Extraction of protein dynamics information from cryo-EM maps using deep learning

Know your molecule: pharmacological characterization of drug candidates to enhance efficacy and reduce late-stage attrition

Despite advances in chemical, computational and biological sciences, the rate of attrition of drug candidates in clinical development is still high. A key point in the small-molecule discovery process that could provide opportunities to help address this challenge is the pharmacological characterization of hit and lead compounds, culminating in the selection of a drug candidate. Deeper characterization is increasingly important, because the 'quality' of drug efficacy, at least for G protein-coupled receptors (GPCRs), is now understood to be much more than activation of commonly evaluated pathways such as cAMP signalling, with many more 'efficacies' of ligands that could be harnessed therapeutically. Such characterization is being enabled by novel assays to characterize the complex behaviour of GPCRs, such as biased signalling and allosteric modulation, as well as advances in structural biology, such as cryo-electron microscopy. This article discusses key factors in the assessments of the pharmacology of hit and lead compounds in the context of GPCRs as a target class, highlighting opportunities to identify drug candidates with the potential to address limitations of current therapies and to improve the probability of them succeeding in clinical development.

Natural triterpenoid-aided identification of the druggable interface of HMGB1 occupied by TLR4

HMGB1 interacts with TLR4 to activate the inflammatory cascade response, contributing to the pathogenesis of endogenous tissue damage and infection. The immense importance of HMGB1-TLR4 interaction in the immune system has made its binding interface an area of significant interest. To map the binding interface of HMGB1 occupied by TLR4, triterpenoids that disrupt the HMGB1-TLR4 interaction and interfere with HMGB1-induced inflammation were developed. Using the unique triterpenoid PT-22 as a probe along with photoaffinity labeling and site-directed mutagenesis, we found that the binding interface of HMGB1 was responsible for the recognition of TLR4 located on the "L" shaped B-box with K114 as a crucial hot-spot residue. Amazingly, this highly conserved interaction surface overlapped with the antigen-recognition epitope of an anti-HMGB1 antibody. Our findings propose a novel strategy for better understanding the druggable interface of HMGB1 that interacts with TLR4 and provide insights for the rational design of HMGB1-TLR4 PPI inhibitors to fine tune immune responses.

Exploring Conformational Landscapes Along Anharmonic Low-Frequency Vibrations

We aim to automatize the identification of collective variables to simplify and speed up enhanced sampling simulations of conformational dynamics in biomolecules. We focus on anharmonic low-frequency vibrations that exhibit fluctuations on time scales faster than conformational transitions but describe a path of least resistance toward structural change. A key challenge is that harmonic approximations are ill-suited to characterize these vibrations, which are observed at far-infrared frequencies and are easily excited by thermal collisions at room temperature. Here, we approached this problem with a frequency-selective anharmonic (FRESEAN) mode analysis that does not rely on harmonic approximations and successfully isolates anharmonic low-frequency vibrations from short molecular dynamics simulation trajectories. We applied FRESEAN mode analysis to simulations of alanine dipeptide, a common test system for enhanced sampling simulation protocols, and compared the performance of isolated low-frequency vibrations to conventional user-defined collective variables (here backbone dihedral angles) in enhanced sampling simulations. The comparison shows that enhanced sampling along anharmonic low-frequency vibrations not only reproduces known conformational dynamics but can even further improve the sampling of slow transitions compared to user-defined collective variables. Notably, free energy surfaces spanned by low-frequency anharmonic vibrational modes exhibit lower barriers associated with conformational transitions relative to representations in backbone dihedral space. We thus conclude that anharmonic low-frequency vibrations provide a promising path for highly effective and fully automated enhanced sampling simulations of conformational dynamics in biomolecules.

Accurate Prediction of Protein Structural Flexibility by Deep Learning Integrating Intricate Atomic Structures and Cryo-EM Density Information

The dynamics of proteins are crucial for understanding their mechanisms. However, computationally predicting protein dynamic information has proven challenging. Here, we propose a neural network model, RMSF-net, which outperforms previous methods and produces the best results in a large-scale protein dynamics dataset; this model can accurately infer the dynamic information of a protein in only a few seconds. By learning effectively from experimental protein structure data and cryo-electron microscopy (cryo-EM) data integration, our approach is able to accurately identify the interactive bidirectional constraints and supervision between cryo-EM maps and PDB models in maximizing the dynamic prediction efficacy. Rigorous 5-fold cross-validation on the dataset demonstrates that RMSF-net achieves test correlation coefficients of 0.746 ± 0.127 at the voxel level and 0.765 ± 0.109 at the residue level, showcasing its ability to deliver dynamic predictions closely approximating molecular dynamics simulations. Additionally, it offers real-time dynamic inference with minimal storage overhead on the order of megabytes. RMSF-net is a freely accessible tool and is anticipated to play an essential role in the study of protein dynamics.

The dynamic duo: cryo-EM teams up with machine learning to visualize biomolecules in motion

Cryo-EM has been a key technique in our understanding of biomolecular structures. Now, machine learning techniques are being used to put these structures in motion, revealing dynamic interactions and processes happening on a molecular and cellular level.[Formula: see text].

Cryo-EM and Molecular Dynamics Simulations Reveal Hidden Conformational Dynamics Controlling Ammonia Transport in Human Asparagine Synthetase

How motions in enzymes might be linked to catalytic function is of considerable general interest. Recent advances in X-ray crystallography and cryogenic electron microscopy offer the promise of elucidating functionally relevant motions in proteins that are not easily amenable to study by other biophysical methods. Here we use 3D variability analysis (3DVA) on cryo-EM maps for wild type (WT) human asparagine synthetase (ASNS) and the R142I ASNS variant to identify conformational changes in the Arg-142 side chain, which mediates the formation of a catalytically relevant intramolecular tunnel. Our 3DVA results for WT ASNS are consistent with independent molecular dynamics (MD) simulations on a model generated from the X-ray structure of human ASNS. Moreover, MD simulations of computational models for the ASNS/β-aspartyl-AMP/MgPPi and R142I/β-aspartyl-AMP/MgPPi ternary complexes, suggest that the structural integrity of the tunnel is impaired in the R142I variant when β-aspartyl-AMP is present in the synthetase active site. The kinetic properties of the R142I ASNS variant support the proposed function of Arg-142. These studies illustrate the power of cryo-EM to identify localized motions and dissect the conformational landscape of large proteins. When combined with MD simulations, 3DVA is a powerful approach to understanding how conformational dynamics might regulate function in multi-domain enzymes possessing multiple active sites.

Cellular Applications of DNP Solid-State NMR - State of the Art and a Look to the Future

Sensitivity enhanced dynamic nuclear polarization solid-state NMR is emerging as a powerful technique for probing the structural properties of conformationally homogenous and heterogenous biomolecular species irrespective of size at atomic resolution within their native environments. Herein we detail advancements that have made acquiring such data, specifically within the confines of intact bacterial and eukaryotic cell a reality and further discuss the type of structural information that can presently be garnered by the technique's exploitation. Subsequently, we discuss bottlenecks that have thus far curbed cellular DNP-ssNMR's broader adoption namely due a lack of sensitivity and spectral resolution. We also explore possible solutions ranging from utilization of new pulse sequences, design of better performing polarizing agents, and application of additional biochemical/ cell biological methodologies.

Integrative approaches for characterizing protein dynamics: NMR, CryoEM, and computer simulations

Proteins are inherently dynamic and their internal motions are essential for biological function. Protein motions cover a broad range of timescales: 10-14-10 s, spanning from sub-picosecond vibrational motions of atoms via microsecond loop conformational rearrangements to millisecond large amplitude domain reorientations. Observing protein dynamics over all timescales and connecting motions and structure to biological mechanisms requires integration of multiple experimental and computational techniques. This review reports on state-of-the-art approaches for assessing dynamics in biological systems using recent examples of virus assemblies, enzymes, and molecular machines. By integrating NMR spectroscopy in solution and the solid state, cryo electron microscopy, and molecular dynamics simulations, atomistic pictures of protein motions are obtained, not accessible from any single method in isolation. This information provides fundamental insights into protein behavior that can guide the development of future therapeutics.

In-Cell Dynamics: The Next Focus of All-Atom Simulations

The cell is a crowded space where large biomolecules and metabolites are in continuous motion. Great strides have been made in in vitro studies of protein dynamics, folding, and protein-protein interactions, and much new data are emerging of how they differ in the cell. In this Perspective, we highlight the current progress in atomistic modeling of in-cell environments, both bacteria and mammals, with emphasis on classical all-atom molecular dynamics simulations. These simulations have been recently used to capture and characterize functional and non-functional protein-protein interactions, protein folding dynamics of small proteins with varied topologies, and dynamics of metabolites. We further discuss the challenges and efforts for updating modern force fields critical to the progress of cellular environment simulations. We also briefly summarize developments in relevant state-of-the-art experimental techniques. As computational and experimental methodologies continue to progress and produce more directly comparable data, we are poised to capture the complex atomistic picture of the cell.

Side-chain dynamics of the α1B -adrenergic receptor determined by NMR via methyl relaxation

G protein-coupled receptors (GPCRs) are medically important membrane proteins that sample inactive, intermediate, and active conformational states characterized by relatively slow interconversions (~μs-ms). On a faster timescale (~ps-ns), the conformational landscape of GPCRs is governed by the rapid dynamics of amino acid side chains. Such dynamics are essential for protein functions such as ligand recognition and allostery. Unfortunately, technical challenges have almost entirely precluded the study of side-chain dynamics for GPCRs. Here, we investigate the rapid side-chain dynamics of a thermostabilized α1B -adrenergic receptor (α1B -AR) as probed by methyl relaxation. We determined order parameters for Ile, Leu, and Val methyl groups in the presence of inverse agonists that bind orthosterically (prazosin, tamsulosin) or allosterically (conopeptide ρ-TIA). Despite the differences in the ligands, the receptor's overall side-chain dynamics are very similar, including those of the apo form. However, ρ-TIA increases the flexibility of Ile1764×56 and possibly of Ile2145×49 , adjacent to Pro2155×50 of the highly conserved P5×50 I3×40 F6×44 motif crucial for receptor activation, suggesting differences in the mechanisms for orthosteric and allosteric receptor inactivation. Overall, increased Ile side-chain rigidity was found for residues closer to the center of the membrane bilayer, correlating with denser packing and lower protein surface exposure. In contrast to two microbial membrane proteins, in α1B -AR Leu exhibited higher flexibility than Ile side chains on average, correlating with the presence of Leu in less densely packed areas and with higher protein-surface exposure than Ile. Our findings demonstrate the feasibility of studying receptor-wide side-chain dynamics in GPCRs to gain functional insights.

A robust normalized local filter to estimate compositional heterogeneity directly from cryo-EM maps

Cryo electron microscopy (cryo-EM) is used by biological research to visualize biomolecular complexes in 3D, but the heterogeneity of cryo-EM reconstructions is not easily estimated. Current processing paradigms nevertheless exert great effort to reduce flexibility and heterogeneity to improve the quality of the reconstruction. Clustering algorithms are typically employed to identify populations of data with reduced variability, but lack assessment of remaining heterogeneity. Here we develope a fast and simple algorithm based on spatial filtering to estimate the heterogeneity of a reconstruction. In the absence of flexibility, this estimate approximates macromolecular component occupancy. We show that our implementation can derive reasonable input parameters, that composition heterogeneity can be estimated based on contrast loss, and that the reconstruction can be modified accordingly to emulate altered constituent occupancy. This stands to benefit conventionally employed maximum-likelihood classification methods, whereas we here limit considerations to cryo-EM map interpretation, quantification, and particle-image signal subtraction.

Novel Artificial Intelligence-Based Approaches for Ab Initio Structure Determination and Atomic Model Building for Cryo-Electron Microscopy

Single particle cryo-electron microscopy (cryo-EM) has emerged as the prevailing method for near-atomic structure determination, shedding light on the important molecular mechanisms of biological macromolecules. However, the inherent dynamics and structural variability of biological complexes coupled with the large number of experimental images generated by a cryo-EM experiment make data processing nontrivial. In particular, ab initio reconstruction and atomic model building remain major bottlenecks that demand substantial computational resources and manual intervention. Approaches utilizing recent innovations in artificial intelligence (AI) technology, particularly deep learning, have the potential to overcome the limitations that cannot be adequately addressed by traditional image processing approaches. Here, we review newly proposed AI-based methods for ab initio volume generation, heterogeneous 3D reconstruction, and atomic model building. We highlight the advancements made by the implementation of AI methods, as well as discuss remaining limitations and areas for future development.

Tertiary structure of single-instant RNA molecule reveals folding landscape

The folding of RNA and protein molecules during their synthesis is a crucial self-assembly process that nature employs to convert genetic information into the complex molecular machinery that supports life. Misfolding events are the cause of several diseases, and the folding pathway of central biomolecules, such as the ribosome, is strictly regulated by programmed maturation processes and folding chaperones. However, the dynamic folding processes are challenging to study because current structure determination methods heavily rely on averaging, and existing computational methods do not efficiently simulate non-equilibrium dynamics. Here we utilize individual-particle cryo-electron tomography (IPET) to investigate the folding landscape of a rationally designed RNA origami 6-helix bundle that undergoes slow maturation from a "young" to "mature" conformation. By optimizing the IPET imaging and electron dose conditions, we obtain 3D reconstructions of 120 individual particles at resolutions ranging from 23-35 Å, enabling us first-time to observe individual RNA helices and tertiary structures without averaging. Statistical analysis of 120 tertiary structures confirms the two main conformations and suggests a possible folding pathway driven by helix-helix compaction. Studies of the full conformational landscape reveal both trapped states, misfolded states, intermediate states, and fully compacted states. The study provides novel insight into RNA folding pathways and paves the way for future studies of the energy landscape of molecular machines and self-assembly processes.

Quantitative analysis of protein dynamics using a deep learning technique combined with experimental cryo-EM density data and MD simulations

Protein functions associated with biological activity are precisely regulated by both tertiary structure and dynamic behavior. Thus, elucidating the high-resolution structures and quantitative information on in-solution dynamics is essential for understanding the molecular mechanisms. The main experimental approaches for determining tertiary structures include nuclear magnetic resonance (NMR), X-ray crystallography, and cryogenic electron microscopy (cryo-EM). Among these procedures, recent remarkable advances in the hardware and analytical techniques of cryo-EM have increasingly determined novel atomic structures of macromolecules, especially those with large molecular weights and complex assemblies. In addition to these experimental approaches, deep learning techniques, such as AlphaFold 2, accurately predict structures from amino acid sequences, accelerating structural biology research. Meanwhile, the quantitative analyses of the protein dynamics are conducted using experimental approaches, such as NMR and hydrogen-deuterium mass spectrometry, and computational approaches, such as molecular dynamics (MD) simulations. Although these procedures can quantitatively explore dynamic behavior at high resolution, the fundamental difficulties, such as signal crowding and high computational cost, greatly hinder their application to large and complex biological macromolecules. In recent years, machine learning techniques, especially deep learning techniques, have been actively applied to structural data to identify features that are difficult for humans to recognize from big data. Here, we review our approach to accurately estimate dynamic properties associated with local fluctuations from three-dimensional cryo-EM density data using a deep learning technique combined with MD simulations.

Artificial intelligence-based HDX (AI-HDX) prediction reveals fundamental characteristics to protein dynamics: Mechanisms on SARS-CoV-2 immune escape

Three-dimensional structure and dynamics are essential for protein function. Advancements in hydrogen-deuterium exchange (HDX) techniques enable probing protein dynamic information in physiologically relevant conditions. HDX-coupled mass spectrometry (HDX-MS) has been broadly applied in pharmaceutical industries. However, it is challenging to obtain dynamics information at the single amino acid resolution and time consuming to perform the experiments and process the data. Here, we demonstrate the first deep learning model, artificial intelligence-based HDX (AI-HDX), that predicts intrinsic protein dynamics based on the protein sequence. It uncovers the protein structural dynamics by combining deep learning, experimental HDX, sequence alignment, and protein structure prediction. AI-HDX can be broadly applied to drug discovery, protein engineering, and biomedical studies. As a demonstration, we elucidated receptor-binding domain structural dynamics as a potential mechanism of anti-severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) antibody efficacy and immune escape. AI-HDX fundamentally differs from the current AI tools for protein analysis and may transform protein design for various applications.

Benchmarking applicability of medium-resolution cryo-EM protein structures for structure-based drug design

Cryo-electron microscopy (cryo-EM) is gaining large attention for high-resolution protein structure determination in solutions. However, a very high percentage of cryo-EM structures correspond to resolutions of 3-5 Å, making the structures difficult to be used in in silico drug design. In this study, we analyze how useful cryo-EM protein structures are for in silico drug design by evaluating ligand docking accuracy. From realistic cross-docking scenarios using medium resolution (3-5 Å) cryo-EM structures and a popular docking tool Autodock-Vina, only 20% of docking succeeded, when the success rate doubles in the same kind of cross-docking but using high-resolution (<2 Å) crystal structures instead. We decipher the reason for failures by decomposing the contribution from resolution-dependent and independent factors. The heterogeneity in the protein side-chain and backbone conformations is identified as the major resolution-dependent factor causing docking difficulty from our analysis, while intrinsic receptor flexibility mainly comprises the resolution-independent factor. We demonstrate the flexibility implementation in current ligand docking tools is able to rescue only a portion of failures (10%), and the limited performance was majorly due to potential structural errors than conformational changes. Our work suggests the strong necessity of more robust method developments on ligand docking and EM modeling techniques in order to fully utilize cryo-EM structures for in silico drug design.

Low-data interpretable deep learning prediction of antibody viscosity using a biophysically meaningful representation

Deep learning, aided by the availability of big data sets, has led to substantial advances across many disciplines. However, many scientific problems of practical interest lack sufficiently large datasets amenable to deep learning. Prediction of antibody viscosity is one such problem where deep learning methods have not yet been explored due to the relative scarcity of relevant training data. In this work, we overcome this limitation using a biophysically meaningful representation that enables us to develop generalizable models even under limited training data. We present, PfAbNet-viscosity, a 3D convolutional neural network architecture, to predict high-concentration viscosity of therapeutic antibodies. We show that with the electrostatic potential surface of the antibody variable region as the only input to the network, the models trained on as few as couple dozen datapoints can generalize with high accuracy. Our feature attribution analysis shows that PfAbNet-viscosity has learned key biophysical drivers of viscosity. The applicability of our approach to other biological systems is discussed.

Beyond sequence: Structure-based machine learning

Recent breakthroughs in protein structure prediction demarcate the start of a new era in structural bioinformatics. Combined with various advances in experimental structure determination and the uninterrupted pace at which new structures are published, this promises an age in which protein structure information is as prevalent and ubiquitous as sequence. Machine learning in protein bioinformatics has been dominated by sequence-based methods, but this is now changing to make use of the deluge of rich structural information as input. Machine learning methods making use of structures are scattered across literature and cover a number of different applications and scopes; while some try to address questions and tasks within a single protein family, others aim to capture characteristics across all available proteins. In this review, we look at the variety of structure-based machine learning approaches, how structures can be used as input, and typical applications of these approaches in protein biology. We also discuss current challenges and opportunities in this all-important and increasingly popular field.

Integrative Bioinformatics Approaches Indicate a Particular Pattern of Some SARS-CoV-2 and Non-SARS-CoV-2 Proteins

Pattern recognition plays a critical role in integrative bioinformatics to determine the structural patterns of proteins of viruses such as SARS-CoV-2. This study identifies the pattern of SARS-CoV-2 proteins to depict the structure-function relationships of the protein alphabets of SARS-CoV-2 and COVID-19. The assembly enumeration algorithm, Anisotropic Network Model, Gaussian Network Model, Markovian Stochastic Model, and image comparison protein-like alphabets were used. The distance score was the lowest with 22 for "I" and highest with 40 for "9". For post-processing and decision, two protein alphabets "C" (PDB ID: 6XC3) and "S" (PDB ID: 7OYG) were evaluated to understand the structural, functional, and evolutionary relationships, and we found uniqueness in the functionality of proteins. Here, models were constructed using "SARS-CoV-2 proteins" (12 numbers) and "non-SARS-CoV-2 proteins" (14 numbers) to create two words, "SARS-CoV-2" and "COVID-19". Similarly, we developed two slogans: "Vaccinate the world against COVID-19" and "Say no to SARS-CoV-2", which were made with the proteins structure. It might generate vaccine-related interest to broad reader categories. Finally, the evolutionary process appears to enhance the protein structure smoothly to provide suitable functionality shaped by natural selection.

G protein gamma subunit, a hidden master regulator of GPCR signaling

Heterotrimeric G proteins (αβγ subunits) that are activated by G protein-coupled receptors (GPCRs) mediate the biological responses of eukaryotic cells to extracellular signals. The α subunits and the tightly bound βγ subunit complex of G proteins have been extensively studied and shown to control the activity of effector molecules. In contrast, the potential roles of the large family of γ subunits have been less studied. In this review, we focus on present knowledge about these proteins. Induced loss of individual γ subunit types in animal and plant models result in strikingly distinct phenotypes indicating that γ subtypes play important and specific roles. Consistent with these findings, downregulation or upregulation of particular γ subunit types result in various types of cancers. Clues about the mechanistic basis of γ subunit function have emerged from imaging the dynamic behavior of G protein subunits in living cells. This shows that in the basal state, G proteins are not constrained to the plasma membrane but shuttle between membranes and on receptor activation βγ complexes translocate reversibly to internal membranes. The translocation kinetics of βγ complexes varies widely and is determined by the membrane affinity of the associated γ subtype. On translocating, some βγ complexes act on effectors in internal membranes. The variation in translocation kinetics determines differential sensitivity and adaptation of cells to external signals. Membrane affinity of γ subunits is thus a parsimonious and elegant mechanism that controls information flow to internal cell membranes while modulating signaling responses.

Visualizing Intramolecular Dynamics of Membrane Proteins

Membrane proteins play important roles in biological functions, with accompanying allosteric structure changes. Understanding intramolecular dynamics helps elucidate catalytic mechanisms and develop new drugs. In contrast to the various technologies for structural analysis, methods for analyzing intramolecular dynamics are limited. Single-molecule measurements using optical microscopy have been widely used for kinetic analysis. Recently, improvements in detectors and image analysis technology have made it possible to use single-molecule determination methods using X-rays and electron beams, such as diffracted X-ray tracking (DXT), X-ray free electron laser (XFEL) imaging, and cryo-electron microscopy (cryo-EM). High-speed atomic force microscopy (HS-AFM) is a scanning probe microscope that can capture the structural dynamics of biomolecules in real time at the single-molecule level. Time-resolved techniques also facilitate an understanding of real-time intramolecular processes during chemical reactions. In this review, recent advances in membrane protein dynamics visualization techniques were presented.

Learning physics-consistent particle interactions

Interacting particle systems play a key role in science and engineering. Access to the governing particle interaction law is fundamental for a complete understanding of such systems. However, the inherent system complexity keeps the particle interaction hidden in many cases. Machine learning methods have the potential to learn the behavior of interacting particle systems by combining experiments with data analysis methods. However, most existing algorithms focus on learning the kinetics at the particle level. Learning pairwise interaction, e.g., pairwise force or pairwise potential energy, remains an open challenge. Here, we propose an algorithm that adapts the Graph Networks framework, which contains an edge part to learn the pairwise interaction and a node part to model the dynamics at particle level. Different from existing approaches that use neural networks in both parts, we design a deterministic operator in the node part that allows to precisely infer the pairwise interactions that are consistent with underlying physical laws by only being trained to predict the particle acceleration. We test the proposed methodology on multiple datasets and demonstrate that it achieves superior performance in inferring correctly the pairwise interactions while also being consistent with the underlying physics on all the datasets. While the previously proposed approaches are able to be applied as simulators, they fail to infer physically consistent particle interactions that satisfy Newton's laws. Moreover, the proposed physics-induced graph network for particle interaction also outperforms the other baseline models in terms of generalization ability to larger systems and robustness to significant levels of noise. The developed methodology can support a better understanding and discovery of the underlying particle interaction laws, and hence, guide the design of materials with targeted properties.

Visualizing Conformational Space of Functional Biomolecular Complexes by Deep Manifold Learning

The cellular functions are executed by biological macromolecular complexes in nonequilibrium dynamic processes, which exhibit a vast diversity of conformational states. Solving the conformational continuum of important biomolecular complexes at the atomic level is essential to understanding their functional mechanisms and guiding structure-based drug discovery. Here, we introduce a deep manifold learning framework, named AlphaCryo4D, which enables atomic-level cryogenic electron microscopy (cryo-EM) reconstructions that approximately visualize the conformational space of biomolecular complexes of interest. AlphaCryo4D integrates 3D deep residual learning with manifold embedding of pseudo-energy landscapes, which simultaneously improves 3D classification accuracy and reconstruction resolution via an energy-based particle-voting algorithm. In blind assessments using simulated heterogeneous datasets, AlphaCryo4D achieved 3D classification accuracy three times those of alternative methods and reconstructed continuous conformational changes of a 130-kDa protein at sub-3 Å resolution. By applying this approach to analyze several experimental datasets of the proteasome, ribosome and spliceosome, we demonstrate its potential generality in exploring hidden conformational space or transient states of macromolecular complexes that remain hitherto invisible. Integration of this approach with time-resolved cryo-EM further allows visualization of conformational continuum in a nonequilibrium regime at the atomic level, thus potentially enabling therapeutic discovery against highly dynamic biomolecular targets.

Potential Therapeutic Candidates against Chlamydia pneumonia Discovered and Developed In Silico Using Core Proteomics and Molecular Docking and Simulation-Based Approaches

Chlamydia pneumonia, a species of the family Chlamydiacea, is a leading cause of pneumonia. Failure to eradicate C. pneumoniae can lead to chronic infection, which is why it is also considered responsible for chronic inflammatory disorders such as asthma, arthritis, etc. There is an urgent need to tackle the major concerns arising due to persistent infections caused by C. pneumoniae as no FDA-approved drug is available against this chronic infection. In the present study, an approach named subtractive proteomics was employed to the core proteomes of five strains of C. pneumonia using various bioinformatic tools, servers, and software. However, 958 non-redundant proteins were predicted from the 4754 core proteins of the core proteome. BLASTp was used to analyze the non-redundant genes against the proteome of humans, and the number of potential genes was reduced to 681. Furthermore, based on subcellular localization prediction, 313 proteins with cytoplasmic localization were selected for metabolic pathway analysis. Upon subsequent analysis, only three cytoplasmic proteins, namely 30S ribosomal protein S4, 4-hydroxybenzoate decarboxylase subunit C, and oligopeptide binding protein, were identified, which have the potential to be novel drug target candidates. The Swiss Model server was used to predict the target proteins’ three-dimensional (3D) structure. The molecular docking technique was employed using MOE software for the virtual screening of a library of 15,000 phytochemicals against the interacting residues of the target proteins. Molecular docking experiments were also evaluated using molecular dynamics simulations and the widely used MM-GBSA and MM-PBSA binding free energy techniques. The findings revealed a promising candidate as a novel target against C. pneumonia infections.

Triangulating variation in the population to define mechanisms for precision management of genetic disease

To understand mechanistically how the protein fold is shaped by therapeutics to inform precision management of disease, we developed variation-capture (VarC) mapping. VarC triangulates sparse sequence variation information found in the population using Gaussian process regression (GPR)-based machine learning to define the combined pairwise-residue interactions contributing to dynamic protein function in the individual in response to therapeutics. Using VarC mapping, we now reveal the pairwise-residue covariant relationships across the entire protein fold of cystic fibrosis (CF) transmembrane conductance regulator (CFTR) to define the molecular mechanisms of clinically approved CF chemical modulators. We discover an energetically destabilized covariant core containing a di-acidic YKDAD endoplasmic reticulum (ER) exit code that is only weakly corrected by current therapeutics. Our results illustrate that VarC provides a generalizable tool to triangulate information from genetic variation in the population to mechanistically discover therapeutic strategies that guide precision management of the individual.

Integration of machine learning with computational structural biology of plants

Computational structural biology of proteins has developed rapidly in recent decades with the development of new computational tools and the advancement of computing hardware. However, while these techniques have widely been used to make advancements in human medicine, these methods have seen less utilization in the plant sciences. In the last several years, machine learning methods have gained popularity in computational structural biology. These methods have enabled the development of new tools which are able to address the major challenges that have hampered the wide adoption of the computational structural biology of plants. This perspective examines the remaining challenges in computational structural biology and how the development of machine learning techniques enables more in-depth computational structural biology of plants.

Probing Structural Perturbation of Biomolecules by Extracting Cryo-EM Data Heterogeneity

Single-particle cryogenic electron microscopy (cryo-EM) has become an indispensable tool to probe high-resolution structural detail of biomolecules. It enables direct visualization of the biomolecules and opens a possibility for averaging molecular images to reconstruct a three-dimensional Coulomb potential density map. Newly developed algorithms for data analysis allow for the extraction of structural heterogeneity from a massive and low signal-to-noise-ratio (SNR) cryo-EM dataset, expanding our understanding of multiple conformational states, or further implications in dynamics, of the target biomolecule. This review provides an overview that briefly describes the workflow of single-particle cryo-EM, including imaging and data processing, and new methods developed for analyzing the data heterogeneity to understand the structural variability of biomolecules.

Exploring cryo-electron microscopy with molecular dynamics

Single particle analysis cryo-electron microscopy (EM) and molecular dynamics (MD) have been complimentary methods since cryo-EM was first applied to the field of structural biology. The relationship started by biasing structural models to fit low-resolution cryo-EM maps of large macromolecular complexes not amenable to crystallization. The connection between cryo-EM and MD evolved as cryo-EM maps improved in resolution, allowing advanced sampling algorithms to simultaneously refine backbone and sidechains. Moving beyond a single static snapshot, modern inferencing approaches integrate cryo-EM and MD to generate structural ensembles from cryo-EM map data or directly from the particle images themselves. We summarize the recent history of MD innovations in the area of cryo-EM modeling. The merits for the myriad of MD based cryo-EM modeling methods are discussed, as well as, the discoveries that were made possible by the integration of molecular modeling with cryo-EM. Lastly, current challenges and potential opportunities are reviewed.

Applications of deep learning in electron microscopy

We review the growing use of machine learning in electron microscopy (EM) driven in part by the availability of fast detectors operating at kiloHertz frame rates leading to large data sets that cannot be processed using manually implemented algorithms. We summarize the various network architectures and error metrics that have been applied to a range of EM-related problems including denoising and inpainting. We then provide a review of the application of these in both physical and life sciences, highlighting how conventional networks and training data have been specifically modified for EM.

Integrating dynamics into enzyme engineering

Enzyme engineering has become a widely adopted practice in research labs and industry. In parallel, the past decades have seen tremendous strides in characterizing the dynamics of proteins, using a growing array of methodologies. Importantly, links have been established between the dynamics of proteins and their function. Characterizing the dynamics of an enzyme prior to, and following, its engineering is beginning to inform on the potential of 'dynamic engineering', i.e. the rational modification of protein dynamics to alter enzyme function. Here we examine the state of knowledge at the intersection of enzyme engineering and protein dynamics, describe current challenges and highlight pioneering work in the nascent area of dynamic engineering.

Machine Learning at the Interface of Polymer Science and Biology: How Far Can We Go?

This Perspective outlines recent progress and future directions for using machine learning (ML), a data-driven method, to address critical questions in the design, synthesis, processing, and characterization of biomacromolecules. The achievement of these tasks requires the navigation of vast and complex chemical and biological spaces, difficult to accomplish with reasonable speed. Using modern algorithms and supercomputers, quantum physics methods are able to examine systems containing a few hundred interacting species and determine the probability of finding them in a particular region of phase space, thereby anticipating their properties. Likewise, modern approaches in chemistry and biomolecular simulation, supported by high performance computing, have culminated in producing data sets of escalating size and intrinsically high complexity. Hence, using ML to extract relevant information from these fields is of paramount importance to advance our understanding of chemical and biomolecular systems. At the heart of ML approaches lie statistical algorithms, which by evaluating a portion of a given data set, identify, learn, and manipulate the underlying rules that govern the whole data set. The assembly of a quality model to represent the data followed by the predictions and elimination of error sources are the key steps in ML. In addition to a growing infrastructure of ML tools to address complex problems, an increasing number of aspects related to our understanding of the fundamental properties of biomacromolecules are exposed to ML. These fields, including those residing at the interface of polymer science and biology (i.e., structure determination, de novo design, folding, and dynamics), strive to adopt and take advantage of the transformative power offered by approaches in the ML domain, which clearly has the potential of accelerating research in the field of biomacromolecules.

Enzymology and Dynamics by Cryogenic Electron Microscopy

Cryogenic electron microscopy (cryo-EM) has revolutionized the field of structural biology, particularly in solving the structures of large protein complexes or cellular machineries that play important biological functions. This review focuses on the contribution and future potential of cryo-EM in related emerging applications-enzymatic mechanisms and dynamic processes. Work on these subjects can benefit greatly from the capability of cryo-EM to solve the structures of specific protein complexes in multiple conditions, including variations in the buffer condition, ligands, and temperature, and to capture multiple conformational states, conformational change intermediates, and reaction intermediates. These studies can expand the structural landscape of specific proteins or protein complexes in multiple dimensions and drive new advances in the fields of enzymology and dynamic processes. The advantages and complementarity of cryo-EM relative to X-ray crystallography and nuclear magnetic resonance with regard to these applications are also addressed. Expected final online publication date for the Annual Review of Biophysics, Volume 51 is May 2022. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.

Cryo-EM Analyses Permit Visualization of Structural Polymorphism of Biological Macromolecules

The functions of biological macromolecules are often associated with conformational malleability of the structures. This phenomenon of chemically identical molecules with different structures is coined structural polymorphism. Conventionally, structural polymorphism is observed directly by structural determination at the density map level from X-ray crystal diffraction. Although crystallography approach can report the conformation of a macromolecule with the position of each atom accurately defined in it, the exploration of structural polymorphism and interpreting biological function in terms of crystal structures is largely constrained by the crystal packing. An alternative approach to studying the macromolecule of interest in solution is thus desirable. With the advancement of instrumentation and computational methods for image analysis and reconstruction, cryo-electron microscope (cryo-EM) has been transformed to be able to produce “in solution” structures of macromolecules routinely with resolutions comparable to crystallography but without the need of crystals. Since the sample preparation of single-particle cryo-EM allows for all forms co-existing in solution to be simultaneously frozen, the image data contain rich information as to structural polymorphism. The ensemble of structure information can be subsequently disentangled through three-dimensional (3D) classification analyses. In this review, we highlight important examples of protein structural polymorphism in relation to allostery, subunit cooperativity and function plasticity recently revealed by cryo-EM analyses, and review recent developments in 3D classification algorithms including neural network/deep learning approaches that would enable cryo-EM analyese in this regard. Finally, we brief the frontier of cryo-EM structure determination of RNA molecules where resolving the structural polymorphism is at dawn.

Machine-Learning Microstructure for Inverse Material Design

Metallurgy and material design have thousands of years' history and have played a critical role in the civilization process of humankind. The traditional trial-and-error method has been unprecedentedly challenged in the modern era when the number of components and phases in novel alloys keeps increasing, with high-entropy alloys as the representative. New opportunities emerge for alloy design in the artificial intelligence era. Here, a successful machine-learning (ML) method has been developed to identify the microstructure images with eye-challenging morphology for a number of martensitic and ferritic steels. Assisted by it, a new neural-network method is proposed for the inverse design of alloys with 20 components, which can accelerate the design process based on microstructure. The method is also readily applied to other material systems given sufficient microstructure images. This work lays the foundation for inverse alloy design based on microstructure images with extremely similar features.

Conquer by cryo-EM without physically dividing

This mini-review provides an update regarding the substantial progress that has been made in using single-particle cryo-EM to obtain high-resolution structures for proteins and other macromolecules whose particle sizes are smaller than 100 kDa. We point out that establishing the limits of what can be accomplished, both in terms of particle size and attainable resolution, serves as a guide for what might be expected when attempting to improve the resolution of small flexible portions of a larger structure using focused refinement approaches. These approaches, which involve computationally ignoring all but a specific, targeted region of interest on the macromolecules, is known as 'masking and refining,' and it thus is the computational equivalent of the 'divide and conquer' approach that has been used so successfully in X-ray crystallography. The benefit of masked refinement, however, is that one is able to determine structures in their native architectural context, without physically separating them from the biological connections that they require for their function. This mini-review also compares where experimental achievements currently stand relative to various theoretical estimates for the smallest particle size that can be successfully reconstructed to high resolution. Since it is clear that a substantial gap still remains between the two, we briefly recap the areas in which further improvement seems possible, both in equipment and in methods.

New Frontiers for Machine Learning in Protein Science

Protein function is fundamentally reliant on inter-molecular interactions that underpin the ability of proteins to form complexes driving biological processes in living cells. Increasingly, such interactions are recognised as being formed between proteins that exist on a broad spectrum of dynamic conformational states and levels of intrinsic disorder. Additionally, the sizes of the structures formed can range from simple binary complexes to large dynamic biomolecular condensates measuring 100 nm or more. Understanding the parameters that govern such interactions, how they form, how they lead to function and what happens when they take place in unintended manners and lead to disease, represent some of the core questions for molecular biosciences. In light of recent advances made in solving the protein folding problem by machine learning methods, we discuss here the challenges and opportunities brought by these new data-driven approaches for the next frontiers of biomolecular science.

Comprehensive strategies of machine-learning-based quantitative structure-activity relationship models

Early quantitative structure-activity relationship (QSAR) technologies have unsatisfactory versatility and accuracy in fields such as drug discovery because they are based on traditional machine learning and interpretive expert features. The development of Big Data and deep learning technologies significantly improve the processing of unstructured data and unleash the great potential of QSAR. Here we discuss the integration of wet experiments (which provide experimental data and reliable verification), molecular dynamics simulation (which provides mechanistic interpretation at the atomic/molecular levels), and machine learning (including deep learning) techniques to improve QSAR models. We first review the history of traditional QSAR and point out its problems. We then propose a better QSAR model characterized by a new iterative framework to integrate machine learning with disparate data input. Finally, we discuss the application of QSAR and machine learning to many practical research fields, including drug development and clinical trials.

Experimental demonstration of the novel "van-Hove integral method (vHI)" for measuring diffusive dynamics by elastic neutron scattering

Quasi-elastic neutron scattering (QENS)-based on the seminal work of Nobel Laureate Brockhouse-has been one of the major methods for studying pico-second to nano-second diffusive dynamics over the past 70 years. This is regarded as an "inelastic" method for dynamics. In contrast, we recently proposed a new neutron-scattering method for dynamics, which uses the elastic line of the scattering to access system dynamics directly in the time domain (Benedetto and Kearley in Sci Rep 9:11284, 2019). This new method has been denoted "vHI" that stands for "van Hove Integral". The reason is that, under certain conditions, the measured elastic intensity corresponds to the running-time integral of the intermediate scattering function, [Formula: see text], up to a time that is inversely proportional to the energy band-width incident on the sample. As a result, [Formula: see text] is accessed from the time derivative of the measured vHI profile. vHI has been supported by numerical and Monte-Carlo simulations, but has been difficult to validate experimentally due to the lack of a suitable instrument. Here we show that vHI works in practice, which we achieved by using a simple modification to the standard QENS backscattering spectrometer methodology. Basically, we varied the neutron-energy band-widths incident at the sample via a step-wise variation of the frequency of the monochromator Doppler-drive. This provides a measurement of the vHI profile at the detectors. The same instrument and sample were also used in standard QENS mode for comparison. The intermediate scattering functions, [Formula: see text], obtained by the two methods-vHI and QENS-are strikingly similar providing a direct experimental validation of the vHI method. Perhaps surprisingly, the counting statistics of the two methods are comparable even though the instrument used was expressly designed for QENS. This shows that the methodology modification adopted here can be used in practice to access vHI profiles at many of the backscattering spectrometers worldwide. We also show that partial integrations of the measured QENS spectrum cannot provide the vHI profile, which clarifies a common misconception. At the same time, we show a novel approach which does access [Formula: see text] from QENS spectra.

Deep Learning-Based Advances in Protein Structure Prediction

Obtaining an accurate description of protein structure is a fundamental step toward understanding the underpinning of biology. Although recent advances in experimental approaches have greatly enhanced our capabilities to experimentally determine protein structures, the gap between the number of protein sequences and known protein structures is ever increasing. Computational protein structure prediction is one of the ways to fill this gap. Recently, the protein structure prediction field has witnessed a lot of advances due to Deep Learning (DL)-based approaches as evidenced by the success of AlphaFold2 in the most recent Critical Assessment of protein Structure Prediction (CASP14). In this article, we highlight important milestones and progresses in the field of protein structure prediction due to DL-based methods as observed in CASP experiments. We describe advances in various steps of protein structure prediction pipeline viz. protein contact map prediction, protein distogram prediction, protein real-valued distance prediction, and Quality Assessment/refinement. We also highlight some end-to-end DL-based approaches for protein structure prediction approaches. Additionally, as there have been some recent DL-based advances in protein structure determination using Cryo-Electron (Cryo-EM) microscopy based, we also highlight some of the important progress in the field. Finally, we provide an outlook and possible future research directions for DL-based approaches in the protein structure prediction arena.