Structure Prediction Modeling Research Articles

Prediction of specific structural damage to the knee joint using qualitative isokinetic analysis

BackgroundAn isokinetic moment curve (IMC) pattern-damaged structure prediction model may be of considerable value in assisting the diagnosis of knee injuries in clinical scenarios. This study aimed to explore the association between irregular IMC patterns and specific structural damages in the knee, including anterior cruciate ligament (ACL) rupture, meniscus (MS) injury, and patellofemoral joint (PFJ) lesions, and to develop an IMC pattern-damaged structure prediction model.MethodsA total of 94 subjects were enrolled in this study and underwent isokinetic testing of the knee joint (5 consecutive flexion-extension movements within the range of motion of 90°-10°, 60°/s). Qualitative analysis of the IMCs for all subjects was completed by two blinded examiners. A multinomial logistic regression analysis was used to investigate whether a specific abnormal curve pattern was associated with specific knee structural injuries and to test the predictive effectiveness of IMC patterns for specific structural damage in the knee.ResultsThe results of the multinomial logistic regression revealed a significant association between the irregular IMC patterns of the knee extensors and specific structural damages (“Valley” - ACL, PFJ, and ACL + MS, “Drop” - ACL, and ACL + MS, “Shaking” - ACL, MS, PFJ, and ACL + MS). The accuracy and Macro-averaged F1 score of the predicting model were 56.1% and 0.426, respectively.ConclusionThe associations between irregular IMC patterns and specific knee structural injuries were identified. However, the accuracy and Macro-averaged F1 score of the established predictive model indicated its relatively low predictive efficacy. For the development of a more accurate predictive model, it may be essential to incorporate angle-specific and/or speed-specific analyses of qualitative and quantitative data in isokinetic testing. Furthermore, the utilization of artificial intelligence image recognition technology may prove beneficial for analyzing large datasets in the future.

An improved multi-scale convolutional neural network with gated recurrent neural network model for protein secondary structure prediction

An improved multi-scale convolutional neural network with gated recurrent neural network model for protein secondary structure prediction

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

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

RNA3DB: A structurally-dissimilar dataset split for training and benchmarking deep learning models for RNA structure prediction

With advances in protein structure prediction thanks to deep learning models like AlphaFold, RNA structure prediction has recently received increased attention from deep learning researchers. RNAs introduce substantial challenges due to the sparser availability and lower structural diversity of the experimentally resolved RNA structures in comparison to protein structures. These challenges are often poorly addressed by the existing literature, many of which report inflated performance due to using training and testing sets with significant structural overlap. Further, the most recent Critical Assessment of Structure Prediction (CASP15) has shown that deep learning models for RNA structure are currently outperformed by traditional methods.In this paper we present RNA3DB, a dataset of structured RNAs, derived from the Protein Data Bank (PDB), that is designed for training and benchmarking deep learning models. The RNA3DB method arranges the RNA 3D chains into distinct groups (Components) that are non-redundant both with regard to sequence as well as structure, providing a robust way of dividing training, validation, and testing sets. Any split of these structurally-dissimilar Components are guaranteed to produce test and validations sets that are distinct by sequence and structure from those in the training set. We provide the RNA3DB dataset, a particular train/test split of the RNA3DB Components (in an approximate 70/30 ratio) that will be updated periodically. We also provide the RNA3DB methodology along with the source-code, with the goal of creating a reproducible and customizable tool for producing structurally-dissimilar dataset splits for structural RNAs.

Abstract 4909: An RNA-based model for tertiary lymphoid structure (TLS) prediction and classification in pancreatic adenocarcinoma (PDAC)

Abstract Although TLS status possesses prognostic significance in PDAC and can potentially affect chemotherapy outcomes, there is currently a notable lack of RNA sequencing (RNA-seq) models that specialize in TLS identification and classification in PDAC. Here, we developed a model for predicting TLS status (high or low) based on RNA-seq data. Design: Hematoxylin and eosin (H&E) whole slide images of PDAC samples from The Cancer Genome Atlas (TCGA, n = 118) and Clinical Proteomic Tumor Analysis Consortium (CPTAC, n = 129) were used to detect intratumoral and borderline TLSs followed by TLS density measurements (units/mm2) by an experienced pathologist. The samples were then stratified into TLS-high and TLS-low groups based on median density values. Next, we used deconvolution by Kassandra algorithm to identify cell subtypes enriched in each TLS group based on gene expression (RNA-seq) data. Calculation of ssGSEA scores for gene signatures corresponding to cell subtypes and TLS structures was performed, along with survival analysis. Differential expression analysis between TLS-high and TLS-low samples, followed by functional enrichment (|logFC| > 2; padj < 0.01), was conducted. The LightGBM gradient boosting classifier was then trained on ranked expression data with sequential feature selection to predict TLS-high and TLS-low groups. We trained the model with H&E staining annotations and ranked RNA expression data from TCGA or CPTAC samples (total n = 167). The remaining 80 were designated as hold-out samples. The weighted F1 score was computed as a performance metric. Findings: Median density of detected TLSs in the TCGA and CPTAC samples was 0.012 units/mm2. Kassandra deconvolution revealed B-cell enrichment, but fibroblast and M2 macrophage depletion in the TLS-high group. Our calculated ssGSEA scores of previously described TLS gene signatures, along with those of different B-cell subtypes and follicular dendritic cells, showed significant association with the TLS-high group. Genes associated with B-cell proliferation, differentiation, and signaling (CD19, CD22, CD79A, CD79B, and CR2) were also upregulated in this group. Comparing the performance of our RNA-based model on the validation dataset with manual TLS classification by a pathologist, we obtained an F1 weighted score of 0.72 and ROC-AUC score of 0.77. Thus, the TLS predictions by our model concurred with the TLS classification based on H&E annotations and pathological evaluation. Moreover, patients in the predicted TLS-low group had worse overall survival (OS) compared to the TLS-high group (Log(HR) = 0.76; 95% CI [0.04; 1.48]; p<0.05). We present an RNA-based model that stratifies PDAC samples as TLS-high or TLS-low, with predictions that conform to pathological findings. We also found TLS-low samples to associate with worse OS, thus offering an objective means to predict prognoses of PDAC patients based on TLS status. Citation Format: Alexandra Livanova, Andrey Tyshevich, Andrey Kravets, Stanislav Kurpe, Nadezhda Lukashevich, Dmitry Ivchenkov, Daniil Dymov, Anna Belozerova, Kirill Kryukov, Aleksandr Sarachakov, Viktor Svekolkin, Vladimir Kushnarev. An RNA-based model for tertiary lymphoid structure (TLS) prediction and classification in pancreatic adenocarcinoma (PDAC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4909.

Spatially resolved proteomics of the Arabidopsis stomatal lineage identifies polarity complexes for cell divisions and stomatal pores

Cell polarity is used to guide asymmetric divisions and create morphologically diverse cells. We find that two oppositely oriented cortical polarity domains present during the asymmetric divisions in the Arabidopsis stomatal lineage are reconfigured into polar domains marking ventral (pore-forming) and outward-facing domains of maturing stomatal guard cells. Proteins that define these opposing polarity domains were used as baits in miniTurboID-based proximity labeling. Among differentially enriched proteins, we find kinases, putative microtubule-interacting proteins, and polar SOSEKIs with their effector ANGUSTIFOLIA. Using AI-facilitated protein structure prediction models, we identify potential protein-protein interaction interfaces among them. Functional and localization analyses of the polarity protein OPL2 and its putative interaction partners suggest a positive interaction with mitotic microtubules and a role in cytokinesis. This combination of proteomics and structural modeling with live-cell imaging provides insights into how polarity is rewired in different cell types and cell-cycle stages.

Protein Design Using Structure-Prediction Networks: AlphaFold and RoseTTAFold as Protein Structure Foundation Models.

Designing proteins with tailored structures and functions is a long-standing goal in bioengineering. Recently, deep learning advances have enabled protein structure prediction at near-experimental accuracy, which has catalyzed progress in protein design as well. We review recent studies that use structure-prediction neural networks to design proteins, via approaches such as activation maximization, inpainting, or denoising diffusion. These methods have led to major improvements over previous methods in wet-lab success rates for designing protein binders, metalloproteins, enzymes, and oligomeric assemblies. These results show that structure-prediction models are a powerful foundation for developing protein-design tools and suggest that continued improvement of their accuracy and generality will be key to unlocking the full potential of protein design.

The time revolution in macromolecular crystallography.

Macromolecular crystallography has historically provided the atomic structures of proteins fundamental to cellular functions. However, the advent of cryo-electron microscopy for structure determination of large and increasingly smaller and flexible proteins signaled a paradigm shift in structural biology. The extensive structural and sequence data from crystallography and advanced sequencing techniques have been pivotal for training computational models for accurate structure prediction, unveiling the general fold of most proteins. Here, we present a perspective on the rise of time-resolved crystallography as the new frontier of macromolecular structure determination. We trace the evolution from the pioneering time-resolved crystallography methods to modern serial crystallography, highlighting the synergy between rapid detection technologies and state-of-the-art x-ray sources. These innovations are redefining our exploration of protein dynamics, with high-resolution crystallography uniquely positioned to elucidate rapid dynamic processes at ambient temperatures, thus deepening our understanding of protein functionality. We propose that the integration of dynamic structural data with machine learning advancements will unlock predictive capabilities for protein kinetics, revolutionizing dynamics like macromolecular crystallography revolutionized structural biology.

A Structural Equation Modeling-Based Maternal Mortality Prediction Model with Three Levels of Determinants

A Structural Equation Modeling-Based Maternal Mortality Prediction Model with Three Levels of Determinants

Biochemical and in silico structural characterization of a cold-active arginase from the psychrophilic yeast, Glaciozyma antarctica PI12.

Glaciozyma antarctica PI12 is a psychrophilic yeast isolated from Antarctica. In this work, we describe the heterologous production, biochemical properties and in silico structure analysis of an arginase from this yeast (GaArg). GaArg is a metalloenzyme that catalyses thehydrolysis ofL-arginine to L-ornithine and urea. The cDNA of GaArg was reversed transcribed, cloned, expressed and purified as a recombinant protein in Escherichia coli. The purified protein was active against L-arginine as its substrate in a reaction at 20°C, pH 9. At 10-35°C and pH 7-9, the catalytic activity of the protein was still present around 50%. Mn2+, Ni2+, Co2+ and K+ were able to enhance the enzyme activity more than two-fold, while GaArg is most sensitive to SDS, EDTA and DTT. The predicted structure model of GaArg showed a very similar overall fold with other known arginases. GaArg possesses predominantly smaller and uncharged amino acids, fewer salt bridges, hydrogen bonds and hydrophobic interactions compared to the other counterparts. GaArg is the first reported arginase that is cold-active, facilitated by unique structural characteristics for its adaptation of catalytic functions at low-temperature environments. The structure and function of cold-active GaArg provide insights into the potentiality of new applications in various biotechnology and pharmaceutical industries.

Protein language models meet reduced amino acid alphabets.

Protein language models (PLMs), which borrowed ideas for modelling and inference from natural language processing, have demonstrated the ability to extract meaningful representations in an unsupervised way. This led to significant performance improvement in several downstream tasks. Clustering amino acids based on their physical-chemical properties to achieve reduced alphabets has been of interest in past research, but their application to PLMs or folding models is unexplored. Here, we investigate the efficacy of PLMs trained on reduced amino acid alphabets in capturing evolutionary information, and we explore how the loss of protein sequence information impacts learned representations and downstream task performance. Our empirical work shows that PLMs trained on the full alphabet and a large number of sequences capture fine details that are lost in alphabet reduction methods. We further show the ability of a structure prediction model(ESMFold) to fold CASP14 protein sequences translated using a reduced alphabet. For 10 proteins out of the 50 targets, reduced alphabets improve structural predictions with LDDT-Cα differences of up to 19%. Trained models and code are available at github.com/Ieremie/reduced-alph-PLM.

Flexible protein-protein docking with a multitrack iterative transformer.

Conventional protein-protein docking algorithms usually rely on heavy candidate sampling and reranking, but these steps are time-consuming and hinder applications that require high-throughput complex structure prediction, for example, structure-based virtual screening. Existing deep learning methods for protein-protein docking, despite being much faster, suffer from low docking success rates. In addition, they simplify the problem to assume no conformational changes within any protein upon binding (rigid docking). This assumption precludes applications when binding-induced conformational changes play a role, such as allosteric inhibition or docking from uncertain unbound model structures. To address these limitations, we present GeoDock, a multitrack iterative transformer network to predict a docked structure from separate docking partners. Unlike deep learning models for protein structure prediction that input multiple sequence alignments, GeoDock inputs just the sequences and structures of the docking partners, which suits the tasks when the individual structures are given. GeoDock is flexible at the protein residue level, allowing the prediction of conformational changes upon binding. On the Database of Interacting Protein Structures (DIPS) test set, GeoDock achieves a 43% top-1 success rate, outperforming all other tested methods. However, in the standard DIPS train/test splits, we discovered contamination of close homologs in the training set. After decontaminating the training set, the success rate is 31%. On the DB5.5 test set and a benchmark dataset of antibody-antigen complexes, GeoDock outperforms the deep learning models trained using the same dataset but falls behind most of the conventional methods and AlphaFold-Multimer. GeoDock attains an average inference speed of under 1 s on a single GPU, enabling its application in large-scale structure screening. Although binding-induced conformational changes are still a challenge owing to limited training and evaluation data, our architecture sets up the foundation to capture this backbone flexibility. Code and a demonstration Jupyter notebook are available at https://github.com/Graylab/GeoDock.

Fabrication of high aspect ratio grooves on aluminium nitride by laser and chemical milling enhanced micro milling

Aluminium nitride (AlN) ceramic is a typically difficult-to-machine material used in electronic packaging. Laser and chemical milling enhanced micro milling (LCMEM), a high-quality and efficient processing method for AlN, is proposed in this study. Rough machining is completed by repeated alternating laser ablation and chemical removal; the final finish is achieved by micro milling (MM). To achieve precise laser and chemical milling parameters based on the target structure, a laser and chemical milling (LCM) prediction model is established and verified; this can be used to predict the morphology of the grooves after LCM. The change trend of the grooves after LCM with the laser parameters is investigated. It was found that LCMEM can improve the surface quality by 60 %, and an optimal finishing surface with a surface roughness Ra value of 54 nm can be obtained when the feed per tooth is 0.4 µm/z. Lastly, this study uses LCMEM to produce a high aspect ratio groove with aspect ratio of 2.5 and depth of 1250 µm. Compared with conventional MM, LCMEM reduces the allowance of MM by 71.95 %, improves the groove accuracy, and reduces tool wear.

Coal structure identification based on geophysical logging data: Insights from Wavelet Transform (WT) and Particle Swarm Optimization Support Vector Machine (PSO-SVM) algorithms

Coal structure is closely related to microscopic and macroscopic properties of coal, and its accurate identification is of great significance to coalbed methane (CBM) reservoir evaluation, hydraulic fracturing performance and production efficiency prediction. The identification of coal structure based on geophysical logging data has become a popular topic as well as a challenge. The emerging Machine Learning (ML) provides convenience for this issue. Under this background, in the case of the Shanxi Formation No. 3 coal seam and the Taiyuan Formation No. 15 coal seam in Mabidong Block, Qinshui Basin, China, this paper proposed a new coal structure identification method based on geophysical logging data using Wavelet Transform (WT) and Particle Swarm Optimization Support Vector Machine (PSO-SVM) algorithms. Our results showed that the vertical resolution of logging data can be effectively improved when sym5 wavelet basis and third level decomposition were selected in wavelet decomposition, and 4.5 was selected as the weighting coefficient in wavelet reconstruction. The coal structure prediction model based on the logging data processed by WT was established by PSO-SVM algorithm, where PSO was used for parameter optimization (optimal penalty factor C and width parameter σ) and SVM with Radial Basis Function (RBF) kernel was used for model establishment. The Hold-out Cross Validation (HO CV) method was used to test the generalization ability of the prediction model, and the accuracy (ACC) of coal structure prediction in the training set and testing set was 94.26% and 88.46%, respectively. The prediction model was applied to identify the coal structure of two coring wells and the predicted coal structure class was consistent with the true coal structure class, confirming the validity of the model. The coal structure prediction results in the whole study area showed that the tectonic conditions control the coal structure. This work provides new insights for coal structure identification.

Identification of clean energy development routes under carbon emission constraints: A path towards structural adjustment of the power system

Clarifying the clean energy (CE) development path is significant for realizing energy system transformation, coping with climate risks, and ensuring energy and ecological security. This paper uses national data from China from 1978 to 2021 as a sample. It considers the year-by-year dynamic indicators of changes in technology, population, industry, and economic factors in the short term. It takes the mandatory carbon emission reduction targets set by the country as the basis for carbon constraints. The Markov chain model was introduced to construct a power demand structure prediction model under scenarios to more accurately quantify the energy-saving and emission-reduction paths under different carbon constraints and economic development scenarios. The results show that China's electricity demand will rapidly grow in the short term. The upper limit constraint of thermal power generation under the high and normal economic growth models still has room to grow in the short term under the target carbon emission constraint scenario. During 2025–2030, the overall lower limit growth rate of the CE power generation structure under the target emission reduction scenario must exceed 8%, and the overall growth rate under the high-efficiency emission reduction scenario must exceed 12%. By 2030, the proportion of hydropower in CE generation is estimated to decrease to 33.7%, while nuclear power, wind power, and photovoltaic power generation will increase to 16.3%, 30.5% and 19.5%, respectively. These results provide a reference for the government to formulate a CE development path against the background of carbon constraints.

Interpreting forces as deep learning gradients improves quality of predicted protein structures

Protein structure predictions from deep learning models like AlphaFold2, despite their remarkable accuracy, are likely insufficient for direct use in downstream tasks like molecular docking. The functionality of such models could be improved with a combination of increased accuracy and physical intuition. We propose a new method to train deep learning protein structure prediction models using molecular dynamics force fields to work toward these goals. Our custom PyTorch loss function, OpenMM-Loss, represents the potential energy of a predicted structure. OpenMM-Loss can be applied to any all-atom representation of a protein structure capable of mapping into our software package, SidechainNet. We demonstrate our method’s efficacy by finetuning OpenFold. We show that subsequently predicted protein structures, both before and after a relaxation procedure, exhibit comparable accuracy while displaying lower potential energy and improved structural quality as assessed by MolProbity metrics.

Rendering protein mutation movies with MutAmore

BackgroundThe success of AlphaFold2 in reliable protein three-dimensional (3D) structure prediction, assists the move of structural biology toward studies of protein dynamics and mutational impact on structure and function. This transition needs tools that qualitatively assess alternative 3D conformations.ResultsWe introduce MutAmore, a bioinformatics tool that renders individual images of protein 3D structures for, e.g., sequence mutations into a visually intuitive movie format. MutAmore streamlines a pipeline casting single amino-acid variations (SAVs) into a dynamic 3D mutation movie providing a qualitative perspective on the mutational landscape of a protein. By default, the tool first generates all possible variants of the sequence reachable through SAVs (L*19 for proteins with L residues). Next, it predicts the structural conformation for all L*19 variants using state-of-the-art models. Finally, it visualizes the mutation matrix and produces a color-coded 3D animation. Alternatively, users can input other types of variants, e.g., from experimental structures.ConclusionMutAmore samples alternative protein configurations to study the dynamical space accessible from SAVs in the post-AlphaFold2 era of structural biology. As the field shifts towards the exploration of alternative conformations of proteins, MutAmore aids in the understanding of the structural impact of mutations by providing a flexible pipeline for the generation of protein mutation movies using current and future structure prediction models.

SASA-Net: A Spatial-Aware Self-Attention Mechanism for Building Protein 3D Structure Directly From Inter- Residue Distances.

Protein functions are tightly related to the fine details of their 3D structures. To understand protein structures, computational prediction approaches are highly needed. Recently, protein structure prediction has achieved considerable progresses mainly due to the increased accuracy of inter-residue distance estimation and the application of deep learning techniques. Most of the distance-based ab initio prediction approaches adopt a two-step diagram: constructing a potential function based on the estimated inter-residue distances, and then build a 3D structure that minimizes the potential function. These approaches have proven very promising; however, they still suffer from several limitations, especially the inaccuracies incurred by the handcrafted potential function. Here, we present SASA-Net, a deep learning-based approach that directly learns protein 3D structure from the estimated inter-residue distances. Unlike the existing approach simply representing protein structures as coordinates of atoms, SASA-Net represents protein structures using pose of residues, i.e., the coordinate system of each individual residue in which all backbone atoms of this residue are fixed. The key element of SASA-Net is a spatial-aware self-attention mechanism, which is able to adjust a residue's pose according to all other residues' features and the estimated distances between residues. By iteratively applying the spatial-aware self-attention mechanism, SASA-Net continuously improves the structure and finally acquires a structure with high accuracy. Using the CATH35 proteins as representatives, we demonstrate that SASA-Net is able to accurately and efficiently build structures from the estimated inter-residue distances. The high accuracy and efficiency of SASA-Net enables an end-to-end neural network model for protein structure prediction through combining SASA-Net and an neural network for inter-residue distance prediction. Source code of SASA-Net is available at https://github.com/gongtiansu/SASA-Net/.

Structural seismic response prediction based on convolutional neural networks

The seismic response of buildings is crucial for structural performance analysis. For structures with complete design data, the seismic response can be predicted using finite element analysis. However, for structures lacking necessary information, building finite element models and predicting their seismic response can be challenging. Compared to finite element analysis, convolutional neural networks (CNNs) can establish a neural network mapping relationship between the structure and the seismic response to predict the structural response without design data. In this paper, a structural response prediction model based on CNNs is established, aiming to analyze the effect of natural frequency reduction on the structural response after the Tohoku earthquake. The successful prediction of the structural acceleration and displacement response provides a new analytical method for predicting the seismic response of buildings lacking design data.

Challenges in bridging the gap between protein structure prediction and functional interpretation.

The rapid evolution of protein structure prediction tools has significantly broadened access to protein structural data. Although predicted structure models have the potential to accelerate and impact fundamental and translational research significantly, it is essential to note that they are not validated and cannot be considered the ground truth. Thus, challenges persist, particularly in capturing protein dynamics, predicting multi-chain structures, interpreting protein function, and assessing model quality. Interdisciplinary collaborations are crucial to overcoming these obstacles. Databases like the AlphaFold Protein Structure Database, the ESM Metagenomic Atlas, and initiatives like the 3D-Beacons Network provide FAIR access to these data, enabling their interpretation and application across a broader scientific community. Whilst substantial advancements have been made in protein structure prediction, further progress is required to address the remaining challenges. Developing training materials, nurturing collaborations, and ensuring open data sharing will be paramount in this pursuit. The continued evolution of these tools and methodologies will deepen our understanding of protein function and accelerate disease pathogenesis and drug development discoveries.