Structure Prediction Modeling Research Articles

VarGibbs Usage in the Optimization of Nearest-Neighbor Parameters and Prediction of Melting Temperature of RNA Duplexes.

The nearest-neighbor (NN) model is a general tool for the evaluation for oligonucleotide thermodynamic stability. It is primarily used for the prediction of melting temperatures but has also found use in RNA secondary structure prediction and theoretical models of hybridization kinetics. One of the key problems is to obtain the NN parameters from melting temperatures, and VarGibbs was designed to obtain those parameters directly from melting temperatures. Here we will describe the basic workflow from RNA melting temperatures to NN parameters with the use of VarGibbs. We start by a brief revision of the basic concepts of RNA hybridization and of the NN model and then show how to prepare the data files, run the parameter optimization, and interpret the results.

Improved protein structure prediction with trRosettaX2, AlphaFold2, and optimized MSAs in CASP15.

We present the monomer and multimer structure prediction results of our methods in CASP15. We first designed an elaborate pipeline that leverages complementary sequence databases and advanced database searching algorithms to generate high-quality multiple sequence alignments (MSAs). Top MSAs were then selected for the subsequent step of structure prediction. We utilized trRosettaX2 and AlphaFold2 for monomer structure prediction (group name Yang-Server), and AlphaFold-Multimer for multimer structure prediction (group name Yang-Multimer). Yang-Server and Yang-Multimer are ranked at the top and the fourth, respectively, for monomer and multimer structure prediction. For 94 monomers, the average TM-score of the predicted structure models by Yang-Server is 0.876, compared to 0.798 by the default AlphaFold2 (i.e., the group NBIS-AF2-standard). For 42 multimers, the average DockQ score of the predicted structure models by Yang-Multimer is 0.464, compared to 0.389 by the default AlphaFold-Multimer (i.e., the group NBIS-AF2-multimer). Detailed analysis of the results shows that several factors contribute to the improvement, including improved MSAs, iterated modeling for large targets, interplay between monomer and multimer structure prediction for intertwined structures, etc. However, the structure predictions for orphan proteins and multimers remain challenging, and breakthroughs in this area are anticipated in the future.

Application of BPNN Optimized by Fireworks Algorithm on Structural Bearing Capacity Analysis of Transmission Tower under Icing Conditions

With the increasing power demand in city development, the construction and application of transmission towers need to meet higher requirements. Icing is an extreme meteorological condition in the world, and icing disasters cause various accidents such as tower collapse and line disconnection of large-area lines every year. Therefore, it is of great significance to study the ultimate bearing capacity of transmission towers under icing conditions. The structural bearing capacity under icing is greatly affected by meteorological conditions and environmental factors, which has strong randomness and complexity, and brings a series of urgent problems to the structural stability of transmission towers. In the paper, the ultimate bearing capacity analysis and failure path research are conducted on the 500 kV linear cathead tower structure, and a structural bearing capacity prediction model based on the fireworks algorithm (FWA) is established. In order to accurately predict the structural bearing capacity, reduce the adverse effects of line icing, and ensure the structural stability of the transmission tower, the FWA is introduced into the neural network model. The self-adjusting mechanism of the local search ability and global search ability of the FWA is used to optimize the optimization process of weights and thresholds in the neural network, and a structural bearing capacity prediction model based on the FWA improved backpropagation neural network (BPNN) is proposed. Through the analysis and calculation of the measured data of the transmission tower in a certain area, and the comparative study with the traditional BPNN and the BPNN optimized by other common algorithms, the analysis and calculation results of the FWA-BPNN model used in this paper are closer to the actual value, the multiple indicators are more superior, and the model is more stable, which can realize the rapid and accurate analysis of the axial stress of tower members and maximum displacement between member nodes of the transmission tower under the icing condition.

Clinical and molecular characterization of 10 Chinese children with congenital adrenal hyperplasia due to 11beta-hydroxylase deficiency

BackgroundThe clinical manifestations of nonclassical 11beta-hydroxylase deficiency are very similar to those of non-classical 21-hydroxylase deficiency. For this study, we investigated the relationship between the clinical and molecular features of congenital adrenal hyperplasia caused by 11beta-hydroxylase deficiency and reviewed the related literature, which are expected to provide assistance for the clinical diagnosis and analysis of congenital adrenal hyperplasia.MethodsClinical data for 10 Chinese patients diagnosed with congenital adrenal hyperplasia in our hospital from 2018 to 2022 were retrospectively analyzed. We examined the effects of gene mutations on protease activity and constructed three-dimensional structure prediction models of proteins.ResultsWe describe 10 patients with 11beta-hydroxylase gene mutations (n = 5, 46,XY; n = 5, 46,XX), with 10 novel mutations were reported. Female patients received treatment at an early stage, with an average age of 2.08 ± 1.66 years, whereas male patients received treatment significantly later, at an average age of 9.77 ± 3.62 years. The most common CYP11B1 pathogenic variant in the Chinese population was found to be c.1360C > T. All mutations lead to spatial conformational changes that affect protein stability.ConclusionsOur study found that there was no significant correlation between each specific mutation and the severity of clinical manifestations. Different patients with the same gene pathogenic variant may have mild or severe clinical manifestations. The correlation between genotype and phenotype needs further study. Three-dimensional protein simulations may provide additional support for the physiopathological mechanism of genetic mutations.

Microstructural changes precede depression in patients with relapsing-remitting Multiple Sclerosis

BackgroundMultiple Sclerosis lesions in the brain and spinal cord can lead to different symptoms, including cognitive and mood changes. In this study we explore the temporal relationship between early microstructural changes in subcortical volumes and cognitive and emotional function in a longitudinal cohort study of patients with relapsing-remitting Multiple Sclerosis.MethodsIn vivo imaging in forty-six patients with relapsing-remitting Multiple Sclerosis was performed annually over 3 years magnetic resonance imaging. Microstructural changes were estimated in subcortical structures using the free water fraction, a diffusion-based MRI metric. In parallel, patients were assessed with the Hospital Anxiety and Depression Scale amongst other tests. Predictive structural equation modeling was set up to further explore the relationship between imaging and the assessment scores. In a general linear model analysis, the cohort was split into patients with higher and lower depression scores.ResultsNearly all subcortical diffusion microstructure estimates at the baseline visit correlate with the depression score at the 2 years follow-up. The predictive nature of baseline free water estimates and depression subscores after 2 years are confirmed in the predictive structural equation modeling analysis with the thalamus showing the greatest effect size. The general linear model analysis shows patterns of MRI free water differences in the thalamus and amygdala/hippocampus area between participants with high and low depression score.ConclusionsOur data suggests a relationship between higher levels of free-water in the subcortical structures in an early stage of Multiple Sclerosis and depression symptoms at a later stage of the disease.

Kaposi's Sarcoma-Associated Herpesvirus ORF67.5 Functions as a Component of the Terminase Complex.

Kaposi's sarcoma-associated herpesvirus (KSHV) is a double-stranded DNA (dsDNA) gammaherpesvirus with a poorly characterized lytic replication cycle. However, the lytic replication cycle of the alpha- and betaherpesviruses are well characterized. During lytic infection of alpha- and betaherpesviruses, the viral genome is replicated as a precursor form, which contains tandem genomes linked via terminal repeats (TRs). One genomic unit of the precursor form is packaged into a capsid and is cleaved at the TR by the terminase complex. While the alpha- and betaherpesvirus terminases are well characterized, the KSHV terminase remains poorly understood. KSHV open reading frame 7 (ORF7), ORF29, and ORF67.5 are presumed to be components of the terminase complex based on their homology to other terminase proteins. We previously reported that ORF7-deficient KSHV formed numerous immature soccer ball-like capsids and failed to cleave the TRs. ORF7 interacted with ORF29 and ORF67.5; however, ORF29 and ORF67.5 did not interact with each other. While these results suggested that ORF7 is important for KSHV terminase function and capsid formation, the function of ORF67.5 was completely unknown. Therefore, to analyze the function of ORF67.5, we constructed ORF67.5-deficient BAC16. ORF67.5-deficient KSHV failed to produce infectious virus and cleave the TRs, and numerous soccer ball-like capsids were observed in ORF67.5-deficient KSHV-harboring cells. Furthermore, ORF67.5 promoted the interaction between ORF7 and ORF29, and ORF29 increased the interaction between ORF67.5 and ORF7. Thus, our data indicated that ORF67.5 functions as a component of the KSHV terminase complex by contributing to TR cleavage, terminase complex formation, capsid formation, and virus production. IMPORTANCE Although the formation and function of the alpha- and betaherpesvirus terminase complexes are well understood, the Kaposi's sarcoma-associated herpesvirus (KSHV) terminase complex is still largely uncharacterized. This complex presumably contains KSHV open reading frame 7 (ORF7), ORF29, and ORF67.5. We were the first to report the presence of soccer ball-like capsids in ORF7-deficient KSHV-harboring lytic-induced cells. Here, we demonstrated that ORF67.5-deficient KSHV also formed soccer ball-like capsids in lytic-induced cells. Moreover, ORF67.5 was required for terminal repeat (TR) cleavage, infectious virus production, and enhancement of the interaction between ORF7 and ORF29. ORF67.5 has several highly conserved regions among its human herpesviral homologs. These regions were necessary for virus production and for the interaction of ORF67.5 with ORF7, which was supported by the artificial intelligence (AI)-predicted structure model. Importantly, our results provide the first evidence showing that ORF67.5 is essential for terminase complex formation and TR cleavage.

A Simplified Prediction Model of Structural Seismic Vulnerability Considering a Multivariate Fuzzy Membership Algorithm

ABSTRACT Fuzzy decision-making and analytic hierarchical processes are ubiquitously used to predict the seismic damage and vulnerability of building clusters. However, the factors affecting the seismic vulnerability of building structures are diversification, cognitive uncertainty, and complex fuzziness. To probe the impact of multiple fuzzy influencing factors on the vulnerability of regional buildings and the degree of membership between them, the development of a structural vulnerability prediction model based on fuzzy decision-making and a hierarchical system with multiple factors has strong theoretical and practical significance. This study proposes a novel method for rapidly predicting structural vulnerability based on a multivariate fuzzy membership index. Rapid fragility prediction models of fiv4e typical structures considering the multivariate fuzzy membership index are established. A new approach is proposed based on the relationship model between the empirical vulnerability index and seven fuzzy membership parameters. Five types of typical structural vulnerability index rapid prediction innovation models are developed. The new prediction model is compared and verified using the quantitative value of China’s macrointensity standard and the structural seismic loss observation data (98,050 × 104 m2 and 995,000 buildings) of 213 typical earthquakes to establish an empirical vulnerability index belt model. The proposed prediction model comprehensively considers multiple fuzzy membership parameters and the empirical structural earthquake damage database. The model analysis and validation results indicate that the proposed model can be further used for the seismic damage evaluation of typical structures and rapid fragility prediction.

Quinolone Resistance in Gallibacterium anatis Determined by Mutations in Quinolone Resistance-Determining Region.

Control of the important pathogen, Gallibacterium anatis, which causes salpingitis and peritonitis in poultry, relies on treatment using antimicrobial compounds. Among these, quinolones and fluoroquinolones have been used extensively, leading to a rise in the prevalence of resistant strains. The molecular mechanisms leading to quinolone resistance, however, have not previously been described for G. anatis, which is the aim of this study. The present study combines phenotypic antimicrobial resistance data with genomic sequence data from a collection of G. anatis strains isolated from avian hosts between 1979 and 2020. Minimum inhibitory concentrations were determined for nalidixic acid, as well as for enrofloxacin for each included strain. In silico analyses included genome-wide queries for genes known to convey resistance towards quinolones, identification of variable positions in the primary structure of quinolone protein targets and structural prediction models. No resistance genes known to confer resistance to quinolones were identified. Yet, a total of nine positions in the quinolone target protein subunits (GyrA, GyrB, ParC and ParE) displayed substantial variation and were further analyzed. By combining variation patterns with observed resistance patterns, positions 83 and 87 in GyrA, as well as position 88 in ParC, appeared to be linked to increased resistance towards both quinolones included. As no notable differences in tertiary structure were observed between subunits of resistant and sensitive strains, the mechanism behind the observed resistance is likely due to subtle shifts in amino acid side chain properties.

A predicted structure of NADPH Oxidase 1 identifies key components of ROS generation and strategies for inhibition.

NADPH oxidase 1 (NOX1) is primarily expressed in epithelial cells and responsible for local generation of reactive oxygen species (ROS). By specifically manipulating the local redox microenvironment, NOX1 actively engages in epithelial immunity, especially in colorectal and pulmonary epithelia. To unravel the structural basis of NOX1 engaged epithelial immune processes, a predicted structure model was established using RaptorX deep learning models. The predicted structure model illustrates a 6-transmembrane domain structure, a FAD binding domain, and an NADPH binding/NOXO1 interacting region. The substrate/cofactor binding scheme with respect to this proposed model highly correlates with published reports and is verified in our site-directed mutagenesis assays. An electron transport chain, from NADPH to FAD and the two heme groups, was well supported by the predicted model. Through molecular docking analysis of various small molecule NOX1 inhibitors and subsequent experimental validation, we identified pronounced active sites for potent NOX1 inhibition. Specifically, LEU60, VAL71, MET181, LEU185, HIS208, PHE211, TYR214, and TYR280 in the transmembrane domain form an active pocket for insertion of the small molecule inhibitors to inhibit electron transfer between the heme groups, thus affecting extracellular ROS generation. Altogether, our study provides structural information to help elucidate the role of NOX1 in epithelial generation of ROS and sheds light on the development of therapeutics for NOX1 related illnesses.

Binding Site Detection Remastered: Enabling Fast, Robust, and Reliable Binding Site Detection and Descriptor Calculation with DoGSite3

Binding site prediction on protein structures is a crucial step in early phase drug discovery whenever experimental or predicted structure models are involved. DoGSite belongs to the widely used tools for this task. It is a grid-based method that uses a Difference-of-Gaussian filter to detect cavities on the protein surface. We recently reimplemented the first version of this method, released in 2010, focusing on improved binding site detection in the presence of ligands and optimized parameters for more robust, reliable, and fast predictions and binding site descriptor calculations. Here, we introduce the new version, DoGSite3, compare it to its predecessor, and re-evaluate DoGSite on published data sets for a large-scale comparative performance evaluation.

Sperm activate TLR2/TLR1 heterodimerization to induce a weak proinflammatory response in the bovine uterus.

Toll-like receptor 2 (TLR2) signaling pathway is involved in the sperm-triggered uterine inflammatory response at insemination, but its precise mechanism at molecular-level remains unknown. According to the ligand specificity, TLR2 forms a heterodimer with TLR1 or TLR6 as an initial step to mediate intracellular signaling, leading to a specific type of immune response. Hence, the present study aimed to identify the active TLR2 heterodimer (TLR2/1 or TLR2/6) that is involved in sperm-uterine immune crosstalk in bovine using various models. First, in-vitro (bovine endometrial epithelial cells, BEECs) and ex-vivo (bovine uterine explant) models were employed to test different TLR2 dimerization pathways in endometrial epithelia after exposure to sperm or TLR2 agonists; PAM3 (TLR2/1 agonist), and PAM2 (TLR2/6 agonist). Additionally, in-silico approaches were performed to confirm the dimer stability using de novo protein structure prediction model for bovine TLRs. The in-vitro approach revealed that sperm triggered the mRNA and protein expression of TLR1 and TLR2 but not TLR6 in BEECs. Moreover, this model disclosed that activation of TLR2/6 heterodimer, triggers a much stronger inflammatory response than TLR2/1 and sperm in bovine uterine epithelia. In the ex-vivo model that mimics the intact uterine tissue at insemination, sperm also induced the protein expression of both TLR1 and TLR2, but not TLR6, in bovine endometrium, particularly in uterine glands. Importantly, PAM3 and sperm induced similar and low mRNA expression of pro-inflammatory cytokines and TNFA protein to a lesser extent than PAM2 in endometrial epithelia. This implied that sperm might trigger a weak inflammatory response via TLR2/TLR1 activation which is similar to that of PAM3. Additionally, the in-silico analyses showed that the existence of bridging ligands is essential for heterodimer stability in bovine TLR2 with either TLR1 or TLR6. Altogether, the present findings revealed that sperm utilize TLR2/1, but not TLR2/6, heterodimerization to trigger a weak physiological inflammatory response in the bovine uterus. This might be the way to remove excess dead sperm remaining in the uterine lumen without tissue damage for providing an ideal uterine environment for early embryo reception and implantation.

A multimode structured prediction model based on dynamic attribution graph attention network for complex industrial processes

Complex industrial processes with dynamic and time-varying characteristics, as well as diverse operating conditions, pose challenges in developing accurate real-time online prediction models for the key performance indicator (KPI). To address these challenges, a KPI prediction framework is proposed in this study, which utilizes a multimode structured prediction model based on a dynamic-attribution graph attention network (DAGAT). First, a multimode clustering method based on a Gaussian mixture model is designed to divide and locate industrial multimode operational data. Second, to address the limitations of traditional deep network models in accurately capturing the complex relationships between process variables and KPI, a graph structure is used to model multi-sensor time series data. Subsequently, a novel DAGAT is proposed to improve node prediction performance and interpretation of industrial graph data. Considering the cumulative effect of large-scale equipment, the DAGAT incorporates knowledge of the industrial process, enabling it to capture the influence of chemical-reaction variables. Simultaneously, the Huber loss function is used to further optimize the prediction results. Moreover, the model can calculate in real-time the contribution of each individual process variable to KPI, providing guidance for optimal control. Finally, the effectiveness and superiority of the proposed method are validated through an industrial case.

Solubilization, purification, and characterization of the hexameric form of phosphatidylserine synthase from Candida albicans

Phosphatidylserine (PS) synthase from Candida albicans, encoded by the CHO1 gene, has been identified as a potential drug target for new antifungals against systemic candidiasis. Rational drug design or small molecule screening are effective ways to identify specific inhibitors of Cho1, but both will be facilitated by protein purification. Due to the transmembrane nature of Cho1, methods were needed to solubilize and purify the native form of Cho1. Here, we used six non-ionic detergents and three styrene maleic acids (SMAs) to solubilize an HA-tagged Cho1 protein from the total microsomal fractions. Blue native PAGE and immunoblot analysis revealed a single band corresponding to Cho1 in all detergent-solubilized fractions, while two bands were present in the SMA2000-solubilized fraction. Our enzymatic assay suggests that digitonin-or DDM-solubilized enzyme has the most PS synthase activity. Pull-downs of HA-tagged Cho1 from the digitonin-solubilized fraction reveal an apparent MW of Cho1 consistent with a hexamer. Furthermore, negative-staining electron microscopy analysis and AlphaFold2 structure prediction modeling suggest the hexamer is composed of a trimer of dimers. We purified Cho1 protein to near-homogeneity as a hexamer using affinity chromatography and TEV protease treatment, and optimized Cho1 enzyme activity for manganese and detergent concentrations, temperature (24 °C), and pH (8.0). The purified Cho1 has a Km for its substrate CDP-diacylglycerol of 72.20μM with a Vmax of 0.079nmol/(μg∗min) while exhibiting a sigmoidal kinetic curve for its other substrate serine, indicating cooperative binding. Purified hexameric Cho1 can potentially be used in downstream structure determination and small drug screening.

Sequence similarity governs generalizability of de novo deep learning models for RNA secondary structure prediction.

Making no use of physical laws or co-evolutionary information, de novo deep learning (DL) models for RNA secondary structure prediction have achieved far superior performances than traditional algorithms. However, their statistical underpinning raises the crucial question of generalizability. We present a quantitative study of the performance and generalizability of a series of de novo DL models, with a minimal two-module architecture and no post-processing, under varied similarities between seen and unseen sequences. Our models demonstrate excellent expressive capacities and outperform existing methods on common benchmark datasets. However, model generalizability, i.e., the performance gap between the seen and unseen sets, degrades rapidly as the sequence similarity decreases. The same trends are observed from several recent DL and machine learning models. And an inverse correlation between performance and generalizability is revealed collectively across all learning-based models with wide-ranging architectures and sizes. We further quantitate how generalizability depends on sequence and structure identity scores via pairwise alignment, providing unique quantitative insights into the limitations of statistical learning. Generalizability thus poses a major hurdle for deploying de novo DL models in practice and various pathways for future advances are discussed.

Exploration of the Atomic Pathway of Seed-Mediated Growth from Icosahedral [Au25(SR)18]− to Bi-Icosahedral Au38(SR)24 and Au44(SR)26 Clusters Based on the 2e– Hopping Mechanism

Size growth is ubiquitous in the gold nanocluster synthesis. However, the atomic-level mechanism of seed-mediated growth of gold clusters remains mysterious. In this study, the seed-mediated growth pathway from the icosahedral [Au25(SR)18]- cluster to the bi-icosahedral Au38(SR)24 and Au44(SR)26 clusters is studied based on the two-electron (2e-) hopping mechanism. First, atomic structures of three key intermediate clusters, [Au29(SR)20]-, [Au33(SR)22]-, and Au41(SR)25, are predicted based on the 2e--unit decomposition strategy. The theoretically simulated UV-Vis spectra based on the predicted structure model of [Au29(SR)20]- and [Au33(SR)22]- matched well with the experimental curves reported previously. Based on the predicted intermediate cluster structures, the size growth pathway from the eight-electron (8e-) [Au25(SR)18]- cluster to 14-electron (14e-) Au38(SR)24 and 18-electron (18e-) Au44(SR)26 clusters is determined. In the step of formation of bi-icosahedral Au38(SR)24 from icosahedral [Au25(SR)18]-, two Au4 units are first formed. The third 2e- hopping step results in formation of an icosahedron unit. The present studies offered new insights into the formation and size conversion mechanism of ligand-protected gold nanoclusters containing icosahedral cores.

Recent Advances in Quantum Computing for Drug Discovery and Development

Drug discovery and development is a time-consuming and cost-intensive process. Computer-aided drug design can speed up the timeline and reduce costs by decreasing the number of necessary biochemical experiments. The number of studies using quantum computing to solve problems in drug development has been increasing in recent years. In this review, we briefly introduce the main steps in drug discovery and development and how computers help to find potential drug candidates. Recent studies of quantum computing in drug development based on the structure of target proteins are listed chronologically. They include protein structure prediction, molecular docking, quantum simulation, and quantitative structure-activity relationship (QSAR) models. Current quantum devices are still susceptible to noise and error but are well suited for hybrid quantum-classical algorithms. The quantum advantage is demonstrated on hybrid systems and quantum-inspired devices such as quantum annealers. We hope to see more applications of quantum computing in the field of drug discovery and development.

Effective Molecular Dynamics from Neural Network-Based Structure Prediction Models.

Recent breakthroughs in neural network-based structure prediction methods, such as AlphaFold2 and RoseTTAFold, have dramatically improved the quality of computational protein structure prediction. These models also provide statistical confidence scores that can estimate uncertainties in the predicted structures, but it remains unclear to what extent these scores are related to the intrinsic conformational dynamics of proteins. Here, we compare AlphaFold2 prediction scores with explicit large-scale molecular dynamics simulations of 28 one- and two-domain proteins with varying degrees of flexibility. We demonstrate a strong correlation between the statistical prediction scores and the explicit motion derived from extensive atomistic molecular dynamics simulations and further derive an elastic network model based on the statistical scores of AlphFold2 (AF-ENM), which we benchmark in combination with coarse-grained molecular dynamics simulations. We show that our AF-ENM method reproduces the global protein dynamics with improved accuracy, providing a powerful way to derive effective molecular dynamics using neural network-based structure prediction models.

Optimal design of γʹ-strengthened high-entropy alloys via machine learning multilayer structural model

γʹ-strengthened high-entropy alloys (HEAs) have been widely studied in recent years because of their excellent mechanical properties at room- and elevated-temperature. The element diversity of HEAs leads to its vast composition and preparation process space and accelerating the design of γʹ-strengthened HEAs by determining phase and mechanical properties remains a prominent challenge. In this study, by building a multi-layer structure prediction model, which includes accurate prediction models of microstructure and mechanical property, aiming to find HEAs with γʹ phase high-volume fraction and high strength. Four γʹ-strengthened alloys were selected from 800,000 candidate alloys by the multilayer structural prediction model, and then it was verified that all four HEAs have a high γʹ phase volume fraction and high strength by experiment. Furthermore, the mathematical relationship between the different metal elements, heat treatment processes, and γ′ phase volume fraction by resolving the machine learning model with the shapely additive algorithm (SHAP). A mathematical relationship model for the strengthening mechanism of HEAs was established to analyze the strengthening relationship of different strengthening mechanisms. The multilayer structural model can be used for the efficient design of γʹ-strengthened high-entropy alloys, and analyze multiple potential relationships that influence the properties of alloys through the underlying data of the model.

Study on novel steel slag foam concrete pore structure and bp neural network prediction model

In this paper, a new type of steel slag foamed concrete was prepared. Its pore structure was analyzed by MATLAB image processing technique, and the influence of steel slag on the compressive strength of foamed concrete was investigated. In addition, a BP neural network prediction model for the pore structure of foam concrete was established. The research results show that when the content of steel slag is 30%, the pore structure of steel slag foamed concrete is appropriate. The compressive strength of steel slag foamed concrete increases with the decrease of the density and the proportion of irregular pores. With the density remaining unchanged, the compressive strength of foamed concrete increases with the increase of the ratio of round and small pores. The prediction errors of the BP neural network model in the average pore diameter and average pore roundness factor were less than 8.5% and 5%, respectively.This research can provide a theoretical basis for the production of steel slag foam concrete.

Ensemble deep learning models for protein secondary structure prediction using bidirectional temporal convolution and bidirectional long short-term memory.

Protein secondary structure prediction (PSSP) is a challenging task in computational biology. However, existing models with deep architectures are not sufficient and comprehensive for deep long-range feature extraction of long sequences. This paper proposes a novel deep learning model to improve Protein secondary structure prediction. In the model, our proposed bidirectional temporal convolutional network (BTCN) can extract the bidirectional deep local dependencies in protein sequences segmented by the sliding window technique, the bidirectional long short-term memory (BLSTM) network can extract the global interactions between residues, and our proposed multi-scale bidirectional temporal convolutional network (MSBTCN) can further capture the bidirectional multi-scale long-range features of residues while preserving the hidden layer information more comprehensively. In particular, we also propose that fusing the features of 3-state and 8-state Protein secondary structure prediction can further improve the prediction accuracy. Moreover, we also propose and compare multiple novel deep models by combining bidirectional long short-term memory with temporal convolutional network (TCN), reverse temporal convolutional network (RTCN), multi-scale temporal convolutional network (multi-scale bidirectional temporal convolutional network), bidirectional temporal convolutional network and multi-scale bidirectional temporal convolutional network, respectively. Furthermore, we demonstrate that the reverse prediction of secondary structure outperforms the forward prediction, suggesting that amino acids at later positions have a greater impact on secondary structure recognition. Experimental results on benchmark datasets including CASP10, CASP11, CASP12, CASP13, CASP14, and CB513 show that our methods achieve better prediction performance compared to five state-of-the-art methods.