Structure Prediction Modeling Research Articles

Deep learning generative model for crystal structure prediction

Deep learning generative model for crystal structure prediction

Benchmark Investigation of SARS-CoV-2 Mutants’ Immune Escape with 2B04 Murine Antibody: A Step Towards Unraveling a Larger Picture

Even though COVID-19 is no longer the primary focus of the global scientific community, its high mutation rate (nearly 30 substitutions per year) poses a threat of a potential comeback. Effective vaccines have been developed and administered to the population, ending the pandemic. Nonetheless, reinfection by newly emerging subvariants, particularly the latest JN.1 strain, remains common. The rapid mutation of this virus demands a fast response from the scientific community in case of an emergency. While the immune escape of earlier variants was extensively investigated, one still needs a comprehensive understanding of how specific mutations, especially in the newest subvariants, influence the antigenic escape of the pathogen. Here, we tested comprehensive in silico approaches to identify methods for fast and accurate prediction of antibody neutralization by various mutants. As a benchmark, we modeled the complexes of the murine antibody 2B04, which neutralizes infection by preventing the SARS-CoV-2 spike glycoprotein’s association with angiotensin-converting enzyme (ACE2). Complexes with the wild-type, B.1.1.7 Alpha, and B.1.427/429 Epsilon SARS-CoV-2 variants were used as positive controls, while complexes with the B.1.351 Beta, P.1 Gamma, B.1.617.2 Delta, B.1.617.1 Kappa, BA.1 Omicron, and the newest JN.1 Omicron variants were used as decoys. Three essentially different algorithms were employed: forced placement based on a template, followed by two steps of extended molecular dynamics simulations; protein–protein docking utilizing PIPER (an FFT-based method extended for use with pairwise interaction potentials); and the AlphaFold 3.0 model for complex structure prediction. Homology modeling was used to assess the 3D structure of the newly emerged JN.1 Omicron subvariant, whose crystallographic structure is not yet available in the Protein Database. After a careful comparison of these three approaches, we were able to identify the pros and cons of each method. Protein–protein docking yielded two false-positive results, while manual placement reinforced by molecular dynamics produced one false positive and one false negative. In contrast, AlphaFold resulted in only one doubtful result and a higher overall accuracy-to-time ratio. The reasons for inaccuracies and potential pitfalls of various approaches are carefully explained. In addition to a comparative analysis of methods, some mechanisms of immune escape are elucidated herein. This provides a critical foundation for improving the predictive accuracy of vaccine efficacy against new viral subvariants, introducing accurate methodologies, and pinpointing potential challenges.

De novo protein design with a denoising diffusion network independent of pretrained structure prediction models.

The recent success of RFdiffusion, a method for protein structure design with a denoising diffusion probabilistic model, has relied on fine-tuning the RoseTTAFold structure prediction network for protein backbone denoising. Here, we introduce SCUBA-diffusion (SCUBA-D), a protein backbone denoising diffusion probabilistic model freshly trained by considering co-diffusion of sequence representation to enhance model regularization and adversarial losses to minimize data-out-of-distribution errors. While matching the performance of the pretrained RoseTTAFold-based RFdiffusion in generating experimentally realizable protein structures, SCUBA-D readily generates protein structures with not-yet-observed overall folds that are different from those predictable with RoseTTAFold. The accuracy of SCUBA-D was confirmed by the X-ray structures of 16 designed proteins and a protein complex, and by experiments validating designed heme-binding proteins and Ras-binding proteins. Our work shows that deep generative models of images or texts can be fruitfully extended to complex physical objects like protein structures by addressing outstanding issues such as the data-out-of-distribution errors.

ABodyBuilder3: improved and scalable antibody structure predictions.

In this article, we introduce ABodyBuilder3, an improved and scalable antibody structure prediction model based on ABodyBuilder2. We achieve a new state-of-the-art accuracy in the modelling of CDR loops by leveraging language model embeddings, and show how predicted structures can be further improved through careful relaxation strategies. Finally, we incorporate a predicted Local Distance Difference Test into the model output to allow for a more accurate estimation of uncertainties. The software package is available at https://github.com/Exscientia/ABodyBuilder3 with model weights and data at https://zenodo.org/records/11354577.

Plant BCL-Domain Homologues play a conserved role in SWI/SNF complex stability.

The SWItch/Sucrose Non-Fermenting (SWI/SNF) complexes are evolutionarily conserved, ATP-dependent chromatin remodelers crucial for multiple nuclear functions in eukaryotes. Recently, plant BCL-Domain Homolog (BDH) proteins were identified as shared subunits of all plant SWI/SNF complexes, significantly impacting chromatin accessibility and various developmental processes in Arabidopsis. In this study, we performed a comprehensive characterization of bdh mutants, revealing a previously overlooked impact on hypocotyl cell elongation. Through detailed analysis of BDH domains, we identified a plant-specific N-terminal domain that facilitates the interaction between BDH and the rest of the complex. Additionally, we uncovered the critical role of the BDH β-hairpin domain, which is phylogenetically related to metazoan BCL7 SWI/SNF subunits. While phylogenetic analyses did not identify BDH/BCL7 orthologs in fungi, structure prediction modeling demonstrated strong similarities between the SWI/SNF catalytic modules of plants, animals, and fungi, and revealed the yeast Rtt102 protein as a structural homolog of BDH and BCL7. This finding is supported by the ability of Rtt102 to interact with the Arabidopsis catalytic module subunit ARP7 and partially rescue the bdh mutant phenotypes. Further experiments revealed that BDH promotes the stability of the ARP4-ARP7 heterodimer, leading to the partial destabilization of ARP4 in the SWI/SNF complexes. In summary, our study unveils the molecular function of BDH proteins in plant SWI/SNF complexes and suggests that β-hairpin-containing proteins are evolutionarily conserved subunits crucial for ARP heterodimer stability and SWI/SNF activity across eukaryotes.

Structural link prediction model with multi-view text semantic feature extraction

The exponential expansion of information has made text feature extraction based on simple semantic information insufficient for the multidimensional recognition of textual data. In this study, we construct a text semantic structure graph based on various perspectives and introduce weight coefficients and node clustering coefficients of co-occurrence granularity to enhance the link prediction model, in order to comprehensively capture the structural information of the text. Firstly, we jointly build the semantic structure graph based on three proposed perspectives (i.e., scene semantics, text weight, and graph structure), and propose a candidate keyword set in conjunction with an information probability retrieval model. Subsequently, we propose weight coefficients of co-occurrence granularity and node clustering coefficients to improve the link prediction model based on the semantic structure graph, enabling a more comprehensive acquisition of textual structural information. Experimental results demonstrate that our research method can reveal potential correlations and obtain more complete semantic structure information, while the WPAA evaluation index validates the effectiveness of our model.

Impact of Multi-Factor Features on Protein Secondary Structure Prediction.

Protein secondary structure prediction (PSSP) plays a crucial role in resolving protein functions and properties. Significant progress has been made in this field in recent years, and the use of a variety of protein-related features, including amino acid sequences, position-specific score matrices (PSSM), amino acid properties, and secondary structure trend factors, to improve prediction accuracy is an important technical route for it. However, a comprehensive evaluation of the impact of these factor features in secondary structure prediction is lacking in the current work. This study quantitatively analyzes the impact of several major factors on secondary structure prediction models using a more explanatory four-class machine learning approach. The applicability of each factor in the different types of methods, the extent to which the different methods work on each factor, and the evaluation of the effect of multi-factor combinations are explored in detail. Through experiments and analyses, it was found that PSSM performs best in methods with strong high-dimensional features and complex feature extraction capabilities, while amino acid sequences, although performing poorly overall, perform relatively well in methods with strong linear processing capabilities. Also, the combination of amino acid properties and trend factors significantly improved the prediction performance. This study provides empirical evidence for future researchers to optimize multi-factor feature combinations and apply them to protein secondary structure prediction models, which is beneficial in further optimizing the use of these factors to enhance the performance of protein secondary structure prediction models.

Experimental investigation on the residual strength of concrete under the multi-factor coupling action of chloride penetration, freeze-thaw, and fatigue effect

With the ongoing process of globalization and the rapid expansion of the global economy, the proliferation of coastal constructions and infrastructures has witnessed a notable surge. These structures often encounter a confluence of environmental stressors in practical engineering, including cyclic fatigue loading, chloride salt erosion, and freeze-thaw cycles. These synergistic factors accelerate concrete’s failure, resulting in accelerated deterioration and significantly reduced durability. A comprehensive understanding of the damage characteristics of concrete materials in offshore environments, under the multi-factor coupled action of freeze-thaw phenomena, along with an assessment of their residual strength, is crucial for the development of relevant structural analysis and lifetime prediction models. To address this, twenty-seven dedicated specimens were designed and fabricated for this investigation. They were employed to scrutinize the residual strength of concrete under the combined influences of fatigue, freeze-thaw, and chloride penetration. Nine distinct multi-factor coupling actions were meticulously devised and characterized by different parameters. Following the multi-factor coupling action, tests were conducted to assess the residual flexural and compressive strengths, as well as the chloride ion content near the specimen edges. The findings revealed a substantial negative correlation between the residual flexural and compressive strengths with both the number of freeze-thaw cycles and fatigue loading iterations. In contrast, chloride ion concentration and the stress level of fatigue loading (stress ratio: 0.1–0.3) exhibited a comparatively marginal influence. Despite the relatively short test duration, it was evident that both freeze-thaw cycles and fatigue loading exerted a discernible promotional effect on chloride ion ingress in the concrete of the specimens.

An energy-based framework for predicting vehicle noise source intensity: From energy consumption to noise

The vehicle noise source strength prediction model is a crucial component in the field of traffic noise prediction. Despite the establishment of noise source strength localized models in various countries, the theoretical underpinnings of the sound power level models within these frameworks remains unclear. This study addresses this gap by analyzing the correlation between vehicle noise and energy consumption. An energy-based source strength model framework (E-SSIM) is proposed, focusing on developing nonlinear models for basic noise level. E-SSIM is built on acoustical principles and the energy flow of vehicles, integrating noise and energy consumption through the application of multivariate regression theory, characterized by a transient or simplified mathematical framework. Furthermore, sensitivity analysis and road experiments are conducted to validate the proposed framework. The findings reveal that E-SSIM effectively integrates vehicle energy flow and principles of acoustics, thereby providing a theoretical foundation for the logarithmic mathematical structure in classical noise source strength models. The study reveals that in low-speed driving conditions (17–40 km/h), the sensitivity of noise energy to aerodynamic drag energy consumption reaches its peak. Specifically, the sensitivity of E-SSIM, as assessed by the A-weighted sound level, progressively decreases with increasing speed. On the contrary, for the Z-weighted sound level, the sensitivity initially decreases before rising again, reaching its peak stability and robustness at a speed of 23.8 km/h. E-SSIM exhibits superior precision in predicting A/Z-weighted sound pressure levels. Compared to classic logarithmic structural prediction models, the mean absolute percentage error of E-SSIM was reduced by 4.19% and 0.07%. Compared to typical models such as ASJ developed by the Acoustical Society of Japan and CNOSSOS-EU used by the European Commission, E-SSIM yielded a mean absolute percentage error reduction of 68% and 67%. Interestingly, as vehicle internal energy consumption increases, the prediction deviations of E-SSIM, ASJ, and CNOSSOS-EU gradually decrease, possibly because vehicle operating conditions approach stability. E-SSIM can utilize abundant vehicle data to develop generic models, promoting the advancement of noise prediction.

Deep learning methods for protein structure prediction

AbstractProtein structure prediction (PSP) has been a prominent topic in bioinformatics and computational biology, aiming to predict protein function and structure from sequence data. The three‐dimensional conformation of proteins is pivotal for their intricate biological roles. With the advancement of computational capabilities and the adoption of deep learning (DL) technologies (especially Transformer network architectures), the PSP field has ushered in a brand‐new era of “neuralization.” Here, we focus on reviewing the evolution of PSP from traditional to modern deep learning‐based approaches and the characteristics of various structural prediction methods. This emphasizes the advantages of deep learning‐based hybrid prediction methods over traditional approaches. This study also provides a summary analysis of widely used bioinformatics databases and the latest structure prediction models. It discusses deep learning networks and algorithmic optimization for model training, validation, and evaluation. In addition, a summary discussion of the major advances in deep learning‐based protein structure prediction is presented. The update of AlphaFold 3 further extends the boundaries of prediction models, especially in protein‐small molecule structure prediction. This marks a key shift toward a holistic approach in biomolecular structure elucidation, aiming at solving almost all sequence‐to‐structure puzzles in various biological phenomena.

An end-to-end framework for the prediction of protein structure and fitness from single sequence

Significant research progress has been made in the field of protein structure and fitness prediction. Particularly, single-sequence-based structure prediction methods like ESMFold and OmegaFold achieve a balance between inference speed and prediction accuracy, showing promise for many downstream prediction tasks. Here, we propose SPIRED, a single-sequence-based structure prediction model that exhibits comparable performance to the state-of-the-art methods but with approximately 5-fold acceleration in inference and at least one order of magnitude reduction in training consumption. By integrating SPIRED with downstream neural networks, we compose an end-to-end framework named SPIRED-Fitness for the rapid prediction of both protein structure and fitness from single sequence with satisfactory accuracy. Moreover, SPIRED-Stab, the derivative of SPIRED-Fitness, achieves state-of-the-art performance in predicting the mutational effects on protein stability.

AlphaFold2 in biomedical research: facilitating the development of diagnostic strategies for disease.

Proteins, as the primary executors of physiological activity, serve as a key factor in disease diagnosis and treatment. Research into their structures, functions, and interactions is essential to better understand disease mechanisms and potential therapies. DeepMind's AlphaFold2, a deep-learning protein structure prediction model, has proven to be remarkably accurate, and it is widely employed in various aspects of diagnostic research, such as the study of disease biomarkers, microorganism pathogenicity, antigen-antibody structures, and missense mutations. Thus, AlphaFold2 serves as an exceptional tool to bridge fundamental protein research with breakthroughs in disease diagnosis, developments in diagnostic strategies, and the design of novel therapeutic approaches and enhancements in precision medicine. This review outlines the architecture, highlights, and limitations of AlphaFold2, placing particular emphasis on its applications within diagnostic research grounded in disciplines such as immunology, biochemistry, molecular biology, and microbiology.

UPC2 mutations and development of azole resistance in Candida albicans hospital isolates from Lebanon

ObjectivesThis study evaluated the role of Upc2 in the development of azole resistance in Candida albicans isolates from Lebanese hospitalized patients and determined a correlation between resistance and virulence. MethodsThe UPC2 gene which codes for an ergosterol biosynthesis regulator was sequenced and analysed in two azole-resistant and one azole-susceptible C. albicans isolates. An amino acid substitution screening was carried out on Upc2 with a focus on its ligand binding domain (LBD) known to interact with ergosterol. Then, Upc2 protein secondary structure prediction and homology modelling were conducted, followed by total plasma membrane ergosterol and cell wall chitin quantifications. For virulence, mouse models of systemic infection were generated and an agar adhesion and invasion test was performed. ResultsAzole-resistant isolates harboured novel amino acid substitutions in the LBD of Upc2 and changes in protein secondary structures were observed. In addition, these isolates exhibited a significant increase in plasma membrane ergosterol content. Resistance and virulence were inversely correlated while increased cell wall chitin concentration does not seem to be linked to resistance since even though we observed an increase in chitin concentration, it was not statistically significant. ConclusionsThe azole-resistant C. albicans isolates harboured novel amino acid substitutions in the LBD of Upc2 which are speculated to induce an increase in plasma membrane ergosterol content, preventing the binding of azoles to their target, resulting in resistance.

Erratum: Preliminary study on the design of a predictive structural model for daily quality assurance in proton beams

Erratum: Preliminary study on the design of a predictive structural model for daily quality assurance in proton beams

Validation of de novo designed water-soluble and transmembrane β-barrels by in silico folding and melting.

In silico validation of de novo designed proteins with deep learning (DL)-based structure prediction algorithms has become mainstream. However, formal evidence of the relationship between a high-quality predicted model and the chance of experimental success is lacking. We used experimentally characterized de novo water-soluble and transmembrane β-barrel designs to show that AlphaFold2 and ESMFold excel at different tasks. ESMFold can efficiently identify designs generated based on high-quality (designable) backbones. However, only AlphaFold2 can predict which sequences have the best chance of experimentally folding among similar designs. We show that ESMFold can generate high-quality structures from just a few predicted contacts and introduce a new approach based on incremental perturbation of the prediction ("in silico melting"), which can reveal differences in the presence of favorable contacts between designs. This study provides a new insight on DL-based structure prediction models explainability and on how they could be leveraged for the design of increasingly complex proteins; in particular membrane proteins which have historically lacked basic in silico validation tools.

AlphaKnot 2.0: a web server for the visualization of proteins' knotting and a database of knotted AlphaFold-predicted models.

The availability of 3D protein models is rapidly increasing with the development of structure prediction algorithms. With the expanding availability of data, new ways of analysis, especially topological analysis, of those predictions are becoming necessary. Here, we present the updated version of the AlphaKnot service that provides a straightforward way of analyzing structure topology. It was designed specifically to determine knot types of the predicted structure models, however, it can be used for all structures, including the ones solved experimentally. AlphaKnot 2.0 provides the user's ability to obtain the knowledge necessary to assess the topological correctness of the model. Both probabilistic and deterministic knot detection methods are available, together with various visualizations (including a trajectory of simplification steps to highlight the topological complexities). Moreover, the web server provides a list of proteins similar to the queried model within AlphaKnot's database and returns their knot types for direct comparison. We pre-calculated the topology of high-quality models from the AlphaFold Database (4th version) and there are now more than 680.000 knotted models available in the AlphaKnot database. AlphaKnot 2.0 is available at https://alphaknot.cent.uw.edu.pl/.

A Top-Down Design Approach for Generating a Peptide PROTAC Drug Targeting Androgen Receptor for Androgenetic Alopecia Therapy.

While large-scale artificial intelligence (AI) models for protein structure prediction and design are advancing rapidly, the translation of deep learning models for practical macromolecular drug development remains limited. This investigation aims to bridge this gap by combining cutting-edge methodologies to create a novel peptide-based PROTAC drug development paradigm. Using ProteinMPNN and RFdiffusion, we identified binding peptides for androgen receptor (AR) and Von Hippel-Lindau (VHL), followed by computational modeling with Alphafold2-multimer and ZDOCK to predict spatial interrelationships. Experimental validation confirmed the designed peptide's binding ability to AR and VHL. Transdermal microneedle patching technology was seamlessly integrated for the peptide PROTAC drug delivery in androgenic alopecia treatment. In summary, our approach provides a generic method for generating peptide PROTACs and offers a practical application for designing potential therapeutic drugs for androgenetic alopecia. This showcases the potential of interdisciplinary approaches in advancing drug development and personalized medicine.

Advances in the application of AlphaFold2: a protein structure prediction model

Protein structure prediction is an important research field in life sciences and medicine, and it is also a key application scenario of artificial intelligence in scientific research. AlphaFold2 is a protein structure prediction system developed by DeepMind based on deep learning, capable of efficiently generating the atomic-scale spatial structure of a protein from the amino acid sequence. It has demonstrated superior performance in the prediction of protein structures since its inception, thus attracting much attention and research. This paper introduces the model architecture, highlights, limitations, and application progress of AlphaFold2. Furthermore, it briefs the capabilities, highlights, and limitations of several other types of protein structure prediction models and prospects the future development direction in this field.

Accurate Protein pKa Prediction with Physical Organic Chemistry Guided 3D Protein Representation.

Protein pKa is a fundamental physicochemical parameter that dictates protein structure and function. However, accurately determining protein site-pKa values remains a substantial challenge, both experimentally and theoretically. In this study, we introduce a physical organic approach, leveraging a protein structural and physical-organic-parameter-based representation (P-SPOC), to develop a rapid and intuitive model for protein pKa prediction. Our P-SPOC model achieves state-of-the-art predictive accuracy, with a mean absolute error (MAE) of 0.33 pKa units. Furthermore, we have incorporated advanced protein structure prediction models, like AlphaFold2, to approximate structures for proteins lacking three-dimensional representations, which enhances the applicability of our model in the context of structure-undetermined protein research. To promote broader accessibility within the research community, an online prediction interface was also established at isyn.luoszgroup.com.

Insights into the Mechanism of Catalytic Activity of Plasmodium Parasite Malate-Quinone Oxidoreductase.

Plasmodium malate-quinone oxidoreductase (MQO) is a membrane flavoprotein catalyzing the oxidation of malate to oxaloacetate and the reduction of quinone to quinol. Recently, using a yeast expression system, we demonstrated that MQO, expressed in place of mitochondrial malate dehydrogenase (MDH), contributes to the TCA cycle and the electron transport chain in mitochondria, making MQO attractive as a promising drug target in Plasmodium malaria parasites, which lack mitochondrial MDH. However, there is little information on the structure of MQO and its catalytic mechanism, information that will be required to develop novel drugs. Here, we investigated the catalytic site of P. falciparum MQO (PfMQO) using our yeast expression system. We generated a model structure for PfMQO with the AI tool AlphaFold and used protein footprinting by acetylation with acetic anhydride to analyze the surface topology of the model, confirming the computational prediction to be reasonably accurate. Moreover, a putative catalytic site, which includes a possible flavin-binding site, was identified by this combination of protein footprinting and structural prediction model. This active site was analyzed by site-directed mutagenesis. By measuring enzyme activity and protein expression levels in the PfMQO mutants, we showed that several residues at the active site are essential for enzyme function. In addition, a single substitution mutation near the catalytic site resulted in enhanced sensitivity to ferulenol, an inhibitor of PfMQO that competes with malate for binding to the enzyme. This strongly supports the notion that the substrate binds to the proposed catalytic site. Then, the location of the catalytic site was demonstrated by structural comparison with a homologous enzyme. Finally, we used our results to propose a mechanism for the catalytic activity of MQO by reference to the mechanism of action of structurally or functionally homologous enzymes.