Are we fitting data or noise? Analysing the predictive power of commonly used datasets in drug-, materials-, and molecular-discovery.

Data-driven techniques for establishing quantitative structure property relations are a pillar of modern materials and molecular discovery. Fuelled by the recent progress in deep learning methodology and the abundance of new algorithms, it is tempting to chase benchmarks and incrementally build ever more capable machine learning (ML) models. While model evaluation has made significant progress, the intrinsic limitations arising from the underlying experimental data are often overlooked. In the chemical sciences data collection is costly, thus datasets are small and experimental errors can be significant. These limitations of such datasets affect their predictive power, a fact that is rarely considered in a quantitative way. In this study, we analyse commonly used ML datasets for regression and classification from drug discovery, molecular discovery, and materials discovery. We derived maximum and realistic performance bounds for nine such datasets by introducing noise based on estimated or actual experimental errors. We then compared the estimated performance bounds to the reported performance of leading ML models in the literature. Out of the nine datasets and corresponding ML models considered, four were identified to have reached or surpassed dataset performance limitations and thus, they may potentially be fitting noise. More generally, we systematically examine how data range, the magnitude of experimental error, and the number of data points influence dataset performance bounds. Alongside this paper, we release the Python package NoiseEstimator and provide a web-based application for computing realistic performance bounds. This study and the resulting tools will help practitioners in the field understand the limitations of datasets and set realistic expectations for ML model performance. This work stands as a reference point, offering analysis and tools to guide development of future ML models in the chemical sciences.

Exploring the influence of pore shape on conductance and permeation.

There are increasing numbers of ion channel structures featuring heteromeric subunit assembly, exemplified by synaptic α1ßB glycine and α4ß2 nicotinic receptors. These structures exhibit inherent pore asymmetry, but the relevance of this to function is unknown. Furthermore, molecular dynamics simulations performed on symmetrical homomeric channels often lead to thermal distortion whereby conformations of the resulting ensemble are also asymmetrical. When functionally annotating ion channels, researchers often rely on minimal constrictions determined via radius-profile calculations performed with computer programs, such as HOLE or CHAP, coupled with an assessment of pore hydrophobicity. However, such tools typically employ spherical probe particles, limiting their ability to accurately capture pore asymmetry. Here, we introduce an algorithm that employs ellipsoidal probe particles, enabling a more comprehensive representation of the pore geometry. Our analysis reveals that the use of nonspherical ellipsoids for pore characterization provides a more accurate and easily interpretable depiction of conductance. To quantify the implications of pore asymmetry on conductance, we systematically investigated carbon nanotubes with varying degrees of pore asymmetry as model systems. The conductance through these channels shows surprising effects that would otherwise not be predicted with spherical probes. The results have broad implications not only for the functional annotation of biological ion channels but also for the design of synthetic channel systems for use in areas such as water filtration. Furthermore, we make use of the more accurate characterization of channel pores to refine a physical conductance model to obtain a heuristic estimate for single-channel conductance. The code is freely available, obtainable as pip-installable python package and provided as a web service.

The mechanism of mammalian proton-coupled peptide transporters.

Proton-coupled oligopeptide transporters (POTs) are of great pharmaceutical interest owing to their promiscuous substrate binding site that has been linked to improved oral bioavailability of several classes of drugs. Members of the POT family are conserved across all phylogenetic kingdoms and function by coupling peptide uptake to the proton electrochemical gradient. Cryo-EM structures and alphafold models have recently provided new insights into different conformational states of two mammalian POTs, SLC15A1, and SLC15A2. Nevertheless, these studies leave open important questions regarding the mechanism of proton and substrate coupling, while simultaneously providing a unique opportunity to investigate these processes using molecular dynamics (MD) simulations. Here, we employ extensive unbiased and enhanced-sampling MD to map out the full SLC15A2 conformational cycle and its thermodynamic driving forces. By computing conformational free energy landscapes in different protonation states and in the absence or presence of peptide substrate, we identify a likely sequence of intermediate protonation steps that drive inward-directed alternating access. These simulations identify key differences in the extracellular gate between mammalian and bacterial POTs, which we validate experimentally in cell-based transport assays. Our results from constant-PH MD and absolute binding free energy (ABFE) calculations also establish a mechanistic link between proton binding and peptide recognition, revealing key details underpining secondary active transport in POTs. This study provides a vital step forward in understanding proton-coupled peptide and drug transport in mammals and pave the way to integrate knowledge of solute carrier structural biology with enhanced drug design to target tissue and organ bioavailability.

The cells in our body are sealed by a surrounding membrane that allows them to control which molecules can enter or leave. Desired molecules are often imported via transport proteins that require a source of energy. One way that transporter proteins achieve this is by simultaneously moving positively charged particles called protons across the membrane. Proteins called POTs (short for proton-coupled oligopeptide transporters) use this mechanism to import small peptides and drugsin to the cells of the kidney and small intestine. Sitting in the centre of these transporters is a pocket that binds to the imported peptide which has a gate on either side: an outer gate that opens towards the outside of the cell, and an inner gate that opens towards the cell's interior. The movement of protons from the outer to the inner gate is thought to shift the shape of the transporter from an outwards to an inwards-facing state. However, the molecular details of this energetic coupling are not well understood. To explore this, Lichtinger et al. used computer simulations to pinpoint where protons bind on POTs to trigger the gates to open. The simulations proposed that two sites together make up the outward-facing gate, which opens upon proton binding. Lichtinger et al. then validated these sites experimentally in cultured human cells that produce mutant POTs. After the desired peptide/drug has attached to the binding pocket, the protons then move to two more sites further down the transporter. This triggers the inner gate to open, which ultimately allows the small molecule to move into the cell. These findings represent a significant step towards understanding how POTs transport their cargo. Since POTs can transport a range of drugs from the digestive tract into the body, these results could help researchers design molecules that are better absorbed. This could lead to more orally available medications, making it easier for patients to adhere to their treatment regimen.

Automated Absolute Binding Free Energy Calculation Workflow for Drug Discovery.

Absolute binding free energies play a crucial role in drug development, particularly as part of the lead discovery process. In recent work, we showed how in silico predictions directly could support drug development by ranking and recommending favorable ideas over unfavorable ones. Here, we demonstrate a Python workflow that enables the calculation of ABFEs with minimal manual input effort, such as the receptor PDB and ligand SDF files, and outputs a .tsv file containing the ranked ligands and their corresponding binding free energies. The implementation uses Snakemake to structure and control the execution of tasks, allowing for dynamic control of parameters and execution patterns. We provide an example of a benchmark system that demonstrates the effectiveness of the automated workflow.

Plasticity of the selectivity filter is essential for permeation in lysosomal TPC2 channels.

Two-pore channels are pathophysiologically important Na+- and Ca2+-permeable channels expressed in lysosomes and other acidic organelles. Unlike most other ion channels, their permeability is malleable and ligand-tuned such that when gated by the signaling lipid PI(3,5)P2, they are more Na+-selective than when gated by the Ca2+ mobilizing messenger nicotinic acid adenine dinucleotide phosphate. However, the structural basis that underlies such plasticity and single-channel behavior more generally remains poorly understood. A recent Cryo-electron microscopy (cryo-EM) structure of TPC2 bound to PI(3,5)P2 in a proposed open-channel conformation provided an opportunity to address this via molecular dynamics (MD) simulation. To our surprise, simulations designed to compute conductance through this structure revealed almost no Na+ permeation events even at very high transmembrane voltages. However further MD simulations identified a spontaneous transition to a dramatically different conformation of the selectivity filter that involved expansion and a flip in the orientation of two core asparagine residues. This alternative filter conformation was remarkably stable and allowed Na+ to flow through the channel leading to a conductance estimate that was in very good agreement with direct single-channel measurements. Furthermore, this conformation was more permeable for Na+ over Ca2+. Our results have important ramifications not just for understanding the control of ion selectivity in TPC2 channels but also more broadly in terms of how ion channels discriminate ions.

Cryo-EM structures of prokaryotic ligand-gated ion channel GLIC provide insights into gating in a lipid environment.

GLIC, a proton-activated prokaryotic ligand-gated ion channel, served as a model system for understanding the eukaryotic counterparts due to their structural and functional similarities. Despite extensive studies conducted on GLIC, the molecular mechanism of channel gating in the lipid environment requires further investigation. Here, we present the cryo-EM structures of nanodisc-reconstituted GLIC at neutral and acidic pH in the resolution range of 2.6 - 3.4 Å. In our apo state at pH 7.5, the extracellular domain (ECD) displays conformational variations compared to the existing apo structures. At pH 4.0, three distinct conformational states (C1, C2 and O states) are identified. The protonated structures exhibit a compacted and counter-clockwise rotated ECD compared with our apo state. A gradual widening of the pore in the TMD is observed upon reducing the pH, with the widest pore in O state, accompanied by several layers of water pentagons. The pore radius and molecular dynamics (MD) simulations suggest that the O state represents an open conductive state. We also observe state-dependent interactions between several lipids and proteins that may be involved in the regulation of channel gating. Our results provide comprehensive insights into the importance of lipids impact on gating.

Molecular basis for pH sensing in the KDEL trafficking receptor.

Trafficking receptors control protein localization through the recognition of specific signal sequences that specify unique cellular locations. Differences in luminal pH are important for the vectorial trafficking of cargo receptors. The KDEL receptor is responsible for maintaining the integrity of the ER by retrieving luminally localized folding chaperones in a pH-dependent mechanism. Structural studies have revealed the end states of KDEL receptor activation and the mechanism of selective cargo binding. However, precisely how the KDEL receptor responds to changes in luminal pH remains unclear. To explain the mechanism of pH sensing, we combine analysis of X-ray crystal structures of the KDEL receptor at neutral and acidic pH with advanced computational methods and cell-based assays. We show a critical role for ordered water molecules that allows us to infer a direct connection between protonation in different cellular compartments and the consequent changes in the affinity of the receptor for cargo.

Selective Potentiation of the (α4)3(ß2)2 Nicotinic Acetylcholine Receptor Response by NS9283 Analogues.

NS9283, 3-(3-pyridyl)-5-(3-cyanophenyl)-1,2,4-oxadiazole, is a selective positive allosteric modulator of (α4)3(ß2)2 nicotinic acetylcholine receptors (nAChRs). It has good subtype selective therapeutic potential afforded by its specific binding to the unique α4-α4 subunit interface present in the (α4)3(ß2)2 nAChR. However, there is currently a lack of structure activity relationship (SAR) studies aimed at developing a class of congeners endowed with the same profile of activity that can help consolidate the druggability of the α4-α4 subunit interface. In this study, new NS9283 analogues were designed, synthesized, and characterized for their ability to selectively potentiate the ACh activity at heterologous (α4)3(ß2)2 nAChRs vs nAChR subtypes (α4)2(ß2)3, α5α4ß2, and α7. With few exceptions, all the NS9283 analogues exerted positive modulation of the (α4)3(ß2)2 nAChR ACh-evoked responses. Above all, those modified at the 3-cyanophenyl moiety by replacement with 3-nitrophenyl (4), 4-cyanophenyl (10), and N-formyl-4-piperidinyl (20) showed the same efficacy as NS9283, although with lower potency. Molecular dynamics simulations of NS9283 and some selected analogues highlighted consistency between potentiation activity and pose of the ligand inside the α4-α4 site with the main interaction being with the complementary (-) side and induction of a significant conformational change of the Trp156 residue in the principal (+) side.

Recovery from desensitization in GluA2 AMPA receptors is affected by a single mutation in the N-terminal domain interface.

AMPA-type ionotropic glutamate receptors (AMPARs) are central to various neurological processes, including memory and learning. They assemble as homo- or heterotetramers of GluA1, GluA2, GluA3, and GluA4 subunits, each consisting of an N-terminal domain (NTD), a ligand-binding domain, a transmembrane domain, and a C-terminal domain. While AMPAR gating is primarily controlled by reconfiguration in the ligand-binding domain layer, our study focuses on the NTDs, which also influence gating, yet the underlying mechanism remains enigmatic. In this investigation, we employ molecular dynamics simulations to evaluate the NTD interface strength in GluA1, GluA2, and NTD mutants GluA2-H229N and GluA1-N222H. Our findings reveal that GluA1 has a significantly weaker NTD interface than GluA2. The NTD interface of GluA2 can be weakened by a single point mutation in the NTD dimer-of-dimer interface, namely H229N, which renders GluA2 more GluA1-like. Electrophysiology recordings demonstrate that this mutation also leads to slower recovery from desensitization. Moreover, we observe that lowering the pH induces more splayed NTD states and enhances desensitization in GluA2. We hypothesized that H229 was responsible for this pH sensitivity; however, GluA2-H229N was also affected by pH, meaning that H229 is not solely responsible and that protons exert their effect across multiple domains of the AMPAR. In summary, our work unveils an allosteric connection between the NTD interface strength and AMPAR desensitization.

Small-molecule positive allosteric modulation of homomeric kainate receptors GluK1-3: development of screening assays and insight into GluK3 structure.

The kainate receptors GluK1-3 (glutamate receptor ionotropic, kainate receptors 1-3) belong to the family of ionotropic glutamate receptors and are essential for fast excitatory neurotransmission in the brain, and are associated with neurological and psychiatric diseases. How these receptors can be modulated by small-molecule agents is not well understood, especially for GluK3. We show that the positive allosteric modulator BPAM344 can be used to establish robust calcium-sensitive fluorescence-based assays to test agonists, antagonists, and positive allosteric modulators of GluK1-3. The half-maximal effective concentration (EC50) of BPAM344 for potentiating the response of 100 µm kainate was determined to be 26.3 µm for GluK1, 75.4 µm for GluK2, and 639 µm for GluK3. Domoate was found to be a potent agonist for GluK1 and GluK2, with an EC50 of 0.77 and 1.33 µm, respectively, upon co-application of 150 µm BPAM344. At GluK3, domoate acts as a very weak agonist or antagonist with a half-maximal inhibitory concentration (IC50) of 14.5 µm, in presence of 500 µm BPAM344 and 100 µm kainate for competition binding. Using H523A-mutated GluK3, we determined the first dimeric structure of the ligand-binding domain by X-ray crystallography, allowing location of BPAM344, as well as zinc-, sodium-, and chloride-ion binding sites at the dimer interface. Molecular dynamics simulations support the stability of the ion sites as well as the involvement of Asp761, Asp790, and Glu797 in the binding of zinc ions. Using electron microscopy, we show that, in presence of glutamate and BPAM344, full-length GluK3 adopts a dimer-of-dimers arrangement.

Mutate and Conjugate: A Method to Enable Rapid In-Cell Target Validation.

Target validation remains a challenge in drug discovery, which leads to a high attrition rate in the drug discovery process, particularly in Phase II clinical trials. Consequently, new approaches to enhance target validation are valuable tools to improve the drug discovery process. Here, we report the combination of site-directed mutagenesis and electrophilic fragments to enable the rapid identification of small molecules that selectively inhibit the mutant protein. Using the bromodomain-containing protein BRD4 as an example, we employed a structure-based approach to identify the L94C mutation in the first bromodomain of BRD4 [BRD4(1)] as having a minimal effect on BRD4(1) function. We then screened a focused, KAc mimic-containing fragment set and a diverse fragment library against the mutant and wild-type proteins and identified a series of fragments that showed high selectivity for the mutant protein. These compounds were elaborated to include an alkyne click tag to enable the attachment of a fluorescent dye. These clickable compounds were then assessed in HEK293T cells, transiently expressing BRD4(1)WT or BRD4(1)L94C, to determine their selectivity for BRD4(1)L94C over other possible cellular targets. One compound was identified that shows very high selectivity for BRD4(1)L94C over all other proteins. This work provides a proof-of-concept that the combination of site-directed mutagenesis and electrophilic fragments, in a mutate and conjugate approach, can enable rapid identification of small molecule inhibitors for an appropriately mutated protein of interest. This technology can be used to assess the cellular phenotype of inhibiting the protein of interest, and the electrophilic ligand provides a starting point for noncovalent ligand development.

Computed Protein-Protein Enthalpy Signatures as a Tool for Identifying Conformation Sampling Problems.

Understanding the thermodynamic signature of protein-peptide binding events is a major challenge in computational chemistry. The complexity generated by both components possessing many degrees of freedom poses a significant issue for methods that attempt to directly compute the enthalpic contribution to binding. Indeed, the prevailing assumption has been that the errors associated with such approaches would be too large for them to be meaningful. Nevertheless, we currently have no indication of how well the present methods would perform in terms of predicting the enthalpy of binding for protein-peptide complexes. To that end, we carefully assembled and curated a set of 11 protein-peptide complexes where there is structural and isothermal titration calorimetry data available and then computed the absolute enthalpy of binding. The initial "out of the box" calculations were, as expected, very modest in terms of agreement with the experiment. However, careful inspection of the outliers allows for the identification of key sampling problems such as distinct conformations of peptide termini not being sampled or suboptimal cofactor parameters. Additional simulations guided by these aspects can lead to a respectable correlation with isothermal titration calorimetry (ITC) experiments (R2 of 0.88 and an RMSE of 1.48 kcal/mol overall). Although one cannot know prospectively whether computed ITC values will be correct or not, this work shows that if experimental ITC data are available, then this in conjunction with computed ITC, can be used as a tool to know if the ensemble being simulated is representative of the true ensemble or not. That is important for allowing the correct interpretation of the detailed dynamics of the system with respect to the measured enthalpy. The results also suggest that computational calorimetry is becoming increasingly feasible. We provide the data set as a resource for the community, which could be used as a benchmark to help further progress in this area.

Side Groups Convert the α7 Nicotinic Receptor Agonist Ether Quinuclidine into a Type I Positive Allosteric Modulator.

The quinuclidine scaffold has been extensively used for the development of nicotinic acetylcholine receptor (nAChR) agonists, with hydrophobic substituents at position 3 of the quinuclidine framework providing selectivity for α7 nAChRs. In this study, six new ligands (4-9) containing a 3-(pyridin-3-yloxy)quinuclidine moiety (ether quinuclidine) were synthesized to gain a better understanding of the structural-functional properties of ether quinuclidines. To evaluate the pharmacological activity of these ligands, two-electrode voltage-clamp and single-channel recordings were performed. Only ligand 4 activated α7 nAChR. Ligands 5 and 7 had no effects on α7 nAChR, but ligands 6, 8, and 9 potentiated the currents evoked by ACh. Ligand 6 was the most potent and efficacious of the potentiating ligands, with an estimated EC50 for potentiation of 12.6 ± 3.32 µM and a maximal potentiation of EC20 ACh responses of 850 ± 120%. Ligand 6 increased the maximal ACh responses without changing the kinetics of the current responses. At the single-channel level, the potentiation exerted by ligand 6 was evidenced in the low micromolar concentration range by the appearance of prolonged bursts of channel openings. Furthermore, computational studies revealed the preference of ligand 6 for an intersubunit site in the transmembrane domain and highlighted some putative key interactions that explain the different profiles of the synthesized ligands. Notably, Met276 in the 15' position of the transmembrane domain 2 almost abolished the effects of ligand 6 when mutated to Leu. We conclude that ligand 6 is a novel type I positive allosteric modulator (PAM-I) of α7 nAChR.

Correction: Limitations of non-polarizable force fields in describing anion binding poses in non-polar synthetic hosts.

Correction for 'Limitations of non-polarizable force fields in describing anion binding poses in non-polar synthetic hosts' by David Seiferth et al., Phys. Chem. Chem. Phys., 2023, 25, 17596-17608, https://doi.org/10.1039/D3CP00479A.

Tackling Hysteresis in Conformational Sampling: How to Be Forgetful with MEMENTO.

The structure of proteins has long been recognized to hold the key to understanding and engineering their function, and rapid advances in structural biology and protein structure prediction are now supplying researchers with an ever-increasing wealth of structural information. Most of the time, however, structures can only be determined in free energy minima, one at a time. While conformational flexibility may thus be inferred from static end-state structures, their interconversion mechanisms─a central ambition of structural biology─are often beyond the scope of direct experimentation. Given the dynamical nature of the processes in question, many studies have attempted to explore conformational transitions using molecular dynamics (MD). However, ensuring proper convergence and reversibility in the predicted transitions is extremely challenging. In particular, a commonly used technique to map out a path from a starting to a target conformation called steered MD (SMD) can suffer from starting-state dependence (hysteresis) when combined with techniques such as umbrella sampling (US) to compute the free energy profile of a transition. Here, we study this problem in detail on conformational changes of increasing complexity. We also present a new, history-independent approach that we term "MEMENTO" (Morphing End states by Modelling Ensembles with iNdependent TOpologies) to generate paths that alleviate hysteresis in the construction of conformational free energy profiles. MEMENTO utilizes template-based structure modelling to restore physically reasonable protein conformations based on coordinate interpolation (morphing) as an ensemble of plausible intermediates, from which a smooth path is picked. We compare SMD and MEMENTO on well-characterized test cases (the toy peptide deca-alanine and the enzyme adenylate kinase) before discussing its use in more complicated systems (the kinase P38α and the bacterial leucine transporter LeuT). Our work shows that for all but the simplest systems SMD paths should not in general be used to seed umbrella sampling or related techniques, unless the paths are validated by consistent results from biased runs in opposite directions. MEMENTO, on the other hand, performs well as a flexible tool to generate intermediate structures for umbrella sampling. We also demonstrate that extended end-state sampling combined with MEMENTO can aid the discovery of collective variables on a case-by-case basis.

Limitations of non-polarizable force fields in describing anion binding poses in non-polar synthetic hosts.

Transmembrane anion transport by synthetic ionophores has received increasing interest not only because of its relevance for understanding endogenous anion transport, but also because of potential implications for therapeutic routes in disease states where chloride transport is impaired. Computational studies can shed light on the binding recognition process and can deepen our mechanistic understanding of them. However, the ability of molecular mechanics methods to properly capture solvation and binding properties of anions is known to be challenging. Consequently, polarizable models have been suggested to improve the accuracy of such calculations. In this study, we calculate binding free energies for different anions to the synthetic ionophore, biotin[6]uril hexamethyl ester in acetonitrile and to biotin[6]uril hexaacid in water by employing non-polarizable and polarizable force fields. Anion binding shows strong solvent dependency consistent with experimental studies. In water, the binding strengths are iodide > bromide > chloride, and reversed in acetonitrile. These trends are well captured by both classes of force fields. However, the free energy profiles obtained from potential of mean force calculations and preferred binding positions of anions depend on the treatment of electrostatics. Results from simulations using the AMOEBA force-field, which recapitulate the observed binding positions, suggest strong effects from multipoles dominate with a smaller contribution from polarization. The oxidation status of the macrocycle was also found to influence anion recognition in water. Overall, these results have implications for the understanding of anion host interactions not just in synthetic ionophores, but also in narrow cavities of biological ion channels.

The role of loop dynamics in the prediction of ligand-protein binding enthalpy.

The enthalpic and entropic components of ligand-protein binding free energy reflect the interactions and dynamics between ligand and protein. Despite decades of study, our understanding and hence our ability to predict these individual components remains poor. In recent years, there has been substantial effort and success in the prediction of relative and absolute binding free energies, but the prediction of the enthalpic (and entropic) contributions in biomolecular systems remains challenging. Indeed, it is not even clear what kind of performance in terms of accuracy could currently be obtained for such systems. It is, however, relatively straight-forward to compute the enthalpy of binding. We thus evaluated the performance of absolute enthalpy of binding calculations using molecular dynamics simulation for ten inhibitors against a member of the bromodomain family, BRD4-1, against isothermal titration calorimetry data. Initial calculations, with the AMBER force-field showed good agreement with experiment (R2 = 0.60) and surprisingly good accuracy with an average of root-mean-square error (RMSE) = 2.49 kcal mol-1. Of the ten predictions, three were obvious outliers that were all over-predicted compared to experiment. Analysis of various simulation factors, including parameterization, buffer concentration and conformational dynamics, revealed that the behaviour of a loop (the ZA loop on the periphery of the binding site) strongly dictates the enthalpic prediction. Consistent with previous observations, the loop exists in two distinct conformational states and by considering one or the other or both states, the prediction for the three outliers can be improved dramatically to the point where the R2 = 0.95 and the accuracy in terms of RMSE improves to 0.90 kcal mol-1. However, performance across force-fields is not consistent: if OPLS and CHARMM are used, different outliers are observed and the correlation with the ZA loop behaviour is not recapitulated, likely reflecting parameterization as a confounding problem. The results provide a benchmark standard for future study and comparison.

Conformational transitions and allosteric modulation in a heteromeric glycine receptor.

Glycine Receptors (GlyRs) provide inhibitory neuronal input in the spinal cord and brainstem, which is critical for muscle coordination and sensory perception. Synaptic GlyRs are a heteromeric assembly of α and ß subunits. Here we present cryo-EM structures of full-length zebrafish α1ßBGlyR in the presence of an antagonist (strychnine), agonist (glycine), or agonist with a positive allosteric modulator (glycine/ivermectin). Each structure shows a distinct pore conformation with varying degrees of asymmetry. Molecular dynamic simulations found the structures were in a closed (strychnine) and desensitized states (glycine and glycine/ivermectin). Ivermectin binds at all five interfaces, but in a distinct binding pose at the ß-α interface. Subunit-specific features were sufficient to solve structures without a fiduciary marker and to confirm the 4α:1ß stoichiometry recently observed. We also report features of the extracellular and intracellular domains. Together, our results show distinct compositional and conformational properties of α1ßGlyR and provide a framework for further study of this physiologically important channel.

Scoring Functions for Protein-Ligand Binding Affinity Prediction using Structure-Based Deep Learning: A Review.

The rapid and accurate in silico prediction of protein-ligand binding free energies or binding affinities has the potential to transform drug discovery. In recent years, there has been a rapid growth of interest in deep learning methods for the prediction of protein-ligand binding affinities based on the structural information of protein-ligand complexes. These structure-based scoring functions often obtain better results than classical scoring functions when applied within their applicability domain. Here we review structure-based scoring functions for binding affinity prediction based on deep learning, focussing on different types of architectures, featurization strategies, data sets, methods for training and evaluation, and the role of explainable artificial intelligence in building useful models for real drug-discovery applications.

When is a hydrophobic gate not a hydrophobic gate?

The flux of ions through a channel is most commonly regulated by changes that result in steric occlusion of its pore. However, ion permeation can also be prevented by formation of a desolvation barrier created by hydrophobic residues that line the pore. As a result of relatively minor structural changes, confined hydrophobic regions in channels may undergo transitions between wet and dry states to gate the pore closed without physical constriction of the permeation pathway. This concept is referred to as hydrophobic gating, and many examples of this process have been demonstrated. However, the term is also now being used in a much broader context that often deviates from its original meaning. In this Viewpoint, we explore the formal definition of a hydrophobic gate, discuss examples of this process compared with other gating mechanisms that simply exploit hydrophobic residues and/or lipids in steric closure of the pore, and describe the best practice for identification of a hydrophobic gate.