Proteins: Structure, Function, and Bioinformatics

Structure-aware proteolysis prediction often ranks candidate cleavage sites using solvent accessibility from static protein structures, yet this assumption is rarely tested against native-state cleavage data. We examined this problem using AprX, a serralysin-family protease that readily hydrolyses caseins but shows lower activity toward native whey proteins. Across six bovine milk proteins, sequence compatibility and local scissile-bond accessibility did not separate caseins from whey proteins. In contrast, accessibility fell progressively in whey proteins as the required contiguous substrate window increased, reaching 0% across the working P4-P4' footprint, while all evaluable casein windows remained accessible in isolated-chain ensembles. Local remodelling showed that whey segments could still adopt AprX-compatible backbone geometry. An external thermolysin benchmark rejected a rigid whole-protein clash metric, but did not alter the contiguous-accessibility result. These findings identify footprint-scale accessibility as a distinct structural variable and show why single-residue solvent exposure can overestimate protease accessibility in folded proteins.

Designed binders are routinely filtered on predicted interface confidence, which reports how self-consistently a model places an interface, not where it lies. We designed 200 nanobody loops against the survivin BIR-domain groove, co-folded 196, and checked epitope contact in every predicted complex. Confidence showed no detectable association with epitope engagement ({rho} = +0.100, 95% CI -0.04 to +0.24, n = 196); its highest and lowest quartiles failed alike (odds ratio 0.97, 95% CI 0.41 to 2.26). It tracked instead the run's other confidence estimates ({rho} = +0.61). Twelve-seed replication separated two candidates the criteria had not: one reached the complete epitope in 10 of 12 runs, the other never. Across 155 predictions the top-ranked model was no better than one drawn at random from its five (P = 0.610), though only 3 of 31 failed on every model. Nothing was assayed; nb_cand_057 is an in silico candidate only.

In this work, we focus on predicting fold-switching phenomena observed in certain protein structures. Proteins that undergo fold-switching are called metamorphic proteins. We model fold-switch prediction in proteins as a position/residue-wise classification task. We use various features extracted from a protein's sequence as well as structure to train the classifiers. We propose a novel class of handcrafted parametrised features that encode the structural context of a residue. We also use embeddings from existing protein language models (PLMs) to train the classifiers. We train binary and 3-class classifiers, with the 3-class classifiers providing granular information on the type of fold-switch. In our computations, higher classification performance is achieved for binary prediction as compared to 3-class granular prediction. The handcrafted features perform comparably with the PLM embeddings on our standard test set. A rigorous error analysis by us during the review period revealed a critical data-leakage vulnerability in the metamorphic protein test data. This analysis also revealed that the PLM-based classifier suffers sequence-memorisation bias and fails to generalise to novel fold-switching sequences. On the other hand, one of our domain-informed handcrafted feature based classifier detects fold-switching without relying on sequence memory.

Recent advances in AI- RNA structure prediction have increased the availability of candidate models, but their accuracy remains variable. Developing reliable quality assessment (QA) methods requires datasets that represent contemporary prediction errors across diverse targets and molecular contexts. Here, we present RNA3D-Decoy, a large-scale benchmark dataset comprising 103,549 decoys for 677 experimentally determined targets, including 416 RNA-only structures,117 RNA structures with metal ions, and 144 RNA-protein complexes. Decoys were generated using five biomolecular predictors: AlphaFold 3, ESMFold2, Protenix, Chai-1, and Boltz-2.Each decoy is evaluated against its experimental reference us-ing TM-score, lDDT, and RMSD, with global and local quality annotations where applicable. DockQ scores additionally characterize RNA-protein interface quality. Target grouping incorporates both RNA and protein sequence relationships to support dataset partitioning. Analyses reveal distinct quality distributions across dataset levels and target-dependent differences among predictors, with each predictor contributing the highest-TM-score model for a subset of targets. Broad pooled score distributions also coexist with limited within-target score variation, particularly for RNA-protein complexes, emphasizing the importance of target-level characterization. By combining AI-generated decoys, standardized quality labels, and structured metadata, RNA3D-Decoy provides a resource for developing and evaluating methods for global quality estimation, local error prediction, and candidate-model ranking. The data set and accompanying code are available at: https://github.com/HamiltonChenLab/RNA3D-Decoy.

Accurate prediction of mutation-induced changes in protein-protein affinity remains a central challenge in computational biophysics and protein engineering. Here we present a physics-based scoring method for predicting the effects of single amino acid substitutions on protein-protein and protein-peptide binding affinity. The method is based on MMPBSA energy calculations and implemented in a user-friendly, publicly available workflow (https://github.com/Hoecker-Lab/mmpbsa_scoring). To derive the scoring function, MMPBSA energy terms were fitted using a large and diverse dataset of experimentally measured binding affinity changes upon point mutation. The terms were then evaluated on independent systems. Specifically, we assessed the ability to predict mutational effects at the interface of SARS-CoV-2 spike protein with its cellular receptor, as well as computationally predicted Armadillo-repeat protein-peptide complexes. The approach performs well across diverse targets and shows promising predictive performance on both experimentally determined and computationally predicted structures. It matches or exceeds the accuracy of established physics-based protein design software and state-of-the-art deep learning-based scoring functions when applied to data outside their training domain. The method provides a computationally efficient and interpretable framework for prioritizing interface mutations in protein engineering and binding affinity optimization.

Nitrate is an important oxyanion in bacterial physiology, with major roles as a nitrogen source for assimilation and as an alternative electron acceptor in anaerobic respiration. Accordingly, diverse nitrate-responsive regulatory systems sense nitrate availability and control the expression of genes involved in nitrate transport and metabolism. In the NreABC system, the soluble GAF-like domain protein NreA functions as a cellular nitrate sensor. To date, our molecular understanding of nitrate recognition by NreA has been derived primarily from studies of the Staphylococcus carnosus protein, with less known about homologues from other species. Here, we identified 736 NreA homologues through bioinformatic analysis of the GAF-like domain superfamily and selected five staphylococcal representatives for thermodynamic characterization by isothermal titration calorimetry. Despite conservation of the nitrate-binding pocket, these homologues resolved into high- and low-affinity groups with distinct thermodynamic profiles. Comparative structural modeling and chimeragenesis of a low-affinity homologue revealed that the C-terminal region is a non-coordinating determinant of nitrate-binding affinity. Together, these findings show that a conserved nitrate-binding pocket preserves recognition, while distal sequence variation shapes thermodynamics and tunes the nitrate concentration range over which NreA sensors can function.

The sulfur mobilization (SUF) system is the only Fe-S cluster biosynthesis pathway in Mycobacterium tuberculosis (M. tb) and Mycolicibacterium smegmatis (M. smegmatis). In this pathway, SufC protein acts as an ATPase in the SufBC2D scaffold for Fe-S cluster biogenesis. This crucial role positions SufC as a promising target for the development of antimycobacterial therapeutics. However, the SufC ATPases from Mycobacterial species are not yet characterized. This study includes cloning, expression, purification, and characterization of SufC homologs from M. tb (Rv1463) and M. smegmatis (MSMEG_3124) using biochemical, biophysical, and computational approaches. Both SufC proteins were purified using affinity and size-exclusion chromatography and observed to exist as monomers in solution. Although both proteins have the conserved domains, they exhibit different ATP-dependent kinetic efficiencies, demonstrating differences in their local environment and catalytic behavior. Furthermore, CRISPR interference targeting MSMEG_3124 results led to diminished M. smegmatis growth, highlighting the critical role of SufC in cellular viability. Modeling and virtual screening led to the identification of three candidate compounds, of which Z2911048515 and Z66741142 were selected for experimental evaluation. Circular dichroism spectroscopy analysis revealed slight differences in the spectral features of Rv1463 in the presence of Z2911048515 and Z66741142, while significant spectral variations were observed for MSMEG_3124. Importantly, Z66741142 was able to cause changes in the fluorescence spectra of both SufC protein homologs. Surprisingly, both compounds showed only limited inhibition of SufC ATPase activity, exhibited distinct modes of action, and may serve as preliminary chemical scaffolds for optimization and testing as effective inhibitors. This study presents functional diversity of mycobacterial SufC proteins across pathogenic and non-pathogenic species and provides a basis for developing a framework for further mechanistic and structure-guided investigation within the indispensable SUF pathway.

In the post-AlphaFold era, advances in protein complex structure prediction have enabled the generation of thousands of candidate models per target, making the accurate and efficient selection of high-quality models from large candidate pools a critical challenge. Existing model-selection approaches commonly employ estimated model accuracy (EMA) methods, which include single-model and consensus-based methods. The former lack cross-model comparative information, whereas the latter often incur high computational costs from pairwise structural alignments and are influenced by candidate-pool quality. Here, we present DeepUMQA-Selection, a two-stage framework combining single-model accuracy estimators for preliminary selection with Ultrafast Shape Recognition (USR)-based structural alignment for consensus selection. Preliminary selection enriches the candidate pool with high-quality models and reduces subsequent structural comparisons, while USR-based structural alignment enables efficient consensus selection without explicit structural superposition. On the retrospective CASP16 QMODE3 benchmark dataset, DeepUMQA-Selection outperformed all participating methods and predictor-derived confidence metrics, as measured by mean Top-5 weighted penalty. The advantage was particularly pronounced for heteromeric targets, with a 13.9% improvement in model-selection performance relative to the best-performing CASP16 participating method. Furthermore, a confidence-assisted strategy enabled the framework to accommodate larger protein complexes while maintaining superior model-selection performance over competing methods, supporting its scalability. Runtime benchmarking showed that the proposed USR-based structural alignment achieved a substantial 317-fold speedup over US-align for candidate pools containing 1,000 models, demonstrating its computational efficiency for large-scale model selection. Our results demonstrate that DeepUMQA-Selection enables accurate and computationally efficient selection of high-quality protein complex models from large-scale candidate pools.

A critical step in the emergence of a virus is the ability of the viral protein to bind a host receptor and mediate cell entry. For many coronaviruses, this interaction occurs between the Spike S1 subunit and the human ACE2 receptor. Whether this binding interface can be computationally distinguished across unstudied viruses without experimentally resolved protein structures remains an open question. We predicted how 28 emerging coronaviruses may bind to human ACE2 using structural predictions, static interaction prediction programs, and molecular dynamics simulations. To screen the emerging coronaviruses, we predicted a library of S1 structures using AlphaFold. These predicted structures were then used to model the S1-ACE2 interaction with AlphaFold, ClusPro, and HADDOCK. We used known ACE2-binding sarbecoviruses as positive controls and coronaviruses that bind other receptors as negative controls to threshold predicted binding. Contact analysis quantified the predicted binding and revealed that these static interaction prediction methods varied in discriminative power. Less restrained static predictions separated binders from non-binders, whereas heavily restrained docking did not, potentially forcing an interaction where none should exist. This analysis highlighted an emerging coronavirus, Zhejiang2013, as a potential ACE2 binder. We used molecular dynamics simulations to further assess the static predictions and model the interaction over time. Overall, our results indicate that Zhejiang2013 exhibits dynamic interaction patterns consistent with ACE2 binding. Given that two ACE2-binding coronaviruses have caused global pandemics within the past two decades, identifying potential ACE2 binders is critical for early warning and pandemic preparedness.

Carbohydrates are essential molecules involved in a wide range of biological processes. While their structural representation in the Protein Data Bank (PDB) has been hindered due to their highly diverse and flexible nature, advancements on X-ray crystallography and CryoEM are accelerating the obtaining of their experimental structures, which in turn supports improved analyses of their structural properties, such as the role of their non-chair conformations on biological processes. In this context, the current work provides a comprehensive and up to date overview of carbohydrate structures within the PDB through classification, quantification, and characterization of monosaccharides and glycan chains, incorporating Cremer-Pople puckering coordinates to evaluate pyranose conformations. Our findings show that, even with a steep increase in CryoEM structures including more complex oligosaccharides, carbohydrate data remains concentrate dinsmallglycanchains, with a about 4% of non-chair conformation sintopresolution structures. Free energy landscapes for the pyranose ring in the presence and absence of nine different binding proteins, including both enzymes and non-enzymes, reinforce the preferred stability of chairs and the rarity of conformational induction upon binding.

[FeFe]-hydrogenases are molecular hydrogen (H2) converting enzymes that employ a hexanuclear iron-complex, the H-cluster, as catalytic cofactor, and a proton transfer pathway (PTP) that allows efficient proton coupled electron transfer (PCET). Recent phylogenetic analyses revealed different groups of [FeFe]-hydrogenases. Although only very few members from other groups have been characterized, current knowledge suggests high catalytic activity is predominantly associated with group A members. Here, we show that metal ions inhibit group A [FeFe]-hydrogenases by binding to three specific residues at the entrance of the PTP. Exchanging residue H565 of Clostridium pasteurianum CpI results in a metal-insensitive protein variant with wildtype like activity. In contrast, exchanging S320 and H569 in CpI, and their counterparts in additional group A [FeFe]-hydrogenases results in enzymes with strongly decreased activities and large overpotential requirements. These features are consistent with important roles of both residues in catalytic proton transfer. CpI structures reveal that H569, locally anchored by E278, can be present in two conformations so that we propose a histidine-dependent pendulum mechanism for exchanging protons between bulk solvent and the PTP, which is well supported by theoretical calculations. By that, our study widens information on how the mobility of histidine, governed by residues in the vicinity, contributes to the fundamental concept of PCET.

Diffusion-based models have substantially advanced biomolecular structure prediction. Within this framework, restraint-guided inference (RGI) corrects the denoiser output at each reverse-diffusion step using a differentiable restraint loss, allowing available experimental information, stereochemical requirements, or a specified target conformational state to be incorporated into predictions without retraining the model. Here, we present RGI-Toolkit, a general-purpose Python library that makes RGI available for a range of structure predictors. The toolkit consolidates restraint definition, atom selection, loss evaluation, and coordinate optimization into a shared, backend-agnostic engine, limiting predictor-specific code to a thin preprocessing adapter and a single hook in the denoising loop. Restraints on distances, angles, and dihedrals, on deviations from reference structures, and on user-defined reaction coordinates, together with ligand conformer restraints, are all specified in a common configuration format. The toolkit currently supports six predictors, including AlphaFold3. Across all six predictors, conformer restraints eliminated chirality and cis/trans errors in the predicted ligands and improved their local geometry, while geometric and custom restraints systematically shifted the sampled protein conformations toward the specified target states. RGI-Toolkit thus provides a unified framework for integrating experimental information and prior geometric knowledge into structure prediction and for generating biomolecular structures with specified conformational and stereochemical properties.

Although use of AI-based tools has gained ground in molecular biology research, their reliability remains a matter of debate. Here, we combined AlphaFold 3 modelling with microscale thermophoresis to investigate the interactions between glucan-binding protein C and the biofilm-inhibiting aptamer and its five derivatives. Structural modelling identified a binding motif on the aptamer and suggested a mechanism of action; however, the predictions were only partially supported by the experimental results. The AlphaFold generated confidence metrics did not correlate directly with the measured binding affinities, indicating that confidence scores can help identify plausible interaction interfaces, but they are not reliable predictors of binding strength. These results highlight the importance of integrating structural prediction with biophysical validation to dissect aptamer-protein recognition.

Atomistic simulations of intrinsically disordered proteins (IDPs) are notoriously sensitive to force field inaccuracies, either regarding the protein or the water model, yielding inaccurate observables such as compactness, secondary structure propensities, or kinetics. The currently most widely used IDP force fields are Amber99sb-disp (A99disp) and Charmm36m (C36m). A99disp includes a new water model and thus optimized both, the protein and the water interactions. In contrast, C36m used the Tip3p water model and optimized only protein interactions. In many cases, C36m+Tip3p underestimates radii of gyration compared to FRET or SAXS experiments. Such overly compact structural ensembles are believed to arise from an imbalance between protein-protein, protein-water, and water-water interactions, which might be due to Tip3p inaccuracies. Here, we aim at re-balancing these interactions by combining C36m with the Optimal Point Charge (OPC) water model. Recently, this C36m+OPC combination showed improved accuracy for the disordered domain of the measles virus nucleoprotein. Here we present a systematic assessment, comparing C36m+OPC, C36m+Tip3p, C36m+Tip4p, C22*, A03ws, A99sb-ws, and A99disp for five IDPs, as well as a subset of those for five globular proteins, a set of disordered AGQ-repeat peptides, and the fast folding miniprotein CLN025. We compared extensive MD simulations (> 8.5 ms) with SAXS, NMR, circular dichroism, photo-induced electron transfer (PET), T-jump infrared spectroscopy, and X-ray crystallography measurements. We found that combining C36m with OPC improved accuracy over C36m+Tip3p for IDP ensembles without compromising its accuracy for globular proteins. While also the kinetics of the AGQ-peptides were more accurate for C36m+OPC, those of CLN025 folding were less accurate. Overall, C36m+OPC showed similar accuracy as A99sb-ws and A99disp, the latter is currently considered among the most accurate protein force fields.

Structure-guided mutation design requires relating a proposed substitution to its molecular environment to anticipate its effects. Existing evaluations of large language models (LLMs) in protein engineering emphasize sequence-based prediction or tool-integrated workflows, leaving unclear whether LLMs themselves can infer mutation effects from structural evidence without dedicated predictors. We evaluate physical intuition-qualitative reasoning that connects structural observations with molecular interaction principles-using a benchmark that tests the same substitution in the same protein across different binding partners, with predictions based on local atomic coordinates and no dedicated mutation-effect predictors. In the tested comparisons, the best-performing LLM achieved group accuracy comparable to the evaluated energy predictors and Biomni, an LLM-driven tool workflow, while the methods differed in their prediction patterns. Most evaluated LLMs achieved higher group accuracy for larger affinity improvements. Structural interventions support selective dependence on mutation-site and partner information, while assessable structural claims largely agree with the supplied coordinates. These findings support direct LLM structural reasoning as a basis for developing rational mutation-design tools that connect molecular observations with partner-specific effects.

The self-splicing group I intron ribozyme from Tetrahymena thermophillia (THr) is a prototypical example of how RNA overcomes its limited chemical repertoire to form complex, negatively charged junctions and catalyze chemical reactions. Despite over 30 years of biochemical and structural characterization, questions still remain regarding the requirements for THr catalysis that may inform structure based drug design against group I introns, as well as the use of THr as an RNA editing tool. Recent cryo-EM structures have provided an unprecedented window into the first and second steps of splicing, enabling detailed mechanistic investigation at the atomistic level. In this work, we leverage solvation theory, molecular dynamics, alchemical free energy simulations, and machine learning accelerated QM/MM to probe outstanding questions related to 1) the role of a possible third Mg2+ binding in the active site, 2) differences in catalytic requirements between Gexo cofactors and between the first and second steps of splicing, 3) the potential role of an acid or base, and 4) restrictions on the identity of the 5' exon splice site nucleotide, U(-1). We propose that a transient, cofactor dependent, interaction with a third metal participates in the first step, while a higher affinity interaction with this third ion is required in the second step. We also propose that the presence of a phosphate on the Gexo in the first step, and analogously the G{omega} in the second step, forms a metal ion binding site not present when guanosine is bound. Finally, we propose and computationally test design modifications of G22 (isofunctional substitutions of 2,6-diaminopurine that tune the pKa at the N1 position) that we predict will enable the accommodation of C(-1) in a crucial wobble-like pair at physiological pH. This design would provide a potential strategy for expanding the sequence space accessible to programmable trans-splicing ribozymes in RNA editing applications.

Structural model building and refinement need many decisions between different programs. These decisions often depend on an experienced researcher who knows how the programs work together and how to read experimental data. We used Claude Code with Claude Opus 5.5 to supervise this work for 24 cryo-EM and X-ray cases. The work finished in about 36 hours on one workstation. The agent prepared the inputs, ran established programs, read the validation results, compared candidate models and chose the next step. For 15 deposited cryo-EM models with weak starting geometry, the median MolProbity score improved from 2.62 to 1.76. For three crystal structures, the agent models came close to the published Rfree values. Chemical information such as ligands, ions and modifications must be provided by the researcher, in the same way as the sequence. With this information, the agent can place and check these components. These results show that agent supervision is acceptable when every decision can be inspected. The researcher makes the final scientific judgment. We suggest practical requirements for this kind of use. Future AI agents may help researchers obtain high quality and reliable structures of biological macromolecules.

Mitochondrial NADH-ubiquinone (UQ) oxidoreductase (respiratory complex I) reduces UQ within its long tunnel-like cavity, which is coupled to triggering proton translocation at the membrane domain. Although mammalian complex I adopts distinct "open" and "closed" conformations, the architecture of the UQ-accessing tunnel in the open conformation remains poorly defined owing to partial disorder of the UQ catalytic region near Fe-S cluster N2 in the tunnel. Here, we applied cryo-electron microscopy (EM) single particle analysis to determine the structures of complex I in bovine heart submitochondrial particles in the presence or absence of a mycothiazole-type inhibitor. Preceding photoaffinity labeling experiments using a synthetic photoreactive derivative ([125I]pMT) revealed that mycothiazole specifically labels the 49-kDa and ND1 subunits, both of which form a part of the UQ-accessing tunnel. Cryo-EM analysis demonstrated that bound mycothiazole reorganizes and stabilizes the UQ catalytic region in both open and closed conformations, enabling the first complete structural modeling of the entire UQ-accessing tunnel in the open conformation. Unexpectedly, in the presence of bound mycothiazole, loops of the PSST subunit forming the catalytic region in the tunnel adopted closed conformation-like features, whereas open conformation-like features were determined for the transmembrane helices of ND1 and ND6 subunits. The present study revealed that bound mycothiazole can reorganize the UQ catalytic region in the UQ-accessing tunnel and lead to a hybrid conformational state beyond the canonical open-versus-closed conformational model.

Protein-protein interactions (PPIs) govern fundamental biological processes and are central to therapeutic discovery, yet computational prediction methods remain severely limited by a lack of high-quality negative data (non-interacting PPIs). Existing models are trained almost exclusively on positive interactions or random pairs designated as negatives, while legacy negative datasets are limited in scale and contain correctness errors. To address this, we introduce Flock, a negative-enriched PPI dataset of unprecedented scale, containing 26,934 positive and 377,643 negative pairs. Flock integrates biological assemblies from the PDB with highly confident, curated negatives mined from scientific literature using a state-of-the-art agentic LLM workflow. To address the evaluation leakage seen in protein structure modelling, we also present Flock's Leakage-Free Set, a benchmark that strictly controls for sequence and interface similarity, and historical date-based cutoffs used in the training of co-folding models. Our evaluation reveals that the co-folding model ESMFold2 shows performance degradation on the Leakage-Free set, whereas the protein language model ESMC demonstrates stronger generalisation. Flock provides the most comprehensive and challenging PPI benchmark to date, establishing a rigorous new standard for evaluating the generalisability of protein structure and language models on this task.

Designing RNA sequences that reliably adopt functional three-dimensional structures remains a central challenge in RNA engineering because folding depends on cooperative interactions beyond canonical base pairing. Here we present DS3dRNA, an interaction-based framework for de novo RNA sequence design that combines a three-body statistical potential with physics-guided sequence sampling and supports design against multiple conformations. Across the evaluated benchmarks, DS3dRNA outperformed representative RNA inverse-design methods in native-sequence recovery and agreement between predicted and target structures. Energy-sequence-quality analyses further showed that lower design energies generally accompanied higher sequence recovery and macro-averaged F1 scores (MacroF1). Experimentally tested Mango II designs retained high-affinity fluorogenic activity, and five twister ribozyme designs yielded mean endpoint cleavage fractions of 37.7-50.6%, compared with 23.5% for the wild type. These results establish explicit higher-order interaction scoring as a complementary approach to emerging data-driven RNA design methods and provide a framework for designing functional RNAs from experimental or predicted structural ensembles.