PPIs included in POINeT were merged from various sources, including DIP 24 , MINT 25 , BIND 26 , HPRD 27 , MIPS 28 , CYGD 29 , BioGRID 30 and NCBI interaction ftp://ftp.ncbi.nlm.nih.gov/gene/GeneRIF/interactions.gz. Since different PPI databases use different ID systems, these disparate IDs have been mapped to the NCBI Gene IDs. Therefore, PPIs with different designations from various sources may map to the same interactions. The components of POINeT are combined systematically to meet the needs of the users. For each PPI, additional information was provided, including PubMed IDs, links to the literature reporting this PPI, and Gene Ontology (GO) annotations 31 . Users may input a set of proteins using their corresponding Gene Symbols, Gene IDs or UniProt accession numbers to query the PPI data. Table 2 lists the numbers of PPIs collected from various data sources by POINeT. Interologs can be incorporated into the query result to enrich the potential PPIs in the output networks.

The workflow for querying, filtering and downloading PPIs is depicted in Figure 2A. Briefly, the user inputs the query terms (genes or proteins), which will be recorded as attr-Query, into POINeT to search for all available PPIs, referred to as ppi-AllPPI. If a query has no available PPI, POINeT stores it as attr-noInteractionQuery. If certain filtering criteria are set in the query page, such as 'Number of iterations' or 'Number of literature references', the number of PPIs included in ppi-AllPPI will change accordingly. Subsequently, the nodes involved in ppi-AllPPI will be in the attr-Interactor table and the degrees of these nodes will be calculated. Since the proteins outside of the query protein set could serve as a mediator in PPI network, such as a regulator or an adapter protein, nodes with a degree >= 2 are defined as mediator and recorded in the attr-Mediator table. The mediators are nodes (query and/or non-query) connecting any two query proteins. This will form another network, which removes all nodes with a degree = 1 and is denoted as ppi-Degree2. This network can reduce the complexity of network visualization and illustrate how queries are connected through these mediators. These mediators may be an important member of the sub-network around the query proteins. If a query node interacts with itself and forms a homodimer, this node will be recorded in the attr-HomoDimer table. Furthermore, if two interactors of one interaction were both present in the attr-Query table, this interaction will be documented in ppi-QQPPI. Interactors in the ppi-QQPPI network will be recorded in the attr-QQ table. Figure 2B illustrates various components in a PPI network. Interologs in different species can be inferred systematically using the NCBI HomoloGene database. These interologs' PPI will be recorded in the ppi-InterologsPPI table. Using the gene2go mapping table provided by NCBI, whether two interactors of one PPI share the same GO annotation will be noted, resulting in the ppi-GOPPI network. Finally, if interactors of ppi-QQPPI are present in the attr-Hub table, these interactors will be placed in the attr-QH table, which denotes that a node exhibits both a query and a hub in the network. POINeT will merge ppi-QQPPI, ppi-GOPPI, and ppi-InterologsPPI into ppi-FilteredPPI. This network contains PPIs with relatively reliable and certain biological significances. This network, which is smaller than ppi-AllPPI, can be visualized and analyzed with ease and extended with other selected features. These described tables can be downloaded in multiple formats.

The query results of POINeT can be downloaded in multiple formats, including Excel, sif (simple interaction format), and txt formats. Using the exported sif format, ppi-AllPPI, ppi-Degree2, ppi-FilteredPPI and all attributes can be downloaded. Tissue-specific expression profiles can also be exported into individual attribute files. The query results in sif format can be easily integrated with tissue-specific expression profiles, and visualized in CytoScape 39 . Also, plain text files can be downloaded as well. However, Excel and txt formats do not support the export of tissue-specific expression profiles.

POINeT provides a straightforward viewer with sufficient functionalities. No additional software installations are required. Networks and tissue-specific expression profiles can be visualized directly in the browser. The viewer supports zooming and panning of the networks. The concept of layers in geographic information system (GIS) 40 was adopted. Different output results were defined as different layers. Through the selection of different layers, ppi-QQPPI, ppi-GOPPI and ppi-InterologsPPI can be displayed individually or as a merged network, ppi-FilteredPPI. The labels on each layer can also be turned on/off, as can the labeling of selected nodes. Finally, nodes can be selected to display the associated interactions, PubMed IDs, and Gene Ontology annotations, and provide the links to external databases. Also, tissue-specific expression values are treated as attributes of the nodes in the network. Using the concept of layers adopted from GIS, different tissue expressions can be selected and displayed for the same nodes to facilitate the analysis and comparison of these expression profiles. The network viewer provided by POINeT permits users to observe gene expression levels of the same PPI network in different tissue types.

PPI network analysis is an emerging field for the identification of, for example, disease related genes. By analyzing the gene expression and combining with the integration of omic data sets from different species to construct the PPI network, Migual et al., identified potential genes associated with higher risk of breast cancer 41 . A genes network of disorders linked by known disorder-gene associations led to discover a single graph-theoretic framework in disease gene associations, indicating the common genetic origin of many diseases 42 . These reports suggest that one application of biological networks is the identification of novel marker genes for diseases and the study of interactions among these marker genes. In the case of schizophrenia, a population-based analysis has revealed four genes, DAAO (DAO), DAOA, DTNBP1 and NRG1 43 , to be associated with the schizophrenia. Certainly it would be interesting to discover any associations among these genes in terms of biological networks and their potential involvement in specific biological pathways.

Using these four genes as queries, there are interesting links between DTNBP1 and NRG1 (Figure 3A). DTNBP1 and NRG1 are both involved in fully connected cliques. Two nodes lie between DTNBP1 and NRG1; these are DLG4 and EXOC4. The interactions among DTNBP1, DLG4 and EXOC4 are present in various brain tissues, such as prefrontal cortex and temporal lobe, which are known to be related to schizophrenia etiology (Figure 3B and 3C). Most of the interactions in this sub-network are missing in other un-related tissues, such as adipocyte (Figure 3D). DLG4 is known to be involved in nicotine dependence 44 . There is no known association between DLG4 and schizophrenia in the literature; notwithstanding this, because there are constant controversial debates on the genetic factors contributing to schizophrenia, DLG4 is greatly deserving of further investigation. Similarly, EXOC4 is known to be involved in the exocyst complex, which is critical for the release of neurotransmitters 45 ; at present its functional involvement in schizophrenia is unknown. The roles of DLG4 and EXOC4 in schizophrenia remain to be explored, and the two genes might serve as putative risk markers with potential for further studies.

Recently, various proteomes focusing on specific spatiotemporal conditions have been elucidated, such as the midbody 46 . For these proteome results, it would be interesting to devise the interactions among the protein components, further extending the proteome into the interactome. The midbody is an important organelle formed in the later stage of cytokinesis, and is required for the separation of two daughter cells after cell division. The midbody interactome 46 has been listed as an example on the POINeT website. Based on a literature review, we have extended this set to 190 midbody-related proteins. The first question we asked was whether the limited numbers of midbody proteins, identified in the recent proteomic screen, participate individually in the process of cytokinesis or whether groups of the midbody proteins interact with each other and form a network. The second was how to fill in the missing gaps in the constructed midbody PPI network and identify novel targets participating in the process of cytokinesis. Using POINeT can answer, at least in part, these two questions, i.e. identify the PPI network of the midbody. Besides PPI and network analysis, ranking/prioritization of nodes in networks may also contribute to the identification of novel components of the midbody proteome.

The mediators in the midbody interactome have been ranked using two measurements: the hub degree and the sub-network specificity score (S3). In order to evaluate these two scores, the top 30 proteins ranked by these two scores were listed and analyzed for their ability to enrich midbody-related proteins (Table 3). Figure 4 illustrates the results. Four types of proteins were considered to be putative midbody proteins 46 , including actin-related, cytokinesis-related, membrane-associated, and Rho proteins. The top 30 mediators ranked by S3 contain only 13% unknown proteins, with 87% putative midbody proteins, whereas the top 30 mediators ranked by degree centrality contain 63% unknown proteins, with only 37% putative midbody proteins. Figure 4 illustrates that the sub-network specificity score can effectively enrich the proportion of proteins highly related to the designated sub-network. The involvements of the S3 top-ranked genes in the midbody proteome have also been, at least in part, confirmed experimentally (manuscript in preparation). These results suggest that S3 could be employed to refine the midbody proteome, identify novel midbody proteins and rank these proteins, which may be complement to proteomics studies.

To demonstrate the PPI filtering capability of POINeT, a PPI network specific to liver has been constructed from 173 highly abundant proteins in mass proteomic data of liver 47 48 . Two gene expression profiles from the SymAtlas database were selected: liver and fetal liver. Figure 5 illustrates the results of the tissue-specific gene expression profile filtering. PPIs are shown in the figures when the gene expression levels of the two interacting proteins exceed the specified thresholds.

The threshold selection is dependent on the questions to be addressed. The networks filtered with liver and fetal liver gene expression profiles are largely similar, since two tissues represent the same tissue in different developmental stages. However, some minor differences can be noted. For example, with a higher threshold value of 16,384, it can be noted that interactions between HBA1 (hemoglobin alpha 1) and HBG2 (hemoglobin gamma 2) are less abundant in liver but prominent in fetal liver. It should be noted that this expression threshold is selected to reveal the differences in abundances of genes in adult/fetal livers. Users may set this threshold based on the questions to be addressed. Liver is responsible for the synthesis of hemoglobins in the fetus. In adults, the predominant forms of hemoglobins are composed of 2 beta chains and 2 alpha chains, whereas the fetal hemoglobins are composed of 2 gamma chains and 2 alpha chains. The fetal hemoglobins are replaced by adult hemoglobins after birth. Also, interactions between fibronectin 1 (FN1) and the other three genes, transferrin (TF), albumin (ALB) and apolipoprotein A-I (APOA1), are less abundant in fetal liver. This might be because the expression level of fibronectin 1 in fetal liver is lower than that in adult liver. Up regulation of fibronectin induces hepatic haematopoiesis during the second trimester 49 . Expression level of fibronectin may only become closer to that of the adult liver after that stage. Thus, tissue-specific expression profiles combined with PPI networks are able to capture the subtle differences between different tissues and the interactions therein.