Bayesian network model for identification of pathways by integrating protein interaction with genetic interaction data

Inference of new and useful hypotheses from heterogeneous sources of genome-scale experimental data requires new computational methods that can integrate different types of data. Gene expression and genetic interaction data are two most informative data types, each allowing the identification of genes at different levels of cellular regulatory network hierarchy. We present an integrative data analysis approach, which, rather than correlating the findings from the two data sets, uses each type of data independently to identify the components of molecular pathways and combines them into a single directed network. Our computational genomics approach is based on a set of inference rules traditionally used for reasoning on genetic experiments, which we have formalized and implemented in a software tool. The approach uses chemogenetic interaction and expression data to infer the type of relation between the chemical substance (perturber) and a transcription factor by using previous knowledge on the set of genes whose expression the transcription factor in question regulates. We have used the proposed approach to successfully infer the models for the action of the drug rapamycin and of a DNA damaging agent on their molecular targets and pathways in yeast cells. The developed method is available as a web-based tool at http://www.ailab.si/perturbagen.

Recent studies suggest a protective role of diabetes in the development of aneurysm, but the biological mechanisms behind this are still unknown. This type of association is not present in the case of diabetes and atherosclerosis despite similar risk factors for aneurysm and atherosclerosis. We postulate the existence of genes that disrupt the pathways needed for the onset of aneurysm in the presence of diabetes. Motivated by the significance of genetic interactions in understanding disease-disease associations, we tackle this problem by integrating protein-protein interaction and genetic interaction data, i.e., we examine the biological pathways related to the three diseases that contain genes involved in the following genetic interactions: one gene in a genetic interaction is part of a diabetes pathway, the other gene is part of an aneurysm, or an atherosclerosis pathway. We create a protein-protein interaction sub-network that contains disease pathways described above. We then use a "brokerage" measure - a topological measure that identifies proteins in this sub-network whose removal severely affects the interconnectedness of their neighbourhood, enabling such proteins to disrupt the pathway they are in. We identify a set of proteins with high brokerage values and find this set to be enriched in biological functions, including cell-matrix adhesion, which facilitates mechanisms that have already been suggested as possible causes of diabetes-aneurysm association. We further narrow down our set to 16 proteins that are involved in an aneurysm or an atherosclerosis pathway and are encoded by genes participating in genetic interactions with a gene in a diabetes pathway. This set is enriched in kinases and phosphorylation processes, with two pleiotropic kinases that are involved in both aneurysm and atherosclerosis pathways. Kinases can turn on or off proteins, explaining how functional changes of such proteins could result in the disruption of pathways. So if in an aneurysm-related pathway a gene is turned off, the onset of the disease could be prevented. However, mutations of pleiotropic genes could have effects only on one of the traits, which explains why pleiotropic kinases that are involved in both aneurysm and atherosclerosis pathways could disrupt aneurysm pathways explaining the reduced risk of aneurysm in diabetes patients, but not affect the atherosclerosis pathways.

The Biological General Repository for Interaction Datasets (BioGRID) is a freely available public database that provides the biological and biomedical research communities with curated protein and genetic interaction data. Structured experimental evidence codes, an intuitive search interface, and visualization tools enable the discovery of individual gene, protein, or biological network function. BioGRID houses interaction data for the major model organism species--including yeast, nematode, fly, zebrafish, mouse, and human--with particular emphasis on the budding yeast Saccharomyces cerevisiae and the fission yeast Schizosaccharomyces pombe as pioneer eukaryotic models for network biology. BioGRID has achieved comprehensive curation coverage of the entire literature for these two major yeast models, which is actively maintained through monthly curation updates. As of September 2015, BioGRID houses approximately 335,400 biological interactions for budding yeast and approximately 67,800 interactions for fission yeast. BioGRID also supports an integrated posttranslational modification (PTM) viewer that incorporates more than 20,100 yeast phosphorylation sites curated through its sister database, the PhosphoGRID.

BackgroundSynthetic lethality experiments identify pairs of genes with complementary function. More direct functional associations (for example greater probability of membership in a single protein complex) may be inferred between genes that share synthetic lethal interaction partners than genes that are directly synthetic lethal. Probabilistic algorithms that identify gene modules based on motif discovery are highly appropriate for the analysis of synthetic lethal genetic interaction data and have great potential in integrative analysis of heterogeneous datasets.ResultsWe have developed Genetic Interaction Motif Finding (GIMF), an algorithm for unsupervised motif discovery from synthetic lethal interaction data. Interaction motifs are characterized by position weight matrices and optimized through expectation maximization. Given a seed gene, GIMF performs a nonlinear transform on the input genetic interaction data and automatically assigns genes to the motif or non-motif category. We demonstrate the capacity to extract known and novel pathways for Saccharomyces cerevisiae (budding yeast). Annotations suggested for several uncharacterized genes are supported by recent experimental evidence. GIMF is efficient in computation, requires no training and automatically down-weights promiscuous genes with high degrees.ConclusionGIMF effectively identifies pathways from synthetic lethality data with several unique features. It is mostly suitable for building gene modules around seed genes. Optimal choice of one single model parameter allows construction of gene networks with different levels of confidence. The impact of hub genes the generic probabilistic framework of GIMF may be used to group other types of biological entities such as proteins based on stochastic motifs. Analysis of the strongest motifs discovered by the algorithm indicates that synthetic lethal interactions are depleted between genes within a motif, suggesting that synthetic lethality occurs between-pathway rather than within-pathway.

The first computational interaction network built from Drosophila melanogaster protein-protein and genetic interaction data allows the functional annotation of orphan genes and reveals clusters of functionally-related genes.

An important goal of systems biology is to understand and identify mechanisms of the human body system. Genes play functional roles in the context of complex pathways. Analysis of genes as networks is therefore important to understand whole system mechanisms. Biological activities are governed by various signaling networks. The advent of high-throughput technologies has made it possible to obtain biological information on a genome-wide scale. Genetic interactions have been identified from high-throughput data such as microarray data using Bayesian networks. In this paper, we infer the disease-specific gene interaction network using a Bayesian network, which is robust to noise in the data. We apply a genetic algorithm to learn the Bayesian network. We use heterogeneous data, including microarray, protein-protein interaction (PPI), and HumanNET data to learn and score the network. We also exploit single nucleotide polymorphism (SNP) data to infer disease-specific genetic interaction network. We included SNPs as this data may help detect weak signals related to genetic variation. In this paper, we reconstruct interactions between pathway genes using our method. We confirm that our method has statistically significant reconstruction power by applying it to Type II diabetes data. Importantly, using Alzheimer disease data, we infer an unreported interaction between a SNP and a disease-related gene.

Computational prediction of protein interaction motifs from integrated protein sequence, structure and interaction data

Protein-protein interaction (PPI) data is essential to elucidate the complex molecular relationships in living systems, and thus understand the biological functions at cellular and systems levels. The complete map of PPIs that can occur in a living organism is called the interactome. For animals, PPI data is stored in multiple databases (e.g., BioGRID, CCSB, DroID, FlyBase, HIPPIE, HitPredict, HomoMINT, INstruct, Interactome3D, mentha, MINT, and PINA2) with different formats. This makes PPI comparisons difficult to perform, especially between species, since orthologous proteins may have different names. Moreover, there is only a partial overlap between databases, even when considering a single species. The EvoPPI ( http://evoppi.i3s.up.pt ) web application presented in this paper allows comparison of data from the different databases at the species level, or between species using a BLAST approach. We show its usefulness by performing a comparative study of the interactome of the nine polyglutamine (polyQ) disease proteins, namely androgen receptor (AR), atrophin-1 (ATN1), ataxin 1 (ATXN1), ataxin 2 (ATXN2), ataxin 3 (ATXN3), ataxin 7 (ATXN7), calcium voltage-gated channel subunit alpha1 A (CACNA1A), Huntingtin (HTT), and TATA-binding protein (TBP). Here we show that none of the human interactors of these proteins is common to all nine interactomes. Only 15 proteins are common to at least 4 of these polyQ disease proteins, and 40% of these are involved in ubiquitin protein ligase-binding function. The results obtained in this study suggest that polyQ disease proteins are involved in different functional networks. Comparisons with Mus musculus PPIs are also made for AR and TBP, using EvoPPI BLAST search approach (a unique feature of EvoPPI), with the goal of understanding why there is a significant excess of common interactors for these proteins in humans.

Protein complexes are a cornerstone of many biological processes. Protein-protein interaction (PPI) data enable a number of computational methods for predicting protein complexes. However, the insufficiency of the PPI data significantly lowers the accuracy of computational methods. In the current work, the authors develop a novel method named clustering based on multiple biological information (CMBI) to discover protein complexes via the integration of multiple biological resources including gene expression profiles, essential protein information and PPI data. First, CMBI defines the functional similarity of each pair of interacting proteins based on the edge-clustering coefficient and the Pearson correlation coefficient. Second, CMBI selects essential proteins as seeds to build the protein complexes. A redundancy-filtering procedure is performed to eliminate redundant complexes. In addition to the essential proteins, CMBI also uses other proteins as seeds to expand protein complexes. To check the performance of CMBI, the authors compare the complexes discovered by CMBI with the ones found by other techniques by matching the predicted complexes against the reference complexes. The authors use subsequently GO::TermFinder to analyse the complexes predicted by various methods. Finally, the effect of parameters T and R is investigated. The results from GO functional enrichment and matching analyses show that CMBI performs significantly better than the state-of-the-art methods.

Motivations. Available biomolecular annotations are very valuable, but dispersed and far from being complete. High quality integration of scattered annotation data and reliable identification of new annotations can greatly support unveiling new biomedical knowledge. Biomolecular interactions are one of the main objectives of biomolecular studies, due to the understanding that biological processes are mainly driven by interactions among biomolecular entities, such as proteins and DNA. New powerful high-throughput experimental techniques are providing numerous protein-protein interaction data; they are being collected, together with computational results, in several different databases, which include IntAct, BioGrid, BIND, DIP, HPRD and MINT. Yet, they generally do not include phenotypic or even functional or structural information about the interactors, which in many cases are available in other databases. In particular, no information is available about the association of protein-protein interactions with genetic disorders. This creates the need to integrate the sparsely available data in order to enrich the identified interactions with additional evidence, support their biological interpretation and identify their involvement in inherited pathologies. In addition, integration of gene and protein annotations and protein-protein interaction data provides the base to infer new gene annotations and interaction networks. Methods. We developed an automatic association inference method, based on the transitive closure concept, and applied it on the data from several distributed sources integrated in our Genomic and Proteomic Data Warehouse (GPDW). In particular, by leveraging protein-protein interaction data, provided by the IntAct and MINT databases, and protein encoding gene data form the Entrez Gene database, we inferred gene interaction networks. In addition, by taking advantage of genetic disorder and phenotype data provided by the OMIM database, we inferred associations between proteins and genetic disorders and their phenotypes. Then, in order to identify genetic disorders possibly associated with protein-protein interactions, we looked for those interacting proteins that resulted associated with the same genetic disorder. Results. Our GPDW currently includes 46,154 human protein-protein interactions (out of the 254,048 protein-protein interactions contained), which involve 12,178 different human proteins (out of the 326,766 human proteins in the GPDW) that are encoded by 11,232 different human genes. By applying the above described method, we identified 1,130 gene networks and found 1,136 human protein-protein interactions associated with 628 genetic disorders (6% of all genetic disorders in the GPDW), which are related to 86 clinical synopses (87% of all clinical synopses in the GPDW) and 3,481 phenotypes (10% of the total phenotypes in the GPDW). Among others, we found four interacting proteins (AHSA1_HUMAN, CFTR_HUMAN, DERL1_HUMAN, and RNF5_HUMAN) whose encoding genes are known to be associated with cystic fibrosis, as well as other 43 genes. Mutations of the CFTR human gene are known to be directly involved in different grades and manifestations of cystic fibrosis. The found associations of the AHSA1_HUMAN, CFTR_HUMAN, DERL1_HUMAN, and RNF5_HUMAN interacting proteins with cystic fibrosis could suggest that some types of this multi-variant disorder may be associated with defects in the interactions between these proteins. Possibly, different CFTR_HUMAN protein mutations could alter its functional interaction with one or more of the AHSA1_HUMAN, DERL1_HUMAN and RNF5_HUMAN proteins. If this would be proven, such finding would also suggest, as a possible disease treatment strategy, the engineering of a synthetic protein interacting with the mutated CFTR_HUMAN protein and similar in function to the one of the AHSA1_HUMAN, DERL1_HUMAN or RNF5_HUMAN proteins whose interactions with the mutated CFTR_HUMAN protein result altered. The above discussed findings demonstrate the importance of the transitive closure based inference method developed, as well as of the data integration approach implemented in the GPDW and the relevance of the comprehensive data there integrated. Availability The GPDW constitutes the backend of a Genomic and Proteomic Knowledge Base (GPKB) that is publicly available at http://www.bioinformatics.dei.polimi.it/GPKB/ through a prototype easy-to-use and efficient Web interface.

Biofilm formation by Candida albicans is a key aspect of its pathobiology and is regulated by an integrated network of transcription factors (Bcr1, Brg1, Efg1, Ndt80, Rob1, and Tec1). To understand the details of how the transcription factors function together to regulate biofilm formation, we used a systematic genetic interaction approach based on generating all possible double heterozygous mutants of the network genes and quantitatively analyzing the genetic interactions between them. Overall, the network is highly susceptible to genetic perturbation with the six network heterozygous mutants all showing alterations in biofilm formation (haploinsufficiency). In addition, many double heterozygous mutants are as severely affected as homozygous deletions. As a result, the network shows properties of a highly interdependent ‘small-world’ network that is highly efficient but not robust. In addition, these genetic interaction data indicate that TEC1 represents a network component whose expression is highly sensitive to small perturbations in the function of other networks TFs. We have also found that expression of ROB1 is dependent on both auto-regulation and cooperative interactions with other network TFs. Finally, the heterozygous NDT80 deletion mutant is hyperfilamentous under both biofilm and hyphae-inducing conditions in a TEC1-dependent manner. Taken together, genetic interaction analysis of this network has provided new insights into the functions of individual TFs as well as into the role of the overall network topology in its function.

BackgroundProtein-protein interaction (PPI) data are widely used to generate network models that aim to describe the relationships between proteins in biological systems. The fidelity and completeness of such networks is primarily limited by the paucity of protein interaction information and by the restriction of most of these data to just a few widely studied experimental organisms. In order to extend the utility of existing PPIs, computational methods can be used that exploit functional conservation between orthologous proteins across taxa to predict putative PPIs or 'interologs'. To date most interolog prediction efforts have been restricted to specific biological domains with fixed underlying data sources and there are no software tools available that provide a generalised framework for 'on-the-fly' interolog prediction.ResultsWe introduce Bio::Homology::InterologWalk, a Perl module to retrieve, prioritise and visualise putative protein-protein interactions through an orthology-walk method. The module uses orthology and experimental interaction data to generate putative PPIs and optionally collates meta-data into an Interaction Prioritisation Index that can be used to help prioritise interologs for further analysis. We show the application of our interolog prediction method to the genomic interactome of the fruit fly, Drosophila melanogaster. We analyse the resulting interaction networks and show that the method proposes new interactome members and interactions that are candidates for future experimental investigation.ConclusionsOur interolog prediction tool employs the Ensembl Perl API and PSICQUIC enabled protein interaction data sources to generate up to date interologs 'on-the-fly'. This represents a significant advance on previous methods for interolog prediction as it allows the use of the latest orthology and protein interaction data for all of the genomes in Ensembl. The module outputs simple text files, making it easy to customise the results by post-processing, allowing the putative PPI datasets to be easily integrated into existing analysis workflows. The Bio::Homology::InterologWalk module, sample scripts and full documentation are freely available from the Comprehensive Perl Archive Network (CPAN) under the GNU Public license.

Protein-protein interaction (PPI) data are often stored in publicly available databases that are often queried by the use of simple query interfaces allowing only key-based queries. Such databases enable the retrieval of one or more proteins that interact with a target protein, using a target protein identifier. Nevertheless, a lot of biological information is available and is spread on different sources and encoded in different ontologies (e.g. Gene Ontology). Annotating existing PPI databases with biological information may result in richer querying interfaces and successively could enable the development of novel algorithms that may use biological information. The OntoPIN project showed the effectiveness of the introduction of a framework for the ontology-based management and querying of Protein-Protein Interaction Data. The OntoPIN framework first merges PPI data with annotations extracted from existing ontologies (e.g. Gene Ontology) and stores annotated data into a database. Then, a semantic-based query interface enables users to query these data by using biological concepts. OntoPIN allows: (a) to extend existing PPI databases by using ontologies, (b) to enable a key-based querying of annotated data, and (c) to offer a novel query interface based on semantic similarity among annotations. This paper presents the current state of the OntoPIN project discussing a case study on human protein interaction data.

BackgroundEssential proteins are distinctly important for an organism’s survival and development and crucial to disease analysis and drug design as well. Large-scale protein-protein interaction (PPI) data sets exist in Saccharomyces cerevisiae, which provides us with a valuable opportunity to predict identify essential proteins from PPI networks. Many network topology-based computational methods have been designed to detect essential proteins. However, these methods are limited by the completeness of available PPI data. To break out of these restraints, some computational methods have been proposed by integrating PPI networks and multi-source biological data. Despite the progress in the research of multiple data fusion, it is still challenging to improve the prediction accuracy of the computational methods.ResultsIn this paper, we design a novel iterative model for essential proteins prediction, named Randomly Walking in the Heterogeneous Network (RWHN). In RWHN, a weighted protein-protein interaction network and a domain-domain association network are constructed according to the original PPI network and the known protein-domain association network, firstly. And then, we establish a new heterogeneous matrix by combining the two constructed networks with the protein-domain association network. Based on the heterogeneous matrix, a transition probability matrix is established by normalized operation. Finally, an improved PageRank algorithm is adopted on the heterogeneous network for essential proteins prediction. In order to eliminate the influence of the false negative, information on orthologous proteins and the subcellular localization information of proteins are integrated to initialize the score vector of proteins. In RWHN, the topology, conservative and functional features of essential proteins are all taken into account in the prediction process. The experimental results show that RWHN obviously exceeds in predicting essential proteins ten other competing methods.ConclusionsWe demonstrated that integrating multi-source data into a heterogeneous network can preserve the complex relationship among multiple biological data and improve the prediction accuracy of essential proteins. RWHN, our proposed method, is effective for the prediction of essential proteins.

Signal flow direction is one of the most important features of the protein-protein interactions in signaling networks. However, almost all the outcomes of current high-throughout techniques for protein-protein interactions mapping are usually supposed to be non-directional. Based on the pairwise interaction domains, here we defined a novel parameter protein interaction directional score and then used it to predict the direction of signal flow between proteins in proteome-wide signaling networks. Using 5-fold cross-validation, our approach obtained a satisfied performance with the accuracy 89.79%, coverage 48.08%, and error ratio 16.91%. As an application, we established an integrated human directional protein interaction network, including 2,237 proteins and 5,530 interactions, and inferred a large amount of novel signaling pathways. Directional protein interaction network was strongly supported by the known signaling pathways literature (with the 87.5% accuracy) and further analyses on the biological annotation, subcellular localization, and network topology property. Thus, this study provided an effective method to define the upstream/downstream relations of interacting protein pairs and a powerful tool to unravel the unknown signaling pathways.