A structural biology community assessment of AlphaFold2 applications

Structural biology of single-stranded, positive-sense RNA viruses in the age of accurate atomic-scale predictions of protein structures

For decades atomic structures of proteins could only be determined experimentally and at a very slow pace. This was a particular problem for RNA viruses, for which sequences diverge fast and horizontal transfers are common. This made modeling from known structures difficult and uncertain. Only hard experimental structural data could allow accurate atomic descriptions of viral proteins and subsequent analyses, from mutant phenotype prediction to drug design. This has changed. With the advent of AlphaFold, that allows accurate protein structure prediction from sequence only, it is now possible in most cases to have the structure of a new protein of interest in a matter of minutes. In this mini review we focus on important consequences of this new state of affairs. While most of our conclusions are likely relevant to RNA viruses in general, here we focus on single-stranded, positive-sense RNA viruses. Taking as case studies proteins that are studied in our lab, we highlight why these viruses generally encode proteins that are particularly tough cases, being membrane-associated proteins with alternate conformations, structures, and interactions that may not be conserved even between close relatives. For these proteins AlphaFold may even fail or at least mislead, but with a proper approach it may also allow jump-starting the study of difficult or understudied viruses.

The novel protein C variant p.C101F results in early intracellular degradation that drives type I protein C deficiency

Hereditary protein C (PC) deficiency is an inherited thrombophilic disorder caused by variants in the PC gene (PROC). We identified a novel PROC variant, c.302G>T, p.Cys101Phe (C101F), in a patient with type I PC deficiency. We analyzed the intracellular dynamics of the C101F variant of PC (PC-C101F) to elucidate the pathogenic mechanism underlying this condition. Wild-type PC (PC-WT) and PC-C101F were transiently expressed in HEK293 cells for expression and functional analyses. The PC antigen levels in the cell lysate and culture supernatant of PC-C101F-expressing cells were significantly lower than those of PC-WT-expressing cells. In cycloheximide (CHX) chase experiments, the intracellular PC antigen level gradually decreased in PC-C101F-expressing cells, but remained stable at 0 and 6 h in the presence of CHX/MG132. No significant difference in co-localization with the endoplasmic reticulum was observed between PC-C101F and PC-WT. 101Cys forms a disulfide bond with 106Cys, which is crucial for maintaining the conformation of PC. PC-C101F likely results in protein misfolding and proteasomal degradation, leading to type I PC deficiency. These findings highlight the importance of cysteine residues in the three-dimensional structure of PC and provide insight into the mechanism of type I PC deficiency.

High-Throughput Evaluation of Natural Diversity of F-Type ATP Synthase Rotor Ring Stoichiometries

Adenosine triphosphate (ATP) synthases are large enzymes present in every living cell. They consist of a transmembrane and a soluble domain, each comprising multiple subunits. The transmembrane part contains an oligomeric rotor ring (c-ring), whose stoichiometry defines the ratio between the number of synthesized ATP molecules and the number of ions transported through the membrane. Currently, c-rings of F-Type ATP synthases consisting of 8-17 (except 16) subunits have been experimentally demonstrated, but it is not known whether other stoichiometries are present in natural organisms. Here, we present an easy-to-use high-throughput computational approach based on AlphaFold that allows us to estimate the stoichiometry of all homo-oligomeric c-rings, whose sequences are present in genomic databases. We validate the approach on the available experimental data, obtaining the correlation as high as 0.94 for the reference dataset and use it to predict the existence of c-rings with stoichiometry varying at least from 8 to 27. We then conduct molecular dynamics simulations of two c-rings with stoichiometry above 17 to corroborate the machine learning-based predictions. Our work strongly suggests existence of rotor rings with previously undescribed high stoichiometry in natural organisms and highlights the utility of AlphaFold-based approaches for studying homo-oligomeric proteins.

Hierarchy in regulator interactions with distant transcriptional activation domains empowers rheostatic regulation

Transcription factors carry long intrinsically disordered regions often containing multiple activation domains. Despite numerous recent high-throughput identifications and characterizations of activation domains, the interplay between sequence motifs, activation domains, and regulator binding in intrinsically disordered transcription factor regions remains unresolved. Here, we map sequence motifs and activation domains in an Arabidopsis thaliana NAC transcription factor clade, revealing that although sequence motifs and activation domains often coincide, no systematic overlap exists. Biophysical analyses using NMR spectroscopy show that the long intrinsically disordered region of senescence-associated transcription factor ANAC046 is devoid of residual structure. We identify two activation domain/sequence motif regions, one at each end that both bind a panel of six positive and negative regulator domains from biologically relevant regulators promiscuously. Binding affinities measured using isothermal titration calorimetry reveal a hierarchy for regulator binding of the two ANAC046 activation domain/sequence motif regions defining these as regulatory hotspots. Despite extensive dynamic intramolecular contacts along the disordered chain revealed using paramagnetic relaxation enhancement experiments and simulations, the regions remain uncoupled in binding. Together, the results imply rheostatic regulation by ANAC046 through concentration-dependent regulator competition, a mechanism likely mirrored in other transcription factors with distantly located activation domains.

A quantitative comparison of the deleteriousness of missense and nonsense mutations using the structurally resolved human protein interactome

The complex genotype-to-phenotype relationships in Mendelian diseases can be elucidated by mutation-induced disturbances to the networks of molecular interactions (interactomes) in human cells. Missense and nonsense mutations cause distinct perturbations within the human protein interactome, leading to functional and phenotypic effects with varying degrees of severity. Here, we structurally resolve the human protein interactome at atomic-level resolutions and perform structural and thermodynamic calculations to assess the biophysical implications of these mutations. We focus on a specific type of missense mutation, known as "quasi-null" mutations, which destabilize proteins and cause similar functional consequences (node removal) to nonsense mutations. We propose a "fold difference" quantification of deleteriousness, which measures the ratio between the fractions of node-removal mutations in datasets of Mendelian disease-causing and non-pathogenic mutations. We estimate the fold differences of node-removal mutations to range from 3 (for quasi-null mutations with folding ΔΔG ≥2 kcal/mol) to 20 (for nonsense mutations). We observe a strong positive correlation between biophysical destabilization and phenotypic deleteriousness, demonstrating that the deleteriousness of quasi-null mutations spans a continuous spectrum, with nonsense mutations at the extreme (highly deleterious) end. Our findings substantiate the disparity in phenotypic severity between missense and nonsense mutations and suggest that mutation-induced protein destabilization is indicative of the phenotypic outcomes of missense mutations. Our analyses of node-removal mutations allow for the potential identification of proteins whose removal or destabilization lead to harmful phenotypes, enabling the development of targeted therapeutic approaches, and enhancing comprehension of the intricate mechanisms governing genotype-to-phenotype relationships in clinically relevant diseases.

AI-Assisted Protein-Peptide Complex Prediction in a Practical Setting

Accurate prediction of protein-peptide complex structures plays a critical role in structure-based drug design, including antibody design. Most peptide-docking benchmark studies were conducted using crystal structures of protein-peptide complexes; as such, the performance of the current peptide docking tools in the practical setting is unknown. Here, the practical setting implies there are no crystal or other experimental structures for the complex, nor for the receptor and peptide. In this work, we have developed a practical docking protocol that incorporated two famous machine learning models, AlphaFold 2 for structural prediction and ANI-2x for ab initio potential prediction, to achieve a high success rate in modeling protein-peptide complex structures. The docking protocol consists of three major stages. In the first stage, the 3D structure of the receptor is predicted by AlphaFold 2 using the monomer mode, and that of the peptide is predicted by AlphaFold 2 using the multimer mode. We found that it is essential to include the receptor information to generate a high-quality 3D structure of the peptide. In the second stage, rigid protein-peptide docking is performed using ZDOCK software. In the last stage, the top 10 docking poses are relaxed and refined by ANI-2x in conjunction with our in-house geometry optimization algorithm-conjugate gradient with backtracking line search (CG-BS). CG-BS was developed by us to more efficiently perform geometry optimization, which takes the potential and force directly from ANI-2x machine learning models. The docking protocol achieved a very encouraging performance for a set of 62 very challenging protein-peptide systems which had an overall success rate of 34% if only the top 1 docking poses were considered. This success rate increased to 45% if the top 3 docking poses were considered. It is emphasized that this encouraging protein-peptide docking performance was achieved without using any crystal or experimental structures.

Analysing protein complexes in plant science: insights and limitation with AlphaFold 3

AlphaFold 3 (AF3), an artificial intelligence (AI)-based software for protein complex structure prediction, represents a significant advancement in structural biology. Its flexibility and enhanced scalability have unlocked new applications in various fields, specifically in plant science, including improving crop resilience and predicting the structures of plant-specific proteins involved in stress responses, signalling pathways, and immune responses. Comparisons with existing tools, such as ClusPro and AlphaPulldown, highlight AF3's unique strengths in sequence-based interaction predictions and its greater adaptability to various biomolecular structures. However, limitations persist, including challenges in modelling large complexes, protein dynamics, and structures from underrepresented plant proteins with limited evolutionary data. Additionally, AF3 encounters difficulties in predicting mutation effects on protein interactions and DNA binding, which can be improved with molecular dynamics and experimental validation. This review presents an overview of AF3's advancements, using examples in plant and fungal research, and comparisons with existing tools. It also discusses current limitations and offers perspectives on integrating molecular dynamics and experimental validation to enhance its capabilities.

Prediction of secreted uncharacterized protein structures from Beauveria bassiana ARSEF 2860 unravels novel toxins-like families

Insecticides are toxic substances used to control a wide variety of agricultural insect pests. Most of these are chemicals in nature, and their increasing residues in soil, water, and fruits contribute to environmental pollution, chronic human illnesses, and the emergence of insecticide resistance phenomenon. In the context of a green environment, bioinsecticide metabolites, including proteins, are a safe alternative that mostly has selective toxicity to insects. Thus, this study aimed to predict and identify new toxin-like families through uncharacterized secreted proteins from one of the most potent entomopathogenic fungi, Beauveria bassiana ARSEF 2860, which was selected as a model. In this work, a total of 2483 amino acid sequences of uncharacterized proteins (Ups) were retrieved from the RefSeq database. Among these, 365 UPs were identified as secreted proteins using the SignalP web server. We implemented the integration of well-designed bioinformatic tools to characterize and anticipate their homologous similarities at the sequence (InterPro) and structural (AlphaFold2) levels. The structural function annotation of these proteins was predicted using DeepFRI. With 269 successfully predicted folds, we identified new putative families with pathogenesis functions related to toxins like Janus-faced atracotoxins (insecticidal spider toxin), Cry toxins (commercial insecticide from Bacillus thuringiensis), ARTs-like toxins, and other insecticidal toxins. Furthermore, some proteins that are not homologous to any known experimental data were functionally predicted as cation metal ion binding (Zn, Na, and Co) with potential toxicity. Collectively, computational structural genomics can be used to study host-pathogen interactions and predict novel families.

Unveiling the influence of fastest nobel prize winner discovery: alphafold's algorithmic intelligence in medical sciences

CONTEXT

AlphaFold's advanced AI technology has transformed protein structure interpretation. By predicting three-dimensional protein structures from amino acid sequences, AlphaFold has solved the complex protein-folding problem, previously challenging for experimental methods due to numerous possible conformations. Since its inception, AlphaFold has introduced several versions, including AlphaFold2, AlphaFold DB, AlphaFold Multimer, Alpha Missense, and AlphaFold3, each further enhancing protein structure prediction. Remarkably, AlphaFold is recognized as the fastest Nobel Prize winner in science history. This technology has extensive applications, potentially transforming treatment and diagnosis in medical sciences by reducing drug design costs and time, while elucidating structural pathways of human body systems. Numerous studies have demonstrated how AlphaFold aids in understanding health conditions by providing critical information about protein mutations, abnormal protein-protein interactions, and changes in protein dynamics. Researchers have also developed new technologies and pipelines using different versions of AlphaFold to amplify its potential. However, addressing existing limitations is crucial to maximizing AlphaFold's capacity to redefine medical research. This article reviews AlphaFold's impact on five key aspects of medical sciences: protein mutation, protein-protein interaction, molecular dynamics, drug design, and immunotherapy.

METHODS

This review examines the contributions of various AlphaFold versions AlphaFold2, AlphaFold DB, AlphaFold Multimer, Alpha Missense, and AlphaFold3 to protein structure prediction. The methods include an extensive analysis of computational techniques and software used in interpreting and predicting protein structures, emphasizing advances in AI technology and its applications in medical research.

The human proteome with direct physical access to DNA

In a human cell, DNA is packed with histones, RNA, and chromatin-associated proteins, forming a cohesive gel. At any given moment, only a subset of the proteome has physical access to the DNA and organizes its structure, transcription, replication, repair, and other essential molecular functions. We have developed a "zero-distance" photo-crosslinking approach to quantify proteins in direct contact with DNA in living cells. Collecting DNA interactomes from human breast cancer cells, we present an atlas of over one thousand proteins with physical access to DNA and hundreds of peptide-nucleotide crosslinks pinpointing protein-DNA interfaces with single-amino-acid resolution. Quantitative comparisons of DNA interactomes from differentially treated cells recapitulate the recruitment of key transcription factors as well as DNA repair proteins and uncover fast-acting restrictors of chromatin accessibility on a timescale of minutes. This opens a direct way to explore genomic regulation in a hypothesis-free manner, applicable to many organisms and systems.

Capturing the Conformational Heterogeneity of HSPB1 Chaperone Oligomers at Atomic Resolution

Small heat shock proteins (sHSPs), including HSPB1, are essential regulators of cellular proteostasis that interact with unfolded and partially folded proteins to prevent aberrant misfolding and aggregation. These proteins fulfill a similar role in biological condensates, where they interact with intrinsically disordered proteins to modulate their liquid-liquid and liquid-to-solid phase transitions. Characterizing the sHSP structure, dynamics, and client interactions is challenging due to their partially disordered nature, their tendency to form polydisperse oligomers, and their diverse range of clients. In this work, we leverage various biophysical methods, including fast 1H-based magic angle spinning (MAS) NMR spectroscopy, molecular dynamics (MD) simulations, and modeling, to shed new light on the structure and dynamics of HSPB1 oligomers. Using split-intein-mediated segmental labeling, we provide unambiguous evidence that in the oligomer context, the N-terminal domain (NTD) of HSPB1 is rigid and adopts an ensemble of heterogeneous conformations, the α-Crystallin domain (ACD) forms dimers and experiences multiple distinct local environments, while the C-terminal domain (CTD) remains highly dynamic. Our computational models suggest that the NTDs participate in extensive NTD-NTD and NTD-ACD interactions and are sequestered within the oligomer interior. We further demonstrate that HSPB1 higher order oligomers disassemble into smaller oligomeric species in the presence of a client protein and that an accessible NTD is essential for HSPB1 partitioning into condensates and interactions with client proteins. Our integrated approach provides a high-resolution view of the complex oligomeric landscape of HSPB1 and sheds light on the elusive network of interactions that underlies the function of HSPB1 in biological condensates.

Structural Biology in the AlphaFold Era: How Far Is Artificial Intelligence from Deciphering the Protein Folding Code?

Proteins are biomolecules characterized by uncommon chemical and physicochemical complexities coupled with extreme responsiveness to even minor chemical modifications or environmental variations. Since the shape that proteins assume is fundamental for their function, understanding the chemical and structural bases that drive their three-dimensional structures represents the central problem for an atomic-level interpretation of biology. Not surprisingly, this question has progressively become the Holy Grail of structural biology (the folding problem). From this perspective, we initially describe and discuss the different formulations of the folding problem. In the present manuscript, the folding problem is framed from a historical perspective, effectively highlighting the progress made in the last lustrum. We chronologically summarize the major contributions that traditional methodologies provide in approaching this multifaceted problem. We then describe the recent advent and evolution of predictive approaches based on machine learning techniques that are revolutionizing the field by pointing out the potentialities and limitations of this approach. In the final part of the perspective, we illustrate the contribution that computational approaches will make in current structural biology to overcome the limitations of the reductionist approach of studying individual molecules to afford the atomic-level characterization of entire cellular compartments.

ACMG/AMP interpretation of BRCA1 missense variants: Structure-informed scores add evidence strength granularity to the PP3/BP4 computational evidence

Classification of missense variants is challenging. Lacking compelling clinical and/or functional data, ACMG/AMP lines of evidence are restricted to PM2 (rarity code applied at supporting level) and PP3/BP4 (computational evidence based mostly on multiple-sequence-alignment conservation tools). Currently, the ClinGen ENIGMA BRCA1/2 Variant Curation Expert Panel uses BayesDel to apply PP3/BP4 to missense variants located in the BRCA1 RING/BRCT domains. The ACMG/AMP framework does not refer explicitly to protein structure as a putative source of pathogenic/benign evidence. Here, we tested the value of incorporating structure-based evidence such as relative solvent accessibility (RSA), folding stability (ΔΔG), and/or AlphaMissense pathogenicity to the classification of BRCA1 missense variants. We used MAVE functional scores as proxies for pathogenicity/benignity. We computed RSA and FoldX5.0 ΔΔG predictions using as alternative input templates for either PDB files or AlphaFold2 models, and we retrieved pre-computed AlphaMissense and BayesDel scores. We calculated likelihood ratios toward pathogenicity/benignity provided by the tools (individually or combined). We performed a clinical validation of major findings using the large-scale BRIDGES case-control dataset. AlphaMissense outperforms ΔΔG and BayesDel, providing similar PP3/BP4 evidence strengths with lower rate of variants in the uninformative score range. AlphaMissense combined with ΔΔG increases evidence strength granularity. AlphaFold2 models perform well as input templates for ΔΔG predictions. Regardless of the tool, BP4 (but not PP3) is highly dependent on RSA, with benignity evidence provided only to variants targeting buried or partially buried residues (RSA ≤ 60%). Stratification by functional domain did not reveal major differences. In brief, structure-based analysis improves PP3/BP4 assessment, uncovering a relevant role for RSA.

Precision proteogenomics reveals pan-cancer impact of germline variants

We investigate the impact of germline variants on cancer patients' proteomes, encompassing 1,064 individuals across 10 cancer types. We introduced an approach, "precision peptidomics," mapping 337,469 coding germline variants onto peptides from patients' mass spectrometry data, revealing their potential impact on post-translational modifications, protein stability, allele-specific expression, and protein structure by leveraging the relevant protein databases. We identified rare pathogenic and common germline variants in cancer genes potentially affecting proteomic features, including variants altering protein abundance and structure and variants in kinases (ERBB2 and MAP2K2) impacting phosphorylation. Precision peptidome analysis predicted destabilizing events in signal-regulatory protein alpha (SIRPA) and glial fibrillary acid protein (GFAP), relevant to immunomodulation and glioblastoma diagnostics, respectively. Genome-wide association studies identified quantitative trait loci for gene expression and protein levels, spanning millions of SNPs and thousands of proteins. Polygenic risk scores correlated with distal effects from risk variants. Our findings emphasize the contribution of germline genetics to cancer heterogeneity and high-throughput precision peptidomics.

Structural biology in variant interpretation: Perspectives and practices from two studies

Structural biology offers a powerful lens through which to assess genetic variants by providing insights into their impact on clinically relevant protein structure and function. Due to the availability of new, user-friendly, web-based tools, structural analyses by wider audiences have become more mainstream. These new tools, including AlphaMissense and AlphaFold, have recently been in the limelight due to their initial success and projected future promise; however, the intricacies and limitations of using these tools still need to be disseminated to the more general audience that is likely to use them in variant analysis. Here, we expound on frameworks applying structural biology to variant interpretation by examining two accompanying articles. To this end, we explore the nuances of choosing the correct protein model, compare and contrast various structural approaches, and highlight both the advantages and limitations of employing structural biology in variant interpretation. Using two articles published in this issue of The American Journal of Human Genetics as a baseline, we focus on case studies in TP53 and BRCA1 to illuminate gene-specific differences in the applications of structural information, which illustrate the complexities inherent in this field. Additionally, we discuss the implications of recent advancements, such as AlphaFold, and provide practical guidance for researchers navigating variant interpretation using structural biology.

Leveraging AlphaFold2 and residual dipolar couplings for side-chain methyl group assignment: A case study with S. cerevisiae Xrs2

Side-chain methyl group NMR spectroscopy provides invaluable insights into macromolecular structure, dynamics, and function, particularly for large biomolecular complexes. Accurate assignment of methyl group resonances in two-dimensional spectra is essential for structural and dynamics studies. Traditional methyl group assignment strategies rely on either transferring assignments from backbone resonance data or NOESY data and high-resolution experimental structures; however, these methods are often limited by molecular size or availability of structural information, respectively. Here, we describe the use of AlphaFold2 structural models as a basis for the manual, distance-based assignment of side-chain methyl group resonances in the folded domains of S. cerevisiae Xrs2. While AlphaFold2 models facilitated initial assignments for the methyl resonances, inaccuracies in the side-chain coordinates highlighted the need for improved structural models. By generating >500 ColabFold-derived models and filtering with methyl residual dipolar couplings (RDCs), we identified structural models with superior agreement to experimental data. These refined models enabled additional methyl group assignments while suggesting an iterative approach to simultaneously improve structure prediction and resonance assignment. Our findings outline a workflow that integrates machine learning-based structural predictions with experimental NMR data, offering a pathway for advancing methyl group assignment in systems lacking high-resolution experimental structures.

Engineering cardiolipin binding to an artificial membrane protein reveals determinants for lipid-mediated stabilization

Integral membrane proteins carry out essential functions in the cell, and their activities are often modulated by specific protein-lipid interactions in the membrane. Here, we elucidate the intricate role of cardiolipin (CDL), a regulatory lipid, as a stabilizer of membrane proteins and their complexes. Using the in silico-designed model protein TMHC4_R (ROCKET) as a scaffold, we employ a combination of molecular dynamics simulations and native mass spectrometry to explore the protein features that facilitate preferential lipid interactions and mediate stabilization. We find that the spatial arrangement of positively charged residues as well as local conformational flexibility are factors that distinguish stabilizing from non-stabilizing CDL interactions. However, we also find that even in this controlled, artificial system, a clear-cut distinction between binding and stabilization is difficult to attain, revealing that overlapping lipid contacts can partially compensate for the effects of binding site mutations. Extending our insights to naturally occurring proteins, we identify a stabilizing CDL site within the E. coli rhomboid intramembrane protease GlpG and uncover its regulatory influence on enzyme substrate preference. In this work, we establish a framework for engineering functional lipid interactions, paving the way for the design of proteins with membrane-specific properties or functions.

NMR Pure Shift Spectroscopy and Its Potential Applications in the Pharmaceutical Industry

1H nuclear magnetic resonance (NMR) spectroscopy plays an important role in the pharmaceutical industry, but for complex substances, spectral analysis is challenging due to the narrow chemical shift range and signal splitting caused by scalar coupling. Pure shift techniques can suppress scalar coupling, improving spectral resolution. This article provides a review of pure shift techniques, including the main homonuclear broadband decoupling experiments and the methods for obtaining optimal pure shift spectra with the assistance of deep learning. Furthermore, it explores the potential application directions of pure shift techniques in the pharmaceutical industry, supported by relevant scientific examples. By summarizing recent advances and application opportunities, this article aims to promote the development and practical implementation of pure shift NMR techniques in the pharmaceutical industry.

Assessing Structural Classification Using AlphaFold2 Models Through ECOD-Based Comparative Analysis

Identifying homologous proteins is a fundamental task in structural bioinformatics. While AlphaFold2 has revolutionized protein structure prediction, the extent to which structure comparison of its models can reliably detect homologs remains unclear. In this study, we evaluate the feasibility of homology detection using AlphaFold2-predicted structures through structural comparisons. We considered the classification of the ECOD database for experimental structures as the correct standard and obtained their corresponding predicted models from AlphaFoldDB. To ensure blind assessment, we divided the structures into test and train sets according to their release date. Predicted and experimental 3D structures in the test and train sets were compared using 3D structure comparisons (MATRAS, Dali, and Foldseek) and sequence comparisons (BLAST and HHsearch). The results were evaluated based on the homology annotations in the ECOD database. For top-1 accuracy, the performance of structural comparisons was comparable to that of HHsearch. However, when considering metrics that included all structural pairs, including more remote homology, structural comparisons outperformed HHsearch. No significant differences were observed between comparisons of experimental versus experimental, predicted versus experimental, and predicted versus predicted structures with pLDDT (prediction confidence) values greater than 60. We also demonstrate that predicted protein structures, determined by NMR, had lower pLDDT values and contained fewer coils than their experimental counterparts. These findings highlight the potential of AlphaFold2 models in structural classification and suggest that 3D structural searches should be conducted not only against the PDB but also against AlphaFoldDB to identify more potential homologs.

Widespread variation in molecular interactions and regulatory properties among transcription factor isoforms

Most human transcription factor (TF) genes encode multiple protein isoforms differing in DNA-binding domains, effector domains, or other protein regions. The global extent to which this results in functional differences between isoforms remains unknown. Here, we systematically compared 693 isoforms of 246 TF genes, assessing DNA binding, protein binding, transcriptional activation, subcellular localization, and condensate formation. Relative to reference isoforms, two-thirds of alternative TF isoforms exhibit differences in one or more molecular activities, which often could not be predicted from sequence. We observed two primary categories of alternative TF isoforms: "rewirers" and "negative regulators," both of which were associated with differentiation and cancer. Our results support a model wherein the relative expression levels of, and interactions involving, TF isoforms add an understudied layer of complexity to gene regulatory networks, demonstrating the importance of isoform-aware characterization of TF functions and providing a rich resource for further studies.

Bias in, bias out - AlphaFold-Multimer and the structural complexity of protein interfaces

A structural understanding of protein-protein interactions is a key component of many facets of applied molecular biology research. AlphaFold-Multimer (AF-MM) provided a breakthrough in the ability to predict protein-protein interface structure. However, the available training data for this model and the resulting benchmarking and validation efforts show a bias toward interactions between more ordered regions of proteins. Here we highlight some of the successes and limitations of AF-MM and discuss available methods and future directions to enable balanced prediction of all interface types.

Deep learning and its applications in nuclear magnetic resonance spectroscopy

Nuclear Magnetic Resonance (NMR), as an advanced technology, has widespread applications in various fields like chemistry, biology, and medicine. However, issues such as long acquisition times for multidimensional spectra and low sensitivity limit the broader application of NMR. Traditional algorithms aim to address these issues but have limitations in speed and accuracy. Deep Learning (DL), a branch of Artificial Intelligence (AI) technology, has shown remarkable success in many fields including NMR. This paper presents an overview of the basics of DL and current applications of DL in NMR, highlights existing challenges, and suggests potential directions for improvement.

Chromatin structures from integrated AI and polymer physics model

The physical organization of the genome in three-dimensional space regulates many biological processes, including gene expression and cell differentiation. Three-dimensional characterization of genome structure is critical to understanding these biological processes. Direct experimental measurements of genome structure are challenging; computational models of chromatin structure are therefore necessary. We develop an approach that combines a particle-based chromatin polymer model, molecular simulation, and machine learning to efficiently and accurately estimate chromatin structure from indirect measures of genome structure. More specifically, we introduce a new approach where the interaction parameters of the polymer model are extracted from experimental Hi-C data using a graph neural network (GNN). We train the GNN on simulated data from the underlying polymer model, avoiding the need for large quantities of experimental data. The resulting approach accurately estimates chromatin structures across all chromosomes and across several experimental cell lines despite being trained almost exclusively on simulated data. The proposed approach can be viewed as a general framework for combining physical modeling with machine learning, and it could be extended to integrate additional biological data modalities. Ultimately, we achieve accurate and high-throughput estimations of chromatin structure from Hi-C data, which will be necessary as experimental methodologies, such as single-cell Hi-C, improve.

Structural Biology of RNA and Protein-RNA Complexes after AlphaFold3

Recent breakthroughs in AI-mediated protein structure prediction have significantly accelerated research and generated valuable hypotheses within the field of structural biology and beyond. Notably, AlphaFold2 has facilitated the determination of larger protein complexes for which only limited experimental data are available. De novo predictions can now be experimentally validated with relative ease compared to the pre-AlphaFold2 era. In May 2024, AlphaFold3 was launched with high expectations, promising the capability to accurately predict RNA structures and protein-RNA complexes - features that were absent in AlphaFold2. This review evaluates the extent to which AlphaFold3 fulfills this promise through specific examples. At present, AlphaFold3 falls short in reliably predicting RNA and protein-RNA complex structures, particularly for non-canonical interactions where training data remain scarce. As a result, users should exercise caution when using AlphaFold3 predictions as hypotheses generators for RNA and protein-RNA complex structures. In the interim, integrating AI-based predictors with data-driven docking tools is recommended to address these limitations. This approach can help bridge the gap until sufficient training data are available to enable the development of more reliable predictive algorithms.

Navigating the unstructured by evaluating alphafold's efficacy in predicting missing residues and structural disorder in proteins

The study investigated regions with undefined structures, known as "missing" segments in X-ray crystallography and cryo-electron microscopy (Cryo-EM) data, by assessing their predicted structural confidence and disorder scores. Utilizing a comprehensive dataset from the Protein Data Bank (PDB), residues were categorized as "modeled", "hard missing" and "soft missing" based on their visibility in structural datasets. Key features were determined, including a confidence score predicted local distance difference test (pLDDT) from AlphaFold2, an advanced structural prediction tool, and a disorder score from IUPred, a traditional disorder prediction method. To enhance prediction performance for unstructured residues, we employed a Long Short-Term Memory (LSTM) model, integrating both scores with amino acid sequences. Notable patterns such as composition, region lengths and prediction scores were observed in unstructured residues and regions identified through structural experiments over our studied period. Our findings also indicate that "hard missing" residues often align with low confidence scores, whereas "soft missing" residues exhibit dynamic behavior that can complicate predictions. The incorporation of pLDDT, IUPred scores, and sequence data into the LSTM model has improved the differentiation between structured and unstructured residues, particularly for shorter unstructured regions. This research elucidates the relationship between established computational predictions and experimental structural data, enhancing our ability to target structurally significant areas for research and guiding experimental designs toward functionally relevant regions.

High resolution profiling of cell cycle-dependent protein and phosphorylation abundance changes in non-transformed cells

The cell cycle governs a precise series of molecular events, regulated by coordinated changes in protein and phosphorylation abundance, that culminates in the generation of two daughter cells. Here, we present a proteomic and phosphoproteomic analysis of the human cell cycle in hTERT-RPE-1 cells using deep quantitative mass spectrometry by isobaric labelling. By analysing non-transformed cells and improving the temporal resolution and coverage of key cell cycle regulators, we present a dataset of cell cycle-dependent protein and phosphorylation site oscillation that offers a foundational reference for investigating cell cycle regulation. These data reveal regulatory intricacies including proteins and phosphorylation sites exhibiting cell cycle-dependent oscillation, and proteins targeted for degradation during mitotic exit. Integrated with complementary resources, our data link cycle-dependent abundance dynamics to functional changes and are accessible through the Cell Cycle database (CCdb), an interactive web-based resource for the cell cycle community.

Deep generative modeling of temperature-dependent structural ensembles of proteins

Deep learning has revolutionized protein structure prediction, but capturing conformational ensembles and structural variability remains an open challenge. While molecular dynamics (MD) is the foundation method for simulating biomolecular dynamics, it is computationally expensive. Recently, deep learning models trained on MD have made progress in generating structural ensembles at reduced cost. However, they remain limited in modeling atomistic details and, crucially, incorporating the effect of environmental factors. Here, we present aSAM (atomistic structural autoencoder model), a latent diffusion model trained on MD to generate heavy atom protein ensembles. Unlike most methods, aSAM models atoms in a latent space, greatly facilitating accurate sampling of side chain and backbone torsion angle distributions. Additionally, we extended aSAM into the first reported transferable generator conditioned on temperature, named aSAMt. Trained on the large and open mdCATH dataset, aSAMt captures temperature-dependent ensemble properties and demonstrates generalization beyond training temperatures. By comparing aSAMt ensembles to long MD simulations of fast folding proteins, we find that high-temperature training enhances the ability of deep generators to explore energy landscapes. Finally, we also show that our MD-based aSAMt can already capture experimentally observed thermal behavior of proteins. Our work is a step towards generalizable ensemble generation to complement physics-based approaches.

DeepPath: Overcoming data scarcity for protein transition pathway prediction using physics-based deep learning

The structural dynamics of proteins play a crucial role in their function, yet most experimental and deep learning methods produce only static models. While molecular dynamics (MD) simulations provide atomistic insight into conformational transitions, they remain computationally prohibitive, particularly for large-scale motions. Here, we introduce DeepPath, a deep-learning-based framework that rapidly generates physically realistic transition pathways between known protein states. Unlike conventional supervised learning approaches, DeepPath employs active learning to iteratively refine its predictions, leveraging molecular mechanical force fields as an oracle to guide pathway generation. We validated DeepPath on three biologically relevant test cases: SHP2 activation, CdiB H1 secretion, and the BAM complex lateral gate opening. DeepPath accurately predicted the transition pathways for all test cases, reproducing key intermediate structures and transient interactions observed in previous studies. Notably, DeepPath also predicted an intermediate between the BAM inward- and outward-open states that closely aligns with an experimentally observed hybrid-barrel structure (TMscore = 0.91). Across all cases, DeepPath achieved accurate pathway predictions within hours, showcasing an efficient alternative to MD simulations for exploring protein conformational transitions.

Exploring the effects of missense mutations on protein thermodynamics through structure-based approaches: findings from the CAGI6 challenges

Missense mutations are known contributors to diverse genetic disorders, due to their subtle, single amino acid changes imparted on the resultant protein. Because of this, understanding the impact of these mutations on protein stability and function is crucial for unravelling disease mechanisms and developing targeted therapies. The Critical Assessment of Genome Interpretation (CAGI) provides a valuable platform for benchmarking state-of-the-art computational methods in predicting the impact of disease-related mutations on protein thermodynamics. Here we report the performance of our comprehensive platform of structure-based computational approaches to evaluate mutations impacting protein structure and function on 3 challenges from CAGI6: Calmodulin, MAPK1 and MAPK3. Our stability predictors have achieved correlations of up to 0.74 and AUCs of 1 when predicting changes in ΔΔG for MAPK1 and MAPK3, respectively, and AUC of up to 0.75 in the Calmodulin challenge. Overall, our study highlights the importance of structure-based approaches in understanding the effects of missense mutations on protein thermodynamics. The results obtained from the CAGI6 challenges contribute to the ongoing efforts to enhance our understanding of disease mechanisms and facilitate the development of personalised medicine approaches.

Transition to Piscivory Seen Through Brain Transcriptomics in a Juvenile Percid Fish: Complex Interplay of Differential Gene Transcription, Alternative Splicing, and ncRNA Activity

Pikeperch (Sander Lucioperca) belongs to main predatory fish species in freshwater bodies throughout Europe playing the key role by reducing planktivorous fish abundance. Two size classes of the young-of-the-year (YOY) pikeperch are known in Europe and North America. Our long-term fish survey elucidates late-summer size distribution of YOY pikeperch in the Lipno Reservoir (Czechia) and recognizes two distinct subcohorts: smaller pelagic planktivores heavily outnumber larger demersal piscivores. To explore molecular mechanisms accompanying the switch from planktivory to piscivory, we compared brain transcriptomes of both subcohorts and identified 148 differentially transcribed genes. The pathway enrichment analyses identified the piscivorous phase to be associated with genes involved in collagen and extracellular matrix generation with numerous Gene Ontology (GO), while the planktivorous phase was associated with genes for non-muscle-myosins (NMM) with less GO terms. Transcripts further upregulated in planktivores from the periphery of the NMM network were Pmchl, Pomcl, and Pyyb, all involved also in appetite control and producing (an)orexigenic neuropeptides. Noncoding RNAs were upregulated in transcriptomes of planktivores including three transcripts of snoRNA U85. Thirty genes mostly functionally unrelated to those differentially transcribed were alternatively spliced between the subcohorts. Our results indicate planktivores as potentially driven by voracity to initiate the switch to piscivory, while piscivores undergo a dynamic brain development. We propose a spatiotemporal spreading of juvenile development over a longer period and larger spatial scales through developmental plasticity as an adaptation to exploiting all types of resources and decreasing the intraspecific competition.

VarMeter2: An enhanced structure-based method for predicting pathogenic missense variants through Mahalanobis distance

Various computational methods have been developed to predict the pathogenicity of missense variants, which is crucial for diagnosing rare diseases. Recently, we introduced VarMeter, a diagnostic tool for predicting variant pathogenicity based on normalized solvent-accessible surface area (nSASA) and mutation energy calculated from AlphaFold 3D models, and validated it on arylsulfatase L. To evaluate the broader applicability of VarMeter and enhance its predictive accuracy, here we analyzed 296 pathogenic and 240 benign variants extracted from the ClinVar database. By comparing structural features including nSASA, mutation energy, and predicted local distance difference test (pLDDT) score, we identified distinct characteristics between pathogenic and benign variants. These features were used to develop VarMeter2, which classifies variants based on Mahalanobis distance. VarMeter2 achieved a prediction accuracy of 82 % for the ClinVar dataset, a marked improvement over the original VarMeter (74 %), and 84 % for published missense variants of N-sulphoglucosamine sulphohydrolase (SGSH), an enzyme associated with Sanfillippo syndrome A. Application of VarMeter 2 to SGSH variants in our clinical database identified a novel SGSH variant, Q365P, as pathogenic. The recombinant Q365P protein lacked enzymatic activity as compared with wild-type SGSH. Furthermore, it was largely retained in the endoplasmic reticulum and failed to reach the Golgi, probably due to misfolding. Protein stability assays confirmed reduced stability of the variant, further explaining its loss of function. Consistently, the patient homozygous for this variant was diagnosed with Sanfilippo syndrome A. These results underscore the predictive power and versatility of VarMeter2 in assessing the pathogenicity of missense variants.

Macromolecular crystallography from an industrial perspective - the impact of synchrotron radiation on structure-based drug discovery

Structure-based drug design has been an integral part of drug discovery for over three decades, contributing to the development of numerous approved drugs. Here we discuss the evolution, as well as the current state, of structure-based drug design within the pharmaceutical industry, using data from AstraZeneca's internal repository for crystal structures to provide additional context. Over the past 20 years, the company has transitioned from a mixed in-house and synchrotron data collection model to a `synchrotron-only' approach, enabled by technological advancements at synchrotron facilities. We provide real-world examples of structure delivery to projects, including a high-throughput project and a case where a single structure was pivotal for discovering a candidate drug. We conclude that, despite recent developments in single-particle cryo-EM and deep-learning structure prediction methods, macromolecular crystallography remains a critical tool for drug discovery.

Recent advances in AI-driven protein-ligand interaction predictions

Structure-based drug discovery is a fundamental approach in modern drug development, leveraging computational models to predict protein-ligand interactions. AI-driven methodologies are significantly improving key aspects of the field, including ligand binding site prediction, protein-ligand binding pose estimation, scoring function development, and virtual screening. In this review, we summarize the recent AI-driven advances in various protein-ligand interaction prediction tasks. Traditional docking methods based on empirical scoring functions often lack accuracy, whereas AI models, including graph neural networks, mixture density networks, transformers, and diffusion models, have enhanced predictive performance. Ligand binding site prediction has been refined using geometric deep learning and sequence-based embeddings, aiding in the identification of potential druggable target sites. Binding pose prediction has evolved with sampling-based and regression-based models, as well as protein-ligand co-generation frameworks. AI-powered scoring functions now integrate physical constraints and deep learning techniques to improve binding affinity estimation, leading to more robust virtual screening strategies. Despite these advances, generalization across diverse protein-ligand pairs remains a challenge. As AI technologies continue to evolve, they are expected to revolutionize molecular docking and affinity prediction, increasing both the accuracy and efficiency of structure-based drug discovery.

Systematic characterization of indel variants using a yeast-based protein folding sensor

Gene variants resulting in insertions or deletions of amino acid residues (indels) have important consequences for evolution and are often linked to disease, yet, compared to missense variants, the effects of indels are poorly understood and predicted. We developed a sensitive protein folding sensor based on the complementation of uracil auxotrophy in yeast by circular permutated orotate phosphoribosyltransferase (CPOP). The sensor reports on the folding of disease-linked missense variants and de-novo-designed proteins. Applying the folding sensor to a saturated library of single-residue indels in human dihydrofolate reductase (DHFR) revealed that most regions that tolerate indels are confined to internal loops, the termini, and a central α helix. Several indels are temperature sensitive, and folding is rescued upon binding to methotrexate. Rosetta and AlphaFold2 predictions correlate with the observed effects, suggesting that most indels destabilize the native fold and that these computational tools are useful for the classification of indels observed in population sequencing.

Hypoxia favors tumor growth in colorectal cancer in an integrin αDβ1/hemoglobin δ-dependent manner

Low oxygen tension (PO2), characterizes the tissue environment of tumors. The colorectal tumor line Colo205, grown under reduced oxygen tension expresses a novel αDβ1 integrin, which forms a cell surface complex with hemoglobin δ. This resulted in high local affinity for oxygen, which increased cell adhesion as compared with cells grown under normal oxygen tension. Staining with antibodies to the integrin αD polypeptide and hemoglobin δ, and transfection with cDNAs for GFP-hemoglobin δ and mCherry-αD, showed co-localization of αD and hemoglobin δ. Antibodies to αD and β1 integrins, an RGD peptide, and an αDβ1 binding peptide from hemoglobin δ, blocked the αDβ1-hemoglobin interaction and lowered oxygen consumption. Downregulation of integrin αD or hemoglobin δ expression inhibited cell proliferation in hypoxia. The very frequent expression of complexes between αDβ1 and hemoglobin δ on the cell surface offers potential diagnostic and therapeutic targets in colorectal cancer.

AlphaFold and what is next: bridging functional, systems and structural biology

INTRODUCTION

The DeepMind's AlphaFold (AF) has revolutionized biomedical and biocience research by providing both experts and non-experts with an invaluable tool for predicting protein structures. However, while AF is highly effective for predicting structures of rigid and globular proteins, it is not able to fully capture the dynamics, conformational variability, and interactions of proteins with ligands and other biomacromolecules.

AREAS COVERED

In this review, we present a comprehensive overview of the latest advancements in 3D model predictions for biomacromolecules using AF. We also provide a detailed analysis its of strengths and limitations, and explore more recent iterations, modifications, and practical applications of this strategy. Moreover, we map the path forward for expanding the landscape of AF toward predicting structures of every protein and peptide, and their interactions in the proteome in the most physiologically relevant form. This discussion is based on an extensive literature search performed using PubMed and Google Scholar.

EXPERT OPINION

While significant progress has been made to enhance AF's modeling capabilities, we argue that a combined approach integrating both various in silico and in vitro methods will be most beneficial for the future of structural biology, bridging the gaps between static and dynamic features of proteins and their functions.

DisorderUnetLM: Validating ProteinUnet for efficient protein intrinsic disorder prediction

The prediction of intrinsic disorder regions has significant implications for understanding protein functions and dynamics. It can help to discover novel protein-protein interactions essential for designing new drugs and enzymes. Recently, a new generation of predictors based on protein language models (pLMs) is emerging. These algorithms reach state-of-the-art accuracy without calculating time-consuming multiple sequence alignments (MSAs). This article introduces the new DisorderUnetLM disorder predictor, which builds upon the idea of ProteinUnet. It uses the Attention U-Net convolutional network and incorporates features from the ProtTrans pLM. DisorderUnetLM achieves top results in the direct comparison with recent predictors exploiting MSAs and pLMs. Moreover, among 43 predictors on the latest CAID-2 benchmark, it ranks 1st for the NOX subset in terms of the ROC-AUC metric (0.844) and 2nd for the AP metric (0.596). For the CAID-2 PDB subset, it ranks in the top 10 (ROC-AUC of 0.924 and AP of 0.862). The code and model are publicly available and fully reproducible at doi.org/10.24433/CO.7350682.v1.

AFFIPred: AlphaFold2 structure-based Functional Impact Prediction of missense variations

Protein structure holds immense potential for pathogenicity prediction, albeit structure-based predictors are limited compared to the sequence-based counterparts due to the "structure knowledge gap" between large number of available protein sequences and relatively limited number of structures. Leveraging the highly accurate protein structures predicted by AlphaFold2 (AF2), we introduce AFFIPred, an ensemble machine learning classifier that combines sequence and AF2-based structural characteristics to predict missense variant pathogenicity. Based on the assessments on unseen datasets, AFFIPred reached a comparable level of performance with the state-of-the-art predictors such as AlphaMissense. We also showed that the recruitment of AF2 structures that are full-length and represent the unbound states ensures more precise SASA calculations compared to the recruitment of experimental structures. In line with the completeness of the AF2 structures, their use provide a more comprehensive view of the structural characteristics of the missense variation datasets by capturing all variants. AFFIPred maintains high-level accuracy without the limitations of PDB-based classifiers. AFFIPred has predicted over 210 million variations of the human proteome, which are accessible at https://affipred.timucinlab.com/.

Massively parallel barcode sequencing revealed the interchangeability of capsule transporters in Streptococcus pneumoniae

Multidrug/oligosaccharidyl-lipid/polysaccharide (MOP) family transporters are essential in glycan synthesis, flipping lipid-linked precursors across cell membranes. Yet, how they select their substrates remains enigmatic. Here, we investigate the substrate specificity of the MOP transporters in the capsular polysaccharide (CPS) synthesis pathway in Streptococcus pneumoniae. These capsule flippases collectively transport more than 100 types of capsule precursors. To determine whether they can substitute for one another, we developed a high-throughput approach to systematically examine nearly 6000 combinations of flippases and substrates. CPS flippases fall into three groups: relaxed, type-specific, and strictly specific. Cargo size and CPS acetylation affect transport, and we isolated additional gain-of-function flippase variants that can substitute for the peptidoglycan flippase YtgP (MurJ). We also showed that combining flippase variants in a single cassette allows various CPS precursors to be flipped, which may aid glycoengineering. This study reveals that MOP flippases exhibit broad specificity, shaping the evolution of glycan synthesis.

Fe(II) and 2-oxoglutarate-dependent dioxygenases for natural product synthesis: molecular insights into reaction diversity

Covering: up to 2024Fe(II) and 2-oxoglutarate-dependent dioxygenases (Fe/2OG DOs) are a superfamily of enzymes that play important roles in a variety of catalytic reactions, including hydroxylation, ring formation, ring reconstruction, desaturation, and demethylation. Each member of this family has similarities in their overall structure, but they have varying specific differences, making Fe/2OG DOs attractive for catalytic diversity. With the advancement of current research, more Fe/2OG DOs have been discovered, and their catalytic scope has been further broadened; however, apart from hydroxylation, many reaction mechanisms have not been accurately demonstrated, and there is a lack of a systematic understanding of their molecular basis. Recently, an increasing number of X-ray structures of Fe/2OG DOs have provided new insights into the structural basis of their function and substrate-binding properties. This structural information is essential for understanding catalytic mechanisms and mining potential catalytic reactions. In this review, we summarize most of the Fe/2OG DOs whose structures have been resolved in recent years, focus on their structural features, and explore the relationships between various structural elements and unique catalytic mechanisms and their associated reaction type classification.

Codeine 3-O-demethylase catalyzed biotransformation of morphinan alkaloids in Escherichia coli: site directed mutagenesis of terminal residues improves enzyme expression, stability and biotransformation yield

The cultivation of opium poppy is the only commercially viable source of most morphinan alkaloids. Bioproduction of morphinan alkaloids in recombinant whole-cell systems provides a promising alternate source of these valuable compounds. The enzyme codeine 3-O-demethylase can transform morphinan alkaloids by O-demethylation and has been applied in single step biotransformation reactions or as part of larger biosynthetic cascade, however, the productivity for these reactions remains low and suboptimal enzyme properties could be improved. This mutagenesis study targeted non-conserved N-and C-terminal residues, which were replaced with the equivalent residues from enzyme thebaine 6-O-demethylase. Whole cell biotransformation performance was significantly improved in Escherichia coli expressing codeine 3-O-demethylase mutants, with a ~ 2.8-fold increase in the production of oripavine from thebaine and ~ 1.3-fold increase in the production of morphine from codeine. Statistical analysis of biotransformation yield, enzyme expression and stability, predicted using changes in Gibbs free energy (ΔΔG) with deep-learning-based model DDmut, suggested that altered enzyme stability and/or expression of soluble protein may contribute to the observed improvements in biotransformation. This approach could be beneficial for screening future codeine 3-O-demethylase mutations and for other enzymes.

Gradations in protein dynamics captured by experimental NMR are not well represented by AlphaFold2 models and other computational metrics

The advent of accurate methods to predict the fold of proteins initiated by AlphaFold2 is rapidly changing our understanding of proteins and helping their design. However, these methods are mainly trained on protein structures determined with X-ray diffraction, where the protein is packed in crystals at often cryogenic temperatures. They can therefore only reliably cover well-folded parts of proteins that experience few, if any, conformational changes. Experimentally, solution nuclear magnetic resonance (NMR) is the experimental method of choice to gain insight into protein dynamics at near physiological conditions. Computationally, methods such as molecular dynamics (MD) simulations and Normal Mode Analysis (NMA) allow the estimation of a protein's intrinsic flexibility based on a single protein structure. This work addresses, on a large scale, the relationships for proteins between the AlphaFold2 pLDDT metric, the observed dynamics in solution from NMR metrics, interpreted MD simulations, and the computed dynamics with NMA from single AlphaFold2 models and NMR ensembles. We observe that these metrics agree well for rigid residues that adopt a single well-defined conformation, which are clearly distinct from residues that exhibit dynamic behavior and adopt multiple conformations. This direct order/disorder categorisation is reflected in the correlations observed between the parameters, but becomes very limited when considering only the likely dynamic residues. The gradations of dynamics observed by NMR in flexible protein regions are therefore not represented by these computational approaches. Our results are interactively available for each protein from https://bio2byte.be/af_nmr_nma/.

Structure-based self-supervised learning enables ultrafast protein stability prediction upon mutation

Predicting free energy changes (ΔΔG) is essential for enhancing our understanding of protein evolution and plays a pivotal role in protein engineering and pharmaceutical development. While traditional methods offer valuable insights, they are often constrained by computational speed and reliance on biased training datasets. These constraints become particularly evident when aiming for accurate ΔΔG predictions across a diverse array of protein sequences. Herein, we introduce Pythia, a self-supervised graph neural network specifically designed for zero-shot ΔΔG predictions. Our comparative benchmarks demonstrate that Pythia outperforms other self-supervised pretraining models and force field-based approaches while also exhibiting competitive performance with fully supervised models. Notably, Pythia shows strong correlations and achieves a remarkable increase in computational speed of up to 105-fold. We further validated Pythia's performance in predicting the thermostabilizing mutations of limonene epoxide hydrolase, leading to higher experimental success rates. This exceptional efficiency has enabled us to explore 26 million high-quality protein structures, marking a significant advancement in our ability to navigate the protein sequence space and enhance our understanding of the relationships between protein genotype and phenotype. In addition, we established a web server at https://pythia.wulab.xyz to allow users to easily perform such predictions.

MFIB 2.0: a major update of the database of protein complexes formed by mutual folding of the constituting protein chains

While the majority of proteins with available structures are able to fold independently and mediate interactions only after acquiring their folded state, a subset of the known protein complexes contains protein chains that are intrinsically disordered in isolation. The Mutual Folding Induced by Binding (MFIB) database collects and classifies protein complexes, wherein all constituent protein chains would be unstable/disordered in isolation but fold into a well-defined 3D complex structure upon binding. This phenomenon is often termed as cooperative folding and binding or mutual synergistic folding (MSF). Here we present a major update to the database: we collected and annotated hundreds of new protein complexes fulfilling the criteria of MSF, leading to an almost six-fold increase in the size of the database. Many novel features have also been introduced, such as clustering of the complexes based on structural similarity and domain types, assigning different evidence levels to each entry and adding the evidence coverage label that allowed us to include complexes of multi(sub)domain monomers with partial MSF. The MFIB 2.0 database is available at https://mfib.pbrg.hu.

BFVD-a large repository of predicted viral protein structures

The AlphaFold Protein Structure Database (AFDB) is the largest repository of accurately predicted structures with taxonomic labels. Despite providing predictions for over 214 million UniProt entries, the AFDB does not cover viral sequences, severely limiting their study. To address this, we created the Big Fantastic Virus Database (BFVD), a repository of 351 242 protein structures predicted by applying ColabFold to the viral sequence representatives of the UniRef30 clusters. By utilizing homology searches across two petabases of assembled sequencing data, we improved 36% of these structure predictions beyond ColabFold's initial results. BFVD holds a unique repertoire of protein structures as over 62% of its entries show no or low structural similarity to existing repositories. We demonstrate how a substantial fraction of bacteriophage proteins, which remained unannotated based on their sequences, can be matched with similar structures from BFVD. In that, BFVD is on par with the AFDB, while holding nearly three orders of magnitude fewer structures. BFVD is an important virus-specific expansion to protein structure repositories, offering new opportunities to advance viral research. BFVD can be freely downloaded at bfvd.steineggerlab.workers.dev and queried using Foldseek and UniProt labels at bfvd.foldseek.com.

Structural basis of signaling complex inhibition by IL-6 domain-swapped dimers

Interleukin-6 (IL-6) is a multifaceted cytokine essential in many immune system processes and their regulation. It also plays a key role in hematopoiesis, and in triggering the acute phase reaction. IL-6 overproduction is critical in chronic inflammation associated with autoimmune diseases like rheumatoid arthritis and contributes to cytokine storms in COVID-19 patients. Over 20 years ago, researchers proposed that IL-6, which is typically monomeric, can also form dimers via a domain-swap mechanism, with indirect evidence supporting their existence. The physiological significance of IL-6 dimers was shown in B-cell chronic lymphocytic leukemia. However, no structures have been reported so far. Here, we present the crystal structure of an IL-6 domain-swapped dimer that computational approaches could not predict. The structure explains why the IL-6 dimer is antagonistic to the IL-6 monomer in signaling complex formation and provides insights for IL-6 targeted therapies.

An in silico framework to visualize how cancer-associated mutations influence structural plasticity of the chemokine receptor CCR3

G protein Coupled Receptors (GPCRs) are the largest family of cell surface receptors in humans. Somatic mutations in GPCRs are implicated in cancer progression and metastasis, but mechanisms are poorly understood. Emerging evidence implicates perturbation of intra-receptor activation pathway motifs whereby extracellular signals are transmitted intracellularly. Recently, sufficiently sensitive methodology was described to calculate structural strain as a function of missense mutations in AlphaFold-predicted model structures, which was extensively validated on experimental and predicted structural datasets. When paired with Molecular Dynamics (MD) simulations, these tools provide a facile approach to screen mutations in silico. We applied this framework to calculate the structural and dynamic effects of cancer-associated mutations in the chemokine receptor CCR3, a Class A GPCR involved in cancer and autoimmune disorders. Residue-residue contact scoring refined effective strain results, highlighting significant remodeling of inter- and intra-motif contacts along the highly conserved GPCR activation pathway network. We then integrated AlphaFold-derived predicted Local Distance Difference Test scores with per-residue Root Mean Square Fluctuations and activation pathway Contact Analysis (CONAN) from coarse grain MD simulations to identify statistically significant changes in receptor dynamics upon mutation. Finally, analysis of negative control mutants suggests false positive results in AlphaFold pipelines should be considered but can be mitigated with stricter control of statistical analysis. Our results indicate selected mutants influence structural plasticity of CCR3 related to ligand interaction, activation, and G protein coupling, using a framework that could be applicable to a wide range of biochemically relevant protein targets following further validation.

Predicting absolute protein folding stability using generative models

While there has been substantial progress in our ability to predict changes in protein stability due to amino acid substitutions, progress has been slower in methods to predict the absolute stability of a protein. Here, we show how a generative model for protein sequence can be leveraged to predict absolute protein stability. We benchmark our predictions across a broad set of proteins and find a mean error of 1.5 kcal/mol and a correlation coefficient of 0.7 for the absolute stability across a range of natural, small- to medium-sized proteins up to ca. 150 amino acid residues. We analyze current limitations and future directions including how such a model may be useful for predicting conformational free energies. Our approach is simple to use and freely available at an online implementation available via https://github.com/KULL-Centre/_2024_cagiada_stability.

Applicability of AlphaFold2 in the modeling of dimeric, trimeric, and tetrameric coiled-coil domains

Coiled coils are a common protein structural motif involved in cellular functions ranging from mediating protein-protein interactions to facilitating processes such as signal transduction or regulation of gene expression. They are formed by two or more alpha helices that wind around a central axis to form a buried hydrophobic core. Various forms of coiled-coil bundles have been reported, each characterized by the number, orientation, and degree of winding of the constituent helices. This variability is underpinned by short sequence repeats that form coiled coils and whose properties determine both their overall topology and the local geometry of the hydrophobic core. The strikingly repetitive sequence has enabled the development of accurate sequence-based coiled-coil prediction methods; however, the modeling of coiled-coil domains remains a challenging task. In this work, we evaluated the accuracy of AlphaFold2 in modeling coiled-coil domains, both in modeling local geometry and in predicting global topological properties. Furthermore, we show that the prediction of the oligomeric state of coiled-coil bundles can be achieved by using the internal representations of AlphaFold2, with a performance better than any previous state-of-the-art method (code available at https://github.com/labstructbioinf/dc2_oligo).

Computational approach based on freely accessible tools for antimicrobial drug design

Antimicrobial drug development is crucial for public health, especially with the emergence of pandemics and drug resistance that prompts the search for new therapeutic resources. In this context, in silico assays consist of a valuable approach in the rational drug design because they enable a faster and more cost-effective identification of drug candidates compared to in vitro screening. However, once a potential drug is identified, in vitro and in vivo assays are essential to verify the expected activity of the compound and advance it through the subsequent stages of drug development. This work aims to outline an in silico protocol that utilizes only freely available computational tools for identifying new potential antimicrobial agents, which is also suitable in the broad spectrum of drug design. Additionally, this paper reviews relevant computational methods in this context and provides a summary of information concerning the protein-ligand interaction.