Machine learning (ML) has revolutionised the field of structure-based drug design (SBDD) in recent years. During the training stage, ML techniques typically analyse large amounts of experimentally determined data to create predictive models in order to inform the drug discovery process. Deep learning (DL) is a subfield of ML, that relies on multiple layers of a neural network to extract significantly more complex patterns from experimental data, and has recently become a popular choice in SBDD. This review provides a thorough summary of the recent DL trends in SBDD with a particular focus on de novo drug design, binding site prediction, and binding affinity prediction of small molecules.

While RNA folding was originally seen as a simple problem to solve, it has been shown that the promiscuous interactions of the nucleobases result in structural polymorphism, with several competing structures generally observed for non-coding RNA. This inherent complexity limits our understanding of these molecules from experiments alone, and computational methods are commonly used to study RNA. Here, we discuss three advanced sampling schemes, namely Hamiltonian-replica exchange molecular dynamics (MD), ratchet-and-pawl MD and discrete path sampling, as well as the HiRE-RNA coarse-graining scheme, and highlight how these approaches are complementary with reference to recent case studies. While all computational methods have their shortcomings, the plurality of simulation methods leads to a better understanding of experimental findings and can inform and guide experimental work on RNA polymorphism.

Chapter 1: COVID-19 pathogenesis poses paradoxes difficult to explain with traditional physiology. For instance, since type II pneumocytes are considered the primary cellular target of SARS-CoV-2; as these produce pulmonary surfactant (PS), the possibility that insufficient PS plays a role in COVID-19 pathogenesis has been raised. However, the opposite of predicted high alveolar surface tension is found in many early COVID-19 patients: paradoxically normal lung volumes and high compliance occur, with profound hypoxemia. That ‘COVID anomaly’ was quickly rationalised by invoking traditional vascular mechanisms–mainly because of surprisingly preserved alveolar surface in early hypoxemic cases. However, that quick rejection of alveolar damage only occurred because the actual mechanism of gas exchange has long been presumed to be non-problematic, due to diffusion through the alveolar surface. On the contrary, we provide physical chemical evidence that gas exchange occurs by an process of expansion and contraction of the three-dimensional structures of PS and its associated proteins. This view explains anomalous observations from the level of cryo-TEM to whole individuals. It encompasses results from premature infants to the deepest diving seals. Once understood, the COVID anomaly dissolves and is straightforwardly explained as covert viral damage to the 3D structure of PS, with direct treatment implications. As a natural experiment, the SARS-CoV-2 virus itself has helped us to simplify and clarify not only the nature of dyspnea and its relationship to pulmonary compliance, but also the fine detail of the PS including such features as water channels which had heretofore been entirely unexpected.

Magnesium ions (Mg2+) are vital for RNA structure and cellular functions. Present efforts in RNA structure determination and understanding of RNA functions are hampered by the inability to accurately locate Mg2+ ions in an RNA. Here we present a machine-learning method, originally developed for computer visual recognition, to predict Mg2+ binding sites in RNA molecules. By incorporating geometrical and electrostatic features of RNA, we capture the key ingredients of Mg2+-RNA interactions, and from deep learning, predict the Mg2+ density distribution. Five-fold cross-validation on a dataset of 177 selected Mg2+-containing structures and comparisons with different methods validate the approach. This new approach predicts Mg2+ binding sites with notably higher accuracy and efficiency. More importantly, saliency analysis for eight different Mg2+ binding motifs indicates that the model can reveal critical coordinating atoms for Mg2+ ions and ion-RNA inner/outer-sphere coordination. Furthermore, implementation of the model uncovers new Mg2+ binding motifs. This new approach may be combined with X-ray crystallography structure determination to pinpoint the metal ion binding sites.

When life arose from prebiotic molecules 3.5 billion years ago, what came first? Informational molecules (RNA, DNA), functional ones (proteins), or something else? We argue here for a different logic: rather than seeking a molecule type, we seek a dynamical process. Biology required an ability to evolve before it could choose and optimise materials. We hypothesise that the evolution process was rooted in the peptide folding process. Modelling shows how short random peptides can collapse in water and catalyse the elongation of others, powering both increased folding stability and emergent autocatalysis through a disorder-to-order process.

Models of insulin secretory vesicles from pancreatic beta cells have been created using the cellPACK suite of tools to research, curate, construct and visualise the current state of knowledge. The model integrates experimental information from proteomics, structural biology, cryoelectron microscopy and X-ray tomography, and is used to generate models of mature and immature vesicles. A new method was developed to generate a confidence score that reconciles inconsistencies between three available proteomes using expert annotations of cellular localisation. The models are used to simulate soft X-ray tomograms, allowing quantification of features that are observed in experimental tomograms, and in turn, allowing interpretation of X-ray tomograms at the molecular level.

Biomolecular recognition including binding of small molecules, peptides and proteins to their target receptors plays a key role in cellular function and has been targeted for therapeutic drug design. However, the high flexibility of biomolecules and slow binding and dissociation processes have presented challenges for computational modelling. Here, we review the challenges and computational approaches developed to characterise biomolecular binding, including molecular docking, molecular dynamics simulations (especially enhanced sampling) and machine learning. Further improvements are still needed in order to accurately and efficiently characterise binding structures, mechanisms, thermodynamics and kinetics of biomolecules in the future.

α-Synuclein is a small neuronal protein that reversibly associates with lipid membranes. The membrane interactions are believed to be central to the healthy function of this protein involved in synaptic plasticity and neurotransmitter release. α-Synuclein has been speculated to induce vesicle fusion as well as fission, processes which are analogous to each other but proceed in different directions and involve different driving forces. In the current work, we analyse α-synuclein-induced small unilamellar vesicle deformation from a thermodynamics point of view. We show that the structures interpreted in the literature as fusion intermediates are in fact a stable deformed state and neither fusion nor vesicle clustering occurs. We speculate on the driving force for the observed deformation and put forward a hypothesis that α-synuclein self-assembly on the lipid membrane precedes and induces membrane remodelling.

Interactions between proteins and single-stranded DNA (ssDNA) are crucial for many fundamental biological processes, including DNA replication and genetic recombination. Thus, understanding detailed mechanisms of these interactions is necessary to uncover regulatory rules occurring in all living cells. The RNA-binding Hfq is a pleiotropic bacterial regulator that mediates many aspects of nucleic acid metabolism. The protein notably mediates mRNA stability and translation efficiency by using stress-related small regulatory RNA as cofactors. In addition, Hfq helps to compact double-stranded DNA. In this paper, we focused on the action of Hfq on ssDNA. A combination of experimental methodologies, including spectroscopy and molecular imaging, has been used to probe the interactions of Hfq and its amyloid C-terminal region with ssDNA. Our analysis revealed that Hfq binds to ssDNA. Moreover, we demonstrate for the first time that Hfq drastically changes the structure and helical parameters of ssDNA, mainly due to its C-terminal amyloid-like domain. The formation of the nucleoprotein complexes between Hfq and ssDNA unveils important implications for DNA replication and recombination.

Recent breakthroughs in deep learning-based protein structure prediction show that it is possible to obtain highly accurate models for a wide range of difficult protein targets for which only the amino acid sequence is known. The availability of accurately predicted models from sequences can potentially revolutionise many modelling approaches in structural biology, including the interpretation of cryo-EM density maps. Although atomic structures can be readily solved from cryo-EM maps of better than 4 Å resolution, it is still challenging to determine accurate models from lower-resolution density maps. Here, we report on the benefits of models predicted by AlphaFold2 (the best-performing structure prediction method at CASP14) on cryo-EM refinement using the Phenix refinement suite for AlphaFold2 models. To study the robustness of model refinement at a lower resolution of interest, we introduced hybrid maps (i.e. experimental cryo-EM maps) filtered to lower resolutions by real-space convolution. The AlphaFold2 models were refined to attain good accuracies above 0.8 TM scores for 9 of the 13 cryo-EM maps. TM scores improved for AlphaFold2 models refined against all 13 cryo-EM maps of better than 4.5 Å resolution, 8 hybrid maps of 6 Å resolution, and 3 hybrid maps of 8 Å resolution. The results show that it is possible (at least with the Phenix protocol) to extend the refinement success below 4.5 Å resolution. We even found isolated cases in which resolution lowering was slightly beneficial for refinement, suggesting that high-resolution cryo-EM maps might sometimes trap AlphaFold2 models in local optima.

Copper is a trace element vital to many cellular functions. Yet its abnormal levels are toxic to cells, provoking a variety of severe diseases. The high affinity copper transporter 1 (CTR1), being the main in-cell copper [Cu(I)] entry route, tightly regulates its cellular uptake via a still elusive mechanism. Here, all-atoms simulations unlock the molecular terms of Cu(I) transport in eukaryotes disclosing that the two methionine (Met) triads, forming the selectivity filter, play an unprecedented dual role both enabling selective Cu(I) transport and regulating its uptake rate thanks to an intimate coupling between the conformational plasticity of their bulky side chains and the number of bound Cu(I) ions. Namely, the Met residues act as a gate reducing the Cu(I) import rate when two ions simultaneously bind to CTR1. This may represent an elegant autoregulatory mechanism through which CTR1 protects the cells from excessively high, and hence toxic, in-cell Cu(I) levels. Overall, our outcomes resolve fundamental questions in CTR1 biology and open new windows of opportunity to tackle diseases associated with an imbalanced copper uptake.

SARS-CoV-2 nucleocapsid (N) protein plays the essential roles in key steps of the viral life cycle, thus representing a top drug target. Functionality of N protein including liquid–liquid phase separation (LLPS) depends on its interaction with nucleic acids. Only the variants with N proteins functional in binding nucleic acids might survive and spread in evolution and indeed, the residues critical for binding nucleic acids are highly conserved. Hydroxychloroquine (HCQ) was shown to prevent the transmission in a large-scale clinical study in Singapore but so far, no specific SARS-CoV-2 protein was experimentally identified to be targeted by HCQ. Here by NMR, we unambiguously decode that HCQ specifically binds NTD and CTD of N protein with Kd of 112.1 and 57.1 μM, respectively to inhibit their interaction with nucleic acid, as well as to disrupt LLPS. Most importantly, HCQ-binding residues are identical in SARS-CoV-2 variants and therefore HCQ is likely effective to different variants. The results not only provide a structural basis for the anti-SARS-CoV-2 activity of HCQ, but also renders HCQ to be the first known drug capable of targeting LLPS. Furthermore, the unique structure of the HCQ-CTD complex suggests a promising strategy for design of better anti-SARS-CoV-2 drugs from HCQ.

Modafinil is a mild psychostimulant-like drug enhancing wakefulness, improving attention and developing performance in various cognitive tasks, but its mechanism of action is not completely understood. This is the first combination of amperometry, electrochemical cytometry and mass spectrometry to interrogate the mechanism of action of a drug, here modafinil, at cellular and sub-cellular level. We employed single-cell amperometry (SCA) and intracellular vesicle impact electrochemical cytometry (IVIEC) to investigate the alterations in exocytotic release and vesicular catecholamine storage following modafinil treatment. The SCA results reveal that modafinil slows down the exocytosis process so that, the number of catecholamines released per exocytotic event is enhanced in the modafinil-treated cells. Also, IVIEC results offer an upregulation effect of modafinil on the vesicular catecholamine storage. Mass spectrometry imaging by time-of-flight secondary ion mass spectrometry (ToF-SIMS) illustrates that treatment with modafinil reduces the cylindrical-shaped phosphatidylcholine at the cellular membrane, while the high curvature lipids with conical structures such as phosphatidylethanolamine and phosphatidylinositol are elevated after modafinil treatment. Combining the results obtained by SCA, IVIEC and ToF-SIMS suggests that modafinil-treated cells release a larger portion of their vesicular content at least in part by changing the lipid composition of the cell membrane, suggesting regulation of cognition.

The human chaperone DNAJB6b increases the solubility of proteins involved in protein aggregation diseases and suppresses the nucleation of amyloid structures. Due to such favourable properties, DNAJB6b has gained increasing attention over the past decade. The understanding of how DNAJB6b operates on a molecular level may aid the design of inhibitors against amyloid formation. In this work, fundamental aspects of DNAJB6b self-assembly have been examined, providing a basis for future experimental designs and conclusions. The results imply the formation of large chaperone clusters in a concentration-dependent manner. Microfluidic diffusional sizing (MDS) was used to evaluate how DNAJB6b average hydrodynamic radius varies with concentration. We found that, in 20 mM sodium phosphate buffer, 0.2 mM EDTA, at pH 8.0 and room temperature, DNAJB6b displays a micellar behaviour, with a critical micelle concentration (CMC) of around 120 nM. The average hydrodynamic radius appears to be concentration independent between ∼10 μM and 100 μM, with a mean radius of about 12 nm. The CMC found by MDS is supported by native agarose gel electrophoresis and the size distribution appears bimodal in the DNAJB6b concentration range ∼100 nM to 4 μM.