Curr Opin Struct Biol
· 2026 Jun · PMID 41825109
·
Full text
G-protein-coupled receptors (GPCRs) are a large family of membrane proteins that mediate cellular responses to diverse stimuli and serve as targets for ∼35 % of Food and Drug Administration-approved drugs. Their structur...G-protein-coupled receptors (GPCRs) are a large family of membrane proteins that mediate cellular responses to diverse stimuli and serve as targets for ∼35 % of Food and Drug Administration-approved drugs. Their structural complexity, conformational heterogeneity, and membrane embedding have historically hindered experimental characterization, although advances in crystallization and cryogenic electron microscopy have expanded access to high-resolution receptor structures. In parallel, artificial intelligence (AI) has transformed protein modeling and drug discovery as recognized by the 2024 Nobel Prize in Chemistry. This minireview highlights recent applications of AI to GPCR research (2023-2025), including structure prediction, virtual screening, generative design of small molecules and protein binders, mechanistic studies using molecular dynamics, and systems-level analyses. Together, these approaches are reshaping GPCR biology and accelerating next-generation drug discovery.
A small number of residues at protein-protein interfaces, commonly referred to as hotspots, dominate binding free energy and play a decisive role in stabilizing protein complexes. Identifying these residues is central to...A small number of residues at protein-protein interfaces, commonly referred to as hotspots, dominate binding free energy and play a decisive role in stabilizing protein complexes. Identifying these residues is central to understanding the energetic architecture of protein-protein interactions and to developing strategies for therapeutic intervention. Although experimental approaches such as alanine scanning have provided critical insights, they are often impractical for large or dynamic systems. This has positioned computational approaches at the forefront of hotspot analysis. This review highlights recent developments in molecular dynamics simulations and machine-learning-based predictors for hotspot identification, discusses current challenges, and outlines emerging directions in the field. Finally, we suggest that combining these complementary approaches could offer a powerful framework for capturing the dynamic and energetic complexity of protein interfaces, making hotspot predictions more robust and interpretable.
Macromolecular assemblies underpin essential cellular processes, yet their structural characterization remains challenging. Integrative modeling provides an approach for determining structures of macromolecular assemblie...Macromolecular assemblies underpin essential cellular processes, yet their structural characterization remains challenging. Integrative modeling provides an approach for determining structures of macromolecular assemblies, combining diverse experimental data with physical principles, the statistics of previous structures, and prior models. There is a growing interest in leveraging the implicit structural knowledge learned by artificial intelligence-based structure-prediction methods such as AlphaFold (AF), for integrative modeling. Here, we discuss recent methods that combine AF with experimental data for integrative modeling in four ways: validating AF-based ensembles with experimental data; combining structural priors from AF with experimental data; fine-tuning AF with experimental data; and incorporating experimental data at inference time. We also outline key challenges in integrative structure determination using AF.
Single-molecule experiments have become an integral part of modern structural biology. Unlike other methods, single-molecule Förster resonance energy transfer (smFRET) spectroscopy opens direct access to distance-based t...Single-molecule experiments have become an integral part of modern structural biology. Unlike other methods, single-molecule Förster resonance energy transfer (smFRET) spectroscopy opens direct access to distance-based temporal trajectories of protein motions. Recent innovations in analysing smFRET experiments with correlation and photon-trajectory based methods have pushed the time resolution of dynamics to much faster than milliseconds. Here, we review these methods, together with their most recent applications and their impact on our understanding of the function of proteins. Important current topics range from the dynamics of intrinsically disordered proteins in complex with their binding partners or in biomolecular condensates, to the conformational dynamics of proteins during their function, from enzymes to molecular machines. We focus particularly on the determination of the timescales of motions and how the utmost information can be gleaned from single-molecule data at the single-photon level.
Intrinsically disordered proteins (IDPs) and intrinsically disordered regions (IDRs) are critical regulators in health and disease but remain underexploited as drug targets. Unlike folded proteins, they populate dynamic...Intrinsically disordered proteins (IDPs) and intrinsically disordered regions (IDRs) are critical regulators in health and disease but remain underexploited as drug targets. Unlike folded proteins, they populate dynamic ensembles where interactions can be transient or multivalent, and both enthalpic and entropic contributions shape binding, complicating ligand discovery. Here, we analyze three key barriers hindering progress: (1) nontraditional binding mechanisms that challenge classical drug design, (2) experimental and computational limitations for studying disorder, and (3) a lack of systematic datasets. Our analysis of the Biological Magnetic Resonance Data Bank (BMRB) and BindingDB highlights the extreme underrepresentation of IDPs and IDRs, underscoring the need for community-driven data resources. By integrating new binding paradigms, tailored methodologies, and standardized datasets, drug discovery can begin to harness IDPs as a new therapeutic frontier.
Chemical cross-linking coupled with mass spectrometry (XL-MS) has become a powerful tool for probing residue-level proximities within macromolecular assemblies. By providing sparse but informative distance restraints, XL...Chemical cross-linking coupled with mass spectrometry (XL-MS) has become a powerful tool for probing residue-level proximities within macromolecular assemblies. By providing sparse but informative distance restraints, XL-MS can be integrated with electron microscopy and domain-level high-resolution structures to model the architecture of protein complexes. Unlike X-ray crystallography, electron microscopy, or solid-state Nuclear Magnetic Resonance (NMR), XL-MS can be applied under near-physiological conditions, scaled to large modular systems, and performed at higher throughput. In this review, we highlight recent advances in the field, with particular emphasis on the impact of AI-driven structure prediction. As an illustration, we describe a hybrid protocol that combines the Integrative Modeling Platform (IMP) with the deep neural network Chai-1 to dock and refine the helicase Dbp10 on a transient ribosome biogenesis intermediate using XL-MS restraints.
New deep learning-based methods for modeling and generation of protein structures have opened a new chapter in the field of protein design, transforming many previously unattainable challenges into routine tasks. Protein...New deep learning-based methods for modeling and generation of protein structures have opened a new chapter in the field of protein design, transforming many previously unattainable challenges into routine tasks. Protein-binder design, an important and challenging task in protein engineering, has also experienced significant progress, promising to provide solutions to many therapeutic and bioengineering problems. Novel protein folds of tailored surface complementarity to their target can be generated and stabilized by amino acid sequences with unprecedentedly high experimental success rates. These advancements can be largely attributed to the power of the new structure prediction models, such as AlphaFold, as well as deep generative models that learn data distributions and allow sampling of new molecules conditioned on function-related features. In this review, we will discuss the development of binder design approaches, focusing on the state-of-the-art methods and their applications as well as new challenges.
Graph neural networks (GNNs) are emerging as powerful tools for advancing molecular dynamics (MD) simulations, providing data-driven frameworks to complement traditional physics-based approaches. By representing atoms an...Graph neural networks (GNNs) are emerging as powerful tools for advancing molecular dynamics (MD) simulations, providing data-driven frameworks to complement traditional physics-based approaches. By representing atoms and their interactions as graphs, GNNs naturally encode chemical and structural information, enabling accurate neural network force fields trained on quantum data, automated discovery of collective variables for enhanced sampling, and efficient prediction of atomic forces to extend simulation timescales. Beyond driving MD, GNNs facilitate the analysis of high-dimensional trajectories, offering interpretable insights through attention mechanisms or transferable embeddings. Applications such as protein-DNA assembly, pretrained featurizers, and cryptic pocket discovery illustrate the breadth of GNNs, underscoring their potential to transform biomolecular simulations and accelerate mechanistic and translational discoveries.
Nearly a quarter of the proteins encoded in most organisms are transmembrane proteins. Contrary to textbook description, many feature a hydrophilic groove which is laterally exposed to the hydrophobic region of the lipid...Nearly a quarter of the proteins encoded in most organisms are transmembrane proteins. Contrary to textbook description, many feature a hydrophilic groove which is laterally exposed to the hydrophobic region of the lipid membrane. This cavity is stabilized by neighboring lipid headgroups that sink deep into the membrane and consequently move bidirectionally from one leaflet to the other, in a process nicknamed lipid 'scrambling.' These proteins, called scramblases, have been reported to serve in many cellular functions, ranging from lipid redistribution during organelle growth to cellular apoptosis. Despite their importance, the identity of most scramblases has remained a mystery for many years. In the last few years, in silico techniques have accelerated the discovery of dozens of new scramblases. Nonetheless, together with these discoveries, key questions have emerged. In this review, we highlight some open questions in this emerging field and showcase how modern computational techniques can help addressing them.
Membrane lipids can bind to specific sites on membrane proteins in a ligand-like manner and modulate protein structure and function. Molecular dynamics simulations encompass a suite of approaches to identify, characteris...Membrane lipids can bind to specific sites on membrane proteins in a ligand-like manner and modulate protein structure and function. Molecular dynamics simulations encompass a suite of approaches to identify, characterise, and explain the atomic-level mechanisms that underlie the functional effects of ligand-like lipids on membrane proteins. Simulations have shown good agreement with available structural data on lipid-protein interactions. Building on successes, simulations are now used to identify new interactions and mechanisms de novo for a given membrane protein. In this age of abundance, it is increasingly possible to analyse patterns across large groups of proteins and in ever more complex membrane environments. The dawn of machine learning approaches in lipid-protein cofolding holds considerable promise to synergistically capitalise on this availability of simulation data and uncover new facets of ligand-like lipid biology.
Proteins often rely on conformational dynamics to perform their biological functions. A detailed understanding of protein dynamics is fundamental to revealing the biophysical principles of life and to accelerating therap...Proteins often rely on conformational dynamics to perform their biological functions. A detailed understanding of protein dynamics is fundamental to revealing the biophysical principles of life and to accelerating therapeutic discovery. However, purely data-driven artificial intelligence (AI) methods face significant challenges in capturing the full spectrum of protein conformational dynamics. This review highlights recent advances in overcoming these challenges through the integration of biophysical constraints with AI-driven approaches. By combining fundamental biophysical principles, experimentally measured biophysical data, and physics-based methodologies into AI models, the integrated approaches show promise in enhancing both the performance and interpretability of protein dynamics predictions. Several key perspectives and future directions in the field are also discussed.
Advances in machine-learning techniques and the availability of high-quality pK databases have promoted the development of AI-driven pK predictors. This review surveys recent advances in AI-driven pK prediction for both...Advances in machine-learning techniques and the availability of high-quality pK databases have promoted the development of AI-driven pK predictors. This review surveys recent advances in AI-driven pK prediction for both proteins and small molecules, and reveals that methodology has evolved along divergent trajectories for the two molecular classes, giving rise to largely independent lineages. Finally, open challenges in pK prediction, including data scarcity, thermodynamic consistency, and a general-purpose model, are highlighted for future development.
Membrane protein complexes are essential for cellular functions, which rely on both constituent protein structures and their interactions within native membranes. While in vitro methods have successfully yielded high-res...Membrane protein complexes are essential for cellular functions, which rely on both constituent protein structures and their interactions within native membranes. While in vitro methods have successfully yielded high-resolution structures of individual proteins and subcomplexes, these approaches typically require detergent extraction and extensive purification, which can disrupt the native membrane environment and potentially alter the supramolecular organization. In situ structural biology has therefore emerged as an effective strategy to overcome these limitations by directly visualizing macromolecular machines within their physiological context. With continuous technological advancements, several recent studies have resolved in situ structures of large protein complexes at high or even near-atomic resolution. This review focuses on recent in situ high-resolution studies of membrane protein megacomplexes, highlighting key technical innovations, structural insights, and the remaining challenges and opportunities in the field.
Drug-target residence time is a crucial determinant of pharmacological efficacy, complementing traditional equilibrium affinity measures. Variations in residence time influence drug selectivity, therapeutic windows, and...Drug-target residence time is a crucial determinant of pharmacological efficacy, complementing traditional equilibrium affinity measures. Variations in residence time influence drug selectivity, therapeutic windows, and resistance development, yet its molecular underpinnings remain incompletely understood. Here we review factors governing residence time, including kinetic parameters and structural influences, and examine how mutations can alter dissociation rates to confer drug resistance. We highlight recent advances in experimental and computational methods, such as molecular dynamics simulations, that enable prediction and rational design of compounds with optimized residence times. These insights underscore the importance of incorporating kinetic considerations into drug discovery to improve efficacy and overcome resistance. Our findings suggest that optimizing residence time offers a promising strategy to enhance therapeutic outcomes for diverse diseases.
T cell receptor (TCR) diversity underpins cellular immunity. While αβ TCRs have been extensively studied in the context of major histocompatibility complex (MHC)-restricted antigen recognition, the γδ TCR system remains...T cell receptor (TCR) diversity underpins cellular immunity. While αβ TCRs have been extensively studied in the context of major histocompatibility complex (MHC)-restricted antigen recognition, the γδ TCR system remains underexplored. Unlike their αβ counterparts, γδ TCRs display versatile, often MHC-independent recognition modes, engaging diverse ligands ranging from butyrophilins (BTNs) and other disparate molecules. Recent advances in cryo-electron microscopy (cryo-EM) paired with crystallographic data have illuminated critical aspects of γδ TCR -ligand interactions, the CD3 complex architecture, and the inherent flexibility underpinning their varied recognition modes. In this review, we compare the classical αβ TCR-MHC paradigm against the backdrop of emerging γδ TCR structures, highlighting the latest cryo-EM findings and their implications for immunobiology.
Protein design enables the creation of novel structures and functions beyond those found in nature, with recent progress accelerated by computational modeling and machine learning. However, many automated methods act as...Protein design enables the creation of novel structures and functions beyond those found in nature, with recent progress accelerated by computational modeling and machine learning. However, many automated methods act as black boxes, limiting mechanistic insight. Here we highlight the continuing importance of rational protein design, defined as an approach rooted in physical principles, chemical intuition, and sequence-structure-function relationships. We outline three complementary strategies: backbone-first, sequence-first, and function-first, which provide interpretable design frameworks and enable robust scaffold generation, motif incorporation, and functional engineering. Looking forward, we argue that hybrid workflows combining rational principles with machine learning offer the most promising route to dynamic, explainable, and generalizable protein design.
Drug-target residence time (τ) is reviewed from two perspectives: mathematics and molecular dynamics. The first focuses on the quantification of τ using a mathematical formalism applicable to different pharmacological me...Drug-target residence time (τ) is reviewed from two perspectives: mathematics and molecular dynamics. The first focuses on the quantification of τ using a mathematical formalism applicable to different pharmacological mechanistic conditions. This formalism is based on the concept of the smallest-modulus eigenvalue of a subsystem of interest, in which the global formation process has been eliminated. The second includes relevant studies of recent years to provide a structural explanation of τ predictions. Special attention is paid to physically supported artificial intelligence methods. The main objective of this minireview is to promote a combined approach in which mathematics and physics work synergistically to describe the complexity associated with τ in G protein-coupled receptors.
Transformers are rapidly reshaping structural biology. We argue the reason is "Emergent Latent Biology" (ELB): transformers place proteins into high-dimensional representations where hidden biophysical patterns become ea...Transformers are rapidly reshaping structural biology. We argue the reason is "Emergent Latent Biology" (ELB): transformers place proteins into high-dimensional representations where hidden biophysical patterns become easier to see. We explore this concept across four key areas: protein folding, variant effects, protein-protein and protein-drug interactions. Highlighting recent gains, we note that traditional, physics-based calculations are still required for the hardest quantitative jobs, like predicting precise binding strength. Furthermore, we draw attention to major pitfalls, arguing progress depends on solving the critical "chemistry gap," modelling chemical modifications, and the "dynamics gap", predicting protein movement, which requires better validation methods and new large-scale experiments.
RNA structure prediction remains one of the most challenging problems in computational biology, with significant implications for understanding gene regulation, drug design, and synthetic biology. While deep learning has...RNA structure prediction remains one of the most challenging problems in computational biology, with significant implications for understanding gene regulation, drug design, and synthetic biology. While deep learning has revolutionized protein structure prediction, RNA presents unique challenges including limited training data, complex noncanonical interactions, and conformational flexibility. This review examines the evolution from traditional physics-based methods to current deep learning approaches for RNA secondary and tertiary structure prediction. After briefly exploring traditional methods, like Direct Coupling Analysis and physics-based simulations, we systematically review three deep learning paradigms: language model-based methods, end-to-end structure predictors, and geometry-distance prediction approaches. Furthermore, we identify critical future research directions focusing on advanced tokenization strategies to address data scarcity and explainable artificial intelligence techniques to improve model interpretability. Despite significant progress, achieving transformative performance requires continued methodological innovation, specifically designed for RNA's unique characteristics, and a substantial expansion of high-quality structural datasets.