Protein-protein Interaction Datasets Research Articles

Abstract 4929: Dissection of protein-protein interaction-mediated cross-talk pathways in hepatocellular carcinoma

Abstract Multiple dys-regulated cell signaling pathways and extensive cross-talks, which can be depicted via protein-protein interactions (PPIs), among individual signaling pathways, characterize cancer development and progression. However, systematic analysis of the cross-talk events from distinct pathways via PPIs remains fragmentary, and the reliability of PPI datasets, particularly at the cellular level, remains to be vigorously examined. Here, we first build the cross-talk pathway map of heptocellular carcinoma (HCC) through overexpression patterns in HCC among 23 pathways. By detecting endogenous PPIs in two HCC cell lines, Huh7 (low invasive cells) and Mahlavu (high invasive cells), we validate 50 PPI events within pathway and 19 PPI-mediated cross-talk events among 23 pathways in HCC. Not surprisingly, many distinct PPIs are present between Huh7 and Mahlavu, providing the opportunity to characterize tumor invasion. By analyzing topology of cross-talk PPI network, CRKL has a pivotal role in cross-talk pathway. This initial attempt to map the cross-talk pathways in HCC may represent the first step in deciphering hepatocarcinogenesis at the systems level and might reveal a novel strategy for HCC treatment. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr 4929. doi:10.1158/1538-7445.AM2011-4929

QiSampler: evaluation of scoring schemes for high-throughput datasets using a repetitive sampling strategy on gold standards

BackgroundHigh-throughput biological experiments can produce a large amount of data showing little overlap with current knowledge. This may be a problem when evaluating alternative scoring mechanisms for such data according to a gold standard dataset because standard statistical tests may not be appropriate.FindingsTo address this problem we have implemented the QiSampler tool that uses a repetitive sampling strategy to evaluate several scoring schemes or experimental parameters for any type of high-throughput data given a gold standard. We provide two example applications of the tool: selection of the best scoring scheme for a high-throughput protein-protein interaction dataset by comparison to a dataset derived from the literature, and evaluation of functional enrichment in a set of tumour-related differentially expressed genes from a thyroid microarray dataset.ConclusionsQiSampler is implemented as an open source R script and a web server, which can be accessed at http://cbdm.mdc-berlin.de/tools/sampler/.

Triathlon for energy functions: Who is the winner for design of protein–protein interactions?

Computational prediction of stabilizing mutations into monomeric proteins has become an almost ordinary task. Yet, computational stabilization of protein–protein complexes remains a challenge. Design of protein–protein interactions (PPIs) is impeded by the absence of an energy function that could reliably reproduce all favorable interactions between the binding partners. In this work, we present three energy functions: one function that was trained on monomeric proteins, while the other two were optimized by different techniques to predict side-chain conformations in a dataset of PPIs. The performances of these energy functions are evaluated in three different tasks related to design of PPIs: predicting side-chain conformations in PPIs, recovering native binding-interface sequences, and predicting changes in free energy of binding due to mutations. Our findings show that both functions optimized on side-chain repacking in PPIs are more suitable for PPI design compared to the function trained on monomeric proteins. Yet, no function performs best at all three tasks. Comparison of the three energy functions and their performances revealed that (1) burial of polar atoms should not be penalized significantly in PPI design as in single-protein design and (2) contribution of electrostatic interactions should be increased several-fold when switching from single-protein to PPI design. In addition, the use of a softer van der Waals potential is beneficial in cases when backbone flexibility is important. All things considered, we define an energy function that captures most of the nuances of the binding energetics and hence, should be used in future for design of PPIs.

Comparative analysis and assessment of M. tuberculosis H37Rv protein-protein interaction datasets.

BackgroundM. tuberculosis is a formidable bacterial pathogen. There is thus an increasing demand on understanding the function and relationship of proteins in various strains of M. tuberculosis. Protein-protein interactions (PPIs) data are crucial for this kind of knowledge. However, the quality of the main available M. tuberculosis PPI datasets is unclear. This hampers the effectiveness of research works that rely on these PPI datasets. Here, we analyze the two main available M. tuberculosis H37Rv PPI datasets. The first dataset is the high-throughput B2H PPI dataset from Wang et al’s recent paper in Journal of Proteome Research. The second dataset is from STRING database, version 8.3, comprising entirely of H37Rv PPIs predicted using various methods. We find that these two datasets have a surprisingly low level of agreement. We postulate the following causes for this low level of agreement: (i) the H37Rv B2H PPI dataset is of low quality; (ii) the H37Rv STRING PPI dataset is of low quality; and/or (iii) the H37Rv STRING PPIs are predictions of other forms of functional associations rather than direct physical interactions.ResultsTo test the quality of these two datasets, we evaluate them based on correlated gene expression profiles, coherent informative GO term annotations, and conservation in other organisms. We observe a significantly greater portion of PPIs in the H37Rv STRING PPI dataset (with score ≥ 770) having correlated gene expression profiles and coherent informative GO term annotations in both interaction partners than that in the H37Rv B2H PPI dataset. Predicted H37Rv interologs derived from non-M. tuberculosis experimental PPIs are much more similar to the H37Rv STRING functional associations dataset (with score ≥ 770) than the H37Rv B2H PPI dataset. H37Rv predicted physical interologs from IntAct also show extremely low similarity with the H37Rv B2H PPI dataset; and this similarity level is much lower than that between the S. aureus MRSA252 predicted physical interologs from IntAct and S. aureus MRSA252 pull-down PPIs. Comparative analysis with several representative two-hybrid PPI datasets in other species further confirms that the H37Rv B2H PPI dataset is of low quality. Next, to test the possibility that the H37Rv STRING PPIs are not purely direct physical interactions, we compare M. tuberculosis H37Rv protein pairs that catalyze adjacent steps in enzymatic reactions to B2H PPIs and predicted PPIs in STRING, which shows it has much lower similarities with the B2H PPIs than with STRING PPIs. This result strongly suggests that the H37Rv STRING PPIs more likely correspond to indirect relationships between protein pairs than to B2H PPIs. For more precise support, we turn to S. cerevisiae for its comprehensively studied interactome. We compare S. cerevisiae predicted PPIs in STRING to three independent protein relationship datasets which respectively comprise PPIs reported in Y2H assays, protein pairs reported to be in the same protein complexes, and protein pairs that catalyze successive reaction steps in enzymatic reactions. Our analysis reveals that S. cerevisiae predicted STRING PPIs have much higher similarity to the latter two types of protein pairs than to two-hybrid PPIs. As H37Rv STRING PPIs are predicted using similar methods as S. cerevisiae predicted STRING PPIs, this suggests that these H37Rv STRING PPIs are more likely to correspond to the latter two types of protein pairs rather than to two-hybrid PPIs as well.ConclusionsThe H37Rv B2H PPI dataset has low quality. It should not be used as the gold standard to assess the quality of other (possibly predicted) H37Rv PPI datasets. The H37Rv STRING PPI dataset also has low quality; nevertheless, a subset consisting of STRING PPIs with score ≥770 has satisfactory quality. However, these STRING “PPIs” should be interpreted as functional associations, which include a substantial portion of indirect protein interactions, rather than direct physical interactions. These two factors cause the strikingly low similarity between these two main H37Rv PPI datasets. The results and conclusions from this comparative analysis provide valuable guidance in using these M. tuberculosis H37Rv PPI datasets in subsequent studies for a wide range of purposes.

Conditional random field approach to prediction of protein-protein interactions using domain information

BackgroundFor understanding cellular systems and biological networks, it is important to analyze functions and interactions of proteins and domains. Many methods for predicting protein-protein interactions have been developed. It is known that mutual information between residues at interacting sites can be higher than that at non-interacting sites. It is based on the thought that amino acid residues at interacting sites have coevolved with those at the corresponding residues in the partner proteins. Several studies have shown that such mutual information is useful for identifying contact residues in interacting proteins.ResultsWe propose novel methods using conditional random fields for predicting protein-protein interactions. We focus on the mutual information between residues, and combine it with conditional random fields. In the methods, protein-protein interactions are modeled using domain-domain interactions. We perform computational experiments using protein-protein interaction datasets for several organisms, and calculate AUC (Area Under ROC Curve) score. The results suggest that our proposed methods with and without mutual information outperform EM (Expectation Maximization) method proposed by Deng et al., which is one of the best predictors based on domain-domain interactions.ConclusionsWe propose novel methods using conditional random fields with and without mutual information between domains. Our methods based on domain-domain interactions are useful for predicting protein-protein interactions.

An improved method for scoring protein-protein interactions using semantic similarity within the gene ontology

BackgroundSemantic similarity measures are useful to assess the physiological relevance of protein-protein interactions (PPIs). They quantify similarity between proteins based on their function using annotation systems like the Gene Ontology (GO). Proteins that interact in the cell are likely to be in similar locations or involved in similar biological processes compared to proteins that do not interact. Thus the more semantically similar the gene function annotations are among the interacting proteins, more likely the interaction is physiologically relevant. However, most semantic similarity measures used for PPI confidence assessment do not consider the unequal depth of term hierarchies in different classes of cellular location, molecular function, and biological process ontologies of GO and thus may over-or under-estimate similarity.ResultsWe describe an improved algorithm, Topological Clustering Semantic Similarity (TCSS), to compute semantic similarity between GO terms annotated to proteins in interaction datasets. Our algorithm, considers unequal depth of biological knowledge representation in different branches of the GO graph. The central idea is to divide the GO graph into sub-graphs and score PPIs higher if participating proteins belong to the same sub-graph as compared to if they belong to different sub-graphs.ConclusionsThe TCSS algorithm performs better than other semantic similarity measurement techniques that we evaluated in terms of their performance on distinguishing true from false protein interactions, and correlation with gene expression and protein families. We show an average improvement of 4.6 times the F1 score over Resnik, the next best method, on our Saccharomyces cerevisiae PPI dataset and 2 times on our Homo sapiens PPI dataset using cellular component, biological process and molecular function GO annotations.

Network analysis of human protein location

BackgroundUnderstanding cellular systems requires the knowledge of a protein's subcellular localization (SCL). Although experimental and predicted data for protein SCL are archived in various databases, SCL prediction remains a non-trivial problem in genome annotation. Current SCL prediction tools use amino-acid sequence features and text mining approaches. A comprehensive analysis of protein SCL in human PPI and metabolic networks for various subcellular compartments is necessary for developing a robust SCL prediction methodology.ResultsBased on protein-protein interaction (PPI) and metabolite-linked protein interaction (MLPI) networks of proteins, we have compared, contrasted and analysed the statistical properties across different subcellular compartments. We integrated PPI and metabolic datasets with SCL information of human proteins from LOCATE and GOA (Gene Ontology Annotation) and estimated three statistical properties: Chi-square (χ2) test, Paired Localisation Correlation Profile (PLCP) and network topological measures. For the PPI network, Pearson's chi-square test shows that for the same SCL category, twice as many interacting protein pairs are observed than estimated when compared to non-interacting protein pairs (χ2 = 1270.19, P-value < 2.2 × 10-16), whereas for MLPI, metabolite-linked protein pairs having the same SCL are observed 20% more than expected, compared to non-metabolite linked proteins (χ2 = 110.02, P-value < 2.2 x10-16). To address the issue of proteins with multiple SCLs, we have specifically used the PLCP (Pair Localization Correlation Profile) measure. PLCP analysis revealed that protein interactions are majorly restricted to the same SCL, though significant cross-compartment interactions are seen for nuclear proteins. Metabolite-linked protein pairs are restricted to specific compartments such as the mitochondrion (P-value < 6.0e-07), the lysosome (P-value < 4.7e-05) and the Golgi apparatus (P-value < 1.0e-15). These findings indicate that the metabolic network adds value to the information in the PPI network for the localisation process of proteins in human subcellular compartments.ConclusionsThe MLPI network differs significantly from the PPI network in its SCL distribution. The PPI network shows passive protein interaction, possibly due to its high false positive rate, across different subcellular compartments, which seem to be absent in the MLPI network, as the MLPI network has evolved to maintain high substrate specificity for proteins.

ProHits: integrated software for mass spectrometry–based interaction proteomics

ProHits: integrated software for mass spectrometry–based interaction proteomics

Semantic and layered protein function prediction from PPI networks

Background The past few years have seen a rapid development in novel high-throughput technologies that have created large-scale data on protein–protein interactions (PPI) across human and most model species. This data is commonly represented as networks, with nodes representing proteins and edges representing the PPIs. A fundamental challenge to bioinformatics is how to interpret this wealth of data to elucidate the interaction of patterns and the biological characteristics of the proteins. One significant purpose of this interpretation is to predict unknown protein functions. Although many approaches have been proposed in recent years, the challenge still remains how to reasonably and precisely measure the functional similarities between proteins to improve the prediction effectiveness. Results We used a Semantic and Layered Protein Function Prediction (SLPFP) framework to more effectively predict unknown protein functions at different functional levels. The framework relies on a new protein similarity measurement and a clustering-based protein function prediction algorithm. The new protein similarity measurement incorporates the topological structure of the PPI network, as well as the protein’s semantic information in terms of known protein functions at different functional layers. Experiments on real PPI datasets were conducted to evaluate the effectiveness of the proposed framework in predicting unknown protein functions. Conclusion The proposed framework has a higher prediction accuracy compared with other similar approaches. The prediction results are stable even for a large number of proteins. Furthermore, the framework is able to predict unknown functions at different functional layers within the Munich Information Center for Protein Sequence (MIPS) hierarchical functional scheme. The experimental results demonstrated that the new protein similarity measurement reflects more reasonably and precisely relationships between proteins.

The systematic annotation of the three main GPCR families in Reactome

Reactome is an open-source, freely available database of human biological pathways and processes. A major goal of our work is to provide an integrated view of cellular signalling processes that spans from ligand–receptor interactions to molecular readouts at the level of metabolic and transcriptional events. To this end, we have built the first catalogue of all human G protein-coupled receptors (GPCRs) known to bind endogenous or natural ligands. The UniProt database has records for 797 proteins classified as GPCRs and sorted into families A/1, B/2 and C/3 on the basis of amino accid sequence. To these records we have added details from the IUPHAR database and our own manual curation of relevant literature to create reactions in which 563 GPCRs bind ligands and also interact with specific G-proteins to initiate signalling cascades. We believe the remaining 234 GPCRs are true orphans. The Reactome GPCR pathway can be viewed as a detailed interactive diagram and can be exported in many forms. It provides a template for the orthology-based inference of GPCR reactions for diverse model organism species, and can be overlaid with protein–protein interaction and gene expression datasets to facilitate overrepresentation studies and other forms of pathway analysis.Database URL: http://www.reactome.org

Incorporating multiple genomic features with the utilization of interacting domain patterns to improve the prediction of protein–protein interactions

Protein–protein interaction (PPI) networks play an outstanding role in the organization of life. Parallel to the growth of experimental techniques on determining PPIs, the emergence of computational methods has greatly accelerated the time needed for the identification of PPIs on a wide genomic scale. Although experimental approaches have limitations that can be complemented by the computational methods, the results from computational methods still suffer from high false positive rates which contribute to the lack of solid PPI information. Our study introduces the PPI-Filter; a computational framework aimed at improving PPI prediction results. It is a post-prediction process which involves filtration, using information based on three different genomic features; (i) gene ontology annotation (GOA), (ii) homologous interactions and (iii) protein families (PFAM) domain interactions. In the study, we incorporated a protein function prediction method, based on interacting domain patterns, the protein function predictor or PFP (), for the purpose of aiding the GOA. The goal is to improve the robustness of predicted PPI pairs by removing the false positive pairs and sustaining as much true positive pairs as possible, thus achieving a high confidence level of PPI datasets. The PPI-Filter has been proven to be applicable based on the satisfactory results obtained using signal-to-noise ratio (SNR) and strength measurements that were applied on different computational PPI prediction methods.

Human Protein Structural Interaction Network: Domain Effects on Network Topology and Protein Function*

Availability of high-throughput protein-protein interaction and domain-domain interaction information make it possible to study human structural interaction network, and uncover the inner relationship between structure and function of proteins in proteomics area.The widely-distributed domains are thought to affect structure and function of proteins.However, it is still a challenge to investigate potential mechanisms of these effects combing the structural information (e.g domain number, domain length and domain coverage).The proteins were classified into single- or multi- domain proteins, then human protein structural interaction network was constructed with classification information by integrating the protein-protein interaction and the domain- domain interaction data.Furthermore, comparing with the protein-protein interaction network, specific structure characteristics of human protein structural interaction network were studied. And with respect to single-/multi- domain proteins, function enrichment analysis was carried out for their functions. With domain-domain interaction considered, human protein structural interaction network could provide more detailed information distinct from networks based on protein-protein interaction datasets only, and might reveal the underlying complexity of protein-protein interaction network from the perspective of protein classification. Human protein structural interaction network could exploit domain information to provide additional and crucial protein interaction details necessary for understanding what human structural interactions imply.

Utilizing shared interacting domain patterns and Gene Ontology information to improve protein–protein interaction prediction

Protein–protein interactions (PPIs) play a significant role in many crucial cellular operations such as metabolism, signaling and regulations. The computational methods for predicting PPIs have shown tremendous growth in recent years, but problem such as huge false positive rates has contributed to the lack of solid PPI information. We aimed at enhancing the overlap between computational predictions and experimental results in an effort to partially remove PPIs falsely predicted. The use of protein function predictor named PFP( ) that are based on shared interacting domain patterns is introduced in this study with the purpose of aiding the Gene Ontology Annotations (GOA). We used GOA and PFP( ) as agents in a filtering process to reduce false positive pairs in the computationally predicted PPI datasets. The functions predicted by PFP( ) were extracted from cross-species PPI data in order to assign novel functional annotations for the uncharacterized proteins and also as additional functions for those that are already characterized by the GO (Gene Ontology). The implementation of PFP( ) managed to increase the chances of finding matching function annotation for the first rule in the filtration process as much as 20%. To assess the capability of the proposed framework in filtering false PPIs, we applied it on the available S. cerevisiae PPIs and measured the performance in two aspects, the improvement made indicated as Signal-to-Noise Ratio (SNR) and the strength of improvement, respectively. The proposed filtering framework significantly achieved better performance than without it in both metrics.

Exploratory analysis of protein translation regulatory networks using hierarchical random graphs

BackgroundProtein translation is a vital cellular process for any living organism. The availability of interaction databases provides an opportunity for researchers to exploit the immense amount of data in silico such as studying biological networks. There has been an extensive effort using computational methods in deciphering the transcriptional regulatory networks. However, research on translation regulatory networks has caught little attention in the bioinformatics and computational biology community.ResultsIn this paper, we present an exploratory analysis of yeast protein translation regulatory networks using hierarchical random graphs. We derive a protein translation regulatory network from a protein-protein interaction dataset. Using a hierarchical random graph model, we show that the network exhibits well organized hierarchical structure. In addition, we apply this technique to predict missing links in the network. ConclusionsThe hierarchical random graph mode can be a potentially useful technique for inferring hierarchical structure from network data and predicting missing links in partly known networks. The results from the reconstructed protein translation regulatory networks have potential implications for better understanding mechanisms of translational control from a system’s perspective.

A Comprehensive Resource of Interacting Protein Regions for Refining Human Transcription Factor Networks

Large-scale data sets of protein-protein interactions (PPIs) are a valuable resource for mapping and analysis of the topological and dynamic features of interactome networks. The currently available large-scale PPI data sets only contain information on interaction partners. The data presented in this study also include the sequences involved in the interactions (i.e., the interacting regions, IRs) suggested to correspond to functional and structural domains. Here we present the first large-scale IR data set obtained using mRNA display for 50 human transcription factors (TFs), including 12 transcription-related proteins. The core data set (966 IRs; 943 PPIs) displays a verification rate of 70%. Analysis of the IR data set revealed the existence of IRs that interact with multiple partners. Furthermore, these IRs were preferentially associated with intrinsic disorder. This finding supports the hypothesis that intrinsically disordered regions play a major role in the dynamics and diversity of TF networks through their ability to structurally adapt to and bind with multiple partners. Accordingly, this domain-based interaction resource represents an important step in refining protein interactions and networks at the domain level and in associating network analysis with biological structure and function.

Prediction of protein functions based on function–function correlation relations

A protein function pair approach, based on protein–protein interaction (PPI) data, is proposed to predict protein functions. Randomization tests are performed on the PPI dataset, which resulted in a protein function correlation scoring value which is used to rank the relative importance of a function pair. It has been found that certain classes of protein functions tend to be correlated together. Scoring values of these correlation pairs allow us to predict the functionality of a protein given that it interacts with proteins having well-defined function annotations. The jackknife test is used to validate the function pair method. The protein function pair approach achieves a prediction sensitivity comparable to an approach using more sophisticated method. The main advantages of this approach are as follows: (i) a set of function–function correlation relations are derived and intuitive biological interpretation can be achieved, and (ii) its simplicity, only two parameters are needed.

Analysis of protein complexes through model‐based biclustering of label‐free quantitative AP‐MS data

Affinity purification followed by mass spectrometry (AP-MS) has become a common approach for identifying protein–protein interactions (PPIs) and complexes. However, data analysis and visualization often rely on generic approaches that do not take advantage of the quantitative nature of AP-MS. We present a novel computational method, nested clustering, for biclustering of label-free quantitative AP-MS data. Our approach forms bait clusters based on the similarity of quantitative interaction profiles and identifies submatrices of prey proteins showing consistent quantitative association within bait clusters. In doing so, nested clustering effectively addresses the problem of overrepresentation of interactions involving baits proteins as compared with proteins only identified as preys. The method does not require specification of the number of bait clusters, which is an advantage against existing model-based clustering methods. We illustrate the performance of the algorithm using two published intermediate scale human PPI data sets, which are representative of the AP-MS data generated from mammalian cells. We also discuss general challenges of analyzing and interpreting clustering results in the context of AP-MS data.

A discriminative approach for identifying domain–domain interactions from protein–protein interactions

Protein domains are functional and structural units of proteins. Therefore, identification of domain-domain interactions (DDIs) can provide insight into the biological functions of proteins. In this article, we propose a novel discriminative approach for predicting DDIs based on both protein-protein interactions (PPIs) and the derived information of non-PPIs. We make a threefold contribution to the work in this area. First, we take into account non-PPIs explicitly and treat the domain combinations that can discriminate PPIs from non-PPIs as putative DDIs. Second, DDI identification is formalized as a feature selection problem, in which it tries to find out a minimum set of informative features (i.e., putative DDIs) that discriminate PPIs from non-PPIs, which is plausible in biology and is able to predict DDIs in a systematic and accurate manner. Third, multidomain combinations including two-domain combinations are taken into account in the proposed method, where multidomain cooperations may help proteins to interact with each other. Numerical results on several DDI prediction benchmark data sets show that the proposed discriminative method performs comparably well with other top algorithms with respect to overall performance, and outperforms other methods in terms of precision. The PPI data sets used for prediction of DDIs and prediction results can be found at http://csb.shu.edu.cn/dipd.

Inferring protein function by domain context similarities in protein-protein interaction networks

BackgroundGenome sequencing projects generate massive amounts of sequence data but there are still many proteins whose functions remain unknown. The availability of large scale protein-protein interaction data sets makes it possible to develop new function prediction methods based on protein-protein interaction (PPI) networks. Although several existing methods combine multiple information resources, there is no study that integrates protein domain information and PPI networks to predict protein functions.ResultsThe domain context similarity can be a useful index to predict protein function similarity. The prediction accuracy of our method in yeast is between 63%-67%, which outperforms the other methods in terms of ROC curves.ConclusionThis paper presents a novel protein function prediction method that combines protein domain composition information and PPI networks. Performance evaluations show that this method outperforms existing methods.

Breaking the hierarchy - a new cluster selection mechanism for hierarchical clustering methods

BackgroundHierarchical clustering methods like Ward's method have been used since decades to understand biological and chemical data sets. In order to get a partition of the data set, it is necessary to choose an optimal level of the hierarchy by a so-called level selection algorithm. In 2005, a new kind of hierarchical clustering method was introduced by Palla et al. that differs in two ways from Ward's method: it can be used on data on which no full similarity matrix is defined and it can produce overlapping clusters, i.e., allow for multiple membership of items in clusters. These features are optimal for biological and chemical data sets but until now no level selection algorithm has been published for this method.ResultsIn this article we provide a general selection scheme, the level independent clustering selection method, called LInCS. With it, clusters can be selected from any level in quadratic time with respect to the number of clusters. Since hierarchically clustered data is not necessarily associated with a similarity measure, the selection is based on a graph theoretic notion of cohesive clusters. We present results of our method on two data sets, a set of drug like molecules and set of protein-protein interaction (PPI) data. In both cases the method provides a clustering with very good sensitivity and specificity values according to a given reference clustering. Moreover, we can show for the PPI data set that our graph theoretic cohesiveness measure indeed chooses biologically homogeneous clusters and disregards inhomogeneous ones in most cases. We finally discuss how the method can be generalized to other hierarchical clustering methods to allow for a level independent cluster selection.ConclusionUsing our new cluster selection method together with the method by Palla et al. provides a new interesting clustering mechanism that allows to compute overlapping clusters, which is especially valuable for biological and chemical data sets.