Evaluation of different biological data and computational classification methods for use in protein interaction prediction

Protein–protein interactions play a key role in many biological systems. High‐throughput methods can directly detect the set of interacting proteins in yeast, but the results are often incomplete and exhibit high false‐positive and false‐negative rates. Recently, many different research groups independently suggested using supervised learning methods to integrate direct and indirect biological data sources for the protein interaction prediction task. However, the data sources, approaches, and implementations varied. Furthermore, the protein interaction prediction task itself can be subdivided into prediction of (1) physical interaction, (2) co‐complex relationship, and (3) pathway co‐membership. To investigate systematically the utility of different data sources and the way the data is encoded as features for predicting each of these types of protein interactions, we assembled a large set of biological features and varied their encoding for use in each of the three prediction tasks. Six different classifiers were used to assess the accuracy in predicting interactions, Random Forest (RF), RF similarity‐based k‐Nearest‐Neighbor, Naïve Bayes, Decision Tree, Logistic Regression, and Support Vector Machine. For all classifiers, the three prediction tasks had different success rates, and co‐complex prediction appears to be an easier task than the other two. Independently of prediction task, however, the RF classifier consistently ranked as one of the top two classifiers for all combinations of feature sets. Therefore, we used this classifier to study the importance of different biological datasets. First, we used the splitting function of the RF tree structure, the Gini index, to estimate feature importance. Second, we determined classification accuracy when only the top‐ranking features were used as an input in the classifier. We find that the importance of different features depends on the specific prediction task and the way they are encoded. Strikingly, gene expression is consistently the most important feature for all three prediction tasks, while the protein interactions identified using the yeast‐2‐hybrid system were not among the top‐ranking features under any condition. Proteins 2006. © 2006 Wiley‐Liss, Inc.     

Joint learning of gene functions--a Bayesian network model approach

In this paper, we develop a machine learning system for determining gene functions from heterogeneous data sources using a Weighted Naive Bayesian network (WNB). The knowledge of gene functions is crucial for understanding many fundamental biological mechanisms such as regulatory pathways, cell cycles and diseases. Our major goal is to accurately infer functions of putative genes or Open Reading Frames (ORFs) from existing databases using computational methods. However, this task is intrinsically difficult since the underlying biological processes represent complex interactions of multiple entities. Therefore, many functional links would be missing when only one or two sources of data are used in the prediction. Our hypothesis is that integrating evidence from multiple and complementary sources could significantly improve the prediction accuracy. In this paper, our experimental results not only suggest that the above hypothesis is valid, but also provide guidelines for using the WNB system for data collection, training and predictions. The combined training data sets contain information from gene annotations, gene expressions, clustering outputs, keyword annotations, and sequence homology from public databases. The current system is trained and tested on the genes of budding yeast Saccharomyces cerevisiae. Our WNB model can also be used to analyze the contribution of each source of information toward the prediction performance through the weight training process. The contribution analysis could potentially lead to significant scientific discovery by facilitating the interpretation and understanding of the complex relationships between biological entities.     

Boosting Probabilistic Graphical Model Inference by Incorporating Prior Knowledge from Multiple Sources

Inferring regulatory networks from experimental data via probabilistic graphical models is a popular framework to gain insights into biological systems. However, the inherent noise in experimental data coupled with a limited sample size reduces the performance of network reverse engineering. Prior knowledge from existing sources of biological information can address this low signal to noise problem by biasing the network inference towards biologically plausible network structures. Although integrating various sources of information is desirable, their heterogeneous nature makes this task challenging. We propose two computational methods to incorporate various information sources into a probabilistic consensus structure prior to be used in graphical model inference. Our first model, called Latent Factor Model (LFM), assumes a high degree of correlation among external information sources and reconstructs a hidden variable as a common source in a Bayesian manner. The second model, a Noisy-OR, picks up the strongest support for an interaction among information sources in a probabilistic fashion. Our extensive computational studies on KEGG signaling pathways as well as on gene expression data from breast cancer and yeast heat shock response reveal that both approaches can significantly enhance the reconstruction accuracy of Bayesian Networks compared to other competing methods as well as to the situation without any prior. Our framework allows for using diverse information sources, like pathway databases, GO terms and protein domain data, etc. and is flexible enough to integrate new sources, if available.     

Augmented transitive relationships with high impact protein distillation in protein interaction prediction

Predicting new protein-protein interactions is important for discovering novel functions of various biological pathways. Predicting these interactions is a crucial and challenging task. Moreover, discovering new protein-protein interactions through biological experiments is still difficult. Therefore, it is increasingly important to discover new protein interactions. Many studies have predicted protein-protein interactions, using biological features such as Gene Ontology (GO) functional annotations and structural domains of two proteins. In this paper, we propose an augmented transitive relationships predictor (ATRP), a new method of predicting potential protein interactions using transitive relationships and annotations of protein interactions. In addition, a distillation of virtual direct protein-protein interactions is proposed to deal with unbalanced distribution of different types of interactions in the existing protein-protein interaction databases. Our results demonstrate that ATRP can effectively predict protein-protein interactions. ATRP achieves an 81% precision, a 74% recall and a 77% F-measure in average rate in the prediction of direct protein-protein interactions. Using the generated benchmark datasets from KUPS to evaluate of all types of the protein-protein interaction, ATRP achieved a 93% precision, a 49% recall and a 64% F-measure in average rate. This article is part of a Special Issue entitled: Computational Methods for Protein Interaction and Structural Prediction.     

DAVID gene ID conversion tool

Our current biological knowledge is spread over many independent bioinformatics databases where many different types of gene and protein identifiers are used. The heterogeneous and redundant nature of these identifiers limits data analysis across different bioinformatics resources. It is an even more serious bottleneck of data analysis for larger datasets, such as gene lists derived from microarray and proteomic experiments. The DAVID Gene ID Conversion Tool (DICT), a web-based application, is able to convert user's input gene or gene product identifiers from one type to another in a more comprehensive and high-throughput manner with a uniquely enhanced ID-ID mapping database.     

Analyzing yeast protein-protein interaction data obtained from different sources

High-throughput methods for detecting protein interactions, such as mass spectrometry and yeast two-hybrid assays, continue to produce vast amounts of data that may be exploited to infer protein function and regulation. As this article went to press, the pool of all published interaction information on Saccharomyces cerevisiae was 15,143 interactions among 4,825 proteins, and power-law scaling supports an estimate of 20,000 specific protein interactions. To investigate the biases, overlaps, and complementarities among these data, we have carried out an analysis of two high-throughput mass spectrometry (HMS)-based protein interaction data sets from budding yeast, comparing them to each other and to other interaction data sets. Our analysis reveals 198 interactions among 222 proteins common to both data sets, many of which reflect large multiprotein complexes. It also indicates that a "spoke" model that directly pairs bait proteins with associated proteins is roughly threefold more accurate than a "matrix" model that connects all proteins. In addition, we identify a large, previously unsuspected nucleolar complex of 148 proteins, including 39 proteins of unknown function. Our results indicate that existing large-scale protein interaction data sets are nonsaturating and that integrating many different experimental data sets yields a clearer biological view than any single method alone.     

The identification of informative genes from multiple datasets with increasing complexity

Background  

In microarray data analysis, factors such as data quality, biological variation, and the increasingly multi-layered nature of more complex biological systems complicates the modelling of regulatory networks that can represent and capture the interactions among genes. We believe that the use of multiple datasets derived from related biological systems leads to more robust models. Therefore, we developed a novel framework for modelling regulatory networks that involves training and evaluation on independent datasets. Our approach includes the following steps: (1) ordering the datasets based on their level of noise and informativeness; (2) selection of a Bayesian classifier with an appropriate level of complexity by evaluation of predictive performance on independent data sets; (3) comparing the different gene selections and the influence of increasing the model complexity; (4) functional analysis of the informative genes.     

Mining Pure,Strict Epistatic Interactions from High-Dimensional Datasets: Ameliorating the Curse of Dimensionality

Background

The interaction between loci to affect phenotype is called epistasis. It is strict epistasis if no proper subset of the interacting loci exhibits a marginal effect. For many diseases, it is likely that unknown epistatic interactions affect disease susceptibility. A difficulty when mining epistatic interactions from high-dimensional datasets concerns the curse of dimensionality. There are too many combinations of SNPs to perform an exhaustive search. A method that could locate strict epistasis without an exhaustive search can be considered the brass ring of methods for analyzing high-dimensional datasets.

Methodology/Findings

A SNP pattern is a Bayesian network representing SNP-disease relationships. The Bayesian score for a SNP pattern is the probability of the data given the pattern, and has been used to learn SNP patterns. We identified a bound for the score of a SNP pattern. The bound provides an upper limit on the Bayesian score of any pattern that could be obtained by expanding a given pattern. We felt that the bound might enable the data to say something about the promise of expanding a 1-SNP pattern even when there are no marginal effects. We tested the bound using simulated datasets and semi-synthetic high-dimensional datasets obtained from GWAS datasets. We found that the bound was able to dramatically reduce the search time for strict epistasis. Using an Alzheimer''s dataset, we showed that it is possible to discover an interaction involving the APOE gene based on its score because of its large marginal effect, but that the bound is most effective at discovering interactions without marginal effects.

Conclusions/Significance

We conclude that the bound appears to ameliorate the curse of dimensionality in high-dimensional datasets. This is a very consequential result and could be pivotal in our efforts to reveal the dark matter of genetic disease risk from high-dimensional datasets.     

Fantastic databases and where to find them: Web applications for researchers in a rush

Public databases are essential to the development of multi-omics resources. The amount of data created by biological technologies needs a systematic and organized form of storage, that can quickly be accessed, and managed. This is the objective of a biological database. Here, we present an overview of human databases with web applications. The databases and tools allow the search of biological sequences, genes and genomes, gene expression patterns, epigenetic variation, protein-protein interactions, variant frequency, regulatory elements, and comparative analysis between human and model organisms. Our goal is to provide an opportunity for exploring large datasets and analyzing the data for users with little or no programming skills. Public user-friendly web-based databases facilitate data mining and the search for information applicable to healthcare professionals. Besides, biological databases are essential to improve biomedical search sensitivity and efficiency and merge multiple datasets needed to share data and build global initiatives for the diagnosis, prognosis, and discovery of new treatments for genetic diseases. To show the databases at work, we present a a case study using ACE2 as example of a gene to be investigated. The analysis and the complete list of databases is available in the following website <https://kur1sutaru.github.io/fantastic_databases_and_where_to_find_them/>.     

A mouse protein interactome through combined literature mining with multiple sources of interaction evidence

Protein–protein interactions (PPIs) play crucial roles in a number of biological processes. Recently, protein interaction networks (PINs) for several model organisms and humans have been generated, but few large-scale researches for mice have ever been made neither experimentally nor computationally. In the work, we undertook an effort to map a mouse PIN, in which protein interactions are hidden in enormous amount of biomedical literatures. Following a co-occurrence-based text-mining approach, a probabilistic model—naïve Bayesian was used to filter false-positive interactions by integrating heterogeneous kinds of evidence from genomic and proteomic datasets. A support vector machine algorithm was further used to choose protein pairs with physical interactions. By comparing with the currently available PPI datasets from several model organisms and humans, it showed that the derived mouse PINs have similar topological properties at the global level, but a high local divergence. The mouse protein interaction dataset is stored in the Mouse protein–protein interaction DataBase (MppDB) that is useful source of information for system-level understanding of gene function and biological processes in mammals. Access to the MppDB database is public available at http://bio.scu.edu.cn/mppi.     

Integrating external biological knowledge in the construction of regulatory networks from time-series expression data

ABSTRACT: BACKGROUND: Inference about regulatory networks from high-throughput genomics data is of great interest in systems biology. We present a Bayesian approach to infer gene regulatory networks from time series expression data by integrating various types of biological knowledge. RESULTS: We formulate network construction as a series of variable selection problems and use linear regression to model the data. Our method summarizes additional data sources with an informative prior probability distribution over candidate regression models. We extend the Bayesian model averaging (BMA) variable selection method to select regulators in the regression framework. We summarize the external biological knowledge by an informative prior probability distribution over the candidate regression models. CONCLUSIONS: We demonstrate our method on simulated data and a set of time-series microarray experiments measuring the effect of a drug perturbation on gene expression levels, and show that it outperforms leading regression-based methods in the literature.     

Estimating gene regulatory networks and protein-protein interactions of Saccharomyces cerevisiae from multiple genome-wide data

MOTIVATION: Biological processes in cells are properly performed by gene regulations, signal transductions and interactions between proteins. To understand such molecular networks, we propose a statistical method to estimate gene regulatory networks and protein-protein interaction networks simultaneously from DNA microarray data, protein-protein interaction data and other genome-wide data. RESULTS: We unify Bayesian networks and Markov networks for estimating gene regulatory networks and protein-protein interaction networks according to the reliability of each biological information source. Through the simultaneous construction of gene regulatory networks and protein-protein interaction networks of Saccharomyces cerevisiae cell cycle, we predict the role of several genes whose functions are currently unknown. By using our probabilistic model, we can detect false positives of high-throughput data, such as yeast two-hybrid data. In a genome-wide experiment, we find possible gene regulatory relationships and protein-protein interactions between large protein complexes that underlie complex regulatory mechanisms of biological processes.     

Mining gene expression profiles: an integrated implementation of kernel principal component analysis and singular value decomposition

The detection of genes that show similar profiles under different experimental conditions is often an initial step in inferring the biological significance of such genes. Visualization tools are used to identify genes with similar profiles in microarray studies. Given the large number of genes recorded in microarray experiments, gene expression data are generally displayed on a low dimensional plot, based on linear methods. However, microarray data show nonlinearity, due to high-order terms of interaction between genes, so alternative approaches, such as kernel methods, may be more appropriate. We introduce a technique that combines kernel principal component analysis (KPCA) and Biplot to visualize gene expression profiles. Our approach relies on the singular value decomposition of the input matrix and incorporates an additional step that involves KPCA. The main properties of our method are the extraction of nonlinear features and the preservation of the input variables (genes) in the output display. We apply this algorithm to colon tumor, leukemia and lymphoma datasets. Our approach reveals the underlying structure of the gene expression profiles and provides a more intuitive understanding of the gene and sample association.     

Ranking candidate disease genes from gene expression and protein interaction: a Katz-centrality based approach

Many diseases have complex genetic causes, where a set of alleles can affect the propensity of getting the disease. The identification of such disease genes is important to understand the mechanistic and evolutionary aspects of pathogenesis, improve diagnosis and treatment of the disease, and aid in drug discovery. Current genetic studies typically identify chromosomal regions associated specific diseases. But picking out an unknown disease gene from hundreds of candidates located on the same genomic interval is still challenging. In this study, we propose an approach to prioritize candidate genes by integrating data of gene expression level, protein-protein interaction strength and known disease genes. Our method is based only on two, simple, biologically motivated assumptions--that a gene is a good disease-gene candidate if it is differentially expressed in cases and controls, or that it is close to other disease-gene candidates in its protein interaction network. We tested our method on 40 diseases in 58 gene expression datasets of the NCBI Gene Expression Omnibus database. On these datasets our method is able to predict unknown disease genes as well as identifying pleiotropic genes involved in the physiological cellular processes of many diseases. Our study not only provides an effective algorithm for prioritizing candidate disease genes but is also a way to discover phenotypic interdependency, cooccurrence and shared pathophysiology between different disorders.     

A multiorganism based method for Bayesian gene network estimation

The primary goal of this article is to infer genetic interactions based on gene expression data. A new method for multiorganism Bayesian gene network estimation is presented based on multitask learning. When the input datasets are sparse, as is the case in microarray gene expression data, it becomes difficult to separate random correlations from true correlations that would lead to actual edges when modeling the gene interactions as a Bayesian network. Multitask learning takes advantage of the similarity between related tasks, in order to construct a more accurate model of the underlying relationships represented by the Bayesian networks. The proposed method is tested on synthetic data to illustrate its validity. Then it is iteratively applied on real gene expression data to learn the genetic regulatory networks of two organisms with homologous genes.     

Combining microarrays and biological knowledge for estimating gene networks via bayesian networks

We propose a statistical method for estimating a gene network based on Bayesian networks from microarray gene expression data together with biological knowledge including protein-protein interactions, protein-DNA interactions, binding site information, existing literature and so on. Microarray data do not contain enough information for constructing gene networks accurately in many cases. Our method adds biological knowledge to the estimation method of gene networks under a Bayesian statistical framework, and also controls the trade-off between microarray information and biological knowledge automatically. We conduct Monte Carlo simulations to show the effectiveness of the proposed method. We analyze Saccharomyces cerevisiae gene expression data as an application.     

Predicting Peptide-Mediated Interactions on a Genome-Wide Scale

We describe a method to predict protein-protein interactions (PPIs) formed between structured domains and short peptide motifs. We take an integrative approach based on consensus patterns of known motifs in databases, structures of domain-motif complexes from the PDB and various sources of non-structural evidence. We combine this set of clues using a Bayesian classifier that reports the likelihood of an interaction and obtain significantly improved prediction performance when compared to individual sources of evidence and to previously reported algorithms. Our Bayesian approach was integrated into PrePPI, a structure-based PPI prediction method that, so far, has been limited to interactions formed between two structured domains. Around 80,000 new domain-motif mediated interactions were predicted, thus enhancing PrePPI’s coverage of the human protein interactome.