Search

Feature fusion and selection strategies have been applied to improve accuracy in the prediction of protein-protein interaction (PPI). In this paper, an embedded feature selection framework is developed by integrating a cost function based on analysis of variance (ANOVA) with the particle swarm optimization (PSO), termed AVPSO. Initially, the features of the protein sequences extracted using pseudo-amino acid composition (PseAAC), conjoint triad composition, and local descriptor are fused. Then, AVPSO is employed to select the optimal set of features. The light gradient boosting machine (LGBM) classifier is used to predict the PPIs using the optimal feature subset. On the five-fold cross-validation analysis, the proposed model (AVPSO-LGBM) achieved an average accuracy of 97.12% and 95.09%, respectively, on the intraspecies PPI datasets Saccharomyces cerevisiae and Helicobacter pylori. On the interspecies, PPI datasets of the Human-Bacillus and Human-Yersinia, an average accuracy of 95.20% and 93.44%, are achieved. Results obtained on independent test datasets, and network datasets show that the prediction accuracy of the AVPSO-LGBM is better than the existing methods, demonstrating its generalization ability. The improved prediction performance obtained by the proposed model makes it a reliable and effective PPI prediction model.

The soybean crop, Glycine max (L.) Merr., is consumed by humans, Homo sapiens, worldwide. While the respective bodies of literature and -omics data for each of these organisms are extensive, comparatively few studies investigate the molecular biological processes occurring between the two. We are interested in elucidating the network of protein-protein interactions (PPIs) involved in human-soybean allergies. To this end, we leverage state-of-the-art sequence-based PPI predictors amenable to predicting the enormous comprehensive interactome between human and soybean. A network-based analytical approach is proposed, leveraging similar interaction profiles to identify candidate allergens and proteins involved in the allergy response. Interestingly, the predicted interactome can be explored from two complementary perspectives: which soybean proteins are predicted to interact with specific human proteins and which human proteins are predicted to interact with specific soybean proteins. A total of eight proteins (six specific to the human proteome and two to the soy proteome) have been identified and supported by the literature to be involved in human health, specifically related to immunological and neurological pathways. This study, beyond generating the most comprehensive human-soybean interactome to date, elucidated a soybean seed interactome and identified several proteins putatively consequential to human health.

Protein-protein interaction (PPI) not only plays a critical role in cell life activities, but also plays an important role in discovering the mechanism of biological activity, protein function, and disease states. Developing computational methods is of great significance for PPIs prediction since experimental methods are time-consuming and laborious. In this paper, we proposed a PPI prediction algorithm called GRNN-PPI only using the amino acid sequence information based on general regression neural network and two feature extraction methods. Specifically, we designed a new feature extraction method named Mutation Spectral Radius (MSR) to extract evolutionary information by the BLOSUM62 matrix. Meanwhile, we integrated another feature extraction method, autocorrelation description, which can completely extract information on physicochemical properties and protein sequences. The principal component analysis was applied to eliminate noise, and the general regression neural network was adopted as a classifier. The prediction accuracy of the yeast, human, and Helicobacter pylori1 (H. pylori1) data sets were 97.47%, 99.63%, and 99.97%, respectively. In addition, we also conducted experiments on two important PPI networks and six independent data sets. All results were significantly higher than some state-of-the-art methods used for comparison, showing that our method is feasible and robust.

This paper tackles the problem of predicting the protein-protein interactions that arise in all living systems. Inference of protein-protein interactions is of paramount importance for understanding fundamental biological phenomena, including cross-species protein-protein interactions, such as those causing the 2020-21 pandemic of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Furthermore, it is relevant also for applications such as drug repurposing, where a known authorized drug is applied to novel diseases. On the other hand, a large fraction of existing protein interactions are not known, and their experimental measurement is resource consuming. To this purpose, we adopt a Graph Signal Processing based approach modeling the protein-protein interaction (PPI) network (a.k.a. the interactome) as a graph and some connectivity related node features as a signal on the graph. We then leverage the signal on graph features to infer links between graph nodes, corresponding to interactions between proteins. Specifically, we develop a Markovian model of the signal on graph that enables the representation of connectivity properties of the nodes, and exploit it to derive an algorithm to infer the graph edges. Performance assessment by several metrics recognized in the literature proves that the proposed approach, named GRAph signal processing Based PPI prediction (GRABP), effectively captures underlying biologically grounded properties of the PPI network.

Background Protein–protein interaction (PPI) prediction is an important task towards the understanding of many bioinformatics functions and applications, such as predicting protein functions, gene-disease associations and disease-drug associations. However, many previous PPI prediction researches do not consider missing and spurious interactions inherent in PPI networks. To address these two issues, we define two corresponding tasks, namely missing PPI prediction and spurious PPI prediction, and propose a method that employs graph embeddings that learn vector representations from constructed Gene Ontology Annotation (GOA) graphs and then use embedded vectors to achieve the two tasks. Our method leverages on information from both term–term relations among GO terms and term-protein annotations between GO terms and proteins, and preserves properties of both local and global structural information of the GO annotation graph. Results We compare our method with those methods that are based on information content (IC) and one method that is based on word embeddings, with experiments on three PPI datasets from STRING database. Experimental results demonstrate that our method is more effective than those compared methods. Conclusion Our experimental results demonstrate the effectiveness of using graph embeddings to learn vector representations from undirected GOA graphs for our defined missing and spurious PPI tasks.

Protein is the primary building block of living organisms. It interacts with other proteins and is then involved in various biological processes. Protein–protein interactions (PPIs) help in predicting and hence help in understanding the functionality of the proteins, causes and growth of diseases, and designing new drugs. However, there is a vast gap between the available protein sequences and the identification of protein–protein interactions. To bridge this gap, researchers proposed several computational methods to reveal the interactions between proteins. These methods merely depend on sequence-based information of proteins. With the advancement of technology, different types of information related to proteins are available such as 3D structure information. Nowadays, deep learning techniques are adopted successfully in various domains, including bioinformatics. So, current work focuses on the utilization of different modalities, such as 3D structures and sequence-based information of proteins, and deep learning algorithms to predict PPIs. The proposed approach is divided into several phases. We first get several illustrations of proteins using their 3D coordinates information, and three attributes, such as hydropathy index, isoelectric point, and charge of amino acids. Amino acids are the building blocks of proteins. A pre-trained ResNet50 model, a subclass of a convolutional neural network, is utilized to extract features from these representations of proteins. Autocovariance and conjoint triad are two widely used sequence-based methods to encode proteins, which are used here as another modality of protein sequences. A stacked autoencoder is utilized to get the compact form of sequence-based information. Finally, the features obtained from different modalities are concatenated in pairs and fed into the classifier to predict labels for protein pairs. We have experimented on the human PPIs dataset and Saccharomyces cerevisiae PPIs dataset and compared our results with the state-of-the-art deep-learning-based classifiers. The results achieved by the proposed method are superior to those obtained by the existing methods. Extensive experimentations on different datasets indicate that our approach to learning and combining features from two different modalities is useful in PPI prediction.

Protein-protein interactions (PPIs) are the physical connections between two or more proteins via electrostatic forces or hydrophobic effects. Identification of the PPIs is pivotal, which contributes to many biological processes including protein function, disease incidence, and therapy design. The experimental identification of PPIs via high-throughput technology is time-consuming and expensive. Bioinformatics approaches are expected to solve such restrictions. In this review, our main goal is to provide an inclusive view of the existing sequence-based computational prediction of PPIs. Initially, we briefly introduce the currently available PPI databases and then review the state-of-the-art bioinformatics approaches, working principles, and their performances. Finally, we discuss the caveats and future perspective of the next generation algorithms for the prediction of PPIs.

Das Protein ist ein wesentlicher Bestandteil der biologischen Zelle. Die verschiedenen chemischen Wechselwirkungen zwischen Proteinen, die sogenannten Protein-Protein-Wechselwirkungen (IPP), stehen mit bestimmten Krankheiten und Therapien in Verbindung. Daher hat ihre Identifizierung wichtige Auswirkungen auf mehrere Prozesse, darunter die Prävention von Krankheiten, die funktionelle Annotation und die Entwicklung von Medikamenten. In den letzten Jahrzehnten wurden viele PPIs durch biologische Experimente entdeckt, aber viele von ihnen sind noch unentdeckt. Darüber hinaus sind diese biologischen Experimente aufgrund von Zeit- und Kostenbeschränkungen begrenzt. Daher wird die Entwicklung von Computerwerkzeugen dringend empfohlen. Dadurch könnte die Entdeckung von Medikamenten beschleunigt werden, indem langsame und teure biologische Experimente durch schnellere und kostengünstigere Computersimulationen reduziert werden und die Funktion von Proteinen anhand von Proteinsequenzen annotiert wird. Angesichts der vielen experimentell nachgewiesenen PPIs, deren Informationen in Proteindatenbanken verfügbar sind, wurden mehrere Inferenz- und maschinelle Lernwerkzeuge (Machine Learning) zur Vorhersage von PPIs vorgeschlagen. Die Konzeption solcher Tools umfasst zwei Schritte: die Extraktion von Merkmalen (Deskriptoren) aus den Sequenzinformationen und die Klassifizierung der Interaktionen mithilfe eines überwachten Lernalgorithmus. Die mit den bestehenden Techniken extrahierten Merkmale ermöglichen es den überwachten Lernalgorithmen jedoch nicht, effizient zu arbeiten und ideale Ergebnisse zu erzielen. Daher ist es unser Ziel, die Vorhersageleistung durch die Entwicklung neuer, ergänzender Techniken zu verbessern, die es den überwachten Lernalgorithmen ermöglichen, Interaktionen korrekt aus Proteinsequenzdaten abzuleiten und ideale Ergebnisse zu erzielen. In dieser Dissertation haben wir erstens eine Technik zur Merkmalsextraktion vorgeschlagen, die mit BP (Bigram-Physicochemical) bezeichnet wird. Diese Technik ermöglicht es, Bigram-Merkmale aus Proteinsequenzen für ein effizientes maschinelles Lernen zu extrahieren. Ein Bigram ist ein Satz von zwei aufeinanderfolgenden Buchstaben (hier Aminosäuren) in einem Textdokument (hier die Proteinsequenz). Für ein bestimmtes Protein berechnet BP zunächst eine Matrix aus Informationen über die physikalisch-chemischen Eigenschaften des Proteins. Diese Matrix kann entweder aus einem Abstand (BP1-Ansatz) oder aus einer Funktion (BP-Ansatz) abgeleitet werden. Anschließend extrahiert BP anhand der berechneten Matrix die Bigramm-Merkmale. Die BP-Technik erzeugt keine streng sparsamen Vektoren und ist nicht von einer Datenbank abhängig wie einige bestehende Techniken zur Extraktion von Bigram-Merkmalen. Zweitens haben wir einen neuen Ansatz zur Auswahl optimaler Werte von Hyperparametern eines Modells für maschinelles Lernen vorgeschlagen, der als SVOH bezeichnet wird. Hyperparameter sind die einflussreichen Parameter des maschinellen Lernmodells. In der Regel wird bei der Suche nach optimalen Werten für Hyperparameter die Technik der Grid Search mit der Technik der k-fachen Kreuzvalidierung kombiniert. Im Gegensatz zur Literatur, die den Wert der Zahl k, die eigentlich der Anzahl der Teilmengen der Stichprobe entspricht, auf der Grundlage von a priori auf 5 oder 10 festlegt, führt SVOH ein Lernen der Zahl k durch, um einen optimalen Wert für die Anzahl der Teilmengen zu bestimmen. Der entwickelte Ansatz ermöglicht somit die rigorose Suche nach optimalen Werten für Hyperparameter in einer angepassten Anzahl k von Teilmengen. Die in der Dissertation beschriebenen Techniken, die mit einer Methode der Support Vector Machines (SVM) kombiniert werden und so die Werkzeuge SVM-BP und SVM-BP1 bilden, werden an drei verschiedenen Sätzen realer PPI-Daten getestet und validiert: menschliche HPRD-PPIs, Hefe-PPIs von Saccharomyces cerevisiae und PPIs des Bakteriums Helicobacter pylori. Die Ergebnisse nach einigen Vergleichsexperimenten zeigten, dass diese Tools und insbesondere das SVM-BP-Tool bei den drei verschiedenen IPP-Datensätzen in den Metriken Richtigkeit, Präzision und Sensitivität eine überlegene Leistung erzielten. Wir können sagen, dass die SVM-BP- und SVM-BP1-Tools die Vorhersageleistung von PPIs deutlich verbessern und somit eine echte Hilfe für Biologen bei der Identifizierung von Protein-Protein-Interaktionen und der Arzneimittelforschung darstellen., Protein is an essential component of the biological cell. The different chemical interactions between proteins called protein-protein interactions (PPIs) are linked to certain diseases as well as to certain therapies. Therefore, their identification has important implications for several processes including disease prevention, functional annotation, and drug design. Many PPIs have been detected by biological experiments over the past decades, but many remain undiscovered. In addition, these biological experiments are limited due to time and cost constraints. Therefore, the development of computational tools is highly recommended. This could accelerate drug discovery by reducing slow and expensive biological experiments with faster and cheaper computer simulations and annotate protein function from protein sequences. Given many experimentally detected PPIs with information available in protein databases, several inference and machine learning tools have been proposed to predict PPIs. The design of such tools involves two steps: feature extraction (descriptors) from sequence information and classification of interactions using a supervised learning algorithm. However, the features extracted by existing techniques do not allow supervised learning algorithms to be efficient and to obtain ideal results. Thus, our goal is to improve predictive performance by designing new techniques that allow supervised learning algorithms to correctly infer interactions from protein sequence data and obtain ideal results. In this thesis, we first proposed a feature extraction technique noted BP (Bigram-Physicochemical). This technique allows to extract bigram features from protein sequences for efficient machine learning. A bigram is a set of two successive letters (here the amino acids) in a text document (here the protein sequence). For a given protein, BP first computes a matrix from information about the physicochemical properties of the protein. This matrix can be obtained either from a distance (BP1 approach) or from a function (BP approach). Then BP extracts bigram features using the calculated matrix. The BP technique does not produce strictly parsimonious vectors and does not depend on a database like some existing bigram feature extraction techniques. Secondly, we proposed a new approach for selecting optimal values of hyperparameters of a machine learning model noted SVOH. Hyperparameters are the influential parameters of the machine learning model. Generally, the grid search technique is combined with the k-fold cross-validation technique to find the optimal values of hyperparameters. Contrary to the literature which fixes the value of the number k, which corresponds to the number of subsets of the sample, to 5 or 10 on a priori bases, SVOH does a learning of the number k to determine an optimal value of the number of subsets. The developed approach thus allows to rigorously search for the optimal values of hyperparameters on an adjusted number k of subsets. The techniques described in the thesis, combined with a support vector machine (SVM) method, and thus forming the SVM-BP and SVM-BP1 tools, are tested and validated on three different real PPI datasets: human HPRD PPI, yeast Saccharomyces cerevisiae PPI and Helicobacter pylori PPI. The results obtained after some comparative experiments showed that these tools and particularly the SVM-BP tool, obtained superior performances on the three different PPI datasets in the accuracy, precision, and sensitivity metrics. We can say that the SVM-BP and SVM-BP1 tools improve well the predictive performance of PPI and thus are a real help for biologists in protein-protein interaction identification and drug discovery., La protéine est un composant essentiel de la cellule biologique. Les différentes interactions chimiques entre les protéines appelées interactions protéine-protéine (IPP) sont liées à certaines maladies ainsi qu’à certaines thérapies. De ce fait, leur identification a des implications importantes pour plusieurs processus parmi lesquels la prévention des maladies, l’annotation fonctionnelle et la conception de médicaments. De nombreuses IPP ont été détectées par des expériences biologiques au cours des dernières décennies, mais beaucoup d’entre elles restent encore non découvertes. En outre, ces expériences biologiques sont limitées en raison des contraintes de temps et de coûts. Par conséquent, le développement d’outils informatiques est vivement recommandé. Cela pourrait permettre d'accélérer la découverte de médicaments en réduisant les expériences biologiques lentes et coûteuses à l’aide de simulations informatiques plus rapides et moins coûteuses, et d'annoter la fonction des protéines à partir des séquences de protéines. Compte tenu de nombreuses IPP détectées expérimentalement et dont les informations sont disponibles dans des banques de données sur les protéines, plusieurs outils d’inférence et d’apprentissage automatique (machine learning) ont été proposés pour prédire les IPP. La conception de tels outils passe par deux étapes : l’extraction de caractéristiques (descripteurs) à partir des informations de la séquence et la classification des interactions à l’aide d’un algorithme d’apprentissage supervisé. Cependant, les caractéristiques extraites par les techniques existantes ne permettent pas aux algorithmes d’apprentissage supervisé d’être efficaces et d’obtenir des résultats idéaux. Ainsi, notre objectif est d'améliorer les performances prédictives par la conception de nouvelles techniques complémentaires permettant aux algorithmes d’apprentissage supervisé d’inférer correctement les interactions à partir des données de séquences de protéines et d’obtenir des résultats idéaux. Dans cette thèse, nous avons premièrement proposé une technique d’extraction de caractéristiques notée BP (Bigram-Physicochemical). Cette technique permet d’extraire des caractéristiques bigrammes à partir des séquences de protéines pour un apprentissage automatique efficace. Un bigramme est un ensemble de deux lettres (ici les acides aminés) successives dans un document texte (ici la séquence de protéine). Pour une protéine donnée, BP calcule d’abord une matrice à partir d’informations de propriétés physicochimiques de la protéine. Cette matrice peut être obtenue soit à partir d’une distance (approche BP1) ou soit à partir d’une fonction (approche BP). Ensuite, BP extrait des caractéristiques bigrammes en se servant de la matrice calculée. La technique BP ne produit pas de vecteurs strictement parcimonieux et ne dépend pas d’une base de données comme certaines techniques d’extraction de caractéristiques bigrammes existantes. Deuxièmement nous avons proposé une nouvelle approche de sélection de valeurs optimales d’hyperparamètres d’un modèle d’apprentissage automatique notée SVOH. Les hyperparamètres sont les paramètres influents du modèle d’apprentissage automatique. Généralement, la technique de la recherche sur gille (Grid search) est combinée à la technique de validation croisée k-fois pour la recherche des valeurs optimales d’hyperparamètres. Contrairement à la littérature qui fixe la valeur du nombre k, correspondant en fait au nombre de sous-ensembles de l’échantillon, à 5 ou 10 sur des bases a priori, SVOH fait un apprentissage du nombre k afin de déterminer une valeur optimale du nombre de sous-ensembles. L’approche développée permet ainsi de rechercher rigoureusement sur un nombre k ajusté de sous-ensembles de valeurs optimales d’hyperparamètres. Les techniques décrites au sein de la thèse, combinées à une méthode des machines à vecteurs de supports (SVM) et formant ainsi les outils SVM-BP et SVM-BP1, sont testées et validées sur trois différents ensembles de données IPP réelles : les IPP humaines HPRD, les IPP de la levure Saccharomyces cerevisiae et les IPP de la bactérie Helicobacter pylori. Les résultats obtenus après certaines expériences comparatives ont montré que ces outils et particulièrement l’outil SVM-BP, ont obtenu des performances supérieures sur les trois différents ensembles de données IPP dans les métriques justesse, précision et sensibilité. Nous pouvons dire que les outils SVM-BP et SVM-BP1 améliorent bien les performances prédictives des IPP et constituent ainsi une véritable aide pour les biologistes dans l’identification des interactions protéine-protéine et la recherche médicamenteuse.

Computational identification of human-virus protein-protein interactions (PHIs) is a worthwhile step towards understanding infection mechanisms. Analysis of the PHI networks is important for the determination of path-ogenic diseases. Prediction of these interactions is a popular problem since experimental detection of PHIs is both time-consuming and expensive. The available methods use biological features like amino acid sequences, molecular structure, or biological activities for prediction. Recent studies show that the topological properties of proteins in protein-protein interaction (PPI) networks increase the performance of the predictions. The basic network projections, random-walk-based models, or graph neural networks are used for generating topologically enriched (hybrid) protein embeddings. In this study, we propose a three-stage machine learning pipeline that generates and uses hybrid embeddings for PHI prediction. In the first stage, numerical features are extracted from the amino acid sequences using the Doc2Vec and Byte Pair Encoding method. The amino acid embeddings are used as node features while training a modified GraphSAGE model, which is an improved version of the graph convolutional network. Lastly, the hybrid protein embeddings are used for training a binary interaction classifier model that predicts whether there is an interaction between the given two proteins or not. The proposed method is evaluated with comprehensive experiments to test its functionality and compare it with the state-of-art methods. The experimental results on the benchmark dataset prove the efficiency of the proposed model by having a 3–23% better area under curve (AUC) score than its competitors. Publisher's Version Q2 WOS:000858873400002 PMID: 36037723

Background In recent years, biomedical ontologies have become important for describing existing biological knowledge in the form of knowledge graphs. Data mining approaches that work with knowledge graphs have been proposed, but they are based on vector representations that do not capture the full underlying semantics. An alternative is to use machine learning approaches that explore semantic similarity. However, since ontologies can model multiple perspectives, semantic similarity computations for a given learning task need to be fine-tuned to account for this. Obtaining the best combination of semantic similarity aspects for each learning task is not trivial and typically depends on expert knowledge. Results We have developed a novel approach, evoKGsim, that applies Genetic Programming over a set of semantic similarity features, each based on a semantic aspect of the data, to obtain the best combination for a given supervised learning task. The approach was evaluated on several benchmark datasets for protein-protein interaction prediction using the Gene Ontology as the knowledge graph to support semantic similarity, and it outperformed competing strategies, including manually selected combinations of semantic aspects emulating expert knowledge. evoKGsim was also able to learn species-agnostic models with different combinations of species for training and testing, effectively addressing the limitations of predicting protein-protein interactions for species with fewer known interactions. Conclusions evoKGsim can overcome one of the limitations in knowledge graph-based semantic similarity applications: the need to expertly select which aspects should be taken into account for a given application. Applying this methodology to protein-protein interaction prediction proved successful, paving the way to broader applications.

Heterosis is the phenomenon in which hybrid progeny exhibits superior traits in comparison with those of their parents. Genomic variations between the two parental genomes may generate epistasis interactions, which is one of the genetic hypotheses explaining heterosis. We postulate that protein-protein interactions specific to F1 hybrids (F1 -specific PPIs) may occur when two parental genomes combine, as the proteome of each parent may supply novel interacting partners. To test our assumption, an inter-subspecies hybrid interactome was simulated by in silico PPI prediction between rice japonica (cultivar Nipponbare) and indica (cultivar 9311). Four-thousand, six-hundred and twelve F1 -specific PPIs accounting for 20.5% of total PPIs in the hybrid interactome were found. Genes participating in F1 -specific PPIs tend to encode metabolic enzymes and are generally localized in genomic regions harboring metabolic gene clusters. To test the genetic effect of F1 -specific PPIs in heterosis, genomic selection analysis was performed for trait prediction with additive, dominant and epistatic effects separately considered in the model. We found that the removal of single nucleotide polymorphisms associated with F1 -specific PPIs reduced prediction accuracy when epistatic effects were considered in the model, but no significant changes were observed when additive or dominant effects were considered. In summary, genomic divergence widely dispersed between japonica and indica rice may generate F1 -specific PPIs, part of which may accumulatively contribute to heterosis according to our computational analysis. These candidate F1 -specific PPIs, especially for those involved in metabolic biosynthesis pathways, are worthy of experimental validation when large-scale protein interactome datasets are generated in hybrid rice in the future.

MotivationSequence-based protein–protein interaction (PPI) prediction represents a fundamental computational biology problem. To address this problem, extensive research efforts have been made to extract predefined features from the sequences. Based on these features, statistical algorithms are learned to classify the PPIs. However, such explicit features are usually costly to extract, and typically have limited coverage on the PPI information.ResultsWe present an end-to-end framework, PIPR (Protein–Protein Interaction Prediction Based on Siamese Residual RCNN), for PPI predictions using only the protein sequences. PIPR incorporates a deep residual recurrent convolutional neural network in the Siamese architecture, which leverages both robust local features and contextualized information, which are significant for capturing the mutual influence of proteins sequences. PIPR relieves the data pre-processing efforts that are required by other systems, and generalizes well to different application scenarios. Experimental evaluations show that PIPR outperforms various state-of-the-art systems on the binary PPI prediction problem. Moreover, it shows a promising performance on more challenging problems of interaction type prediction and binding affinity estimation, where existing approaches fall short.Availability and implementationThe implementation is available at https://github.com/muhaochen/seq_ppi.git.Supplementary informationSupplementary data are available at Bioinformatics online.

Motivation To complement experimental efforts, machine learning-based computational methods are playing an increasingly important role to predict human–virus protein–protein interactions (PPIs). Furthermore, transfer learning can effectively apply prior knowledge obtained from a large source dataset/task to a small target dataset/task, improving prediction performance. Results To predict interactions between human and viral proteins, we combine evolutionary sequence profile features with a Siamese convolutional neural network (CNN) architecture and a multi-layer perceptron. Our architecture outperforms various feature encodings-based machine learning and state-of-the-art prediction methods. As our main contribution, we introduce two transfer learning methods (i.e. ‘frozen’ type and ‘fine-tuning’ type) that reliably predict interactions in a target human–virus domain based on training in a source human–virus domain, by retraining CNN layers. Finally, we utilize the ‘frozen’ type transfer learning approach to predict human–SARS-CoV-2 PPIs, indicating that our predictions are topologically and functionally similar to experimentally known interactions. Availability and implementation The source codes and datasets are available at https://github.com/XiaodiYangCAU/TransPPI/. Supplementary information Supplementary data are available at Bioinformatics online.

Proteins are biological molecules that play a critical role in vital biological processes. Interactions between pathogen proteins and host proteins form pathogen-host interaction (PHI) networks. These bipartite interaction networks have great importance in determining which vital activities are affected by the pathogen and the diseases it may cause. Experimental detection of the protein interactions in wet labs is both timeconsuming and costly. The limited number of experimentally detectable interactions and overlook of some potential interactions lead to development of computational methods. In this study, a graph convolution network (GCN) based method is presented that enables to predict protein-protein interactions in PHI networks. The unsupervised trained GCN model (GraphSAGE) uses amino acid sequences as node features as well as the topological information. This is the first study to the best of our knowledge which provides GCN models to do protein-protein interaction prediction in PHI networks. The experimental results show that the developed model performs 10% better than the state-of-art algorithms on the benchmark PHI dataset and it predicts interactions with 96% accuracy.

Proteinler yaşamsal faaliyetlerin gerçekleşmesinde kritik rol oynayan biyolojik moleküllerdir. Konak canlı proteinleri ile patojen proteinleri arasındaki etkileşimler patojenkonak etkileşim (PHI) ağlarını oluşturmaktadır. Bu iki parçalı etkileşim ağları patojenin hangi yaşamsal faaliyetleri etkilediğini belirlemede ve dolayısıyla sebep olabileceği hastalıkların tespitinde büyük öneme sahiptir. Proteinler arası etkileşimlerin laboratuvar ortamında tespiti hem zaman alıcı hem de maliyetlidir. Deneysel olarak saptanabilen etkileşim sayısının kısıtlı olması ve bazı etkileşimlerin gözden kaçması hesaplamalı tahmin yöntemlerinin geliştirilmesine önayak olmaktadır. Bu çalışmada PHI ağlarında protein etkileşim tahmini yapmayı sağlayan çizge evrişim ağı (GCN) tabanlı bir yöntem sunulmaktadır. Gözetimsiz olarak eğitilen GCN modeli (GraphSAGE) topolojik bilginin yanı sıra temel öznitelik olarak amino asit dizilimlerini kullanmaktadır. Bu çalışma bildiğimiz kadarıyla PHI ağlarında GCN tabanlı etkileşim tahmini sağlayan ilk çalışmadır. Deneysel sonuçlar geliştirilen modelin kıyaslama için kullanılan PHI veri seti üzerinde yüksek performanslı algoritmalardan %10 daha iyi performans göstererek %96 oranında doğrulukla etkileşim tahmini yaptığını göstermektedir. Proteins are biological molecules that play a critical role in vital biological processes. Interactions between pathogen proteins and host proteins form pathogen-host interaction (PHI) networks. These bipartite interaction networks have great importance in determining which vital activities are affected by the pathogen and the diseases it may cause. Experimental detection of the protein interactions in wet labs is both timeconsuming and costly. The limited number of experimentally detectable interactions and overlook of some potential interactions lead to development of computational methods. In this study, a graph convolution network (GCN) based method is presented that enables to predict protein-protein interactions in PHI networks. The unsupervised trained GCN model (GraphSAGE) uses amino acid sequences as node features as well as the topological information. This is the first study to the best of our knowledge which provides GCN models to do protein-protein interaction prediction in PHI networks. The experimental results show that the developed model performs 10% better than the state-of-art algorithms on the benchmark PHI dataset and it predicts interactions with 96% accuracy. Publisher's Version

The perturbations of protein-protein interactions (PPIs) were found to be the main cause of cancer. Previous PPI prediction methods which were trained with non-disease general PPI data were not compatible to map the PPI network in cancer. Therefore, we established a novel cancer specific PPI prediction method dubbed NECARE, which was based on relational graph convolutional network (R-GCN) with knowledge-based features. It achieved the best performance with a Matthews correlation coefficient (MCC) = 0.84±0.03 and an F1 = 91±2% compared with other methods. With NECARE, we mapped the cancer interactome atlas and revealed that the perturbations of PPIs were enriched on 1362 genes, which were named cancer hub genes. Those genes were found to over-represent with mutations occurring at protein-macromolecules binding interfaces. Furthermore, over 56% of cancer treatment-related genes belonged to hub genes and they were significantly related to the prognosis of 32 types of cancers. Finally, by coimmunoprecipitation, we confirmed that the NECARE prediction method was highly reliable with a 90% accuracy. Overall, we provided the novel network-based cancer protein-protein interaction prediction method and mapped the perturbation of cancer interactome. NECARE is available at: https://github.com/JiajunQiu/NECARE., Author summary Protein-protein interaction (PPI) network is the biological foundation for the normal function of cells, while the perturbation of this network can result in the pathological state, such as cancer. Notably, the perturbation of PPI network in cancer not only involves in the destruction of old PPI, but also the reconstruction of new PPIs. However, due to the limit of tools, instead of the real physical interaction between proteins, previous cancer network researches only focus on the co-expression relationships. Now, with the development of computational biology, we established a novel cancer specific physical PPI prediction method dubbed NECARE, which was based on relational graph convolutional network (R-GCN) with knowledge-based features. It can infer the PPI in cancer from a general network. And we reveal the cancer PPI interactome by doing high-throughput analysis with NECARE. Also, many cancer hub genes were identified during the analysis, which were enriched for cancer network perturbations. Future studies can benefit from both our method itself and the results of our analysis.

Predicting protein-protein interactions (PPIs) by learning informative representations from amino acid sequences is a challenging yet important problem in biology. Although various deep learning models in Siamese architecture have been proposed to model PPIs from sequences, these methods are computationally expensive for a large number of PPIs due to the pairwise encoding process. Furthermore, these methods are difficult to interpret because of non-intuitive mappings from protein sequences to their sequence representation. To address these challenges, we present a novel deep framework to model and predict PPIs from sequence alone. Our model incorporates a bidirectional gated recurrent unit to learn sequence representations by leveraging contextualized and sequential information from sequences. We further employ a sparse regularization to model long-range dependencies between amino acids and to select important amino acids (protein motifs), thus enhancing interpretability. Besides, the novel design of the encoding process makes our model computationally efficient and scalable to an increasing number of interactions. Experimental results on up-to-date interaction datasets demonstrate that our model achieves superior performance compared to other state-of-the-art methods. Literature-based case studies illustrate the ability of our model to provide biological insights to interpret the predictions.

Protein-protein interaction (PPI) plays an important role in regulating cells and signals. PPI deregulation will lead to many diseases, including pernicious anemia or cancer. Despite the ongoing efforts of the bioassay group, continued data incompleteness limits our ability to understand the molecular roots of human disease. Therefore, it is urgent to develop a computational method that accurately and quickly detects PPIs. In this paper, a highly efficient model is proposed for predicting PPIs through heterogeneous network by combining local feature with global feature. Heterogeneous network is collected from several valuable datasets, containing five types of nodes and nine interactions among them. Local feature is extracted from protein sequence by k-mer method. Global feature is extracted from heterogeneous network by LINE (Large-scale Information Network Embedding). Protein representation is obtained from local feature and global feature by concatenation. Finally, random forest is trained to classify and predict potential protein pairs. The proposed method is demonstrated on STRING dataset and achieved an average 86.55% prediction accuracy with 0.9308 AUC. Extensive contrast experiments are performed with different protein representations and different classifiers. Obtained experiment results illustrate that proposed method is economically viable, which provides a new perspective for future research.

Numerous computational methods have been proposed to predict protein-protein interactions, none of which however, considers the original DNA loci of the interacting proteins in the perspective of 3D genome. Here, we retrospect the DNA origins of the interacting proteins from the perspective of 3D genome and found that by incorporating 3D genome information into existing PPI prediction methods, i.e., Auto Covariance (AC) and Conjoint triad (CT) coding based support vector machine (SVM) and ensemble extreme learning machine (EELM), the predictions can be further improved in terms of accuracy and area under ROC curve (AUC). Combining our previous discoveries, we conjecture this is due to that co-localized DNA elements may lead to increased probability of the co-localization of their downstream RNA elements and protein elements.

Background Protein–protein interactions (PPIs) are of great importance in cellular systems of organisms, since they are the basis of cellular structure and function and many essential cellular processes are related to that. Most proteins perform their functions by interacting with other proteins, so predicting PPIs accurately is crucial for understanding cell physiology. Results Recently, graph convolutional networks (GCNs) have been proposed to capture the graph structure information and generate representations for nodes in the graph. In our paper, we use GCNs to learn the position information of proteins in the PPIs networks graph, which can reflect the properties of proteins to some extent. Combining amino acid sequence information and position information makes a stronger representation for protein, which improves the accuracy of PPIs prediction. Conclusion In previous research methods, most of them only used protein amino acid sequence as input information to make predictions, without considering the structural information of PPIs networks graph. We first time combine amino acid sequence information and position information to make representations for proteins. The experimental results indicate that our method has strong competitiveness compared with several sequence-based methods.

Motivation Modern bioinformatics is facing increasingly complex problems to solve, and we are indeed rapidly approaching an era in which the ability to seamlessly integrate heterogeneous sources of information will be crucial for the scientific progress. Here, we present a novel non-linear data fusion framework that generalizes the conventional matrix factorization paradigm allowing inference over arbitrary entity-relation graphs, and we applied it to the prediction of protein–protein interactions (PPIs). Improving our knowledge of PPI networks at the proteome scale is indeed crucial to understand protein function, physiological and disease states and cell life in general. Results We devised three data fusion-based models for the proteome-level prediction of PPIs, and we show that our method outperforms state of the art approaches on common benchmarks. Moreover, we investigate its predictions on newly published PPIs, showing that this new data has a clear shift in its underlying distributions and we thus train and test our models on this extended dataset. Supplementary information Supplementary data are available at Bioinformatics online.

The scientific community is rapidly generating protein sequence information, but only a fraction of these proteins can be experimentally characterized. While promising deep learning approaches for protein prediction tasks have emerged, they have computational limitations or are designed to solve a specific task. We present a Transformer neural network that pre-trains task-agnostic sequence representations. This model is fine-tuned to solve two different protein prediction tasks: protein family classification and protein interaction prediction. Our method is comparable to existing state-of-the art approaches for protein family classification, while being much more general than other architectures. Further, our method outperforms all other approaches for protein interaction prediction. These results offer a promising framework for fine-tuning the pre-trained sequence representations for other protein prediction tasks.

Motivation Predicting residue–residue contacts between interacting proteins is an important problem in bioinformatics. The growing wealth of sequence data can be used to infer these contacts through correlated mutation analysis on multiple sequence alignments of interacting homologs of the proteins of interest. This requires correct identification of pairs of interacting proteins for many species, in order to avoid introducing noise (i.e. non-interacting sequences) in the analysis that will decrease predictive performance. Results We have designed Ouroboros, a novel algorithm to reduce such noise in intermolecular contact prediction. Our method iterates between weighting proteins according to how likely they are to interact based on the correlated mutations signal, and predicting correlated mutations based on the weighted sequence alignment. We show that this approach accurately discriminates between protein interaction versus non-interaction and simultaneously improves the prediction of intermolecular contact residues compared to a naive application of correlated mutation analysis. This requires no training labels concerning interactions or contacts. Furthermore, the method relaxes the assumption of one-to-one interaction of previous approaches, allowing for the study of many-to-many interactions. Availability and implementation Source code and test data are available at www.bif.wur.nl/. Supplementary information Supplementary data are available at Bioinformatics online.

The unicellular cyanobacterium Cyanothece ATCC 51142 is capable of oxygenic photosynthesis and biological N(2)-fixation (BNF), a process highly sensitive to oxygen. Previous work has focused on determining protein expression levels under different growth conditions. A major gap of our knowledge is an understanding how these expressed proteins are assembled into complexes and organized into metabolic pathways, an area that has not been thoroughly investigated. Here, we combined size-exclusion chromatography (SEC) with label-free quantitative mass spectrometry (MS) and bioinformatics to characterize many protein complexes in from Cyanothece 51142 cells grown under 12-h light-dark cycle. We identified 1386 proteins in duplicate biological replicates, and 64% of those proteins were identified as putative complexes. Pair-wise computational prediction of protein-protein interaction (PPI) identified 74,822 putative interactions, of which 2337 interactions were highly correlated with published protein co-expressions. Many sequential glycolytic and TCA cycle enzymes were identified as putative complexes. We also identified many membrane complexes that contain cytoplasmic domains. Subunits of NDH-1 complex eluted in a fraction with an approximate mass of ~669 kDa, and subunits composition revealed co-existence of distinct forms of NDH-1 complex subunits responsible for respiration, electron flow and CO(2) uptake. The complex form of the phycocyanin beta subunit was non-phosphorylated and the monomer form was phosphorylated at Ser20, suggesting phosphorylation-dependent de-oligomerization of the phycocyanin beta subunit. This study provides an analytical platform for future studies to reveal how these complexes assemble and disassemble as a function of diurnal and circadian rhythms.

All protein-protein interaction (PPI) predictors require the determination of an operational decision threshold when differentiating positive PPIs from negatives. Historically, a single global threshold, typically optimized via cross-validation testing, is applied to all protein pairs. However, we here use data visualization techniques to show that no single decision threshold is suitable for all protein pairs, given the inherent diversity of protein interaction profiles. The recent development of high throughput PPI predictors has enabled the comprehensive scoring of all possible protein-protein pairs. This, in turn, has given rise to context, enabling us now to evaluate a PPI within the context of all possible predictions. Leveraging this context, we introduce a novel modeling framework called Reciprocal Perspective (RP), which estimates a localized threshold on a per-protein basis using several rank order metrics. By considering a putative PPI from the perspective of each of the proteins within the pair, RP rescores the predicted PPI and applies a cascaded Random Forest classifier leading to improvements in recall and precision. We here validate RP using two state-of-the-art PPI predictors, the Protein-protein Interaction Prediction Engine and the Scoring PRotein INTeractions methods, over five organisms: Homo sapiens, Saccharomyces cerevisiae, Arabidopsis thaliana, Caenorhabditis elegans, and Mus musculus. Results demonstrate the application of a post hoc RP rescoring layer significantly improves classification (p