AlphaFold and the amyloid landscape

Protein aggregation is a widespread phenomenon with important implications in many scientific areas. Although amyloid formation is typically considered as detrimental, functional amyloids that perform physiological roles have been identified in all kingdoms of life. Despite their functional and pathological relevance, the structural details of the majority of molecular species involved in the amyloidogenic process remains elusive. Here, we explore the application of AlphaFold, a highly accurate protein structure predictor, in the field of protein aggregation. While we envision a straightforward application of AlphaFold in assisting the design of globular proteins with improved solubility for biomedical and industrial purposes, the use of this algorithm for predicting the structure of aggregated species seems far from trivial. First, in amyloid diseases, the presence of multiple amyloid polymorphs and the heterogeneity of aggregation intermediates challenges the “one sequence, one structure” paradigm, inherent to sequence-based predictions. Second, aberrant aggregation is not the subject of positive selective pressure, precluding the use of evolutionary-based approaches, which are the core of the AlphaFold pipeline. Instead, amyloid polymorphism seems to be constrained by the need for a defined structure-activity relationship in functional amyloids. They may thus provide a starting point for the application of AlphaFold in the amyloid landscape.     

Identification of Iron-Sulfur (Fe-S) Cluster and Zinc (Zn) Binding Sites Within Proteomes Predicted by DeepMind’s AlphaFold2 Program Dramatically Expands the Metalloproteome

DeepMind’s AlphaFold2 software has ushered in a revolution in high quality, 3D protein structure prediction. In very recent work by the DeepMind team, structure predictions have been made for entire proteomes of twenty-one organisms, with >360,000 structures made available for download. Here we show that thousands of novel binding sites for iron-sulfur (Fe-S) clusters and zinc (Zn) ions can be identified within these predicted structures by exhaustive enumeration of all potential ligand-binding orientations. We demonstrate that AlphaFold2 routinely makes highly specific predictions of ligand binding sites: for example, binding sites that are comprised exclusively of four cysteine sidechains fall into three clusters, representing binding sites for 4Fe-4S clusters, 2Fe-2S clusters, or individual Zn ions. We show further: (a) that the majority of known Fe-S cluster and Zn binding sites documented in UniProt are recovered by the AlphaFold2 structures, (b) that there are occasional disputes between AlphaFold2 and UniProt with AlphaFold2 predicting highly plausible alternative binding sites, (c) that the Fe-S cluster binding sites that we identify in E. coli agree well with previous bioinformatics predictions, (d) that cysteines predicted here to be part of ligand binding sites show little overlap with those shown via chemoproteomics techniques to be highly reactive, and (e) that AlphaFold2 occasionally appears to build erroneous disulfide bonds between cysteines that should instead coordinate a ligand. These results suggest that AlphaFold2 could be an important tool for the functional annotation of proteomes, and the methodology presented here is likely to be useful for predicting other ligand-binding sites.     

High-accuracy protein structures by combining machine-learning with physics-based refinement

Protein structure prediction has long been available as an alternative to experimental structure determination, especially via homology modeling based on templates from related sequences. Recently, models based on distance restraints from coevolutionary analysis via machine learning to have significantly expanded the ability to predict structures for sequences without templates. One such method, AlphaFold, also performs well on sequences where templates are available but without using such information directly. Here we show that combining machine-learning based models from AlphaFold with state-of-the-art physics-based refinement via molecular dynamics simulations further improves predictions to outperform any other prediction method tested during the latest round of CASP. The resulting models have highly accurate global and local structures, including high accuracy at functionally important interface residues, and they are highly suitable as initial models for crystal structure determination via molecular replacement.     

Progress at protein structure prediction,as seen in CASP15

In Dec 2020, the results of AlphaFold version 2 were presented at CASP14, sparking a revolution in the field of protein structure predictions. For the first time, a purely computational method could challenge experimental accuracy for structure prediction of single protein domains. The code of AlphaFold v2 was released in the summer of 2021, and since then, it has been shown that it can be used to accurately predict the structure of most ordered proteins and many protein–protein interactions. It has also sparked an explosion of development in the field, improving AI-based methods to predict protein complexes, disordered regions, and protein design. Here I will review some of the inventions sparked by the release of AlphaFold.     

The structural coverage of the human proteome before and after AlphaFold

The protein structure field is experiencing a revolution. From the increased throughput of techniques to determine experimental structures, to developments such as cryo-EM that allow us to find the structures of large protein complexes or, more recently, the development of artificial intelligence tools, such as AlphaFold, that can predict with high accuracy the folding of proteins for which the availability of homology templates is limited. Here we quantify the effect of the recently released AlphaFold database of protein structural models in our knowledge on human proteins. Our results indicate that our current baseline for structural coverage of 48%, considering experimentally-derived or template-based homology models, elevates up to 76% when including AlphaFold predictions. At the same time the fraction of dark proteome is reduced from 26% to just 10% when AlphaFold models are considered. Furthermore, although the coverage of disease-associated genes and mutations was near complete before AlphaFold release (69% of Clinvar pathogenic mutations and 88% of oncogenic mutations), AlphaFold models still provide an additional coverage of 3% to 13% of these critically important sets of biomedical genes and mutations. Finally, we show how the contribution of AlphaFold models to the structural coverage of non-human organisms, including important pathogenic bacteria, is significantly larger than that of the human proteome. Overall, our results show that the sequence-structure gap of human proteins has almost disappeared, an outstanding success of direct consequences for the knowledge on the human genome and the derived medical applications.     

IPSA-Inductive Protein Structure Analysis.

The Inductive Structure Protein Analysis (IPSA) project presents a new method for investigating protein structure. IPSA includes the creation of a new database which was designed specifically for the analysis of protein structure by statistics and machine learning. The Protein Representation Language (PRL) database includes explicit and symbolic representations of geometrical, topological and chemophysical information about secondary structures and the relationships between secondary structures. The IPSA methodology consists of: the use of PRL information to produce a new database of examples of secondary structures which associate together (examples of possible super-secondary structures); then the use of a variety of clustering techniques to produce a consensus clustering of these examples (super-secondary structures); these super-secondary structures are finally examined to uncover any biological features of significance. We have applied this method to find simple super-secondary structures consisting of pairs of alpha-helices. We found four well-defined super-secondary structures, one formed exclusively by long range interactions, and another in association with an additional element of secondary structure (alpha t alpha-motif). Examinations were carried out using homologous pairs and conformational fits which confirm our clustering.     

Mining folded proteomes in the era of accurate structure prediction

Protein structure fundamentally underpins the function and processes of numerous biological systems. Fold recognition algorithms offer a sensitive and robust tool to detect structural, and thereby functional, similarities between distantly related homologs. In the era of accurate structure prediction owing to advances in machine learning techniques and a wealth of experimentally determined structures, previously curated sequence databases have become a rich source of biological information. Here, we use bioinformatic fold recognition algorithms to scan the entire AlphaFold structure database to identify novel protein family members, infer function and group predicted protein structures. As an example of the utility of this approach, we identify novel, previously unknown members of various pore-forming protein families, including MACPFs, GSDMs and aerolysin-like proteins.     

Secure web book to store structural genomics research data

Recently established collaborative structural genomics programs aim at significantly accelerating the crystal structure analysis of proteins. These large-scale projects require efficient data management systems to ensure seamless collaboration between different groups of scientists working towards the same goal. Within the Berlin-based Protein Structure Factory, the synchrotron X-ray data collection and the subsequent crystal structure analysis tasks are located at BESSY, a third-generation synchrotron source. To organize file-based communication and data transfer at the BESSY site of the Protein Structure Factory, we have developed the web-based BCLIMS, the BESSY Crystallography Laboratory Information Management System. BCLIMS is a relational data management system which is powered by MySQL as the database engine and Apache HTTP as the web server. The database interface routines are written in Python programing language. The software is freely available to academic users. Here we describe the storage, retrieval and manipulation of laboratory information, mainly pertaining to the synchrotron X-ray diffraction experiments and the subsequent protein structure analysis, using BCLIMS.     

Threading a database of protein cores

We present an analysis of 10 blind predictions prepared for a recent conference, “Critical Assessment of Techniques for Protein Structure Prediction.”¹ The sequences of these proteins are not detectably similar to those of any protein in the structure database then available, but we attempted, by a threading method, to recognize similarity to known domain folds. Four of the 10 proteins, as we subsequently learned, do indeed show significant similarity to then-known structures. For 2 of these proteins the predictions were accurate, in the sense that a similar structure was at or near the top of the list of threading scores, and the threading alignment agreed well with the corresponding structural alignment. For the best predicted model mean alignment error relative to the optimal structural alignment was 2.7 residues, arising entirely from small “register shifts” of strands or helices. In the analysis we attempt to identify factors responsible for these successes and failures. Since our threading method does not use gap penalties, we may readily distinguish between errors arising from our prior definition of the “cores” of known structures and errors arising from inherent limitations in the threading potential. It would appear from the results that successful substructure recognition depends most critically on accurate definition of the “fold” of a database protein. This definition must correctly delineate substructures that are, and are not, likely to be conserved during protein evolution. © 1995 Wiley-Liss, Inc.     

Statistical analysis of domains in interacting protein pairs

MOTIVATION: Several methods have recently been developed to analyse large-scale sets of physical interactions between proteins in terms of physical contacts between the constituent domains, often with a view to predicting new pairwise interactions. Our aim is to combine genomic interaction data, in which domain-domain contacts are not explicitly reported, with the domain-level structure of individual proteins, in order to learn about the structure of interacting protein pairs. Our approach is driven by the need to assess the evidence for physical contacts between domains in a statistically rigorous way. RESULTS: We develop a statistical approach that assigns p-values to pairs of domain superfamilies, measuring the strength of evidence within a set of protein interactions that domains from these superfamilies form contacts. A set of p-values is calculated for SCOP superfamily pairs, based on a pooled data set of interactions from yeast. These p-values can be used to predict which domains come into contact in an interacting protein pair. This predictive scheme is tested against protein complexes in the Protein Quaternary Structure (PQS) database, and is used to predict domain-domain contacts within 705 interacting protein pairs taken from our pooled data set.     

Dynameomics: Data‐driven methods and models for utilizing large‐scale protein structure repositories for improving fragment‐based loop prediction

Protein function is intimately linked to protein structure and dynamics yet experimentally determined structures frequently omit regions within a protein due to indeterminate data, which is often due protein dynamics. We propose that atomistic molecular dynamics simulations provide a diverse sampling of biologically relevant structures for these missing segments (and beyond) to improve structural modeling and structure prediction. Here we make use of the Dynameomics data warehouse, which contains simulations of representatives of essentially all known protein folds. We developed novel computational methods to efficiently identify, rank and retrieve small peptide structures, or fragments, from this database. We also created a novel data model to analyze and compare large repositories of structural data, such as contained within the Protein Data Bank and the Dynameomics data warehouse. Our evaluation compares these structural repositories for improving loop predictions and analyzes the utility of our methods and models. Using a standard set of loop structures, containing 510 loops, 30 for each loop length from 4 to 20 residues, we find that the inclusion of Dynameomics structures in fragment‐based methods improves the quality of the loop predictions without being dependent on sequence homology. Depending on loop length, ～25–75% of the best predictions came from the Dynameomics set, resulting in lower main chain root‐mean‐square deviations for all fragment lengths using the combined fragment library. We also provide specific cases where Dynameomics fragments provide better predictions for NMR loop structures than fragments from crystal structures. Online access to these fragment libraries is available at http://www.dynameomics.org/fragments .     

Profile-profile comparisons by COMPASS predict intricate homologies between protein families

Recently we proposed a novel method of alignment-alignment comparison, COMPASS (the tool for COmparison of Multiple Protein Alignments with Assessment of Statistical Significance). Here we present several examples of the relations between PFAM protein families that were detected by COMPASS and that lead to the predictions of presently unresolved protein structures. We discuss relatively straightforward COMPASS predictions that are new and interesting to us, and that would require a substantial time and effort to justify even for a skilled PSI-BLAST user. All of the presented COMPASS hits are independently confirmed by other methods, including the ab initio structure-prediction method ROSETTA. The tertiary structure predictions made by ROSETTA proved to be useful for improving sequence-derived alignments, because they are based on a reasonable folding of the polypeptide chain rather than on the information from sequence databases. The ability of COMPASS to predict new relations within the PFAM database indicates the high sensitivity of COMPASS searches and substantiates its potential value for the discovery of previously unknown similarities between protein families.     

The impact of structural genomics: the first quindecennial

The period 2000–2015 brought the advent of high-throughput approaches to protein structure determination. With the overall funding on the order of $2 billion (in 2010 dollars), the structural genomics (SG) consortia established worldwide have developed pipelines for target selection, protein production, sample preparation, crystallization, and structure determination by X-ray crystallography and NMR. These efforts resulted in the determination of over 13,500 protein structures, mostly from unique protein families, and increased the structural coverage of the expanding protein universe. SG programs contributed over 4400 publications to the scientific literature. The NIH-funded Protein Structure Initiatives alone have produced over 2000 scientific publications, which to date have attracted more than 93,000 citations. Software and database developments that were necessary to handle high-throughput structure determination workflows have led to structures of better quality and improved integrity of the associated data. Organized and accessible data have a positive impact on the reproducibility of scientific experiments. Most of the experimental data generated by the SG centers are freely available to the community and has been utilized by scientists in various fields of research. SG projects have created, improved, streamlined, and validated many protocols for protein production and crystallization, data collection, and functional analysis, significantly benefiting biological and biomedical research.     

3D complex: a structural classification of protein complexes

Most of the proteins in a cell assemble into complexes to carry out their function. It is therefore crucial to understand the physicochemical properties as well as the evolution of interactions between proteins. The Protein Data Bank represents an important source of information for such studies, because more than half of the structures are homo- or heteromeric protein complexes. Here we propose the first hierarchical classification of whole protein complexes of known 3-D structure, based on representing their fundamental structural features as a graph. This classification provides the first overview of all the complexes in the Protein Data Bank and allows nonredundant sets to be derived at different levels of detail. This reveals that between one-half and two-thirds of known structures are multimeric, depending on the level of redundancy accepted. We also analyse the structures in terms of the topological arrangement of their subunits and find that they form a small number of arrangements compared with all theoretically possible ones. This is because most complexes contain four subunits or less, and the large majority are homomeric. In addition, there is a strong tendency for symmetry in complexes, even for heteromeric complexes. Finally, through comparison of Biological Units in the Protein Data Bank with the Protein Quaternary Structure database, we identified many possible errors in quaternary structure assignments. Our classification, available as a database and Web server at http://www.3Dcomplex.org, will be a starting point for future work aimed at understanding the structure and evolution of protein complexes.     

Prediction of protein interactions is essential for studying biomolecular mechanisms

Structural characterization of protein interactions is essential for our ability to understand and modulate physiological processes. Computational approaches to modeling of protein complexes provide structural information that far exceeds capabilities of the existing experimental techniques. Protein structure prediction in general, and prediction of protein interactions in particular, has been revolutionized by the rapid progress in Deep Learning techniques. The work of Schweke et al. (Proteomics 2023, 23, 2200323) presents a community-wide study of an important problem of distinguishing physiological protein–protein complexes/interfaces (experimentally determined or modeled) from non-physiological ones. The authors designed and generated a large benchmark set of physiological and non-physiological homodimeric complexes, and evaluated a large set of scoring functions, as well as AlphaFold predictions, on their ability to discriminate the non-physiological interfaces. The problem of separating physiological interfaces from non-physiological ones is very difficult, largely due to the lack of a clear distinction between the two categories in a crowded environment inside a living cell. Still, the ability to identify key physiologically significant interfaces in the variety of possible configurations of a protein–protein complex is important. The study presents a major data resource and methodological development in this important direction for molecular and cellular biology.     

A Physical Approach to Protein Structure Prediction

We describe our global optimization method called Stochastic Perturbation with Soft Constraints (SPSC), which uses information from known proteins to predict secondary structure, but not in the tertiary structure predictions or in generating the terms of the physics-based energy function. Our approach is also characterized by the use of an all atom energy function that includes a novel hydrophobic solvation function derived from experiments that shows promising ability for energy discrimination against misfolded structures. We present the results obtained using our SPSC method and energy function for blind prediction in the 4^th Critical Assessment of Techniques for Protein Structure Prediction competition, and show that our approach is more effective on targets for which less information from known proteins is available. In fact our SPSC method produced the best prediction for one of the most difficult targets of the competition, a new fold protein of 240 amino acids.     

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.     

The PCDDB (Protein Circular Dichroism Data Bank): A Bioinformatics Resource for Protein Characterisations and Methods Development

The Protein Circular Dichroism Data Bank (PCDDB) [https://pcddb.cryst.bbk.ac.uk] is an established resource for the biological, biophysical, chemical, bioinformatics, and molecular biology communities. It is a freely-accessible repository of validated protein circular dichroism (CD) spectra and associated sample and metadata, with entries having links to other bioinformatics resources including, amongst others, structure (PDB), AlphaFold, and sequence (UniProt) databases, as well as to published papers which produced the data and cite the database entries. It includes primary (unprocessed) and final (processed) spectral data, which are available in both text and pictorial formats, as well as detailed sample and validation information produced for each of the entries. Recently the metadata content associated with each of the entries, as well as the number and structural breadth of the protein components included, have been expanded. The PCDDB includes data on both wild-type and mutant proteins, and because CD studies primarily examine proteins in solution, it also contains examples of the effects of different environments on their structures, plus thermal unfolding/folding series. Methods for both sequence and spectral comparisons are included.The data included in the PCDDB complement results from crystal, cryo-electron microscopy, NMR spectroscopy, bioinformatics characterisations and classifications, and other structural information available for the proteins via links to other databases. The entries in the PCDDB have been used for the development of new analytical methodologies, for interpreting spectral and other biophysical data, and for providing insight into structures and functions of individual soluble and membrane proteins and protein complexes.