AlphaFold Models of Small Proteins Rival the Accuracy of Solution NMR Structures

The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins

Artificial intelligence-driven advances in protein structure prediction in recent years have raised the question: has the protein structure-prediction problem been solved? Here, with a focus on nonglobular proteins, we highlight the many strengths and potential weaknesses of DeepMind's AlphaFold2 in the context of its biological and therapeutic applications. We summarize the subtleties associated with evaluation of AlphaFold2 model quality and reliability using the predicted local distance difference test (pLDDT) and predicted aligned error (PAE) values. We highlight various classes of proteins that AlphaFold2 can be applied to and the caveats involved. Concrete examples of how AlphaFold2 models can be integrated with experimental data in the form of small-angle X-ray scattering (SAXS), solution NMR, cryo-electron microscopy (cryo-EM) and X-ray diffraction are discussed. Finally, we highlight the need to move beyond structure prediction of rigid, static structural snapshots toward conformational ensembles and alternate biologically relevant states. The overarching theme is that careful consideration is due when using AlphaFold2-generated models to generate testable hypotheses and structural models, rather than treating predicted models as de facto ground truth structures.

AlphaFold2 as a replacement for solution NMR structure determination of small proteins: Not so fast!

The determination of a protein's structure is often a first step towards the development of a mechanistic understanding of its function. Considerable advances in computational protein structure prediction have been made in recent years, with AlphaFold2 (AF2) emerging as the primary tool used by researchers for this purpose. While AF2 generally predicts accurate structures of folded proteins, we present here a case where AF2 incorrectly predicts the structure of a small, folded and compact protein with high confidence. This protein, pro-interleukin-18 (pro-IL-18), is the precursor of the cytokine IL-18. Interestingly, the structure of pro-IL-18 predicted by AF2 matches that of the mature cytokine, and not the corresponding experimentally determined structure of the pro-form of the protein. Thus, while computational structure prediction holds immense promise for addressing problems in protein biophysics, there is still a need for experimental structure determination, even in the context of small well-folded, globular proteins.

A predicted model-aided reconstruction algorithm for X-ray free-electron laser single-particle imaging

Ultra-intense, ultra-fast X-ray free-electron lasers (XFELs) enable the imaging of single protein molecules under ambient temperature and pressure. A crucial aspect of structure reconstruction involves determining the relative orientations of each diffraction pattern and recovering the missing phase information. In this paper, we introduce a predicted model-aided algorithm for orientation determination and phase retrieval, which has been tested on various simulated datasets and has shown significant improvements in the success rate, accuracy and efficiency of XFEL data reconstruction.

Hidden Structural States of Proteins Revealed by Conformer Selection with AlphaFold-NMR

Recent advances in molecular modeling using deep learning can revolutionize our understanding of dynamic protein structures. NMR is particularly well-suited for determining dynamic features of biomolecular structures. The conventional process for determining biomolecular structures from experimental NMR data involves its representation as conformation-dependent restraints, followed by generation of structural models guided by these spatial restraints. Here we describe an alternative approach: generating a distribution of realistic protein conformational models using artificial intelligence-(AI-) based methods and then selecting the sets of conformers that best explain the experimental data. We applied this conformational selection approach to redetermine the solution NMR structure of the enzyme Gaussia luciferase. First, we generated a diverse set of conformer models using AlphaFold2 (AF2) with an enhanced sampling protocol. The models that best-fit NOESY and chemical shift data were then selected with a Bayesian scoring metric. The resulting models include features of both the published NMR structure and the standard AF2 model generated without enhanced sampling. This "AlphaFold-NMR" protocol also generated an alternative "open" conformational state that fits nearly as well to the overall NMR data but accounts for some NOESY data that is not consistent with first "closed" conformational state; while other NOESY data consistent with this second state are not consistent with the first conformational state. The structure of this "open" structural state differs from that of the "closed" state primarily by the position of a thumb-shaped loop between α-helices H5 and H6, revealing a cryptic surface pocket. These alternative conformational states of Gluc are supported by "double recall" analysis of NOESY data and AF2 models. Additional structural states are also indicated by backbone chemical shift data indicating partially-disordered conformations for the C-terminal segment. Considered as a multistate ensemble, these multiple states of Gluc together fit the NOESY and chemical shift data better than the "restraint-based" NMR structure and provide novel insights into its structure-dynamic-function relationships. This study demonstrates the potential of AI-based modeling with enhanced sampling to generate conformational ensembles followed by conformer selection with experimental data as an alternative to conventional restraint satisfaction protocols for protein NMR structure determination.

Insights into AlphaFold's breakthrough in neurodegenerative diseases

Neurodegenerative diseases (ND) are disorders of the central nervous system (CNS) characterized by impairment in neurons' functions, and complete loss, leading to memory loss, and difficulty in learning, language, and movement processes. The most common among these NDs are Alzheimer's disease (AD) and Parkinson's disease (PD), although several other disorders also exist. These are frontotemporal dementia (FTD), amyotrophic lateral syndrome (ALS), Huntington's disease (HD), and others; the major pathological hallmark of NDs is the proteinopathies, either of amyloid-β (Aβ), tauopathies, or synucleinopathies. Aggregation of proteins that do not undergo normal configuration, either due to mutations or through some disturbance in cellular pathway contributes to the diseases. Artificial Intelligence (AI) and deep learning (DL) have proven to be successful in the diagnosis and treatment of various congenital diseases. DL approaches like AlphaFold (AF) are a major leap towards success in CNS disorders. This 3D protein geometry modeling algorithm developed by DeepMind has the potential to revolutionize biology. AF has the potential to predict 3D-protein confirmation at an accuracy level comparable to experimentally predicted one, with the additional advantage of precisely estimating protein interactions. This breakthrough will be beneficial to identify diseases' advancement and the disturbance of signaling pathways stimulating impaired functions of proteins. Though AlphaFold has solved a major problem in structural biology, it cannot predict membrane proteins-a beneficial approach for drug designing.

Advancing Molecular Dynamics: Toward Standardization, Integration, and Data Accessibility in Structural Biology

Molecular dynamics (MD) simulations have become a valuable tool in structural biology, offering insights into complex biological systems that are difficult to obtain through experimental techniques alone. The lack of available data sets and structures in most published computational work has limited other researchers' use of these models. In recent years, the emergence of online sharing platforms and MD database initiatives favor the deposition of ensembles and structures to accompany publications, favoring reuse of the data sets. However, the lack of uniform metadata collection, formats, and what data are deposited limits the impact and its use by different communities that are not necessarily experts in MD. This Perspective highlights the need for standardization and better resource sharing for processing and interpreting MD simulation results, akin to efforts in other areas of structural biology. As the field moves forward, we will see an increase in popularity and benefits of MD-based integrative approaches combining experimental data and simulations through probabilistic reasoning, but these too are limited by uniformity in experimental data availability and choices on how the data are modeled that are not trivial to decipher from papers. Other fields have addressed similar challenges comprehensively by establishing task forces with different degrees of success. The large scope and number of communities to represent the breadth of types of MD simulations complicates a parallel approach that would fit all. Thus, each group typically decides what data and which format to upload on servers like Zenodo. Uploading data with FAIR (findable, accessible, interoperable, reusable) principles in mind including optimal metadata collection will make the data more accessible and actionable by the community. Such a wealth of simulation data will foster method development and infrastructure advancements, thus propelling the field forward.

The 100-protein NMR spectra dataset: A resource for biomolecular NMR data analysis

Multidimensional NMR spectra are the basis for studying proteins by NMR spectroscopy and crucial for the development and evaluation of methods for biomolecular NMR data analysis. Nevertheless, in contrast to derived data such as chemical shift assignments in the BMRB and protein structures in the PDB databases, this primary data is in general not publicly archived. To change this unsatisfactory situation, we present a standardized set of solution NMR data comprising 1329 2-4-dimensional NMR spectra and associated reference (chemical shift assignments, structures) and derived (peak lists, restraints for structure calculation, etc.) annotations. With the 100-protein NMR spectra dataset that was originally compiled for the development of the ARTINA deep learning-based spectra analysis method, 100 protein structures can be reproduced from their original experimental data. The 100-protein NMR spectra dataset is expected to help the development of computational methods for NMR spectroscopy, in particular machine learning approaches, and enable consistent and objective comparisons of these methods.

How accurately can one predict drug binding modes using AlphaFold models?

Computational prediction of protein structure has been pursued intensely for decades, motivated largely by the goal of using structural models for drug discovery. Recently developed machine-learning methods such as AlphaFold 2 (AF2) have dramatically improved protein structure prediction, with reported accuracy approaching that of experimentally determined structures. To what extent do these advances translate to an ability to predict more accurately how drugs and drug candidates bind to their target proteins? Here, we carefully examine the utility of AF2 protein structure models for predicting binding poses of drug-like molecules at the largest class of drug targets, the G-protein-coupled receptors. We find that AF2 models capture binding pocket structures much more accurately than traditional homology models, with errors nearly as small as differences between structures of the same protein determined experimentally with different ligands bound. Strikingly, however, the accuracy of ligand-binding poses predicted by computational docking to AF2 models is not significantly higher than when docking to traditional homology models and is much lower than when docking to structures determined experimentally without these ligands bound. These results have important implications for all those who might use predicted protein structures for drug discovery.

Exposing the small protein load of bacterial life

The ever-growing repertoire of genomic techniques continues to expand our understanding of the true diversity and richness of prokaryotic genomes. Riboproteogenomics laid the foundation for dynamic studies of previously overlooked genomic elements. Most strikingly, bacterial genomes were revealed to harbor robust repertoires of small open reading frames (sORFs) encoding a diverse and broadly expressed range of small proteins, or sORF-encoded polypeptides (SEPs). In recent years, continuous efforts led to great improvements in the annotation and characterization of such proteins, yet many challenges remain to fully comprehend the pervasive nature of small proteins and their impact on bacterial biology. In this work, we review the recent developments in the dynamic field of bacterial genome reannotation, catalog the important biological roles carried out by small proteins and identify challenges obstructing the way to full understanding of these elusive proteins.

Blind assessment of monomeric AlphaFold2 protein structure models with experimental NMR data

Recent advances in molecular modeling of protein structures are changing the field of structural biology. AlphaFold-2 (AF2), an AI system developed by DeepMind, Inc., utilizes attention-based deep learning to predict models of protein structures with high accuracy relative to structures determined by X-ray crystallography and cryo-electron microscopy (cryoEM). Comparing AF2 models to structures determined using solution NMR data, both high similarities and distinct differences have been observed. Since AF2 was trained on X-ray crystal and cryoEM structures, we assessed how accurately AF2 can model small, monomeric, solution protein NMR structures which (i) were not used in the AF2 training data set, and (ii) did not have homologous structures in the Protein Data Bank at the time of AF2 training. We identified nine open-source protein NMR data sets for such "blind" targets, including chemical shift, raw NMR FID data, NOESY peak lists, and (for 1 case) 15N-1H residual dipolar coupling data. For these nine small (70-108 residues) monomeric proteins, we generated AF2 prediction models and assessed how well these models fit to these experimental NMR data, using several well-established NMR structure validation tools. In most of these cases, the AF2 models fit the NMR data nearly as well, or sometimes better than, the corresponding NMR structure models previously deposited in the Protein Data Bank. These results provide benchmark NMR data for assessing new NMR data analysis and protein structure prediction methods. They also document the potential for using AF2 as a guiding tool in protein NMR data analysis, and more generally for hypothesis generation in structural biology research.

Solution Structure Ensembles of the Open and Closed Forms of the ∼130 kDa Enzyme I via AlphaFold Modeling, Coarse Grained Simulations, and NMR

Large-scale interdomain rearrangements are essential to protein function, governing the activity of large enzymes and molecular machineries. Yet, obtaining an atomic-resolution understanding of how the relative domain positioning is affected by external stimuli is a hard task in modern structural biology. Here, we show that combining structural modeling by AlphaFold2 with coarse-grained molecular dynamics simulations and NMR residual dipolar coupling data is sufficient to characterize the spatial domain organization of bacterial enzyme I (EI), a ∼130 kDa multidomain oligomeric protein that undergoes large-scale conformational changes during its catalytic cycle. In particular, we solve conformational ensembles for EI at two different experimental temperatures and demonstrate that a lower temperature favors sampling of the catalytically competent closed state of the enzyme. These results suggest a role for conformational entropy in the activation of EI and demonstrate the ability of our protocol to detect and characterize the effect of external stimuli (such as mutations, ligand binding, and post-translational modifications) on the interdomain organization of multidomain proteins. We expect the ensemble refinement protocol described here to be easily transferrable to the investigation of the structure and dynamics of other uncharted multidomain systems and have assembled a Google Colab page (https://potoyangroup.github.io/Seq2Ensemble/) to facilitate implementation of the presented methodology elsewhere.

AlphaFold, allosteric, and orthosteric drug discovery: Ways forward

Drug discovery is arguably a highly challenging and significant interdisciplinary aim. The stunning success of the artificial intelligence-powered AlphaFold, whose latest version is buttressed by an innovative machine-learning approach that integrates physical and biological knowledge about protein structures, raised drug discovery hopes that unsurprisingly, have not come to bear. Even though accurate, the models are rigid, including the drug pockets. AlphaFold's mixed performance poses the question of how its power can be harnessed in drug discovery. Here we discuss possible ways of going forward wielding its strengths, while bearing in mind what AlphaFold can and cannot do. For kinases and receptors, an input enriched in active (ON) state models can better AlphaFold's chance of rational drug design success.

Recent Advances in NMR Protein Structure Prediction with ROSETTA

Nuclear magnetic resonance (NMR) spectroscopy is a powerful method for studying the structure and dynamics of proteins in their native state. For high-resolution NMR structure determination, the collection of a rich restraint dataset is necessary. This can be difficult to achieve for proteins with high molecular weight or a complex architecture. Computational modeling techniques can complement sparse NMR datasets (<1 restraint per residue) with additional structural information to elucidate protein structures in these difficult cases. The Rosetta software for protein structure modeling and design is used by structural biologists for structure determination tasks in which limited experimental data is available. This review gives an overview of the computational protocols available in the Rosetta framework for modeling protein structures from NMR data. We explain the computational algorithms used for the integration of different NMR data types in Rosetta. We also highlight new developments, including modeling tools for data from paramagnetic NMR and hydrogen-deuterium exchange, as well as chemical shifts in CS-Rosetta. Furthermore, strategies are discussed to complement and improve structure predictions made by the current state-of-the-art AlphaFold2 program using NMR-guided Rosetta modeling.

AlphaFold2 and its applications in the fields of biology and medicine

AlphaFold2 (AF2) is an artificial intelligence (AI) system developed by DeepMind that can predict three-dimensional (3D) structures of proteins from amino acid sequences with atomic-level accuracy. Protein structure prediction is one of the most challenging problems in computational biology and chemistry, and has puzzled scientists for 50 years. The advent of AF2 presents an unprecedented progress in protein structure prediction and has attracted much attention. Subsequent release of structures of more than 200 million proteins predicted by AF2 further aroused great enthusiasm in the science community, especially in the fields of biology and medicine. AF2 is thought to have a significant impact on structural biology and research areas that need protein structure information, such as drug discovery, protein design, prediction of protein function, et al. Though the time is not long since AF2 was developed, there are already quite a few application studies of AF2 in the fields of biology and medicine, with many of them having preliminarily proved the potential of AF2. To better understand AF2 and promote its applications, we will in this article summarize the principle and system architecture of AF2 as well as the recipe of its success, and particularly focus on reviewing its applications in the fields of biology and medicine. Limitations of current AF2 prediction will also be discussed.

Ranking Peptide Binders by Affinity with AlphaFold

AlphaFold has revolutionized structural biology by predicting highly accurate structures of proteins and their complexes with peptides and other proteins. However, for protein-peptide systems, we are also interested in identifying the highest affinity binder among a set of candidate peptides. We present a novel competitive binding assay using AlphaFold to predict structures of the receptor in the presence of two peptides. For systems in which the individual structures of the peptides are well predicted, the assay captures the higher affinity binder in the bound state, and the other peptide in the unbound form with statistical significance. We test the application on six protein receptors for which we have experimental binding affinities to several peptides. We find that the assay is best suited for identifying medium to strong peptide binders that adopt stable secondary structures upon binding.

Exploring AlphaFold2's Performance on Predicting Amino Acid Side-Chain Conformations and Its Utility in Crystal Structure Determination of B318L Protein

Recent technological breakthroughs in machine-learning-based AlphaFold2 (AF2) are pushing the prediction accuracy of protein structures to an unprecedented level that is on par with experimental structural quality. Despite its outstanding structural modeling capability, further experimental validations and performance assessments of AF2 predictions are still required, thus necessitating the development of integrative structural biology in synergy with both computational and experimental methods. Focusing on the B318L protein that plays an essential role in the African swine fever virus (ASFV) for viral replication, we experimentally demonstrate the high quality of the AF2 predicted model and its practical utility in crystal structural determination. Structural alignment implies that the AF2 model shares nearly the same atomic arrangement as the B318L crystal structure except for some flexible and disordered regions. More importantly, side-chain-based analysis at the individual residue level reveals that AF2's performance is likely dependent on the specific amino acid type and that hydrophobic residues tend to be more accurately predicted by AF2 than hydrophilic residues. Quantitative per-residue RMSD comparisons and further molecular replacement trials suggest that AF2 has a large potential to outperform other computational modeling methods in terms of structural determination. Additionally, it is numerically confirmed that the AF2 model is accurate enough so that it may well potentially withstand experimental data quality to a large extent for structural determination. Finally, an overall structural analysis and molecular docking simulation of the B318L protein are performed. Taken together, our study not only provides new insights into AF2's performance in predicting side-chain conformations but also sheds light upon the significance of AF2 in promoting crystal structural determination, especially when the experimental data quality of the protein crystal is poor.

Blind Assessment of Monomeric AlphaFold2 Protein Structure Models with Experimental NMR Data

Recent advances in molecular modeling of protein structures are changing the field of structural biology. AlphaFold-2 (AF2), an AI system developed by DeepMind, Inc., utilizes attention-based deep learning to predict models of protein structures with high accuracy relative to structures determined by X-ray crystallography and cryo-electron microscopy (cryoEM). Comparing AF2 models to structures determined using solution NMR data, both high similarities and distinct differences have been observed. Since AF2 was trained on X-ray crystal and cryoEM structures, we assessed how accurately AF2 can model small, monomeric, solution protein NMR structures which (i) were not used in the AF2 training data set, and (ii) did not have homologous structures in the Protein Data Bank at the time of AF2 training. We identified nine open source protein NMR data sets for such "blind" targets, including chemical shift, raw NMR FID data, NOESY peak lists, and (for 1 case) 15 N- 1 H residual dipolar coupling data. For these nine small (70 - 108 residues) monomeric proteins, we generated AF2 prediction models and assessed how well these models fit to these experimental NMR data, using several well-established NMR structure validation tools. In most of these cases, the AF2 models fit the NMR data nearly as well, or sometimes better than, the corresponding NMR structure models previously deposited in the Protein Data Bank. These results provide benchmark NMR data for assessing new NMR data analysis and protein structure prediction methods. They also document the potential for using AF2 as a guiding tool in protein NMR data analysis, and more generally for hypothesis generation in structural biology research.

Highlights

AF2 models assessed against NMR data for 9 monomeric proteins not used in training.AF2 models fit NMR data almost as well as the experimentally-determined structures. RPF-DP, PSVS , and PDBStat software provide structure quality and RDC assessment. RPF-DP analysis using AF2 models suggests multiple conformational states.