Automatic recognition of ligands in electron density by machine learning

The current role and evolution of X-ray crystallography in drug discovery and development

INTRODUCTION

Macromolecular X-ray crystallography and cryo-EM are currently the primary techniques used to determine the three-dimensional structures of proteins, nucleic acids, and viruses. Structural information has been critical to drug discovery and structural bioinformatics. The integration of artificial intelligence (AI) into X-ray crystallography has shown great promise in automating and accelerating the analysis of complex structural data, further improving the efficiency and accuracy of structure determination.

AREAS COVERED

This review explores the relationship between X-ray crystallography and other modern structural determination methods. It examines the integration of data acquired from diverse biochemical and biophysical techniques with those derived from structural biology. Additionally, the paper offers insights into the influence of AI on X-ray crystallography, emphasizing how integrating AI with experimental approaches can revolutionize our comprehension of biological processes and interactions.

EXPERT OPINION

Investing in science is crucially emphasized due to its significant role in drug discovery and advancements in healthcare. X-ray crystallography remains an essential source of structural biology data for drug discovery. Recent advances in biochemical, spectroscopic, and bioinformatic methods, along with the integration of AI techniques, hold the potential to revolutionize drug discovery when effectively combined with robust data management practices.

Crystal structure of a variable region segment of Leptospira host-interacting outer surface protein, LigA, reveals the orientation of Ig-like domains

Leptospiral immunoglobulin-like (Lig) protein family is a surface-exposed protein from the pathogenic Leptospira. The Lig protein family has been identified as an essential virulence factor of L. interrogan. One of the family members, LigA, contains 13 homologous tandem repeats of bacterial Ig-like (Big) domains in its extracellular portion. It is crucial in binding with the host's Extracellular matrices (ECM) and complement factors. However, its vital role in the invasion and evasion of pathogenic Leptospira, structural details, and domain organization of the extracellular portion of this protein are not explored thoroughly. Here, we described the first high-resolution crystal structure of a variable region segment (LigA8-9) of LigA at 1.87 Å resolution. The structure showed some remarkably distinctive aspects compared with the most closely related Immunoglobulin superfamily (IgSF) members. The structure illustrated the relative orientation of two domains and highlighted the role of the linker region in the domain orientation. We also observed an apparent electron density of Ca²⁺ ions coordinated with a proper interacting geometry within the protein. Molecular dynamic simulations demonstrated the involvement of a linker salt bridge in providing rigidity between the two domains. Our study proposes an overall arrangement of Ig-like domains in the LigA protein. The structural understanding of the extracellular portion of LigA and its interaction with the ECM provides insight into developing new therapeutics directed toward leptospirosis.

Continuous Validation Across Macromolecular Structure Determination Process

The overall quality of the experimentally determined structures contained in the PDB is exceptionally high, mainly due to the continuous improvement of model building and structural validation programs. Improving reproducibility on a large scale requires expanding the concept of validation in structural biology and all other disciplines to include a broader framework that encompasses the entire project. A successful approach to science requires diligent attention to detail and a focus on the future. An earnest commitment to data availability and reuse is essential for scientific progress, be that by human minds or artificial intelligence.

Analysis of protein structures containing HEPES and MES molecules

X‐ray crystallography is the main experimental method behind ligand–macromolecule complexes found in the Protein Data Bank (PDB). Applying bioinformatics methods to such structural data can fuel drug discovery, albeit under the condition that the information is correct. Regrettably, a small number of structures in the PDB are of suboptimal quality due to incorrectly identified and modeled ligands in protein–ligand complexes. In this paper, we combine a theoretical‐graph approach, nuclear density estimates, bioinformatics methods, and prior chemical knowledge to analyze two non‐physiological ligands, HEPES and MES, that are frequent components of crystallization and purifications buffers. Our analysis includes quantum mechanics calculations and Cambridge Structure Database (CSD) queries to define the ideal conformation of these ligands, geometry analysis of PDB deposits regarding several quality factors, and a search for homologous structures to identify other small molecules that could bind in place of the parasitic ligand. Our results highlight the need for careful refinement of macromolecule–ligand complexes and better validation tools that integrate results from all relevant resources.

PDB Code(s): 3K4L, 3PYI, 5T6L, 6BB0, 1PJX, 3O4P, 6WCF, 3DKE, 3E10, 6G38, 4E8R, 4Z91, 3E9F, 1MOS, 1MOQ, 2ESB, 1VHR, 4P66 and 6NNI;

Refolding of lid subdomain of SARS-CoV-2 nsp14 upon nsp10 interaction releases exonuclease activity

During RNA replication, coronaviruses require proofreading to maintain the integrity of their large genomes. Nsp14 associates with viral polymerase complex to excise the mismatched nucleotides. Aside from the exonuclease activity, nsp14 methyltransferase domain mediates cap methylation, facilitating translation initiation and protecting viral RNA from recognition by the innate immune sensors. The nsp14 exonuclease activity is modulated by a protein co-factor nsp10. While the nsp10/nsp14 complex structure is available, the mechanistic basis for nsp10-mediated modulation remains unclear in the absence of the nsp14 structure. Here, we provide a crystal structure of nsp14 in an apo-form. Comparative analysis of the apo- and nsp10-bound structures explain the modulatory role of the co-factor protein and reveal the allosteric nsp14 control mechanism essential for drug discovery. Further, the flexibility of the N-terminal lid of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) nsp14 structure presented in this study rationalizes the recently proposed idea of nsp14/nsp10/nsp16 ternary complex.

Molecular structure recognition by blob detection

Molecular structure recognition is fundamental in computational chemistry. The most common approach is to calculate the root mean square deviation (RMSD) between two sets of molecular coordinates. However, this method does not perform well for large molecules. In this work, a new method is proposed for structure comparison. Blob detection is used for recognizing structural features. Fragmentation of molecules is proposed as the pre-treatment. Mapping between blobs and atoms is developed as the post-treatment. A set of key parameters important for blob detections are determined. The dissimilarity is quantified by calculating the Euclidean metric of the blob vectors. The overall algorithm is found to be accurate to distinguish structural dissimilarity. The method has potential to be combined with other pattern recognition techniques for new chemistry discoveries.

Molecular structure recognition is fundamental in computational chemistry.

Recognizing and validating ligands with CheckMyBlob

Structure-guided drug design depends on the correct identification of ligands in crystal structures of protein complexes. However, the interpretation of the electron density maps is challenging and often burdened with confirmation bias. Ligand identification can be aided by automatic methods such as CheckMyBlob, a machine learning algorithm that learns to generalize ligand descriptions from sets of moieties deposited in the Protein Data Bank. Here, we present the CheckMyBlob web server, a platform that can identify ligands in unmodeled fragments of electron density maps or validate ligands in existing models. The server processes PDB/mmCIF and MTZ files and returns a ranking of 10 most likely ligands for each detected electron density blob along with interactive 3D visualizations. Additionally, for each prediction/validation, a plugin script is generated that enables users to conduct a detailed analysis of the server results in Coot. The CheckMyBlob web server is available at https://checkmyblob.bioreproducibility.org.

Detecting anomalies in X-ray diffraction images using convolutional neural networks

Our understanding of life is based upon the interpretation of macromolecular structures and their dynamics. Almost 90% of currently known macromolecular models originated from electron density maps constructed using X-ray diffraction images. Even though diffraction images are critical for structure determination, due to their vast amounts and noisy, non-intuitive nature, their quality is rarely inspected. In this paper, we use recent advances in machine learning to automatically detect seven types of anomalies in X-ray diffraction images. For this purpose, we utilize a novel X-ray beam center detection algorithm, propose three different image representations, and compare the predictive performance of general-purpose classifiers and deep convolutional neural networks (CNNs). In benchmark tests on a set of 6,311 X-ray diffraction images, the proposed CNN achieved between 87% and 99% accuracy depending on the type of anomaly. Experimental results show that the proposed anomaly detection system can be considered suitable for early detection of sub-optimal data collection conditions and malfunctions at X-ray experimental stations.

Rapid response to emerging biomedical challenges and threats

As part of the global mobilization to combat the present pandemic, almost 100 000 COVID-19-related papers have been published and nearly a thousand models of macromolecules encoded by SARS-CoV-2 have been deposited in the Protein Data Bank within less than a year. The avalanche of new structural data has given rise to multiple resources dedicated to assessing the correctness and quality of structural data and models. Here, an approach to evaluate the massive amounts of such data using the resource https://covid19.bioreproducibility.org is described, which offers a template that could be used in large-scale initiatives undertaken in response to future biomedical crises. Broader use of the described methodology could considerably curtail information noise and significantly improve the reproducibility of biomedical research.

Covid-19.bioreproducibility.org: A web resource for SARS-CoV-2-related structural models

The COVID-19 pandemic has triggered numerous scientific activities aimed at understanding the SARS-CoV-2 virus and ultimately developing treatments. Structural biologists have already determined hundreds of experimental X-ray, cryo-EM, and NMR structures of proteins and nucleic acids related to this coronavirus, and this number is still growing. To help biomedical researchers, who may not necessarily be experts in structural biology, navigate through the flood of structural models, we have created an online resource, covid19.bioreproducibility.org, that aggregates expert-verified information about SARS-CoV-2-related macromolecular models. In this article, we describe this web resource along with the suite of tools and methodologies used for assessing the structures presented therein.

Deflating the RNA Mg2+ bubble. Stereochemistry to the rescue!

Proper evaluation of the ionic structure of biomolecular systems through X ray and cryo-EM techniques remains challenging but is essential for advancing our understanding of the underlying structure/activity/solvent relationships. However, numerous studies overestimate the number of Mg2+ in deposited structures due to assignment errors finding their origin in improper consideration of stereochemical rules. Herein, to tackle such issues, we re-evaluate the PDBid 6QNR and 6SJ6 models of the ribosome ionic structure. We establish that stereochemical principles need to be carefully pondered when evaluating ion binding features, even when K+ anomalous signals are available as it is the case for the 6QNR PDB entry. For ribosomes, assignment errors can result in misleading conceptions of their solvent structure. For instance, present stereochemical analysis result in a significant decrease of the number of assigned Mg2+ in 6QNR, suggesting that K+ and not Mg2+ is the prevalent ion in the ribosome 1st solvation shell. We stress that the use of proper stereochemical guidelines in combination or not with other identification techniques, such as those pertaining to the detection of transition metals, of some anions and of K+ anomalous signals, is critical for deflating the current Mg2+ bubble witnessed in many ribosome and other RNA structures. We also stress that for the identification of lighter ions such as Mg2+, Na+, …, for which no anomalous signals can be detected, stereochemistry coupled with high resolution structures (<2.4 Å) remain the best currently available option.

On the evolution of the quality of macromolecular models in the PDB

Crystallographic models of biological macromolecules have been ranked using the quality criteria associated with them in the Protein Data Bank (PDB). The outcomes of this quality analysis have been correlated with time and with the journals that published papers based on those models. The results show that the overall quality of PDB structures has substantially improved over the last ten years, but this period of progress was preceded by several years of stagnation or even depression. Moreover, the study shows that the historically observed negative correlation between journal impact and the quality of structural models presented therein seems to disappear as time progresses.

MemBlob database and server for identifying transmembrane regions using cryo-EM maps

SUMMARY

The identification of transmembrane helices in transmembrane proteins is crucial, not only to understand their mechanism of action but also to develop new therapies. While experimental data on the boundaries of membrane-embedded regions are sparse, this information is present in cryo-electron microscopy (cryo-EM) density maps and it has not been utilized yet for determining membrane regions. We developed a computational pipeline, where the inputs of a cryo-EM map, the corresponding atomistic structure, and the potential bilayer orientation determined by TMDET algorithm of a given protein result in an output defining the residues assigned to the bulk water phase, lipid interface and the lipid hydrophobic core. Based on this method, we built a database involving published cryo-EM protein structures and a server to be able to compute this data for newly obtained structures.

AVAILABILITY AND IMPLEMENTATION

http://memblob.hegelab.org.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

The Integrated Resource for Reproducibility in Macromolecular Crystallography: Experiences of the first four years

It has been increasingly recognized that preservation and public accessibility of primary experimental data are cornerstones necessary for the reproducibility of empirical sciences. In the field of molecular crystallography, many journals now recommend that authors of manuscripts presenting a new crystal structure should deposit their primary experimental data (X-ray diffraction images) to one of the dedicated resources created in recent years. Here, we describe our experiences developing the Integrated Resource for Reproducibility in Molecular Crystallography (IRRMC) and describe several examples of a crucial role that diffraction data can play in improving previously determined protein structures. In its first four years, several hundred crystallographers have deposited data from over 5200 diffraction experiments performed at over 60 different synchrotron beamlines or home sources all over the world. In addition to improving the resource and curating submitted data, we have been building a pipeline for extraction or, in some cases, reconstruction of the metadata necessary for seamless automated processing. Preliminary analysis indicates that about 95% of the archived data can be automatically reprocessed. A high rate of reprocessing success shows the feasibility of using the automated metadata extraction and automated processing as a validation step for the deposition of raw diffraction images. The IRRMC is guided by the Findable, Accessible, Interoperable, and Reusable data management principles.

Molstack: A platform for interactive presentations of electron density and cryo-EM maps and their interpretations

In the Special Issue on Tools for Protein Science in 2018, we presented Molstack: a concept of a cloud-based platform for sharing electron density maps and their interpretations. Molstack is a web platform that allows the interactive visualization of density maps through the simultaneous presentation of multiple datasets and models in a way that allows for easy pairwise comparison. We anticipated that the users of this conceptually simple platform would find many different uses for their projects, and we were not mistaken. We have observed researchers use Molstack to present experimental evidence for their models in the form of electron density maps, omit maps, and anomalous difference density maps. Users also employed Molstack to present alternative interpretations of densities, including rerefinements and speculative interpretations. While we anticipated these types of projects to be the main use cases, we were pleased to see Molstack used to display superpositions of different models, as a tool for story-driven presentations, and for collaboration as well. Here, we present developments in the platform that were driven by user feedback, highlight several cases that used Molstack to enhance the publication, and discuss possible directions for the platform.

Nucleobase carbonyl groups are poor Mg2+ inner-sphere binders but excellent monovalent ion binders-a critical PDB survey

Precise knowledge of Mg²⁺ inner-sphere binding site properties is vital for understanding the structure and function of nucleic acid systems. Unfortunately, the PDB, which represents the main source of Mg²⁺ binding sites, contains a substantial number of assignment issues that blur our understanding of the functions of these ions. Here, following a previous study devoted to Mg²⁺ binding to nucleobase nitrogens, we surveyed nucleic acid X-ray structures from the PDB with resolutions ≤2.9 Å to classify the Mg²⁺ inner-sphere binding patterns to nucleotide carbonyl, ribose hydroxyl, cyclic ether, and phosphodiester oxygen atoms. From this classification, we derived a set of "prior-knowledge" nucleobase Mg²⁺ binding sites. We report that crystallographic examples of trustworthy nucleobase Mg²⁺ binding sites are fewer than expected since many of those are associated with misidentified Na⁺ or K⁺ We also emphasize that binding of Na⁺ and K⁺ to nucleic acids is much more frequent than anticipated. Overall, we provide evidence derived from X-ray structures that nucleobases are poor inner-sphere binders for Mg²⁺ but good binders for monovalent ions. Based on strict stereochemical criteria, we propose an extended set of guidelines designed to help in the assignment and validation of ions directly contacting nucleobase and ribose atoms. These guidelines should help in the interpretation of X-ray and cryo-EM solvent density maps. When borderline Mg²⁺ stereochemistry is observed, alternative placement of Na⁺, K⁺, or Ca²⁺ must be considered. We also critically examine the use of lanthanides (Yb³⁺, Tb³⁺) as Mg²⁺ substitutes in crystallography experiments.

Structural Variability in the RLR-MAVS Pathway and Sensitive Detection of Viral RNAs

Cells need high-sensitivity detection of non-self molecules in order to fight against pathogens. These cellular sensors are thus of significant importance to medicinal purposes, especially for treating novel emerging pathogens. RIG-I-like receptors (RLRs) are intracellular sensors for viral RNAs (vRNAs). Their active forms activate mitochondrial antiviral signaling protein (MAVS) and trigger downstream immune responses against viral infection. Functional and structural studies of the RLR-MAVS signaling pathway have revealed significant supramolecular variability in the past few years, which revealed different aspects of the functional signaling pathway. Here I will discuss the molecular events of RLR-MAVS pathway from the angle of detecting single copy or a very low copy number of vRNAs in the presence of non-specific competition from cytosolic RNAs, and review key structural variability in the RLR / vRNA complexes, the MAVS helical polymers, and the adapter-mediated interactions between the active RLR / vRNA complex and the inactive MAVS in triggering the initiation of the MAVS filaments. These structural variations may not be exclusive to each other, but instead may reflect the adaptation of the signaling pathways to different conditions or reach different levels of sensitivity in its response to exogenous vRNAs.

Refining the macromolecular model - achieving the best agreement with the data from X-ray diffraction experiment

Refinement of macromolecular X-ray crystal structures involves using complex software with hundreds of different settings. The complexity of underlying concepts and the sheer amount sof instructions may make it difficult for less experienced crystallographers to achieve optimal results in their refinements. This tutorial review offers guidelines for choosing the best settings for the reciprocal-space refinement of macromolecular models and provides practical tips for manual model correction. To help aspiring crystallographers navigate the process, some of the most practically important concepts of protein structure refinement are described. Among the topics covered are the use and purpose of R-free, geometrical restraints, restraints on atomic displacement parameters (ADPs), refinement weights, various parametrizations of ADPs (full anisotropic refinement and TLS), and omit maps. We also give practical tips for manual model correction in Coot, modelling of side-chains with poor or missing density, and ligand identification, fitting, and refinement.