Global protein interactome exploration through mining genome-scale data in Arabidopsis thaliana

Abstract

BackgroundMany essential cellular processes, such as cellular metabolism, transport, cellular metabolism and most regulatory mechanisms, rely on physical interactions between proteins. Genome-wide protein interactome networks of yeast, human and several other animal organisms have already been established, but this kind of network reminds to be established in the field of plant.ResultsWe first predicted the protein protein interaction in Arabidopsis thaliana with methods, including ortholog, SSBP, gene fusion, gene neighbor, phylogenetic profile, coexpression, protein domain, and used Naïve Bayesian approach next to integrate the results of these methods and text mining data to build a genome-wide protein interactome network. Furthermore, we adopted the data of GO enrichment analysis, pathway, published literature to validate our network, the confirmation of our network shows the feasibility of using our network to predict protein function and other usage.ConclusionsOur interactome is a comprehensive genome-wide network in the organism plant Arabidopsis thaliana, and provides a rich resource for researchers in related field to study the protein function, molecular interaction and potential mechanism under different conditions.

Prediction of Protein–Protein Interactions by Evidence Combining Methods

“…There are some reported methods for extracting negative datasets, such as: (1) Negative datasets are constructed by using random pairs which exclude the experimentally detected interactions [1], and as there are discordant numbers between high-confidence interactions and random pairs, the scale and structure of networks should be balanced between negative and positive datasets. This method may include undetected PPIs; (2) Negative examples are chosen based on the categories of their distinct functions, such as sub-cellular localization (can be accessed by tools such as LOCATE [38], PSORTdb 3.0 [39], LocDB [40]) and annotations (such as KEGG pathways, gene ontology (GO), and Enzyme Commission (EC)) [22,41]. However, these methods can also lead to biases due to varying definitions of categories [42]; (3) Another alternative approach is based on topological policy: choose pairs of separated proteins in existing PPI networks to represent non-interactions: defining negative samples as the protein pairs with the shortest path lengths exceed the median shortest paths in a GSP network [43], or further construct a GSN network based on the principle of keeping the composition and degree of a node identical to the GSP network [20].…”

“…These features may represent a particular source of information such as correlations of gene expression, phylogenetic profiles, sequence-based signatures, GO functional annotation and chemical properties. There are many modes to encode evidence sources into a featured vector, to choose statistical standard and data dimensions, and to check the normalization affect or the reliability of different computational predictions [22,45]. …”

“…Case 1: Naive Bayes strategy is proposed for exploring a model network in specific species which lack protein structural information [18,22]. Available evidence includes genomic and proteomic assembled data, ortholog interaction in model organisms, coexpression profiles and enriched protein−domain pairs, as well as shared functional annotations from Gene Ontology (identified the smallest shared biological process (SSBP) score).…”

“…An alternative strategy is the integration of evidence sources in a statistical learning framework. Combining evidence exhibits the strength of machine learning and data mining to overcome the limitations of independent predictions and make the results consistent with the increase of prediction coverage and accuracy [1,18,19,20,21,22,23]. Such methods of PPI prediction are referred as “prediction of protein–protein interactions by evidence-combining methods”.…”

Systems Analysis of Plant Functional, Transcriptional, Physical Interaction, and Metabolic Networks

“…Protein interactions can be inferred by homology to known interactions in other organisms (interaction ortholog or interlog) (Geisler- Lee et al, 2007), using indirect evidence or literature (Cui et al, 2008) or integrated methods (Xu et al, 2010). Interlogs can be filtered using functional association data to improve prediction reliability (De Bodt et al, 2009).…”