Search

Background: Deep learning-based drug-target affinity (DTA) prediction methods have shown impressive performance, despite a high number of training parameters relative to the available data. Previous studies have highlighted the presence of dataset bias by suggesting that models trained solely on protein or ligand structures may perform similarly to those trained on complex structures. However, these studies did not propose solutions and focused solely on analyzing complex structure-based models. Even when ligands are excluded, protein-only models trained on complex structures still incorporate some ligand information at the binding sites. Therefore, it is unclear whether binding affinity can be accurately predicted using only compound or protein features due to potential dataset bias. In this study, we expanded our analysis to comprehensive databases and investigated dataset bias through compound and protein feature-based methods using multilayer perceptron models. We assessed the impact of this bias on current prediction models and proposed the binding affinity similarity explorer (BASE) web service, which provides bias-reduced datasets. Results: By analyzing eight binding affinity databases using multilayer perceptron models, we confirmed a bias where the compound-protein binding affinity can be accurately predicted using compound features alone. This bias arises because most compounds show consistent binding affinities due to high sequence or functional similarity among their target proteins. Our Uniform Manifold Approximation and Projection analysis based on compound fingerprints further revealed that low and high variation compounds do not exhibit significant structural differences. This suggests that the primary factor driving the consistent binding affinities is protein similarity rather than compound structure. We addressed this bias by creating datasets with progressively reduced protein similarity between the training and test sets, observing significant changes in model performance. We developed the BASE web service to allow researchers to download and utilize these datasets. Feature importance analysis revealed that previous models heavily relied on protein features. However, using bias-reduced datasets increased the importance of compound and interaction features, enabling a more balanced extraction of key features. Conclusions: We propose the BASE web service, providing both the affinity prediction results of existing models and bias-reduced datasets. These resources contribute to the development of generalized and robust predictive models, enhancing the accuracy and reliability of DTA predictions in the drug discovery process. BASE is freely available online at https://synbi2024.kaist.ac.kr/base. [ABSTRACT FROM AUTHOR]

Abstract Background Deep learning-based drug-target affinity (DTA) prediction methods have shown impressive performance, despite a high number of training parameters relative to the available data. Previous studies have highlighted the presence of dataset bias by suggesting that models trained solely on protein or ligand structures may perform similarly to those trained on complex structures. However, these studies did not propose solutions and focused solely on analyzing complex structure-based models. Even when ligands are excluded, protein-only models trained on complex structures still incorporate some ligand information at the binding sites. Therefore, it is unclear whether binding affinity can be accurately predicted using only compound or protein features due to potential dataset bias. In this study, we expanded our analysis to comprehensive databases and investigated dataset bias through compound and protein feature-based methods using multilayer perceptron models. We assessed the impact of this bias on current prediction models and proposed the binding affinity similarity explorer (BASE) web service, which provides bias-reduced datasets. Results By analyzing eight binding affinity databases using multilayer perceptron models, we confirmed a bias where the compound-protein binding affinity can be accurately predicted using compound features alone. This bias arises because most compounds show consistent binding affinities due to high sequence or functional similarity among their target proteins. Our Uniform Manifold Approximation and Projection analysis based on compound fingerprints further revealed that low and high variation compounds do not exhibit significant structural differences. This suggests that the primary factor driving the consistent binding affinities is protein similarity rather than compound structure. We addressed this bias by creating datasets with progressively reduced protein similarity between the training and test sets, observing significant changes in model performance. We developed the BASE web service to allow researchers to download and utilize these datasets. Feature importance analysis revealed that previous models heavily relied on protein features. However, using bias-reduced datasets increased the importance of compound and interaction features, enabling a more balanced extraction of key features. Conclusions We propose the BASE web service, providing both the affinity prediction results of existing models and bias-reduced datasets. These resources contribute to the development of generalized and robust predictive models, enhancing the accuracy and reliability of DTA predictions in the drug discovery process. BASE is freely available online at https://synbi2024.kaist.ac.kr/base .

Phosphorylation is an essential mechanism for regulating protein activities. Determining kinase-specific phosphorylation sites by experiments involves time-consuming and expensive analyzes. Although several studies proposed computational methods to model kinase-specific phosphorylation sites, they typically required abundant experimentally verified phosphorylation sites to yield reliable predictions. Nevertheless, the number of experimentally verified phosphorylation sites for most kinases is relatively small, and the targeting phosphorylation sites are still unidentified for some kinases. In fact, there is little research related to these understudied kinases in the literature. Thus, this study aims to create predictive models for these understudied kinases. A kinase–kinase similarity network was generated by merging the sequence-, functional-, protein-domain- and 'STRING'-related similarities. Thus, besides sequence data, protein–protein interactions and functional pathways were also considered to aid predictive modelling. This similarity network was then integrated with a classification of kinase groups to yield highly similar kinases to a specific understudied type of kinase. Their experimentally verified phosphorylation sites were leveraged as positive sites to train predictive models. The experimentally verified phosphorylation sites of the understudied kinase were used for validation. Results demonstrate that 82 out of 116 understudied kinases were predicted with adequate performance via the proposed modelling strategy, achieving a balanced accuracy of 0.81, 0.78, 0.84, 0.84, 0.85, 0.82, 0.90, 0.82 and 0.85, for the 'TK', 'Other', 'STE', 'CAMK', 'TKL', 'CMGC', 'AGC', 'CK1' and 'Atypical' groups, respectively. Therefore, this study demonstrates that web-like predictive networks can reliably capture the underlying patterns in such understudied kinases by harnessing relevant sources of similarities to predict their specific phosphorylation sites. [ABSTRACT FROM AUTHOR]

We have developed an alignment-free TSR (Triangular Spatial Relationship)-based computational method for protein structural comparison and motif identification and discovery. To demonstrate the potential applications of the method, we have generated two datasets. One dataset contains five classes: Actin/Hsp70, serine protease (chymotrypsin/trypsin/elastase), ArsC/Prdx2, PKA/PKB/PKC, and AChE/BChE at the hierarchical level 1 and twelve groups at the level 2. The other dataset includes representative proteases and ACE/ACE2. The x,y, z coordinates of the structures were obtained from PDB. We calculated the keys (or features) that represent each structure using the TSR-based method. The dataset and data presented here include additional information that help the readers become aware of specific applications of the TSR-based method in protein clustering, identification and discovery of metal ion binding sites as well as to understand the effect of amino acid grouping on protein 3D structural relationships at both global and local levels.

Protein–protein interactions (PPIs) are crucial for various biological processes, and predicting PPIs is a major challenge. To solve this issue, the most common method is link prediction. Currently, the link prediction methods based on network Paths of Length Three (L3) have been proven to be highly effective. In this paper, we propose a novel link prediction algorithm, named SMS, which is based on L3 and protein similarities. We first design a mixed similarity that combines the topological structure and attribute features of nodes. Then, we compute the predicted value by summing the product of all similarities along the L3. Furthermore, we propose the Max Similarity Multiplied Similarity (maxSMS) algorithm from the perspective of maximum impact. Our computational prediction results show that on six datasets, including S. cerevisiae, H. sapiens, and others, the maxSMS algorithm improves the precision of the top 500, area under the precision–recall curve, and normalized discounted cumulative gain by an average of 26.99%, 53.67%, and 6.7%, respectively, compared to other optimal methods. [Display omitted] • A mixed similarity combining sequence and topological structure is introduced. • A link prediction algorithm based on complementarity and similarity is proposed. • The proposed method performs better than existing L3, Sim and other methods. [ABSTRACT FROM AUTHOR]

The presence of receptors and the specific binding of the ligands determine nearly all cellular responses. Binding of a ligand to its receptor causes conformational changes of the receptor that triggers the subsequent signaling cascade. Therefore, systematically studying structures of receptors will provide insight into their functions. We have developed the triangular spatial relationship (TSR)‐based method where all possible triangles are constructed with Cα atoms of a protein as vertices. Every triangle is represented by an integer denoted as a "key" computed through the TSR algorithm. A structure is thereby represented by a vector of integers. In this study, we have first defined substructures using different types of keys. Second, using different types of keys represents a new way to interpret structure hierarchical relations and differences between structures and sequences. Third, we demonstrate the effects of sequence similarity as well as sample size on the structure‐based classifications. Fourth, we show identification of structure motifs, and the motifs containing multiple triangles connected by either an edge or a vertex are mapped to the ligand binding sites of the receptors. The structure motifs are valuable resources for the researchers in the field of signal transduction. Next, we propose amino‐acid scoring matrices that capture "evolutionary closeness" information based on BLOSUM62 matrix, and present the development of a new visualization method where keys are organized according to evolutionary closeness and shown in a 2D image. This new visualization opens a window for developing tools with the aim of identification of specific and common substructures by scanning pixels and neighboring pixels. Finally, we report a new algorithm called as size filtering that is designed to improve structure comparison of large proteins with small proteins. Collectively, we provide an in‐depth interpretation of structure relations through the detailed analyses of different types of keys and their associated key occurrence frequencies, geometries, and labels. In summary, we consider this study as a new computational platform where keys are served as a bridge to connect sequence and structure as well as structure and function for a deep understanding of sequence, structure, and function relationships of the protein family. [ABSTRACT FROM AUTHOR]

The study of protein–protein interaction and the determination of protein functions are important parts of proteomics. Computational methods are used to study the similarity between proteins based on Gene Ontology (GO) to explore their functions and possible interactions. GO is a series of standardized terms that describe gene products from molecular functions, biological processes, and cell components. Previous studies on assessing the similarity of GO terms were primarily based on Information Content (IC) between GO terms to measure the similarity of proteins. However, these methods tend to ignore the structural information between GO terms. Therefore, considering the structural information of GO terms, we systematically analyze the performance of the GO graph and GO Annotation (GOA) graph in calculating the similarity of proteins using different graph embedding methods. When applied to the actual Human and Yeast datasets, the feature vectors of GO terms and proteins are learned based on different graph embedding methods. To measure the similarity of the proteins annotated by different GO numbers, we used Dynamic Time Warping (DTW) and cosine to calculate protein similarity in GO graph and GOA graph, respectively. Link prediction experiments were then performed to evaluate the reliability of protein similarity networks constructed by different methods. It is shown that graph embedding methods have obvious advantages over the traditional IC-based methods. We found that random walk graph embedding methods, in particular, showed excellent performance in calculating the similarity of proteins. By comparing link prediction experiment results from GO(DTW) and GOA(cosine) methods, it is shown that GO(DTW) features provide highly effective information for analyzing the similarity among proteins.

Long non-coding RNAs (lncRNAs) are widely concerned because of their close associations with many key biological activities. Though precise functions of most lncRNAs are unknown, research works show that lncRNAs usually exert biological function by interacting with the corresponding proteins. The experimental validation of interactions between lncRNAs and proteins is costly and time-consuming. In this study, we developed a weighted graph-regularized matrix factorization (LPI-WGRMF) method to find unobserved lncRNA–protein interactions (LPIs) based on lncRNA similarity matrix, protein similarity matrix, and known LPIs. We compared our proposed LPI-WGRMF method with five classical LPI prediction methods, that is, LPBNI, LPI-IBNRA, LPIHN, RWR, and collaborative filtering (CF). The results demonstrate that the LPI-WGRMF method can produce high-accuracy performance, obtaining an AUC score of 0.9012 and AUPR of 0.7324. The case study showed that SFPQ, SNHG3, and PRPF31 may associate with Q9NUL5, Q9NUL5, and Q9UKV8 with the highest linking probabilities and need to further experimental validation.

The study of protein–protein interaction and the determination of protein functions are important parts of proteomics. Computational methods are used to study the similarity between proteins based on Gene Ontology (GO) to explore their functions and possible interactions. GO is a series of standardized terms that describe gene products from molecular functions, biological processes, and cell components. Previous studies on assessing the similarity of GO terms were primarily based on Information Content (IC) between GO terms to measure the similarity of proteins. However, these methods tend to ignore the structural information between GO terms. Therefore, considering the structural information of GO terms, we systematically analyze the performance of the GO graph and GO Annotation (GOA) graph in calculating the similarity of proteins using different graph embedding methods. When applied to the actual Human and Yeast datasets, the feature vectors of GO terms and proteins are learned based on different graph embedding methods. To measure the similarity of the proteins annotated by different GO numbers, we used Dynamic Time Warping (DTW) and cosine to calculate protein similarity in GO graph and GOA graph, respectively. Link prediction experiments were then performed to evaluate the reliability of protein similarity networks constructed by different methods. It is shown that graph embedding methods have obvious advantages over the traditional IC-based methods. We found that random walk graph embedding methods, in particular, showed excellent performance in calculating the similarity of proteins. By comparing link prediction experiment results from GO(DTW) and GOA(cosine) methods, it is shown that GO(DTW) features provide highly effective information for analyzing the similarity among proteins. [ABSTRACT FROM AUTHOR]

Long non-coding RNAs (lncRNAs) are widely concerned because of their close associations with many key biological activities. Though precise functions of most lncRNAs are unknown, research works show that lncRNAs usually exert biological function by interacting with the corresponding proteins. The experimental validation of interactions between lncRNAs and proteins is costly and time-consuming. In this study, we developed a weighted graph-regularized matrix factorization (LPI-WGRMF) method to find unobserved lncRNA–protein interactions (LPIs) based on lncRNA similarity matrix, protein similarity matrix, and known LPIs. We compared our proposed LPI-WGRMF method with five classical LPI prediction methods, that is, LPBNI, LPI-IBNRA, LPIHN, RWR, and collaborative filtering (CF). The results demonstrate that the LPI-WGRMF method can produce high-accuracy performance, obtaining an AUC score of 0.9012 and AUPR of 0.7324. The case study showed that SFPQ, SNHG3, and PRPF31 may associate with Q9NUL5, Q9NUL5, and Q9UKV8 with the highest linking probabilities and need to further experimental validation. [ABSTRACT FROM AUTHOR]

Abstract Background Identification of homologous genes is fundamental to comparative genomics, functional genomics and phylogenomics. Extensive public homology databases are of great value for investigating homology but need to be continually updated to incorporate new sequences. As new sequences are rapidly being generated, there is a need for efficient standalone tools to detect homologs in novel data. Results To address this, we present a fast method for detecting homology groups across a large number of individuals and/or species. We adopted a k-mer based approach which considerably reduces the number of pairwise protein alignments without sacrificing sensitivity. We demonstrate accuracy, scalability, efficiency and applicability of the presented method for detecting homology in large proteomes of bacteria, fungi, plants and Metazoa. Conclusions We clearly observed the trade-off between recall and precision in our homology inference. Favoring recall or precision strongly depends on the application. The clustering behavior of our program can be optimized for particular applications by altering a few key parameters. The program is available for public use at https://github.com/sheikhizadeh/pantools as an extension to our pan-genomic analysis tool, PanTools.

Background: Identification of homologous genes is fundamental to comparative genomics, functional genomics and phylogenomics. Extensive public homology databases are of great value for investigating homology but need to be continually updated to incorporate new sequences. As new sequences are rapidly being generated, there is a need for efficient standalone tools to detect homologs in novel data. Results: To address this, we present a fast method for detecting homology groups across a large number of individuals and/or species. We adopted a k-mer based approach which considerably reduces the number of pairwise protein alignments without sacrificing sensitivity. We demonstrate accuracy, scalability, efficiency and applicability of the presented method for detecting homology in large proteomes of bacteria, fungi, plants and Metazoa. Conclusions: We clearly observed the trade-off between recall and precision in our homology inference. Favoring recall or precision strongly depends on the application. The clustering behavior of our program can be optimized for particular applications by altering a few key parameters. The program is available for public use at https://github.com/sheikhizadeh/pantools as an extension to our pan-genomic analysis tool, PanTools. [ABSTRACT FROM AUTHOR]

Abstract: Many biological phenomena can be represented as complex networks. Using a protein binding site comparison approach, we generated a network of ion binding sites on the scale of all known protein structures from the Protein Data Bank. We found that this ion binding site similarity network is scale‐free, indicating a network in which a few ion binding site scaffolds are the network hubs, and these are connected to hundreds of nodes, whereas the vast majority of nodes have only a few neighbors. Enrichment and statistical analysis of the network components and communities yielded insights into underlying processes from the functional and the structural perspective. Largest components and communities were observed to be closely related to basic metabolic processes and some of the most common structural folds, which, from the evolutionary point of view, indicates that they may be the oldest ones. Further, we derived the first comprehensive map of ion interchangeability, based on binding site similarity. Several highly interchangeable protein‐ion binding site pairs emerged (e.g., Ca2+ and Mg2+), as well as structurally distinct ones. The constructed network of ion binding site similarities will aid in understanding the general principles of protein‐ion binding sites structure, function and evolution. We demonstrate potential uses of the network on proteins involved in cancer development and immune response, where individual ions play prominent roles in disease development. [ABSTRACT FROM AUTHOR]

We describe SILIRID (Simple Ligand–Receptor Interaction Descriptor), a novel fixed size descriptor characterizing protein–ligand interactions. SILIRID can be obtained from the binary interaction fingerprints (IFPs) by summing up the bits corresponding to identical amino acids. This results in a vector of 168 integer numbers corresponding to the product of the number of entries (20 amino acids and one cofactor) and 8 interaction types per amino acid (hydrophobic, aromatic face to face, aromatic edge to face, H-bond donated by the protein, H-bond donated by the ligand, ionic bond with protein cation and protein anion, and interaction with metal ion). Efficiency of SILIRID to distinguish different protein binding sites has been examined in similarity search in sc-PDB database, a druggable portion of the Protein Data Bank, using various protein–ligand complexes as queries. The performance of retrieval of structurally and evolutionary related classes of proteins was comparable to that of state-of-the-art approaches (ROC AUC ≈ 0.91). SILIRID can efficiently be used to visualize chemogenomic space covered by sc-PDB using Generative Topographic Mapping (GTM): sc-PDB SILIRID data form clusters corresponding to different protein types.

O objetivo deste estudo foi buscar in silico os ESTs do Banco de dados do Genoma de Eucalyptus (FORESTs, 2001) correlacionados com genes de alongamento celular envolvidos com crescimento de forma geral. Utilizando bibliotecas de cDNA, nós identificamos agrupamentos de ESTs semelhantes às proteínas controlando alongamento celular registradas no National Center of Biotechnologies Information – NCBI (2002). A busca mostrou similaridade para os seguintes agrupamentos de ESTs: o agrupamento EGEZLV2207D07.g com a enzima esterol delta redutase, EGEQRT4200A01.g com a proteína LKB biossintética de brassinosteróide, EGJMLV2220C06.g com EMBL-transportador mitocondrial half ABC, EGACST6260F07.g e EGMCFB1086B12.g com esterol 22 alfa hidroxilase (CYP90) (P450) e EGEQSL1051D09.g com EGEZSL5201E08.g para aproteína copine. The objective of study was in silico mining of ESTs from Eucalyptus ESTs database (FORESTs, 2001) correlated with cell elongation genes of growth character. Using cDNA libraries, we identified similar EST clusters to the proteins involved on control of cell elongation have been registered on National Center of Biotechnologies Information – NCBI (2002). The data mining has shown similarities for the following ESTs clusters: EGEZLV2207D07.g with delta sterol reductase, EGEQRT4200A01.g with LKB brassinosteroid biosynthetic protein, EGJMLV2220C06.g with EMBL-transporter of mitochondrial half ABC, EGACST6260F07.g and EGMCFB1086B12.g with steroid 22-α hydroxylase, and EGEQSL1051D09.g with EGEZSL5201E08.g for copine protein.

We present two recent integer programming models in molecular biology and study practical reformulations to compute solutions to some of these problems. In extension of previously tested linearization techniques, we formulate corresponding semidefinite relaxations and discuss practical rounding strategies to find good feasible approximate solutions. Our computational results highlight the possible advantages and remaining challenges of this approach especially on large-scale problems. [ABSTRACT FROM AUTHOR]

Abstract: We introduce a new measure for assessing similarity among chemical structures, based on well-established computational-topology algorithms. We argue that although the method considers geometry, it is more than a mere geometric similarity measure, as it takes into account, on different geometric scales, the important topological features of the compared structures. We prove that our measure is rigorous and complies with the proper mathematical requirements. We validate the method through comparing different configurations of simple zinc finger proteins and present an application on ligands binding to membrane-proteINS extracted from the Directory of Useful Decoys: Enhanced database and corresponding decoys. This article is part of a Special Issue entitled: Viral membrane proteins — Channels for cellular networking. [Copyright &y& Elsevier]

We have developed an alignment-free TSR (Triangular Spatial Relationship)-based computational method for protein structural comparison and motif identification and discovery. To demonstrate the potential applications of the method, we have generated two datasets. One dataset contains five classes: Actin/Hsp70, serine protease (chymotrypsin/trypsin/elastase), ArsC/Prdx2, PKA/PKB/PKC, and AChE/BChE at the hierarchical level 1 and twelve groups at the level 2. The other dataset includes representative proteases and ACE/ACE2. The x,y, z coordinates of the structures were obtained from PDB. We calculated the keys (or features) that represent each structure using the TSR-based method. The dataset and data presented here include additional information that help the readers become aware of specific applications of the TSR-based method in protein clustering, identification and discovery of metal ion binding sites as well as to understand the effect of amino acid grouping on protein 3D structural relationships at both global and local levels., Competing Interests: The authors declare that they have no known competing financial interests or personal relationships which have or could be perceived to have influenced the work reported in this article., (Published by Elsevier Inc.)

Peach and almond have been considered as model species for the family Rosaceae and other woody plants. Consequently, mapping and characterisation of genes in these species has important implications. High-resolution melting (HRM) analysis is a recent development in the detection of SNPs and other markers, and proved to be an efficient and cost-effective approach. In this study, we aimed to map genes corresponding to known proteins in other species using the HRM approach. Prunus unigenes were searched and compared with known proteins in the public databases. We developed single-nucleotide polymorphism (SNP) markers, polymorphic in a mapping population produced from a cross between the cloned cultivars Nonpareil and Lauranne. A total of 12 SNP-anchored putative genes were genotyped in the population using HRM, and mapped to an existing linkage map. These genes were mapped on six linkage groups, and the predicted proteins were compared to putative orthologs in other species. Amongst those genes, four were abiotic stress-responsive genes, which can provide a starting point for construction of an abiotic resistance map. Two allergy and detoxification related genes, respectively, were also mapped and analysed. Most of the investigated genes had high similarities to sequences from closely related species such as apricot, apple and other eudicots, and these are putatively orthologous. In addition, it was shown that HRM can be an effective means of genotyping populations for the purpose of constructing a linkage map. Our work provides basic genomic information for the 12 genes, which can be used for further genetic and functional studies. [ABSTRACT FROM AUTHOR]

Abstract: In the context of protein engineering, we consider the problem of computing an mRNA sequence of maximal codon-wise similarity to a given mRNA (and consequently, to a given protein) that additionally satisfies some secondary structure constraints, the so-called mRNA Structure Optimization (MRSO) problem. Since MRSO is known to be APX -hard, Bongartz [D. Bongartz, Some notes on the complexity of protein similarity search under mRNA structure constraints, in: Proc. of the 30th Conference on Current Trends in Theory and Practice of Computer Science (SOFSEM), 2004, pp. 174–183] suggested to attack the problem using the approach of parameterized complexity. In this paper we propose three fixed-parameter algorithms that apply for several interesting parameters of MRSO. We believe these algorithms to be relevant for practical applications today, as well as for possible future applications. Furthermore, our results extend the known tractability borderline of MRSO, and provide new research horizons for further improvements of this sort. [Copyright &y& Elsevier]

Abstract: Data reduction techniques are now a vital part of numerical analysis and principal component analysis is often used to identify important molecular features from a set of descriptors. We now take a different approach and apply data reduction techniques directly to protein structure. With this we can reduce the three-dimensional structural data into two-dimensions while preserving the correct relationships. With two-dimensional representations, structural comparisons between proteins are accelerated significantly. This means that protein–protein similarity comparisons are now feasible on a large scale. We show how the approach can help to predict the function of kinase structures according to the Hanks'' classification based on their structural similarity to different kinase classes. [Copyright &y& Elsevier]

The superfamily of serine proteinase inhibitors (serpins) are involved in a number of fundamental biological processes such as blood coagulation, complement activation, fibrinolysis, angiogenesis, inflammation and tumor suppression and are expressed in a cell-specific manner. The average protein size of a serpin family member is 350–400 amino acids, but gene structure varies in terms of number and size of exons and introns. Previous studies of all known serpins identified 16 clades and 10 orphan sequences. Vertebrate serpins can be conveniently classified into six sub-groups.We provide additional data that updates the phylogenetic analysis in the context of structural and functional properties of the proteins. From these, we can conclude that the functional classification of serpins relies on their protein structure and not on sequence similarity. [Copyright &y& Elsevier]

A method for the quantitative evaluation of structural similarity between protein pairs is developed that makes use of a Delaunay-based topological mapping. The result of the mapping is a three-dimensional array which is representative of the global structural topology and whose elements can be used to construe an integral scoring scheme. This scoring scheme was tested for its dependence on the protein length difference in a pairwise comparison, its ability to provide a reasonable means for structural similarity comparison within a family of structural neighbors of similar length, and its sensitivity to the differences in protein conformation. It is shown that such a topological evaluation of similarity is capable of providing insight into these points of interest. Protein structure comparison using the method is computationally efficient and the topological scores, although providing different information about protein similarity, correlate well with the distance root-mean-square deviation values calculated by rigid-body structural alignment. [Copyright &y& Elsevier]

The use of a Gaussian-based representation of protein structures for evaluating protein-structure similarities and deriving three-dimensional superpositions is presented. The approach, as implemented in the program GAPS, is applied to three pairs of proteins with different topological characteristics (rich α-helix, mixed α-helix/β-strand, and rich β-strand), low sequence identities (10–30%), and recognized difficulties to define a unique optimum alignment.Validation of the GAPS superpositions is done by comparison with superpositions obtained by the TOP, GA_FIT, and ALIGN programs and those directly extracted from the FSSP database. Results suggest that a Gaussian-based methodology offers an objective means to, depending on the Gaussian-based representation, derive a consensus three-dimensional superposition when alternative superposition solutions exist. [ABSTRACT FROM AUTHOR]

[Display omitted] • Exploring the effectiveness of the TSR-based method for protein clustering and structural motif identification via amino acid grouping. • We have classified the keys into the different categories for better understanding of protein structure relations. • Applying amino acid grouping to the TSR-based method modestly improves the accuracy of protein clustering in certain cases. • Applying amino acid grouping facilitates the process of identification or discovery of conserved structural motifs. • TSR-based protein 3-D structural comparison method has its uniqueness in identification and discovery of structural motifs/binding sites. • The substructures we defined for coronaviruses' nsp16 will help future antiviral drug design for improving therapeutic outcome. • The substructures we have defined will also help to understand how nsp10 interacts with nsp16 to regulate function of nsp16. Development of protein 3-D structural comparison methods is essential for understanding protein functions. Some amino acids share structural similarities while others vary considerably. These structures determine the chemical and physical properties of amino acids. Grouping amino acids with similar structures potentially improves the ability to identify structurally conserved regions and increases the global structural similarity between proteins. We systematically studied the effects of amino acid grouping on the numbers of Specific/specific, Common/common, and statistically different keys to achieve a better understanding of protein structure relations. Common keys represent substructures found in all types of proteins and Specific keys represent substructures exclusively belonging to a certain type of proteins in a data set. Our results show that applying amino acid grouping to the Triangular Spatial Relationship (TSR)-based method, while computing structural similarity among proteins, improves the accuracy of protein clustering in certain cases. In addition, applying amino acid grouping facilitates the process of identification or discovery of conserved structural motifs. The results from the principal component analysis (PCA) demonstrate that applying amino acid grouping captures slightly more structural variation than when amino acid grouping is not used, indicating that amino acid grouping reduces structure diversity as predicted. The TSR-based method uniquely identifies and discovers binding sites for drugs or interacting proteins. The binding sites of nsp16 of SARS-CoV-2, SARS-CoV and MERS-CoV that we have defined will aid future antiviral drug design for improving therapeutic outcome. This approach for incorporating the amino acid grouping feature into our structural comparison method is promising and provides a deeper insight into understanding of structural relations of proteins. [ABSTRACT FROM AUTHOR]