The AlphaFold Database of Protein Structures: A Biologist’s Guide

AlphaFold, the deep learning algorithm developed by DeepMind, recently released the three-dimensional models of the whole human proteome to the scientific community. Here we discuss the advantages, limitations and the still unsolved challenges of the AlphaFold models from the perspective of a biologist, who may not be an expert in structural biology.     

AlphaFold and Implications for Intrinsically Disordered Proteins

Accurate predictions of the three-dimensional structures of proteins from their amino acid sequences have come of age. AlphaFold, a deep learning-based approach to protein structure prediction, shows remarkable success in independent assessments of prediction accuracy. A significant epoch in structural bioinformatics was the structural annotation of over 98% of protein sequences in the human proteome. Interestingly, many predictions feature regions of very low confidence, and these regions largely overlap with intrinsically disordered regions (IDRs). That over 30% of regions within the proteome are disordered is congruent with estimates that have been made over the past two decades, as intense efforts have been undertaken to generalize the structure–function paradigm to include the importance of conformational heterogeneity and dynamics. With structural annotations from AlphaFold in hand, there is the temptation to draw inferences regarding the “structures” of IDRs and their interactomes. Here, we offer a cautionary note regarding the misinterpretations that might ensue and highlight efforts that provide concrete understanding of sequence-ensemble-function relationships of IDRs. This perspective is intended to emphasize the importance of IDRs in sequence-function relationships (SERs) and to highlight how one might go about extracting quantitative SERs to make sense of how IDRs function.     

Benchmarking AlphaFold for protein complex modeling reveals accuracy determinants

High‐resolution experimental structural determination of protein–protein interactions has led to valuable mechanistic insights, yet due to the massive number of interactions and experimental limitations there is a need for computational methods that can accurately model their structures. Here we explore the use of the recently developed deep learning method, AlphaFold, to predict structures of protein complexes from sequence. With a benchmark of 152 diverse heterodimeric protein complexes, multiple implementations and parameters of AlphaFold were tested for accuracy. Remarkably, many cases (43%) had near‐native models (medium or high critical assessment of predicted interactions accuracy) generated as top‐ranked predictions by AlphaFold, greatly surpassing the performance of unbound protein–protein docking (9% success rate for near‐native top‐ranked models), however AlphaFold modeling of antibody–antigen complexes within our set was unsuccessful. We identified sequence and structural features associated with lack of AlphaFold success, and we also investigated the impact of multiple sequence alignment input. Benchmarking of a multimer‐optimized version of AlphaFold (AlphaFold‐Multimer) with a set of recently released antibody–antigen structures confirmed a low rate of success for antibody–antigen complexes (11% success), and we found that T cell receptor–antigen complexes are likewise not accurately modeled by that algorithm, showing that adaptive immune recognition poses a challenge for the current AlphaFold algorithm and model. Overall, our study demonstrates that end‐to‐end deep learning can accurately model many transient protein complexes, and highlights areas of improvement for future developments to reliably model any protein–protein interaction of interest.     

A Fifth of the Protein World: Rossmann-like Proteins as an Evolutionarily Successful Structural unit

The Rossmann-like fold is the most prevalent and diversified doubly-wound superfold of ancient evolutionary origin. Rossmann-like domains are present in a variety of metabolic enzymes and are capable of binding diverse ligands. Discerning evolutionary relationships among these domains is challenging because of their diverse functions and ancient origin. We defined a minimal Rossmann-like structural motif (RLM), identified RLM-containing domains among known 3D structures (20%) and classified them according to their homologous relationships. New classifications were incorporated into our Evolutionary Classification of protein Domains (ECOD) database. We defined 156 homology groups (H-groups), which were further clustered into 123 possible homology groups (X-groups). Our analysis revealed that RLM-containing proteins constitute approximately 15% of the human proteome. We found that disease-causing mutations are more frequent within RLM domains than within non-RLM domains of these proteins, highlighting the importance of RLM-containing proteins for human health.     

PhyreRisk: A Dynamic Web Application to Bridge Genomics,Proteomics and 3D Structural Data to Guide Interpretation of Human Genetic Variants

PhyreRisk is an open-access, publicly accessible web application for interactively bridging genomic, proteomic and structural data facilitating the mapping of human variants onto protein structures. A major advance over other tools for sequence-structure variant mapping is that PhyreRisk provides information on 20,214 human canonical proteins and an additional 22,271 alternative protein sequences (isoforms). Specifically, PhyreRisk provides structural coverage (partial or complete) for 70% (14,035 of 20,214 canonical proteins) of the human proteome, by storing 18,874 experimental structures and 84,818 pre-built models of canonical proteins and their isoforms generated using our in house Phyre2. PhyreRisk reports 55,732 experimentally, multi-validated protein interactions from IntAct and 24,260 experimental structures of protein complexes.Another major feature of PhyreRisk is that, rather than presenting a limited set of precomputed variant-structure mapping of known genetic variants, it allows the user to explore novel variants using, as input, genomic coordinates formats (Ensembl, VCF, reference SNP ID and HGVS notations) and Human Build GRCh37 and GRCh38. PhyreRisk also supports mapping variants using amino acid coordinates and searching for genes or proteins of interest.PhyreRisk is designed to empower researchers to translate genetic data into protein structural information, thereby providing a more comprehensive appreciation of the functional impact of variants. PhyreRisk is freely available at http://phyrerisk.bc.ic.ac.uk     

Structural proteomics: developments in structure-to-function predictions

The major challenge for post-genomic research is to functionally assign and validate a large number of novel target genes and their corresponding proteins. Functional genomics approaches have, therefore, gained considerable attention in the quest to convert this massive data set into useful information. One of the crucial components for the functional understanding of unassigned proteins is the analysis of their experimental or modeled 3D structures. Structural proteomics initiatives are generating protein structures at an unprecedented rate but our current knowledge of 3D-structural space is still limited. Estimates on the completeness of the 3D-structural coverage of proteins vary but it is generally accepted that only a minority of the structural proteome has a template structure from which reliable conclusions can be drawn. Thus, structural proteomics has set out to build a map of protein structures that will represent all protein folds included in the 'global proteome'.     

Membranome 3.0: Database of single‐pass membrane proteins with AlphaFold models

The Membranome database provides comprehensive structural information on single‐pass (i.e., bitopic) membrane proteins from six evolutionarily distant organisms, including protein–protein interactions, complexes, mutations, experimental structures, and models of transmembrane α‐helical dimers. We present a new version of this database, Membranome 3.0, which was significantly updated by revising the set of 5,758 bitopic proteins and incorporating models generated by AlphaFold 2 in the database. The AlphaFold models were parsed into structural domains located at the different membrane sides, modified to exclude low‐confidence unstructured terminal regions and signal sequences, validated through comparison with available experimental structures, and positioned with respect to membrane boundaries. Membranome 3.0 was re‐developed to facilitate visualization and comparative analysis of multiple 3D structures of proteins that belong to a specified family, complex, biological pathway, or membrane type. New tools for advanced search and analysis of proteins, their interactions, complexes, and mutations were included. The database is freely accessible at https://membranome.org.     

Energy Landscapes of Protein Aggregation and Conformation Switching in Intrinsically Disordered Proteins

The protein folding problem was apparently solved recently by the advent of a deep learning method for protein structure prediction called AlphaFold. However, this program is not able to make predictions about the protein folding pathways. Moreover, it only treats about half of the human proteome, as the remaining proteins are intrinsically disordered or contain disordered regions. By definition these proteins differ from natively folded proteins and do not adopt a properly folded structure in solution. However these intrinsically disordered proteins (IDPs) also systematically differ in amino acid composition and uniquely often become folded upon binding to an interaction partner. These factors preclude solving IDP structures by current machine-learning methods like AlphaFold, which also cannot solve the protein aggregation problem, since this meta-folding process can give rise to different aggregate sizes and structures. An alternative computational method is provided by molecular dynamics simulations that already successfully explored the energy landscapes of IDP conformational switching and protein aggregation in multiple cases. These energy landscapes are very different from those of ‘simple’ protein folding, where one energy funnel leads to a unique protein structure. Instead, the energy landscapes of IDP conformational switching and protein aggregation feature a number of minima for different competing low-energy structures. In this review, I discuss the characteristics of these multifunneled energy landscapes in detail, illustrated by molecular dynamics simulations that elucidated the underlying conformational transitions and aggregation processes.     

Disorders of protein biogenesis and stability

The great diversity of structural conformations available to proteins allows this class of molecules to carry out the vast majority of biochemical functions in the cell. In order to function adequately, proteins must be synthesized, folded/assembled and degraded with great temporal and spatial accuracy. Precise coordination of multiple processes, including ribosome assembly and movement along mRNA, charging and recycling of tRNAs, recruitment and action of molecular chaperones, and tight control of the degradation machinery is essential to create and maintain a stable proteome. It has become recently evident that even slight errors in any of these processes may lead to disease states. Accordingly, increasing numbers of human diseases have been identified that are due to mutations in genes encoding proteins involved in this so-called "protein quality control". Since these processes are essential for the production and maintenance of the entire proteome of the cell, the deleterious effects of these mutations often extend far beyond the faulty gene. This review provides an overview of human disorders caused by defects in mechanisms underlying protein biogenesis and stability.     

Functional coverage of the human genome by existing structures, structural genomics targets, and homology models

The bias in protein structure and function space resulting from experimental limitations and targeting of particular functional classes of proteins by structural biologists has long been recognized, but never continuously quantified. Using the Enzyme Commission and the Gene Ontology classifications as a reference frame, and integrating structure data from the Protein Data Bank (PDB), target sequences from the structural genomics projects, structure homology derived from the SUPERFAMILY database, and genome annotations from Ensembl and NCBI, we provide a quantified view, both at the domain and whole-protein levels, of the current and projected coverage of protein structure and function space relative to the human genome. Protein structures currently provide at least one domain that covers 37% of the functional classes identified in the genome; whole structure coverage exists for 25% of the genome. If all the structural genomics targets were solved (twice the current number of structures in the PDB), it is estimated that structures of one domain would cover 69% of the functional classes identified and complete structure coverage would be 44%. Homology models from existing experimental structures extend the 37% coverage to 56% of the genome as single domains and 25% to 31% for complete structures. Coverage from homology models is not evenly distributed by protein family, reflecting differing degrees of sequence and structure divergence within families. While these data provide coverage, conversely, they also systematically highlight functional classes of proteins for which structures should be determined. Current key functional families without structure representation are highlighted here; updated information on the "most wanted list" that should be solved is available on a weekly basis from http://function.rcsb.org:8080/pdb/function_distribution/index.html.     

NMR hawk‐eyed view of AlphaFold2 structures

The prediction of the three‐dimensional (3D) structure of proteins from the amino acid sequence made a stunning breakthrough reaching atomic accuracy. Using the neural network‐based method AlphaFold2, 3D structures of almost the entire human proteome have been predicted and made available (https://www.alphafold.ebi.ac.uk). To gain insight into how well AlphaFold2 structures represent the conformation of proteins in solution, I here compare the AlphaFold2 structures of selected small proteins with their 3D structures that were determined by nuclear magnetic resonance (NMR) spectroscopy. Proteins were selected for which the 3D solution structures were determined on the basis of a very large number of distance restraints and residual dipolar couplings and are thus some of the best‐resolved solution structures of proteins to date. The quality of the backbone conformation of the AlphaFold2 structures is assessed by fitting a large set of experimental residual dipolar couplings (RDCs). The analysis shows that experimental RDCs fit extremely well to the AlphaFold2 structures predicted for GB3, DinI, and ubiquitin. In the case of GB3, the accuracy of the AlphaFold2 structure even surpasses that of a 1.1 Å crystal structure. Fitting of experimental RDCs furthermore allows identification of AlphaFold2 structures that are best representative of the protein''s conformation in solution as seen for the EF hands of the N‐terminal domain of Ca²⁺‐ligated calmodulin. Taken together, the analysis shows that structures predicted by AlphaFold2 can be highly representative of the solution conformation of proteins. The combination of AlphaFold2 structures with RDCs promises to be a powerful approach to study structural changes in proteins.     

Proteomic basis of the antibody response to monkeypox virus infection examined in cynomolgus macaques and a comparison to human smallpox vaccination

Monkeypox is a zoonotic viral disease that occurs primarily in Central and West Africa. A recent outbreak in the United States heightened public health concerns for susceptible human populations. Vaccinating with vaccinia virus to prevent smallpox is also effective for monkeypox due to a high degree of sequence conservation. Yet, the identity of antigens within the monkeypox virus proteome contributing to immune responses has not been described in detail. We compared antibody responses to monkeypox virus infection and human smallpox vaccination by using a protein microarray covering 92-95% (166-192 proteins) of representative proteomes from monkeypox viral clades of Central and West Africa, including 92% coverage (250 proteins) of the vaccinia virus proteome as a reference orthopox vaccine. All viral gene clones were verified by sequencing and purified recombinant proteins were used to construct the microarray. Serum IgG of cynomolgus macaques that recovered from monkeypox recognized at least 23 separate proteins within the orthopox proteome, while only 14 of these proteins were recognized by IgG from vaccinated humans. There were 12 of 14 antigens detected by sera of human vaccinees that were also recognized by IgG from convalescent macaques. The greatest level of IgG binding for macaques occurred with the structural proteins F13L and A33R, and the membrane scaffold protein D13L. Significant IgM responses directed towards A44R, F13L and A33R of monkeypox virus were detected before onset of clinical symptoms in macaques. Thus, antibodies from vaccination recognized a small number of proteins shared with pathogenic virus strains, while recovery from infection also involved humoral responses to antigens uniquely recognized within the monkeypox virus proteome.     

Robust classification of protein variation using structural modelling and large-scale data integration

Existing methods for interpreting protein variation focus on annotating mutation pathogenicity rather than detailed interpretation of variant deleteriousness and frequently use only sequence-based or structure-based information. We present VIPUR, a computational framework that seamlessly integrates sequence analysis and structural modelling (using the Rosetta protein modelling suite) to identify and interpret deleterious protein variants. To train VIPUR, we collected 9477 protein variants with known effects on protein function from multiple organisms and curated structural models for each variant from crystal structures and homology models. VIPUR can be applied to mutations in any organism''s proteome with improved generalized accuracy (AUROC .83) and interpretability (AUPR .87) compared to other methods. We demonstrate that VIPUR''s predictions of deleteriousness match the biological phenotypes in ClinVar and provide a clear ranking of prediction confidence. We use VIPUR to interpret known mutations associated with inflammation and diabetes, demonstrating the structural diversity of disrupted functional sites and improved interpretation of mutations associated with human diseases. Lastly, we demonstrate VIPUR''s ability to highlight candidate variants associated with human diseases by applying VIPUR to de novo variants associated with autism spectrum disorders.     

Modeling conformational states of proteins with AlphaFold

Many proteins exert their function by switching among different structures. Knowing the conformational ensembles affiliated with these states is critical to elucidate key mechanistic aspects that govern protein function. While experimental determination efforts are still bottlenecked by cost, time, and technical challenges, the machine-learning technology AlphaFold showed near experimental accuracy in predicting the three-dimensional structure of monomeric proteins. However, an AlphaFold ensemble of models usually represents a single conformational state with minimal structural heterogeneity. Consequently, several pipelines have been proposed to either expand the structural breadth of an ensemble or bias the prediction toward a desired conformational state. Here, we analyze how those pipelines work, what they can and cannot predict, and future directions.     

Evaluation of AlphaFold2 structures as docking targets

AlphaFold2 is a promising new tool for researchers to predict protein structures and generate high-quality models, with low backbone and global root-mean-square deviation (RMSD) when compared with experimental structures. However, it is unclear if the structures predicted by AlphaFold2 will be valuable targets of docking. To address this question, we redocked ligands in the PDBbind datasets against the experimental co-crystallized receptor structures and against the AlphaFold2 structures using AutoDock-GPU. We find that the quality measure provided during structure prediction is not a good predictor of docking performance, despite accurately reflecting the quality of the alpha carbon alignment with experimental structures. Removing low-confidence regions of the predicted structure and making side chains flexible improves the docking outcomes. Overall, despite high-quality prediction of backbone conformation, fine structural details limit the naive application of AlphaFold2 models as docking targets.     

Complementing machine learning‐based structure predictions with native mass spectrometry

The advent of machine learning‐based structure prediction algorithms such as AlphaFold2 (AF2) and RoseTTa Fold have moved the generation of accurate structural models for the entire cellular protein machinery into the reach of the scientific community. However, structure predictions of protein complexes are based on user‐provided input and may require experimental validation. Mass spectrometry (MS) is a versatile, time‐effective tool that provides information on post‐translational modifications, ligand interactions, conformational changes, and higher‐order oligomerization. Using three protein systems, we show that native MS experiments can uncover structural features of ligand interactions, homology models, and point mutations that are undetectable by AF2 alone. We conclude that machine learning can be complemented with MS to yield more accurate structural models on a small and large scale.     

Can Predicted Protein 3D Structures Provide Reliable Insights into whether Missense Variants Are Disease Associated?

Knowledge of protein structure can be used to predict the phenotypic consequence of a missense variant. Since structural coverage of the human proteome can be roughly tripled to over 50% of the residues if homology-predicted structures are included in addition to experimentally determined coordinates, it is important to assess the reliability of using predicted models when analyzing missense variants. Accordingly, we assess whether a missense variant is structurally damaging by using experimental and predicted structures. We considered 606 experimental structures and show that 40% of the 1965 disease-associated missense variants analyzed have a structurally damaging change in the mutant structure. Only 11% of the 2134 neutral variants are structurally damaging. Importantly, similar results are obtained when 1052 structures predicted using Phyre2 algorithm were used, even when the model shares low (< 40%) sequence identity to the template. Thus, structure-based analysis of the effects of missense variants can be effectively applied to homology models. Our in-house pipeline, Missense3D, for structurally assessing missense variants was made available at http://www.sbg.bio.ic.ac.uk/~missense3d