Semantically assessing the reliability of protein interactions

Protein–protein interactions (PPIs) play very important roles in many cellular processes, and provide rich information for discovering biological facts and knowledge. Although various experimental approaches have been developed to generate large amounts of PPI data for different organisms, high-throughput experimental data usually suffers from high error rates, and as a consequence, the biological knowledge discovered from this data is distorted or incorrect. Therefore, it is vital to assess the quality of protein interaction data and extract reliable protein interactions from the high-throughput experimental data. In this paper, we propose a new Semantic Reliability (SR) method to assess the reliability of each protein interaction and identify potential false-positive protein interactions in a dataset. For each pair of target interacting proteins, the SR method takes into account the semantic influence between proteins that interact with the target proteins, and the semantic influence between the target proteins themselves when assessing the interaction reliability. Evaluations on real protein interaction datasets demonstrated that our method outperformed other existing methods in terms of extracting more reliable interactions from original protein interaction datasets.     

Prediction of missed cleavage sites in tryptic peptides aids protein identification in proteomics

Protein identification via peptide mass fingerprinting (PMF) remains a key component of high-throughput proteomics experiments in post-genomic science. Candidate protein identifications are made using bioinformatic tools from peptide peak lists obtained via mass spectrometry (MS). These algorithms rely on several search parameters, including the number of potential uncut peptide bonds matching the primary specificity of the hydrolytic enzyme used in the experiment. Typically, up to one of these "missed cleavages" are considered by the bioinformatics search tools, usually after digestion of the in silico proteome by trypsin. Using two distinct, nonredundant datasets of peptides identified via PMF and tandem MS, a simple predictive method based on information theory is presented which is able to identify experimentally defined missed cleavages with up to 90% accuracy from amino acid sequence alone. Using this simple protocol, we are able to "mask" candidate protein databases so that confident missed cleavage sites need not be considered for in silico digestion. We show that that this leads to an improvement in database searching, with two different search engines, using the PMF dataset as a test set. In addition, the improved approach is also demonstrated on an independent PMF data set of known proteins that also has corresponding high-quality tandem MS data, validating the protein identifications. This approach has wider applicability for proteomics database searching, and the program for predicting missed cleavages and masking Fasta-formatted protein sequence databases has been made available via http:// ispider.smith.man.ac uk/MissedCleave.     

GeneRAGE: a robust algorithm for sequence clustering and domain detection

MOTIVATION: Efficient, accurate and automatic clustering of large protein sequence datasets, such as complete proteomes, into families, according to sequence similarity. Detection and correction of false positive and negative relationships with subsequent detection and resolution of multi-domain proteins. RESULTS: A new algorithm for the automatic clustering of protein sequence datasets has been developed. This algorithm represents all similarity relationships within the dataset in a binary matrix. Removal of false positives is achieved through subsequent symmetrification of the matrix using a Smith-Waterman dynamic programming alignment algorithm. Detection of multi-domain protein families and further false positive relationships within the symmetrical matrix is achieved through iterative processing of matrix elements with successive rounds of Smith-Waterman dynamic programming alignments. Recursive single-linkage clustering of the corrected matrix allows efficient and accurate family representation for each protein in the dataset. Initial clusters containing multi-domain families, are split into their constituent clusters using the information obtained by the multi-domain detection step. This algorithm can hence quickly and accurately cluster large protein datasets into families. Problems due to the presence of multi-domain proteins are minimized, allowing more precise clustering information to be obtained automatically. AVAILABILITY: GeneRAGE (version 1.0) executable binaries for most platforms may be obtained from the authors on request. The system is available to academic users free of charge under license.     

Integrating a functional proteomic approach into the target discovery process

Functional proteomics is a promising technique for the rational identification of novel therapeutic targets by elucidation of the function of newly identified proteins in disease-relevant cellular pathways. Of the recently described high-throughput approaches for analyzing protein-protein interactions, the yeast two-hybrid (Y2H) system has turned out to be one of the most suitable for genome-wide analysis. However, this system presents a challenging technical problem: the high prevalence of false positives and false negatives in datasets due to intrinsic limitations of the technology and the use of a high-throughput, genetic assay. We discuss here the different experimental strategies applied to Y2H assays, their general limitations and advantages. We also address the issue of the contribution of protein interaction mapping to functional biology, especially when combined with complementary genomic and proteomic analyses. Finally, we illustrate how the combination of protein interaction maps with relevant functional assays can provide biological support to large-scale protein interaction datasets and contribute to the identification and validation of potential therapeutic targets.     

A mixture model-based strategy for selecting sets of genes in multiclass response microarray experiments

MOTIVATION: Multiclass response (MCR) experiments are those in which there are more than two classes to be compared. In these experiments, though the null hypothesis is simple, there are typically many patterns of gene expression changes across the different classes that led to complex alternatives. In this paper, we propose a new strategy for selecting genes in MCR that is based on a flexible mixture model for the marginal distribution of a modified F-statistic. Using this model, false positive and negative discovery rates can be estimated and combined to produce a rule for selecting a subset of genes. Moreover, the method proposed allows calculation of these rates for any predefined subset of genes. RESULTS: We illustrate the performance our approach using simulated datasets and a real breast cancer microarray dataset. In this latter study, we investigate predefined subset of genes and point out interesting differences between three distinct biological pathways. AVAILABILITY: http://www.bgx.org.uk/software.html     

Computational Prediction of Protein–Protein Interactions

Recently a number of computational approaches have been developed for the prediction of protein–protein interactions. Complete genome sequencing projects have provided the vast amount of information needed for these analyses. These methods utilize the structural, genomic, and biological context of proteins and genes in complete genomes to predict protein interaction networks and functional linkages between proteins. Given that experimental techniques remain expensive, time-consuming, and labor-intensive, these methods represent an important advance in proteomics. Some of these approaches utilize sequence data alone to predict interactions, while others combine multiple computational and experimental datasets to accurately build protein interaction maps for complete genomes. These methods represent a complementary approach to current high-throughput projects whose aim is to delineate protein interaction maps in complete genomes. We will describe a number of computational protocols for protein interaction prediction based on the structural, genomic, and biological context of proteins in complete genomes, and detail methods for protein interaction network visualization and analysis.     

GANA--a genetic algorithm for NMR backbone resonance assignment

NMR data from different experiments often contain errors; thus, automated backbone resonance assignment is a very challenging issue. In this paper, we present a method called GANA that uses a genetic algorithm to automatically perform backbone resonance assignment with a high degree of precision and recall. Precision is the number of correctly assigned residues divided by the number of assigned residues, and recall is the number of correctly assigned residues divided by the number of residues with known human curated answers. GANA takes spin systems as input data and uses two data structures, candidate lists and adjacency lists, to assign the spin systems to each amino acid of a target protein. Using GANA, almost all spin systems can be mapped correctly onto a target protein, even if the data are noisy. We use the BioMagResBank (BMRB) dataset (901 proteins) to test the performance of GANA. To evaluate the robustness of GANA, we generate four additional datasets from the BMRB dataset to simulate data errors of false positives, false negatives and linking errors. We also use a combination of these three error types to examine the fault tolerance of our method. The average precision rates of GANA on BMRB and the four simulated test cases are 99.61, 99.55, 99.34, 99.35 and 98.60%, respectively. The average recall rates of GANA on BMRB and the four simulated test cases are 99.26, 99.19, 98.85, 98.87 and 97.78%, respectively. We also test GANA on two real wet-lab datasets, hbSBD and hbLBD. The precision and recall rates of GANA on hbSBD are 95.12 and 92.86%, respectively, and those of hbLBD are 100 and 97.40%, respectively.     

IPM: An integrated protein model for false discovery rate estimation and identification in high-throughput proteomics

In high-throughput mass spectrometry proteomics, peptides and proteins are not simply identified as present or not present in a sample, rather the identifications are associated with differing levels of confidence. The false discovery rate (FDR) has emerged as an accepted means for measuring the confidence associated with identifications. We have developed the Systematic Protein Investigative Research Environment (SPIRE) for the purpose of integrating the best available proteomics methods. Two successful approaches to estimating the FDR for MS protein identifications are the MAYU and our current SPIRE methods. We present here a method to combine these two approaches to estimating the FDR for MS protein identifications into an integrated protein model (IPM). We illustrate the high quality performance of this IPM approach through testing on two large publicly available proteomics datasets. MAYU and SPIRE show remarkable consistency in identifying proteins in these datasets. Still, IPM results in a more robust FDR estimation approach and additional identifications, particularly among low abundance proteins. IPM is now implemented as a part of the SPIRE system.     

Diffprot - software for non-parametric statistical analysis of differential proteomics data

Mass spectrometry-based global proteomics experiments generate large sets of data that can be converted into useful information only with an appropriate statistical approach. We present Diffprot - a software tool for statistical analysis of MS-derived quantitative data. With implemented resampling-based statistical test and local variance estimate, Diffprot allows to draw significant results from small scale experiments and effectively eliminates false positive results. To demonstrate the advantages of this software, we performed two spike-in tests with complex biological matrices, one label-free and one based on iTRAQ quantification; in addition, we performed an iTRAQ experiment on bacterial samples. In the spike-in tests, protein ratios were estimated and were in good agreement with theoretical values; statistical significance was assigned to spiked proteins and single or no false positive results were obtained with Diffprot. We compared the performance of Diffprot with other statistical tests - widely used t-test and non-parametric Wilcoxon test. In contrast to Diffprot, both generated many false positive hits in the spike-in experiment. This proved the superiority of the resampling-based method in terms of specificity, making Diffprot a rational choice for small scale high-throughput experiments, when the need to control the false positive rate is particularly pressing.     

Predicting subcellular localization of proteins by hybridizing functional domain composition and pseudo-amino acid composition

Recent advances in large-scale genome sequencing have led to the rapid accumulation of amino acid sequences of proteins whose functions are unknown. Since the functions of these proteins are closely correlated with their subcellular localizations, many efforts have been made to develop a variety of methods for predicting protein subcellular location. In this study, based on the strategy by hybridizing the functional domain composition and the pseudo-amino acid composition (Cai and Chou [2003]: Biochem. Biophys. Res. Commun. 305:407-411), the Intimate Sorting Algorithm (ISort predictor) was developed for predicting the protein subcellular location. As a showcase, the same plant and non-plant protein datasets as investigated by the previous investigators were used for demonstration. The overall success rate by the jackknife test for the plant protein dataset was 85.4%, and that for the non-plant protein dataset 91.9%. These are so far the highest success rates achieved for the two datasets by following a rigorous cross validation test procedure, further confirming that such a hybrid approach may become a very useful high-throughput tool in the area of bioinformatics, proteomics, as well as molecular cell biology.     

Predicting subcellular localization of proteins in a hybridization space

MOTIVATION: The localization of a protein in a cell is closely correlated with its biological function. With the number of sequences entering into databanks rapidly increasing, the importance of developing a powerful high-throughput tool to determine protein subcellular location has become self-evident. In view of this, the Nearest Neighbour Algorithm was developed for predicting the protein subcellular location using the strategy of hybridizing the information derived from the recent development in gene ontology with that from the functional domain composition as well as the pseudo amino acid composition. RESULTS: As a showcase, the same plant and non-plant protein datasets as investigated by the previous investigators were used for demonstration. The overall success rate of the jackknife test for the plant protein dataset was 86%, and that for the non-plant protein dataset 91.2%. These are the highest success rates achieved so far for the two datasets by following a rigorous cross-validation test procedure, suggesting that such a hybrid approach (particularly by incorporating the knowledge of gene ontology) may become a very useful high-throughput tool in the area of bioinformatics, proteomics, as well as molecular cell biology. AVAILABILITY: The software would be made available on sending a request to the authors.     

Protein complex evolution does not involve extensive network rewiring

The formation of proteins into stable protein complexes plays a fundamental role in the operation of the cell. The study of the degree of evolutionary conservation of protein complexes between species and the evolution of protein-protein interactions has been hampered by lack of comprehensive coverage of the high-throughput (HTP) technologies that measure the interactome. We show that new high-throughput datasets on protein co-purification in yeast have a substantially lower false negative rate than previous datasets when compared to known complexes. These datasets are therefore more suitable to estimate the conservation of protein complex membership than hitherto possible. We perform comparative genomics between curated protein complexes from human and the HTP data in Saccharomyces cerevisiae to study the evolution of co-complex memberships. This analysis revealed that out of the 5,960 protein pairs that are part of the same complex in human, 2,216 are absent because both proteins lack an ortholog in S. cerevisiae, while for 1,828 the co-complex membership is disrupted because one of the two proteins lacks an ortholog. For the remaining 1,916 protein pairs, only 10% were never co-purified in the large-scale experiments. This implies a conservation level of co-complex membership of 90% when the genes coding for the protein pairs that participate in the same protein complex are also conserved. We conclude that the evolutionary dynamics of protein complexes are, by and large, not the result of network rewiring (i.e. acquisition or loss of co-complex memberships), but mainly due to genomic acquisition or loss of genes coding for subunits. We thus reveal evidence for the tight interrelation of genomic and network evolution.     

Edge-based scoring and searching method for identifying condition-responsive protein-protein interaction sub-network

MOTIVATION: Current high-throughput protein-protein interaction (PPI) data do not provide information about the condition(s) under which the interactions occur. Thus, the identification of condition-responsive PPI sub-networks is of great importance for investigating how a living cell adapts to changing environments. RESULTS: In this article, we propose a novel edge-based scoring and searching approach to extract a PPI sub-network responsive to conditions related to some investigated gene expression profiles. Using this approach, what we constructed is a sub-network connected by the selected edges (interactions), instead of only a set of vertices (proteins) as in previous works. Furthermore, we suggest a systematic approach to evaluate the biological relevance of the identified responsive sub-network by its ability of capturing condition-relevant functional modules. We apply the proposed method to analyze a human prostate cancer dataset and a yeast cell cycle dataset. The results demonstrate that the edge-based method is able to efficiently capture relevant protein interaction behaviors under the investigated conditions. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.     

EBprot: Statistical analysis of labeling‐based quantitative proteomics data

Labeling‐based proteomics is a powerful method for detection of differentially expressed proteins (DEPs). The current data analysis platform typically relies on protein‐level ratios, which is obtained by summarizing peptide‐level ratios for each protein. In shotgun proteomics, however, some proteins are quantified with more peptides than others, and this reproducibility information is not incorporated into the differential expression (DE) analysis. Here, we propose a novel probabilistic framework EBprot that directly models the peptide‐protein hierarchy and rewards the proteins with reproducible evidence of DE over multiple peptides. To evaluate its performance with known DE states, we conducted a simulation study to show that the peptide‐level analysis of EBprot provides better receiver‐operating characteristic and more accurate estimation of the false discovery rates than the methods based on protein‐level ratios. We also demonstrate superior classification performance of peptide‐level EBprot analysis in a spike‐in dataset. To illustrate the wide applicability of EBprot in different experimental designs, we applied EBprot to a dataset for lung cancer subtype analysis with biological replicates and another dataset for time course phosphoproteome analysis of EGF‐stimulated HeLa cells with multiplexed labeling. Through these examples, we show that the peptide‐level analysis of EBprot is a robust alternative to the existing statistical methods for the DE analysis of labeling‐based quantitative datasets. The software suite is freely available on the Sourceforge website http://ebprot.sourceforge.net/ . All MS data have been deposited in the ProteomeXchange with identifier PXD001426 ( http://proteomecentral.proteomexchange.org/dataset/PXD001426/ ).     

Protein subcellular localization prediction using a hybrid of similarity search and error-correcting output code techniques that produces interpretable results

In silico prediction of protein subcellular localization based on amino acid sequence can reveal valuable information about the protein's innate roles in the cell. Unfortunately, such prediction is made difficult because of complex protein sorting signals. Some prediction methods are based on searching for similar proteins with known localization, assuming that known homologs exist. However, it may not perform well on proteins with no known homolog. In contrast, machine learning-based approaches attempt to infer a predictive model that describes the protein sorting signals. Alas, in doing so, it does not take advantage of known homologs (if they exist) by doing a simple "table lookup". Here, we capture the best of both worlds by combining both approaches. On a dataset with 12 locations, similarity-based and machine learning independently achieve an accuracy of 83.8% and 72.6%, respectively. Our hybrid approach yields an improved accuracy of 85.9%. We compared our method with three other methods' published results. For two of the methods, we used their published datasets for comparison. For the third we used the 12 location dataset. The Error Correcting Output Code algorithm was used to construct our predictive model. This algorithm gives attention to all the classes regardless of number of instances and led to high accuracy among each of the classes and a high prediction rate overall. We also illustrated how the machine learning classifier we use, built over a meaningful set of features can produce interpretable rules that may provide valuable insights into complex protein sorting mechanisms.     

Predicting co-complexed protein pairs from heterogeneous data

Proteins do not carry out their functions alone. Instead, they often act by participating in macromolecular complexes and play different functional roles depending on the other members of the complex. It is therefore interesting to identify co-complex relationships. Although protein complexes can be identified in a high-throughput manner by experimental technologies such as affinity purification coupled with mass spectrometry (APMS), these large-scale datasets often suffer from high false positive and false negative rates. Here, we present a computational method that predicts co-complexed protein pair (CCPP) relationships using kernel methods from heterogeneous data sources. We show that a diffusion kernel based on random walks on the full network topology yields good performance in predicting CCPPs from protein interaction networks. In the setting of direct ranking, a diffusion kernel performs much better than the mutual clustering coefficient. In the setting of SVM classifiers, a diffusion kernel performs much better than a linear kernel. We also show that combination of complementary information improves the performance of our CCPP recognizer. A summation of three diffusion kernels based on two-hybrid, APMS, and genetic interaction networks and three sequence kernels achieves better performance than the sequence kernels or diffusion kernels alone. Inclusion of additional features achieves a still better ROC(50) of 0.937. Assuming a negative-to-positive ratio of 600ratio1, the final classifier achieves 89.3% coverage at an estimated false discovery rate of 10%. Finally, we applied our prediction method to two recently described APMS datasets. We find that our predicted positives are highly enriched with CCPPs that are identified by both datasets, suggesting that our method successfully identifies true CCPPs. An SVM classifier trained from heterogeneous data sources provides accurate predictions of CCPPs in yeast. This computational method thereby provides an inexpensive method for identifying protein complexes that extends and complements high-throughput experimental data.     

Predicting Classifier Performance with Limited Training Data: Applications to Computer-Aided Diagnosis in Breast and Prostate Cancer

Clinical trials increasingly employ medical imaging data in conjunction with supervised classifiers, where the latter require large amounts of training data to accurately model the system. Yet, a classifier selected at the start of the trial based on smaller and more accessible datasets may yield inaccurate and unstable classification performance. In this paper, we aim to address two common concerns in classifier selection for clinical trials: (1) predicting expected classifier performance for large datasets based on error rates calculated from smaller datasets and (2) the selection of appropriate classifiers based on expected performance for larger datasets. We present a framework for comparative evaluation of classifiers using only limited amounts of training data by using random repeated sampling (RRS) in conjunction with a cross-validation sampling strategy. Extrapolated error rates are subsequently validated via comparison with leave-one-out cross-validation performed on a larger dataset. The ability to predict error rates as dataset size increases is demonstrated on both synthetic data as well as three different computational imaging tasks: detecting cancerous image regions in prostate histopathology, differentiating high and low grade cancer in breast histopathology, and detecting cancerous metavoxels in prostate magnetic resonance spectroscopy. For each task, the relationships between 3 distinct classifiers (k-nearest neighbor, naive Bayes, Support Vector Machine) are explored. Further quantitative evaluation in terms of interquartile range (IQR) suggests that our approach consistently yields error rates with lower variability (mean IQRs of 0.0070, 0.0127, and 0.0140) than a traditional RRS approach (mean IQRs of 0.0297, 0.0779, and 0.305) that does not employ cross-validation sampling for all three datasets.     

Computational analysis of human protein interaction networks

Large amounts of human protein interaction data have been produced by experiments and prediction methods. However, the experimental coverage of the human interactome is still low in contrast to predicted data. To gain insight into the value of publicly available human protein network data, we compared predicted datasets, high-throughput results from yeast two-hybrid screens, and literature-curated protein-protein interactions. This evaluation is not only important for further methodological improvements, but also for increasing the confidence in functional hypotheses derived from predictions. Therefore, we assessed the quality and the potential bias of the different datasets using functional similarity based on the Gene Ontology, structural iPfam domain-domain interactions, likelihood ratios, and topological network parameters. This analysis revealed major differences between predicted datasets, but some of them also scored at least as high as the experimental ones regarding multiple quality measures. Therefore, since only small pair wise overlap between most datasets is observed, they may be combined to enlarge the available human interactome data. For this purpose, we additionally studied the influence of protein length on data quality and the number of disease proteins covered by each dataset. We could further demonstrate that protein interactions predicted by more than one method achieve an elevated reliability.