High-resolution Structure Prediction Research Articles

Evolutionary-scale prediction of atomic-level protein structure with a language model

Recent advances in machine learning have leveraged evolutionary information in multiple sequence alignments to predict protein structure. We demonstrate direct inference of full atomic-level protein structure from primary sequence using a large language model. As language models of protein sequences are scaled up to 15 billion parameters, an atomic-resolution picture of protein structure emerges in the learned representations. This results in an order-of-magnitude acceleration of high-resolution structure prediction, which enables large-scale structural characterization of metagenomic proteins. We apply this capability to construct the ESM Metagenomic Atlas by predicting structures for >617 million metagenomic protein sequences, including >225 million that are predicted with high confidence, which gives a view into the vast breadth and diversity of natural proteins.

Narrowing down the conformational space with solid-state NMR in crystal structure prediction of linezolid cocrystals

Many solids crystallize as microcrystalline powders, thus precluding the application of single crystal X-Ray diffraction in structural elucidation. In such cases, a joint use of high-resolution solid-state NMR and crystal structure prediction (CSP) calculations can be successful. However, for molecules showing significant conformational freedom, the CSP-NMR protocol can meet serious obstacles, including ambiguities in NMR signal assignment and too wide conformational search space to be covered by computational methods in reasonable time. Here, we demonstrate a possible way of avoiding these obstacles and making as much use of the two methods as possible in difficult circumstances. In a simple case, our experiments led to crystal structure elucidation of a cocrystal of linezolid (LIN), a wide-range antibiotic, with 2,3-dihydroxybenzoic acid, while a significantly more challenging case of a cocrystal of LIN with 2,4-dihydroxybenzoic acid led to the identification of the most probable conformations of LIN inside the crystal. Having four rotatable bonds, some of which can assume many discreet values, LIN molecule poses a challenge in establishing its conformation in a solid phase. In our work, a set of 27 conformations were used in CSP calculations to yield model crystal structures to be examined against experimental solid-state NMR data, leading to a reliable identification of the most probable molecular arrangements.

PCPD: Plant cytochrome P450 database and web-based tools for structural construction and ligand docking

Plant cytochrome P450s play key roles in the diversification and functional modification of plant natural products. Although over 200,000 plant P450 gene sequences have been recorded, only seven crystalized P450 genes severely hampered the functional characterization, gene mining and engineering of important P450s. Here, we combined Rosetta homologous modeling and MD-based refinement to construct a high-resolution P450 structure prediction process (PCPCM), which was applied to 181 plant P450s with identified functions. Furthermore, we constructed a ligand docking process (PCPLD) that can be applied for plant P450s virtual screening. 10 examples of virtual screening indicated the process can reduce about 80% screening space for next experimental verification. Finally, we constructed a plant P450 database (PCPD: http://p450.biodesign.ac.cn/), which includes the sequences, structures and functions of the 181 plant P450s, and a web service based on PCPCM and PCPLD. Our study not only developed methods for the P450-specific structure analysis, but also introduced a universal approach that can assist the mining and functional analysis of P450 enzymes.

Template-Guided Protein Structure Prediction and Refinement Using Optimized Folding Landscape Force Fields.

When good structural templates can be identified, template-based modeling is the most reliable way to predict the tertiary structure of proteins. In this study, we combine template-based modeling with a realistic coarse-grained force field, AWSEM, that has been optimized using the principles of energy landscape theory. The Associative memory, Water mediated, Structure and Energy Model (AWSEM) is a coarse-grained force field having both transferable tertiary interactions and knowledge-based local-in-sequence interaction terms. We incorporate template information into AWSEM by introducing soft collective biases to the template structures, resulting in a model that we call AWSEM-Template. Structure prediction tests on eight targets, four of which are in the low sequence identity "twilight zone" of homology modeling, show that AWSEM-Template can achieve high-resolution structure prediction. Our results also confirm that using a combination of AWSEM and a template-guided potential leads to more accurate prediction of protein structures than simply using a template-guided potential alone. Free energy profile analyses demonstrate that the soft collective biases to the template effectively increase funneling toward native-like structures while still allowing significant flexibility so as to allow for correction of discrepancies between the target structure and the template. A further stage of refinement using all-atom molecular dynamics augmented with soft collective biases to the structures predicted by AWSEM-Template leads to a further improvement of both backbone and side-chain accuracy by maintaining sufficient flexibility but at the same time discouraging unproductive unfolding events often seen in unrestrained all-atom refinement simulations. The all-atom refinement simulations also reduce patches of frustration of the initial predictions. Some of the backbones found among the structures produced during the initial coarse-grained prediction step already have CE-RMSD values of less than 3 Å with 90% or more of the residues aligned to the experimentally solved structure for all targets. All-atom structures generated during the following all-atom refinement simulations, which started from coarse-grained structures that were chosen without reference to any knowledge about the native structure, have CE-RMSD values of less than 2.5 Å with 90% or more of the residues aligned for 6 out of 8 targets. Clustering low energy structures generated during the initial coarse-grained annealing picks out reliably structures that are within 1 Å of the best sampled structures in 5 out of 8 cases. After the all-atom refinement, structures that are within 1 Å of the best sampled structures can be selected using a simple algorithm based on energetic features alone in 7 out of 8 cases.

A strategy to improve the identification reliability of the chemical constituents by high-resolution mass spectrometry-based isomer structure prediction combined with a quantitative structure retention relationship analysis: Phthalide compounds in Chuanxiong as a test case

High-resolution mass spectrometry (HRMS) provides a powerful tool for the rapid analysis and identification of compounds in herbs. However, the diversity and large differences in the content of the chemical constituents in herbal medicines, especially isomerisms, are a great challenge for mass spectrometry-based structural identification. In the current study, a new strategy for the structural characterization of potential new phthalide compounds was proposed by isomer structure predictions combined with a quantitative structure-retention relationship (QSRR) analysis using phthalide compounds in Chuanxiong as an example. This strategy consists of three steps. First, the structures of phthalide compounds were reasonably predicted on the basis of the structure features and MS/MS fragmentation patterns: (1) the collected raw HRMS data were preliminarily screened by an in-house database; (2) the MS/MS fragmentation patterns of the analogous compounds were summarized; (3) the reported phthalide compounds were identified, and the structures of the isomers were reasonably predicted. Second, the QSRR model was established and verified using representative phthalide compound standards. Finally, the retention times of the predicted isomers were calculated by the QSRR model, and the structures of these peaks were rationally characterized by matching retention times of the detected chromatographic peaks and the predicted isomers. A multiple linear regression QSRR model in which 6 physicochemical variables were screened was built using 23 phthalide standards. The retention times of the phthalide isomers in Chuanxiong were well predicted by the QSRR model combined with reasonable structure predictions (R2=0.955). A total of 81 peaks were detected from Chuanxiong and assigned to reasonable structures, and 26 potential new phthalide compounds were structurally characterized. This strategy can improve the identification efficiency and reliability of homologues in complex materials.

High-resolution structure prediction of β-barrel membrane proteins

[Formula: see text]-Barrel membrane proteins ([Formula: see text]MPs) play important roles, but knowledge of their structures is limited. We have developed a method to predict their 3D structures. We predict strand registers and construct transmembrane (TM) domains of [Formula: see text]MPs accurately, including proteins for which no prediction has been attempted before. Our method also accurately predicts structures from protein families with a limited number of sequences and proteins with novel folds. An average main-chain rmsd of 3.48 Å is achieved between predicted and experimentally resolved structures of TM domains, which is a significant improvement ([Formula: see text]3 Å) over a recent study. For [Formula: see text]MPs with NMR structures, the deviation between predictions and experimentally solved structures is similar to the difference among the NMR structures, indicating excellent prediction accuracy. Moreover, we can now accurately model the extended [Formula: see text]-barrels and loops in non-TM domains, increasing the overall coverage of structure prediction by [Formula: see text]%. Our method is general and can be applied to genome-wide structural prediction of [Formula: see text]MPs.

Automated RNA 3D Structure Prediction with RNAComposer.

RNAs adopt specific structures to perform their activities and these are critical to virtually all RNA-mediated processes. Because of difficulties in experimentally assessing structures of large RNAs using NMR, X-ray crystallography, or cryo-microscopy, there is currently great demand for new high-resolution 3D structure prediction methods. Recently we reported on RNAComposer, a knowledge-based method for the fully automated RNA 3D structure prediction from a user-defined secondary structure. RNAComposer method is especially suited for structural biology users. Since our initial report in 2012, both servers, freely available at http://rnacomposer.ibch.poznan.pl and http://rnacomposer.cs.put.poznan.pl have been often visited. Therefore this chapter provides guidance for using RNAComposer and discusses points that should be considered when predicting 3D RNA structure. An application example presents current scope and limitations of RNAComposer.

I-TASSER server: new development for protein structure and function predictions.

The I-TASSER server (http://zhanglab.ccmb.med.umich.edu/I-TASSER) is an online resource for automated protein structure prediction and structure-based function annotation. In I-TASSER, structural templates are first recognized from the PDB using multiple threading alignment approaches. Full-length structure models are then constructed by iterative fragment assembly simulations. The functional insights are finally derived by matching the predicted structure models with known proteins in the function databases. Although the server has been widely used for various biological and biomedical investigations, numerous comments and suggestions have been reported from the user community. In this article, we summarize recent developments on the I-TASSER server, which were designed to address the requirements from the user community and to increase the accuracy of modeling predictions. Focuses have been made on the introduction of new methods for atomic-level structure refinement, local structure quality estimation and biological function annotations. We expect that these new developments will improve the quality of the I-TASSER server and further facilitate its use by the community for high-resolution structure and function prediction.

Quantitative Protein Topography Analysis and High-Resolution Structure Prediction Using Hydroxyl Radical Labeling and Tandem-Ion Mass Spectrometry (MS)*[S

Hydroxyl radical footprinting based MS for protein structure assessment has the goal of understanding ligand induced conformational changes and macromolecular interactions, for example, protein tertiary and quaternary structure, but the structural resolution provided by typical peptide-level quantification is limiting. In this work, we present experimental strategies using tandem-MS fragmentation to increase the spatial resolution of the technique to the single residue level to provide a high precision tool for molecular biophysics research. Overall, in this study we demonstrated an eightfold increase in structural resolution compared with peptide level assessments. In addition, to provide a quantitative analysis of residue based solvent accessibility and protein topography as a basis for high-resolution structure prediction; we illustrate strategies of data transformation using the relative reactivity of side chains as a normalization strategy and predict side-chain surface area from the footprinting data. We tested the methods by examination of Ca(+2)-calmodulin showing highly significant correlations between surface area and side-chain contact predictions for individual side chains and the crystal structure. Tandem ion based hydroxyl radical footprinting-MS provides quantitative high-resolution protein topology information in solution that can fill existing gaps in structure determination for large proteins and macromolecular complexes.

Automated 3D RNA structure prediction using the RNAComposer method for riboswitches.

Understanding the numerous functions of RNAs depends critically on the knowledge of their three-dimensional (3D) structure. In contrast to the protein field, a much smaller number of RNA 3D structures have been assessed using X-ray crystallography, NMR spectroscopy, and cryomicroscopy. This has led to a great demand to obtain the RNA 3D structures using prediction methods. The 3D structure prediction, especially of large RNAs, still remains a significant challenge and there is still a great demand for high-resolution structure prediction methods. In this chapter, we describe RNAComposer, a method and server for the automated prediction of RNA 3D structures based on the knowledge of secondary structure. Its applications are supported by other automated servers: RNA FRABASE and RNApdbee, developed to search and analyze secondary and 3D structures. Another method, RNAlyzer, offers new way to analyze and visualize quality of RNA 3D models. Scope and limitations of RNAComposer in application for an automated prediction of riboswitches' 3D structure will be presented and discussed. Analysis of the cyclic di-GMP-II riboswitch from Clostridium acetobutylicum (PDB ID 3Q3Z) as an example allows for 3D structure prediction of related riboswitches from Clostridium difficile 4, Bacillus halodurans 1, and Thermus aquaticus Y5.1 of yet unknown structures.

High-resolution modeling of transmembrane helical protein structures from distant homologues.

Eukaryotic transmembrane helical (TMH) proteins perform a wide diversity of critical cellular functions, but remain structurally largely uncharacterized and their high-resolution structure prediction is currently hindered by the lack of close structural homologues. To address this problem, we present a novel and generic method for accurately modeling large TMH protein structures from distant homologues exhibiting distinct loop and TMH conformations. Models of the adenosine A2AR and chemokine CXCR4 receptors were first ranked in GPCR-DOCK blind prediction contests in the receptor structure accuracy category. In a benchmark of 50 TMH protein homolog pairs of diverse topology (from 5 to 12 TMHs), size (from 183 to 420 residues) and sequence identity (from 15% to 70%), the method improves most starting templates, and achieves near-atomic accuracy prediction of membrane-embedded regions. Unlike starting templates, the models are of suitable quality for computer-based protein engineering: redesigned models and redesigned X-ray structures exhibit very similar native interactions. The method should prove useful for the atom-level modeling and design of a large fraction of structurally uncharacterized TMH proteins from a wide range of structural homologues.

AWSEM-MD: Protein Structure Prediction Using Coarse-Grained Physical Potentials and Bioinformatically Based Local Structure Biasing

The associative memory, water mediated, structure and energy model (AWSEM) is a coarse-grained protein force field. AWSEM contains physically motivated terms, such as hydrogen bonding, as well as a bioinformatically based local structure biasing term, which efficiently takes into account many-body effects that are modulated by the local sequence. When combined with appropriate local or global alignments to choose memories, AWSEM can be used to perform de novo protein structure prediction. Herein we present structure prediction results for a particular choice of local sequence alignment method based on short residue sequences called fragments. We demonstrate the model's structure prediction capabilities for three levels of global homology between the target sequence and those proteins used for local structure biasing, all of which assume that the structure of the target sequence is not known. When there are no homologues in the database of structures used for local structure biasing, AWSEM calculations produce structural predictions that are somewhat improved compared with prior works using related approaches. The inclusion of a small number of structures from homologous sequences improves structure prediction only marginally, but when the fragment search is restricted to only homologous sequences, AWSEM can perform high resolution structure prediction and can be used for kinetics and dynamics studies.

Automated 3D structure composition for large RNAs

Understanding the numerous functions that RNAs play in living cells depends critically on knowledge of their three-dimensional structure. Due to the difficulties in experimentally assessing structures of large RNAs, there is currently great demand for new high-resolution structure prediction methods. We present the novel method for the fully automated prediction of RNA 3D structures from a user-defined secondary structure. The concept is founded on the machine translation system. The translation engine operates on the RNA FRABASE database tailored to the dictionary relating the RNA secondary structure and tertiary structure elements. The translation algorithm is very fast. Initial 3D structure is composed in a range of seconds on a single processor. The method assures the prediction of large RNA 3D structures of high quality. Our approach needs neither structural templates nor RNA sequence alignment, required for comparative methods. This enables the building of unresolved yet native and artificial RNA structures. The method is implemented in a publicly available, user-friendly server RNAComposer. It works in an interactive mode and a batch mode. The batch mode is designed for large-scale modelling and accepts atomic distance restraints. Presently, the server is set to build RNA structures of up to 500 residues.

ROSETTAEPR: An Integrated Tool for Protein Structure Determination From Sparse EPR Data

Membrane proteins remain a particular challenge in structural biology. Only about 1.5% of reported tertiary structures and 60 unique membrane protein topologies consisting of more than one transmembrane span are represented in the PDB. However, these proteins make up an estimated 30-40% of the entire proteome, and over half of all therapeutics target this group. Site-directed spin labeling electron paramagnetic resonance (SDSL-EPR) is often used for the structural characterization of proteins that elude other techniques, such as X-ray crystallography and NMR. However, high-resolution structures are difficult to obtain due to uncertainty in the spin label location and sparseness of experimental data. ROSETTAEPR has been designed to improve high-resolution protein structure prediction using sparse SDSL-EPR distance data. The “motion-on-a-cone” spin label model is converted into a knowledge-based potential, which was implemented as a scoring term in ROSETTA. We have demonstrated the feasibility of using ROSETTAEPR with soluble proteins by benchmarking the method on T4-lysozyme. ROSETTAEPR increased the fractions of correctly folded models (RMSDC < 7.5A) and models accurate at medium resolution (RMSDC < 3.5A) by 25%. After full-atom refinement, ROSETTAEPR yielded a 1.7A model of T4-lysozyme, thus indicating that atomic detail models can be achieved by combining sparse EPR data with ROSETTA. ROSETTAEPR was also benchmarked on a set of membrane proteins of known structure. If EPR experimental data were not available, simulated data were derived from the existing structures. It was generally observed that de novo folding in the presence of EPR restraints enriched the recovery of the proteins’ correct topology compared to when folding with no restraints.

Toward a Structure Determination Method for Biomineral-Associated Protein Using Combined Solid- State NMR and Computational Structure Prediction

Protein-biomineral interactions are paramount to materials production in biology, including the mineral phase of hard tissue. Unfortunately, the structure of biomineral-associated proteins cannot be determined by X-ray crystallography or solution nuclear magnetic resonance (NMR). Here we report a method for determining the structure of biomineral-associated proteins. The method combines solid-state NMR (ssNMR) and ssNMR-biased computational structure prediction. In addition, the algorithm is able to identify lattice geometries most compatible with ssNMR constraints, representing a quantitative, novel method for investigating crystal-face binding specificity. We use this method to determine most of the structure of human salivary statherin interacting with the mineral phase of tooth enamel. Computation and experiment converge on an ensemble of related structures and identify preferential binding at three crystal surfaces. The work represents a significant advance toward determining structure of biomineral-adsorbed protein using experimentally biased structure prediction. This method is generally applicable to proteins that can be chemically synthesized.

RosettaEPR: An integrated tool for protein structure determination from sparse EPR data

Site-directed spin labeling electron paramagnetic resonance (SDSL-EPR) is often used for the structural characterization of proteins that elude other techniques, such as X-ray crystallography and nuclear magnetic resonance (NMR). However, high-resolution structures are difficult to obtain due to uncertainty in the spin label location and sparseness of experimental data. Here, we introduce RosettaEPR, which has been designed to improve de novo high-resolution protein structure prediction using sparse SDSL-EPR distance data. The "motion-on-a-cone" spin label model is converted into a knowledge-based potential, which was implemented as a scoring term in Rosetta. RosettaEPR increased the fractions of correctly folded models ( [Formula: see text] <7.5Å) and models accurate at medium resolution ( [Formula: see text] <3.5Å) by 25%. The correlation of score and model quality increased from 0.42 when using no restraints to 0.51 when using bounded restraints and again to 0.62 when using RosettaEPR. This allowed for the selection of accurate models by score. After full-atom refinement, RosettaEPR yielded a 1.7Å model of T4-lysozyme, thus indicating that atomic detail models can be achieved by combining sparse EPR data with Rosetta. While these results indicate RosettaEPR's potential utility in high-resolution protein structure prediction, they are based on a single example. In order to affirm the method's general performance, it must be tested on a larger and more versatile dataset of proteins.

Beyond rotamers: a generative, probabilistic model of side chains in proteins

BackgroundAccurately covering the conformational space of amino acid side chains is essential for important applications such as protein design, docking and high resolution structure prediction. Today, the most common way to capture this conformational space is through rotamer libraries - discrete collections of side chain conformations derived from experimentally determined protein structures. The discretization can be exploited to efficiently search the conformational space. However, discretizing this naturally continuous space comes at the cost of losing detailed information that is crucial for certain applications. For example, rigorously combining rotamers with physical force fields is associated with numerous problems.ResultsIn this work we present BASILISK: a generative, probabilistic model of the conformational space of side chains that makes it possible to sample in continuous space. In addition, sampling can be conditional upon the protein's detailed backbone conformation, again in continuous space - without involving discretization.ConclusionsA careful analysis of the model and a comparison with various rotamer libraries indicates that the model forms an excellent, fully continuous model of side chain conformational space. We also illustrate how the model can be used for rigorous, unbiased sampling with a physical force field, and how it improves side chain prediction when used as a pseudo-energy term. In conclusion, BASILISK is an important step forward on the way to a rigorous probabilistic description of protein structure in continuous space and in atomic detail.

Accurate Automated Protein NMR Structure Determination Using Unassigned NOESY Data

Conventional NMR structure determination requires nearly complete assignment of the cross peaks of a refined NOESY peak list. Depending on the size of the protein and quality of the spectral data, this can be a time-consuming manual process requiring several rounds of peak list refinement and structure determination. Programs such as Aria, CYANA, and AutoStructure can generate models using unassigned NOESY data but are very sensitive to the quality of the input peak lists and can converge to inaccurate structures if the signal-to-noise of the peak lists is low. Here, we show that models with high accuracy and reliability can be produced by combining the strengths of the high-resolution structure prediction program Rosetta with global measures of the agreement between structure models and experimental data. A first round of models generated using CS-Rosetta (Rosetta supplemented with backbone chemical shift information) are filtered on the basis of their goodness-of-fit with unassigned NOESY peak lists using the DP-score, and the best fitting models are subjected to high resolution refinement with the Rosetta rebuild-and-refine protocol. This hybrid approach uses both local backbone chemical shift and the unassigned NOESY data to direct Rosetta trajectories toward the native structure and produces more accurate models than AutoStructure/CYANA or CS-Rosetta alone, particularly when using raw unedited NOESY peak lists. We also show that when accurate manually refined NOESY peak lists are available, Rosetta refinement can consistently increase the accuracy of models generated using CYANA and AutoStructure.

Computation of Conformational Coupling in Allosteric Proteins

In allosteric regulation, an effector molecule binding a protein at one site induces conformational changes, which alter structure and function at a distant active site. Two key challenges in the computational modeling of allostery are the prediction of the structure of one allosteric state starting from the structure of the other, and elucidating the mechanisms underlying the conformational coupling of the effector and active sites. Here we approach these two challenges using the Rosetta high-resolution structure prediction methodology. We find that the method can recapitulate the relaxation of effector-bound forms of single domain allosteric proteins into the corresponding ligand-free states, particularly when sampling is focused on regions known to change conformation most significantly. Analysis of the coupling between contacting pairs of residues in large ensembles of conformations spread throughout the landscape between and around the two allosteric states suggests that the transitions are built up from blocks of tightly coupled interacting sets of residues that are more loosely coupled to one another.

Protein model refinement using an optimized physics-based all-atom force field

One of the greatest challenges in protein structure prediction is the refinement of low-resolution predicted models to high-resolution structures that are close to the native state. Although contemporary structure prediction methods can assemble the correct topology for a large fraction of protein domains, such approximate models are often not of the resolution required for many important applications, including studies of reaction mechanisms and virtual ligand screening. Thus, the development of a method that could bring those structures closer to the native state is of great importance. We recently optimized the relative weights of the components of the Amber ff03 potential on a large set of decoy structures to create a funnel-shaped energy landscape with the native structure at the global minimum. Such an energy function might be able to drive proteins toward their native structure. In this work, for a test set of 47 proteins, with 100 decoy structures per protein that have a range of structural similarities to the native state, we demonstrate that our optimized potential can drive protein models closer to their native structure. Comparing the lowest-energy structure from each trajectory with the starting decoy, structural improvement is seen for 70% of the models on average. The ability to do such systematic structural refinements by using a physics-based all-atom potential represents a promising approach to high-resolution structure prediction.