Predicting protein phenotypes based on protein-protein interaction network

Background

Identifying associated phenotypes of proteins is a challenge of the modern genetics since the multifactorial trait often results from contributions of many proteins. Besides the high-through phenotype assays, the computational methods are alternative ways to identify the phenotypes of proteins.

Methodology/Principal Findings

Here, we proposed a new method for predicting protein phenotypes in yeast based on protein-protein interaction network. Instead of only the most likely phenotype, a series of possible phenotypes for the query protein were generated and ranked acording to the tethering potential score. As a result, the first order prediction accuracy of our method achieved 65.4% evaluated by Jackknife test of 1,267 proteins in budding yeast, much higher than the success rate (15.4%) of a random guess. And the likelihood of the first 3 predicted phenotypes including all the real phenotypes of the proteins was 70.6%.

Conclusions/Significance

The candidate phenotypes predicted by our method provided useful clues for the further validation. In addition, the method can be easily applied to the prediction of protein associated phenotypes in other organisms.     

Structure of the Bro1 domain protein BROX and functional analyses of the ALIX Bro1 domain in HIV-1 budding

Background

Bro1 domains are elongated, banana-shaped domains that were first identified in the yeast ESCRT pathway protein, Bro1p. Humans express three Bro1 domain-containing proteins: ALIX, BROX, and HD-PTP, which function in association with the ESCRT pathway to help mediate intraluminal vesicle formation at multivesicular bodies, the abscission stage of cytokinesis, and/or enveloped virus budding. Human Bro1 domains share the ability to bind the CHMP4 subset of ESCRT-III proteins, associate with the HIV-1 NC^Gag protein, and stimulate the budding of viral Gag proteins. The curved Bro1 domain structure has also been proposed to mediate membrane bending. To date, crystal structures have only been available for the related Bro1 domains from the Bro1p and ALIX proteins, and structures of additional family members should therefore aid in the identification of key structural and functional elements.

Methodology/Principal Findings

We report the crystal structure of the human BROX protein, which comprises a single Bro1 domain. The Bro1 domains from BROX, Bro1p and ALIX adopt similar overall structures and share two common exposed hydrophobic surfaces. Surface 1 is located on the concave face and forms the CHMP4 binding site, whereas Surface 2 is located at the narrow end of the domain. The structures differ in that only ALIX has an extended loop that projects away from the convex face to expose the hydrophobic Phe105 side chain at its tip. Functional studies demonstrated that mutations in Surface 1, Surface 2, or Phe105 all impair the ability of ALIX to stimulate HIV-1 budding.

Conclusions/Significance

Our studies reveal similarities in the overall folds and hydrophobic protein interaction sites of different Bro1 domains, and show that a unique extended loop contributes to the ability of ALIX to function in HIV-1 budding.     

Constitutive patterns of gene expression regulated by RNA-binding proteins

Background

RNA-binding proteins regulate a number of cellular processes, including synthesis, folding, translocation, assembly and clearance of RNAs. Recent studies have reported that an unexpectedly large number of proteins are able to interact with RNA, but the partners of many RNA-binding proteins are still uncharacterized.

Results

We combined prediction of ribonucleoprotein interactions, based on catRAPID calculations, with analysis of protein and RNA expression profiles from human tissues. We found strong interaction propensities for both positively and negatively correlated expression patterns. Our integration of in silico and ex vivo data unraveled two major types of protein–RNA interactions, with positively correlated patterns related to cell cycle control and negatively correlated patterns related to survival, growth and differentiation. To facilitate the investigation of protein–RNA interactions and expression networks, we developed the catRAPID express web server.

Conclusions

Our analysis sheds light on the role of RNA-binding proteins in regulating proliferation and differentiation processes, and we provide a data exploration tool to aid future experimental studies.     

Predicting Drug-Target Interaction Networks Based on Functional Groups and Biological Features

Background

Study of drug-target interaction networks is an important topic for drug development. It is both time-consuming and costly to determine compound-protein interactions or potential drug-target interactions by experiments alone. As a complement, the in silico prediction methods can provide us with very useful information in a timely manner.

Methods/Principal Findings

To realize this, drug compounds are encoded with functional groups and proteins encoded by biological features including biochemical and physicochemical properties. The optimal feature selection procedures are adopted by means of the mRMR (Maximum Relevance Minimum Redundancy) method. Instead of classifying the proteins as a whole family, target proteins are divided into four groups: enzymes, ion channels, G-protein- coupled receptors and nuclear receptors. Thus, four independent predictors are established using the Nearest Neighbor algorithm as their operation engine, with each to predict the interactions between drugs and one of the four protein groups. As a result, the overall success rates by the jackknife cross-validation tests achieved with the four predictors are 85.48%, 80.78%, 78.49%, and 85.66%, respectively.

Conclusion/Significance

Our results indicate that the network prediction system thus established is quite promising and encouraging.     

Computational prediction of protein interactions related to the invasion of erythrocytes by malarial parasites

Background

The invasion of red blood cells (RBCs) by malarial parasites is an essential step in the life cycle of Plasmodium falciparum. Human-parasite surface protein interactions play a critical role in this process. Although several interactions between human and parasite proteins have been discovered, the mechanism related to invasion remains poorly understood because numerous human-parasite protein interactions have not yet been identified. High-throughput screening experiments are not feasible for malarial parasites due to difficulty in expressing the parasite proteins. Here, we performed computational prediction of the PPIs involved in malaria parasite invasion to elucidate the mechanism by which invasion occurs.

Results

In this study, an expectation maximization algorithm was used to estimate the probabilities of domain-domain interactions (DDIs). Estimates of DDI probabilities were then used to infer PPI probabilities. We found that our prediction performance was better than that based on the information of D. melanogaster alone when information related to the six species was used. Prediction performance was assessed using protein interaction data from S. cerevisiae, indicating that the predicted results were reliable. We then used the estimates of DDI probabilities to infer interactions between 490 parasite and 3,787 human membrane proteins. A small-scale dataset was used to illustrate the usability of our method in predicting interactions between human and parasite proteins. The positive predictive value (PPV) was lower than that observed in S. cerevisiae. We integrated gene expression data to improve prediction accuracy and to reduce false positives. We identified 80 membrane proteins highly expressed in the schizont stage by fast Fourier transform method. Approximately 221 erythrocyte membrane proteins were identified using published mass spectral datasets. A network consisting of 205 interactions was predicted. Results of network analysis suggest that SNARE proteins of parasites and APP of humans may function in the invasion of RBCs by parasites.

Conclusions

We predicted a small-scale PPI network that may be involved in parasite invasion of RBCs by integrating DDI information and expression profiles. Experimental studies should be conducted to validate the predicted interactions. The predicted PPIs help elucidate the mechanism of parasite invasion and provide directions for future experimental investigations.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-014-0393-z) contains supplementary material, which is available to authorized users.     

Pathway activity inference for multiclass disease classification through a mathematical programming optimisation framework

Background

Applying machine learning methods on microarray gene expression profiles for disease classification problems is a popular method to derive biomarkers, i.e. sets of genes that can predict disease state or outcome. Traditional approaches where expression of genes were treated independently suffer from low prediction accuracy and difficulty of biological interpretation. Current research efforts focus on integrating information on protein interactions through biochemical pathway datasets with expression profiles to propose pathway-based classifiers that can enhance disease diagnosis and prognosis. As most of the pathway activity inference methods in literature are either unsupervised or applied on two-class datasets, there is good scope to address such limitations by proposing novel methodologies.

Results

A supervised multiclass pathway activity inference method using optimisation techniques is reported. For each pathway expression dataset, patterns of its constituent genes are summarised into one composite feature, termed pathway activity, and a novel mathematical programming model is proposed to infer this feature as a weighted linear summation of expression of its constituent genes. Gene weights are determined by the optimisation model, in a way that the resulting pathway activity has the optimal discriminative power with regards to disease phenotypes. Classification is then performed on the resulting low-dimensional pathway activity profile.

Conclusions

The model was evaluated through a variety of published gene expression profiles that cover different types of disease. We show that not only does it improve classification accuracy, but it can also perform well in multiclass disease datasets, a limitation of other approaches from the literature. Desirable features of the model include the ability to control the maximum number of genes that may participate in determining pathway activity, which may be pre-specified by the user. Overall, this work highlights the potential of building pathway-based multi-phenotype classifiers for accurate disease diagnosis and prognosis problems.

Electronic supplementary material

The online version of this article (doi:10.1186/s12859-014-0390-2) contains supplementary material, which is available to authorized users.     

Protein function assignment through mining cross-species protein-protein interactions

Background

As we move into the post genome-sequencing era, an immediate challenge is how to make best use of the large amount of high-throughput experimental data to assign functions to currently uncharacterized proteins. We here describe CSIDOP, a new method for protein function assignment based on shared interacting domain patterns extracted from cross-species protein-protein interaction data.

Methodology/Principal Findings

The proposed method is assessed both biologically and statistically over the genome of H. sapiens. The CSIDOP method is capable of making protein function prediction with accuracy of 95.42% using 2,972 gene ontology (GO) functional categories. In addition, we are able to assign novel functional annotations for 181 previously uncharacterized proteins in H. sapiens. Furthermore, we demonstrate that for proteins that are characterized by GO, the CSIDOP may predict extra functions. This is attractive as a protein normally executes a variety of functions in different processes and its current GO annotation may be incomplete.

Conclusions/Significance

It can be shown through experimental results that the CSIDOP method is reliable and practical in use. The method will continue to improve as more high quality interaction data becomes available and is readily scalable to a genome-wide application.     

Inference of Functional Relations in Predicted Protein Networks with a Machine Learning Approach

Background

Molecular biology is currently facing the challenging task of functionally characterizing the proteome. The large number of possible protein-protein interactions and complexes, the variety of environmental conditions and cellular states in which these interactions can be reorganized, and the multiple ways in which a protein can influence the function of others, requires the development of experimental and computational approaches to analyze and predict functional associations between proteins as part of their activity in the interactome.

Methodology/Principal Findings

We have studied the possibility of constructing a classifier in order to combine the output of the several protein interaction prediction methods. The AODE (Averaged One-Dependence Estimators) machine learning algorithm is a suitable choice in this case and it provides better results than the individual prediction methods, and it has better performances than other tested alternative methods in this experimental set up. To illustrate the potential use of this new AODE-based Predictor of Protein InterActions (APPIA), when analyzing high-throughput experimental data, we show how it helps to filter the results of published High-Throughput proteomic studies, ranking in a significant way functionally related pairs. Availability: All the predictions of the individual methods and of the combined APPIA predictor, together with the used datasets of functional associations are available at http://ecid.bioinfo.cnio.es/.

Conclusions

We propose a strategy that integrates the main current computational techniques used to predict functional associations into a unified classifier system, specifically focusing on the evaluation of poorly characterized protein pairs. We selected the AODE classifier as the appropriate tool to perform this task. AODE is particularly useful to extract valuable information from large unbalanced and heterogeneous data sets. The combination of the information provided by five prediction interaction prediction methods with some simple sequence features in APPIA is useful in establishing reliability values and helpful to prioritize functional interactions that can be further experimentally characterized.     

Histidine domain-protein tyrosine phosphatase interacts with Grb2 and GrpL

Background

Histidine domain-protein tyrosine phosphatase (HD-PTP) plays a key role in vesicle trafficking and biogenesis. Although it is a large protein with at least five distinct structural domains, only a few of its interactors are presently known, and the significance of these interactions is largely obscure.

Methodology and Results

In this study we performed a yeast two-hybrid screening using a human colon cDNA library and found that Grb2 and GrpL are binding partners of HD-PTP. Co-immunoprecipitation, pull-down and immunocytochemistry experiments confirmed the interactions. We also discovered that the central proline-rich and histidine-rich domain of HD-PTP is responsible for these interactions.

Significance

The interaction of HD-PTP with two adapters of the Grb2 family, essential for numerous signaling pathways, suggests that HD-PTP might be important for signaling through a plethora of receptors.     

The Human-Bacterial Pathogen Protein Interaction Networks of Bacillus anthracis,Francisella tularensis,and Yersinia pestis

Background

Bacillus anthracis, Francisella tularensis, and Yersinia pestis are bacterial pathogens that can cause anthrax, lethal acute pneumonic disease, and bubonic plague, respectively, and are listed as NIAID Category A priority pathogens for possible use as biological weapons. However, the interactions between human proteins and proteins in these bacteria remain poorly characterized leading to an incomplete understanding of their pathogenesis and mechanisms of immune evasion.

Methodology

In this study, we used a high-throughput yeast two-hybrid assay to identify physical interactions between human proteins and proteins from each of these three pathogens. From more than 250,000 screens performed, we identified 3,073 human-B. anthracis, 1,383 human-F. tularensis, and 4,059 human-Y. pestis protein-protein interactions including interactions involving 304 B. anthracis, 52 F. tularensis, and 330 Y. pestis proteins that are uncharacterized. Computational analysis revealed that pathogen proteins preferentially interact with human proteins that are hubs and bottlenecks in the human PPI network. In addition, we computed modules of human-pathogen PPIs that are conserved amongst the three networks. Functionally, such conserved modules reveal commonalities between how the different pathogens interact with crucial host pathways involved in inflammation and immunity.

Significance

These data constitute the first extensive protein interaction networks constructed for bacterial pathogens and their human hosts. This study provides novel insights into host-pathogen interactions.     

Identification of Essential Proteins Based on a New Combination of Local Interaction Density and Protein Complexes

Background

Computational approaches aided by computer science have been used to predict essential proteins and are faster than expensive, time-consuming, laborious experimental approaches. However, the performance of such approaches is still poor, making practical applications of computational approaches difficult in some fields. Hence, the development of more suitable and efficient computing methods is necessary for identification of essential proteins.

Method

In this paper, we propose a new method for predicting essential proteins in a protein interaction network, local interaction density combined with protein complexes (LIDC), based on statistical analyses of essential proteins and protein complexes. First, we introduce a new local topological centrality, local interaction density (LID), of the yeast PPI network; second, we discuss a new integration strategy for multiple bioinformatics. The LIDC method was then developed through a combination of LID and protein complex information based on our new integration strategy. The purpose of LIDC is discovery of important features of essential proteins with their neighbors in real protein complexes, thereby improving the efficiency of identification.

Results

Experimental results based on three different PPI(protein-protein interaction) networks of Saccharomyces cerevisiae and Escherichia coli showed that LIDC outperformed classical topological centrality measures and some recent combinational methods. Moreover, when predicting MIPS datasets, the better improvement of performance obtained by LIDC is over all nine reference methods (i.e., DC, BC, NC, LID, PeC, CoEWC, WDC, ION, and UC).

Conclusions

LIDC is more effective for the prediction of essential proteins than other recently developed methods.     

Predicting functions of proteins in mouse based on weighted protein-protein interaction network and protein hybrid properties

Background

With the huge amount of uncharacterized protein sequences generated in the post-genomic age, it is highly desirable to develop effective computational methods for quickly and accurately predicting their functions. The information thus obtained would be very useful for both basic research and drug development in a timely manner.

Methodology/Principal Findings

Although many efforts have been made in this regard, most of them were based on either sequence similarity or protein-protein interaction (PPI) information. However, the former often fails to work if a query protein has no or very little sequence similarity to any function-known proteins, while the latter had similar problem if the relevant PPI information is not available. In view of this, a new approach is proposed by hybridizing the PPI information and the biochemical/physicochemical features of protein sequences. The overall first-order success rates by the new predictor for the functions of mouse proteins on training set and test set were 69.1% and 70.2%, respectively, and the success rate covered by the results of the top-4 order from a total of 24 orders was 65.2%.

Conclusions/Significance

The results indicate that the new approach is quite promising that may open a new avenue or direction for addressing the difficult and complicated problem.     

Reducing false-positive prediction of minimotifs with a genetic interaction filter

Background

Minimotifs are short contiguous peptide sequences in proteins that have known functions. At its simplest level, the minimotif sequence is present in a source protein and has an activity relationship with a target, most of which are proteins. While many scientists routinely investigate new minimotif functions in proteins, the major web-based discovery tools have a high rate of false-positive prediction. Any new approach that reduces false-positives will be of great help to biologists.

Methods and Findings

We have built three filters that use genetic interactions to reduce false-positive minimotif predictions. The basic filter identifies those minimotifs where the source/target protein pairs have a known genetic interaction. The HomoloGene genetic interaction filter extends these predictions to predicted genetic interactions of orthologous proteins and the node-based filter identifies those minimotifs where proteins that have a genetic interaction with the source or target have a genetic interaction. Each filter was evaluated with a test data set containing thousands of true and false-positives. Based on sensitivity and selectivity performance metrics, the basic filter had the best discrimination for true positives, whereas the node-based filter had the highest sensitivity. We have implemented these genetic interaction filters on the Minimotif Miner 2.3 website. The genetic interaction filter is particularly useful for improving predictions of posttranslational modifications such as phosphorylation and proteolytic cleavage sites.

Conclusions

Genetic interaction data sets can be used to reduce false-positive minimotif predictions. Minimotif prediction in known genetic interactions can help to refine the mechanisms behind the functional connection between genes revealed by genetic experimentation and screens.     

Insulator function and topological domain border strength scale with architectural protein occupancy

Background

Chromosome conformation capture studies suggest that eukaryotic genomes are organized into structures called topologically associating domains. The borders of these domains are highly enriched for architectural proteins with characterized roles in insulator function. However, a majority of architectural protein binding sites localize within topological domains, suggesting sites associated with domain borders represent a functionally different subclass of these regulatory elements. How topologically associating domains are established and what differentiates border-associated from non-border architectural protein binding sites remain unanswered questions.

Results

By mapping the genome-wide target sites for several Drosophila architectural proteins, including previously uncharacterized profiles for TFIIIC and SMC-containing condensin complexes, we uncover an extensive pattern of colocalization in which architectural proteins establish dense clusters at the borders of topological domains. Reporter-based enhancer-blocking insulator activity as well as endogenous domain border strength scale with the occupancy level of architectural protein binding sites, suggesting co-binding by architectural proteins underlies the functional potential of these loci. Analyses in mouse and human stem cells suggest that clustering of architectural proteins is a general feature of genome organization, and conserved architectural protein binding sites may underlie the tissue-invariant nature of topologically associating domains observed in mammals.

Conclusions

We identify a spectrum of architectural protein occupancy that scales with the topological structure of chromosomes and the regulatory potential of these elements. Whereas high occupancy architectural protein binding sites associate with robust partitioning of topologically associating domains and robust insulator function, low occupancy sites appear reserved for gene-specific regulation within topological domains.     

Protein Complex Identification by Integrating Protein-Protein Interaction Evidence from Multiple Sources

Background

Understanding protein complexes is important for understanding the science of cellular organization and function. Many computational methods have been developed to identify protein complexes from experimentally obtained protein-protein interaction (PPI) networks. However, interaction information obtained experimentally can be unreliable and incomplete. Reconstructing these PPI networks with PPI evidences from other sources can improve protein complex identification.

Results

We combined PPI information from 6 different sources and obtained a reconstructed PPI network for yeast through machine learning. Some popular protein complex identification methods were then applied to detect yeast protein complexes using the new PPI networks. Our evaluation indicates that protein complex identification algorithms using the reconstructed PPI network significantly outperform ones on experimentally verified PPI networks.

Conclusions

We conclude that incorporating PPI information from other sources can improve the effectiveness of protein complex identification.     

Assessing the utility of thermodynamic features for microRNA target prediction under relaxed seed and no conservation requirements

Background

Many computational microRNA target prediction tools are focused on several key features, including complementarity to 5′seed of miRNAs and evolutionary conservation. While these features allow for successful target identification, not all miRNA target sites are conserved and adhere to canonical seed complementarity. Several studies have propagated the use of energy features of mRNA:miRNA duplexes as an alternative feature. However, different independent evaluations reported conflicting results on the reliability of energy-based predictions. Here, we reassess the usefulness of energy features for mammalian target prediction, aiming to relax or eliminate the need for perfect seed matches and conservation requirement.

Methodology/Principal Findings

We detect significant differences of energy features at experimentally supported human miRNA target sites and at genome-wide sites of AGO protein interaction. This trend is confirmed on datasets that assay the effect of miRNAs on mRNA and protein expression changes, and a simple linear regression model leads to significant correlation of predicted versus observed expression change. Compared to 6-mer seed matches as baseline, application of our energy-based model leads to ∼3–5-fold enrichment on highly down-regulated targets, and allows for prediction of strictly imperfect targets with enrichment above baseline.

Conclusions/Significance

In conclusion, our results indicate significant promise for energy-based miRNA target prediction that includes a broader range of targets without having to use conservation or impose stringent seed match rules.     

A comparison of performance of plant miRNA target prediction tools and the characterization of features for genome-wide target prediction

Background

Deep-sequencing has enabled the identification of large numbers of miRNAs and siRNAs, making the high-throughput target identification a main limiting factor in defining their function. In plants, several tools have been developed to predict targets, majority of them being trained on Arabidopsis datasets. An extensive and systematic evaluation has not been made for their suitability for predicting targets in species other than Arabidopsis. Nor, these have not been evaluated for their suitability for high-throughput target prediction at genome level.

Results

We evaluated the performance of 11 computational tools in identifying genome-wide targets in Arabidopsis and other plants with procedures that optimized score-cutoffs for estimating targets. Targetfinder was most efficient [89% ‘precision’ (accuracy of prediction), 97% ‘recall’ (sensitivity)] in predicting ‘true-positive’ targets in Arabidopsis miRNA-mRNA interactions. In contrast, only 46% of true positive interactions from non-Arabidopsis species were detected, indicating low ‘recall’ values. Score optimizations increased the ‘recall’ to only 70% (corresponding ‘precision’: 65%) for datasets of true miRNA-mRNA interactions in species other than Arabidopsis. Combining the results of Targetfinder and psRNATarget delivers high true positive coverage, whereas the intersection of psRNATarget and Tapirhybrid outputs deliver highly ‘precise’ predictions. The large number of ‘false negative’ predictions delivered from non-Arabidopsis datasets by all the available tools indicate the diversity in miRNAs-mRNA interaction features between Arabidopsis and other species. A subset of miRNA-mRNA interactions differed significantly for features in seed regions as well as the total number of matches/mismatches.

Conclusion

Although, many plant miRNA target prediction tools may be optimized to predict targets with high specificity in Arabidopsis, such optimized thresholds may not be suitable for many targets in non-Arabidopsis species. More importantly, non-conventional features of miRNA-mRNA interaction may exist in plants indicating alternate mode of miRNA target recognition. Incorporation of these divergent features would enable next-generation of algorithms to better identify target interactions.

Electronic supplementary material

The online version of this article (doi:10.1186/1471-2164-15-348) contains supplementary material, which is available to authorized users.     

IDDI: integrated domain-domain interaction and protein interaction analysis system

Background

Deciphering protein-protein interaction (PPI) in domain level enriches valuable information about binding mechanism and functional role of interacting proteins. The 3D structures of complex proteins are reliable source of domain-domain interaction (DDI) but the number of proven structures is very limited. Several resources for the computationally predicted DDI have been generated but they are scattered in various places and their prediction show erratic performances. A well-organized PPI and DDI analysis system integrating these data with fair scoring system is necessary.

Method

We integrated three structure-based DDI datasets and twenty computationally predicted DDI datasets and constructed an interaction analysis system, named IDDI, which enables to browse protein and domain interactions with their relationships. To integrate heterogeneous DDI information, a novel scoring scheme is introduced to determine the reliability of DDI by considering the prediction scores of each DDI and the confidence levels of each prediction method in the datasets, and independencies between predicted datasets. In addition, we connected this DDI information to the comprehensive PPI information and developed a unified interface for the interaction analysis exploring interaction networks at both protein and domain level.

Result

IDDI provides 204,705 DDIs among total 7,351 Pfam domains in the current version. The result presents that total number of DDIs is increased eight times more than that of previous studies. Due to the increment of data, 50.4% of PPIs could be correlated with DDIs which is more than twice of previous resources. Newly designed scoring scheme outperformed the previous system in its accuracy too. User interface of IDDI system provides interactive investigation of proteins and domains in interactions with interconnected way. A specific example is presented to show the efficiency of the systems to acquire the comprehensive information of target protein with PPI and DDI relationships. IDDI is freely available at http://pcode.kaist.ac.kr/iddi/.