. We present the results of a detailed analysis of patterns of amino acid specificity in adopting local structural conformations and show that the sequence-structure correlation is not very strong compared with a random assignment of sequence to structure. Yet, our analysis may be useful to determine an effective scoring rubric for quantifying the match of an amino acid to its putative local structure.

ProteinNet: a standardized data set for machine learning of protein structure

BackgroundRapid progress in deep learning has spurred its application to bioinformatics problems including protein structure prediction and design. In classic machine learning problems like computer vision, progress has been driven by standardized data sets that facilitate fair assessment of new methods and lower the barrier to entry for non-domain experts. While data sets of protein sequence and structure exist, they lack certain components critical for machine learning, including high-quality multiple sequence alignments and insulated training/validation splits that account for deep but only weakly detectable homology across protein space.ResultsWe created the ProteinNet series of data sets to provide a standardized mechanism for training and assessing data-driven models of protein sequence-structure relationships. ProteinNet integrates sequence, structure, and evolutionary information in programmatically accessible file formats tailored for machine learning frameworks. Multiple sequence alignments of all structurally characterized proteins were created using substantial high-performance computing resources. Standardized data splits were also generated to emulate the difficulty of past CASP (Critical Assessment of protein Structure Prediction) experiments by resetting protein sequence and structure space to the historical states that preceded six prior CASPs. Utilizing sensitive evolution-based distance metrics to segregate distantly related proteins, we have additionally created validation sets distinct from the official CASP sets that faithfully mimic their difficulty.ConclusionProteinNet represents a comprehensive and accessible resource for training and assessing machine-learned models of protein structure.

Sequence statistics of tertiary structural motifs reflect protein stability.

The Protein Data Bank (PDB) has been a key resource for learning general rules of sequence-structure relationships in proteins. Quantitative insights have been gained by defining geometric descriptors of structure (e.g., distances, dihedral angles, solvent exposure, etc.) and observing their distributions and sequence preferences. Here we argue that as the PDB continues to grow, it may become unnecessary to reduce structure into a set of elementary descriptors. Instead, it could be possible to deduce quantitative sequence-structure relationships in the context of precisely-defined complex structural motifs by mining the PDB for closely matching backbone geometries. To validate this idea, we turned to the the task of predicting changes in protein stability upon amino-acid substitution—a difficult problem of broad significance. We defined non-contiguous tertiary motifs (TERMs) around a protein site of interest and extracted sequence preferences from ensembles of closely-matching substructures in the PDB to predict mutational stability changes at the site, ΔΔGm. We demonstrate that these ensemble statistics predict ΔΔGm on par with state-of-the-art statistical and machine-learning methods on large thermodynamic datasets, and outperform these, along with a leading structure-based modeling approach, when tested in the context of unbiased diverse mutations. Further, we show that the performance of the TERM-based method is directly related to the amount of available relevant structural data, automatically improving with the growing PDB. This enables a means of estimating prediction accuracy. Our results clearly demonstrate that: 1) statistics of non-contiguous structural motifs in the PDB encode fundamental sequence-structure relationships related to protein thermodynamic stability, and 2) the PDB is now large enough that such statistics are already useful in practice, with their accuracy expected to continue increasing as the database grows. These observations suggest new ways of using structural data towards addressing problems of computational structural biology.

On the origin of protein superfamilies and superfolds.

Distributions of protein families and folds in genomes are highly skewed, having a small number of prevalent superfamiles/superfolds and a large number of families/folds of a small size. Why are the distributions of protein families and folds skewed? Why are there only a limited number of protein families? Here, we employ an information theoretic approach to investigate the protein sequence-structure relationship that leads to the skewed distributions. We consider that protein sequences and folds constitute an information theoretic channel and computed the most efficient distribution of sequences that code all protein folds. The identified distributions of sequences and folds are found to follow a power law, consistent with those observed for proteins in nature. Importantly, the skewed distributions of sequences and folds are suggested to have different origins: the skewed distribution of sequences is due to evolutionary pressure to achieve efficient coding of necessary folds, whereas that of folds is based on the thermodynamic stability of folds. The current study provides a new information theoretic framework for proteins that could be widely applied for understanding protein sequences, structures, functions, and interactions.

Designability landscape reveals sequence features that define axial helix rotation in four-helical homo-oligomeric antiparallel coiled-coil structures

Coiled coils are widespread protein domains comprising α-helices wound around each other in a regular fashion. Owing to their regularity, coiled-coil structures can be fully described by parametric equations. This in turn makes them an excellent model for studying sequence-structure relationships in proteins. Here, we used computational design to identify sequence features that determine the degree of helix axial rotation in four-helical homo-oligomeric antiparallel coiled coils. We designed 135,000 artificial sequences for a repertoire of backbone models representing all theoretically possible axial rotation states. Analysis of the designed sequences revealed features that precisely define the rotation of the helices. Based on these features we implemented a bioinformatic tool, which given a coiled-coil sequence, predicts the rotation of the helices in its structure. Moreover, we showed that another structural parameter, helix axial shift, is coupled to helix axial rotation and that dependence between these two parameters narrows the number of possible axial rotation states.

Fragment-free approach to protein folding using conditional neural fields

Motivation: One of the major bottlenecks with ab initio protein folding is an effective conformation sampling algorithm that can generate native-like conformations quickly. The popular fragment assembly method generates conformations by restricting the local conformations of a protein to short structural fragments in the PDB. This method may limit conformations to a subspace to which the native fold does not belong because (i) a protein with really new fold may contain some structural fragments not in the PDB and (ii) the discrete nature of fragments may prevent them from building a native-like fold. Previously we have developed a conditional random fields (CRF) method for fragment-free protein folding that can sample conformations in a continuous space and demonstrated that this CRF method compares favorably to the popular fragment assembly method. However, the CRF method is still limited by its capability of generating conformations compatible with a sequence.Results: We present a new fragment-free approach to protein folding using a recently invented probabilistic graphical model conditional neural fields (CNF). This new CNF method is much more powerful than CRF in modeling the sophisticated protein sequence-structure relationship and thus, enables us to generate native-like conformations more easily. We show that when coupled with a simple energy function and replica exchange Monte Carlo simulation, our CNF method can generate decoys much better than CRF on a variety of test proteins including the CASP8 free-modeling targets. In particular, our CNF method can predict a correct fold for T0496_D1, one of the two CASP8 targets with truly new fold. Our predicted model for T0496 is significantly better than all the CASP8 models.Contact: jinboxu@gmail.com

Local descriptors of protein structure: A systematic analysis of the sequence‐structure relationship in proteins using short‐ and long‐range interactions

Local protein structure representations that incorporate long-range contacts between residues are often considered in protein structure comparison but have found relatively little use in structure prediction where assembly from single backbone fragments dominates. Here, we introduce the concept of local descriptors of protein structure to characterize local neighborhoods of amino acids including short- and long-range interactions. We build a library of recurring local descriptors and show that this library is general enough to allow assembly of unseen protein structures. The library could on average re-assemble 83% of 119 unseen structures, and showed little or no performance decrease between homologous targets and targets with folds not represented among domains used to build it. We then systematically evaluate the descriptor library to establish the level of the sequence signal in sets of protein fragments of similar geometrical conformation. In particular, we test whether that signal is strong enough to facilitate correct assignment and alignment of these local geometries to new sequences. We use the signal to assign descriptors to a test set of 479 sequences with less than 40% sequence identity to any domain used to build the library, and show that on average more than 50% of the backbone fragments constituting descriptors can be correctly aligned. We also use the assigned descriptors to infer SCOP folds, and show that correct predictions can be made in many of the 151 cases where PSI-BLAST was unable to detect significant sequence similarity to proteins in the library. Although the combinatorial problem of simultaneously aligning several fragments to sequence is a major bottleneck compared with single fragment methods, the advantage of the current approach is that correct alignments imply correct long range distance constraints. The lack of these constraints is most likely the major reason why structure prediction methods fail to consistently produce adequate models when good templates are unavailable or undetectable. Thus, we believe that the current study offers new and valuable insight into the prediction of sequence-structure relationships in proteins.

Predicting residue-wise contact orders in proteins by support vector regression.

BackgroundThe residue-wise contact order (RWCO) describes the sequence separations between the residues of interest and its contacting residues in a protein sequence. It is a new kind of one-dimensional protein structure that represents the extent of long-range contacts and is considered as a generalization of contact order. Together with secondary structure, accessible surface area, the B factor, and contact number, RWCO provides comprehensive and indispensable important information to reconstructing the protein three-dimensional structure from a set of one-dimensional structural properties. Accurately predicting RWCO values could have many important applications in protein three-dimensional structure prediction and protein folding rate prediction, and give deep insights into protein sequence-structure relationships.ResultsWe developed a novel approach to predict residue-wise contact order values in proteins based on support vector regression (SVR), starting from primary amino acid sequences. We explored seven different sequence encoding schemes to examine their effects on the prediction performance, including local sequence in the form of PSI-BLAST profiles, local sequence plus amino acid composition, local sequence plus molecular weight, local sequence plus secondary structure predicted by PSIPRED, local sequence plus molecular weight and amino acid composition, local sequence plus molecular weight and predicted secondary structure, and local sequence plus molecular weight, amino acid composition and predicted secondary structure. When using local sequences with multiple sequence alignments in the form of PSI-BLAST profiles, we could predict the RWCO distribution with a Pearson correlation coefficient (CC) between the predicted and observed RWCO values of 0.55, and root mean square error (RMSE) of 0.82, based on a well-defined dataset with 680 protein sequences. Moreover, by incorporating global features such as molecular weight and amino acid composition we could further improve the prediction performance with the CC to 0.57 and an RMSE of 0.79. In addition, combining the predicted secondary structure by PSIPRED was found to significantly improve the prediction performance and could yield the best prediction accuracy with a CC of 0.60 and RMSE of 0.78, which provided at least comparable performance compared with the other existing methods.ConclusionThe SVR method shows a prediction performance competitive with or at least comparable to the previously developed linear regression-based methods for predicting RWCO values. In contrast to support vector classification (SVC), SVR is very good at estimating the raw value profiles of the samples. The successful application of the SVR approach in this study reinforces the fact that support vector regression is a powerful tool in extracting the protein sequence-structure relationship and in estimating the protein structural profiles from amino acid sequences.

Thermodynamic analysis of protein sequence-structure relationships in monomer and dimer forms

Two-dimensional lattice model is employed to analyze the sequence–structure relationships between HP and YhHX protein sequences and compact structures in monomer and homodimer forms. Our results show that most structures are encoded by normal and prion protein sequences, and about one-third of the protein sequences are potentially able to refold from its native monomer conformation to alternative ones upon dimerization thermodynamically. Such findings partially reflect the existences of the dimer and oligomer protein structures in living organism and the occurrence of pathologic dimerization or aggregation of monomer proteins. It is observed that protein sequences differ drastically in the number of strains. Statistically, the number of protein sequences decreases exponentially with the number of strains. Calculations using various energy parameters show that the expected value of maximum number of strains for a 16-residue sequence is about its length.

Nonlinear signal analysis methods in the elucidation of protein sequence-structure relationships.

Istituto Superiore di Sanita`, TCE Laboratory, V.le R. Elena 299, 00161 Roma, Italy, Department of Molecular Biophysics and Physiology,Rush University, 1653 West Congress Parkway, Chicago, Illinois 60612, Department of Physiology, Loyola University Medical Center,2160 South First Avenue, Maywood, Illinois 60153, and Department of Biochemical Sciences, University of Rome, “La Sapienza”,P.le A. Moro, 5-00185 Roma, ItalyReceived July 30, 2001

Deciphering globular protein sequence-structure relationships: from observation to prediction

Careful comparison of proteins sharing a same fold but only low or no sequence identity should allow a better understanding of the coding of three-dimensional structures by amino acid sequences. It has already been shown that positions of a given fold occupied mainly by hydrophobic residues in the different proteins of a structural family share very specific physical properties and participate in stabilization of the protein domain. They probably also play a crucial role in the very first steps of folding [ Poupon A, Mornon J.-P (1999) FEBS Lett. 452: 283–289; Mirny LA, Shaknovich EI (1999) J. Mol. Biol. 291: 177–196]. To further understand the sequence–structure relationship, we studied the correlation between allowed mutations at a given three-dimensional position and some of its physical properties. The different amino acids were divided in three groups (hydrophobic, nonpolar or weakly polar and polar or charged), and a correlation was established between the occupation rate of each group at a given position in the fold and the burying, the side-chain dispersion, the interposition distances and the ability to form a network of directly interacting residues. The results are then applied to predict some solvent accessibility. We show that this property can be accurately predicted for about 70% of the residues, providing precious information concerning the corresponding three-dimensional structures. The results are used to predict other structural features, as secondary structures, compactness or long-range interactions between residues remote in sequence. This information will allow the number of possible structures for a given sequence to be reduced considerably, simplifying the ab initio modelling problem to a level where it might be solved by computing methods.

Algorithmic complexity and thermodynamics of sequence-structure relationships in proteins

The information contained in a protein's amino acid sequence dictates its three-dimensional structure. In this situation a frozen or embedded structure, the sequence, contains information that ultimately influences a thermodynamic entity, the protein structure. The interplay between information and thermodynamics is explored by considering the algorithmic complexity and Kolmogorov's universal probability of the sequence and of the structure. It is shown that the algorithmic complexity of a microstate of a polymer is given by its configurational entropy. Using this result and a lattice protein model, a quantitative estimate of the information contained in a protein's structure is made. This is compared to the information content of the sequence. The information content of the sequence is approximately 2.5 bits per amino acid, while the content in the structure is approximately 0.5 bits per amino acid. It is estimated that virtually all the information contained in the protein structure is shared with the sequence. A deeper connection can be made between the shared information content and the thermodynamic entropy governing the system. Using Kolmogorov's universal probability, it is possible to establish statistical-mechanical relationships for objects without resorting to a probabilistic ensemble formalism. This allows the thermodynamics of microstates of objects of known configurations to be determined. Using this formalism, the connection between sequence information and the structural thermodynamics of a protein can be made. This connection has strong implications for how protein sequences evolve over evolutionary time and demonstrates that this evolution is constrained by the thermodynamic evolution of the protein structure.

Sequence-structure relationships in proteins and copolymers.

We model proteins as copolymer chains of H (hydrophobic) and P (polar) monomers configured as self-avoiding flights on three-dimensional simple-cubic lattices. The HH interaction is favorable. The folding problem is to find the ``native'' conformation(s) (lowest free energy) for an HP sequence. Using geometric proofs for self-avoiding lattice chains, we develop equations relating a monomer sequence to its native structures. These constraint relations can be used for two purposes: (1) to compute a tight lower bound on the free energy of the native state for HP sequences of any length, which is useful for testing conformational search strategies, and (2) to develop a search strategy. In its present implementation, the search strategy finds native states for HP lattice chains up to 36 monomers in length, which is a speedup of 5\char21{}15 orders of magnitude over existing brute-force exhaustive-search methods.