SPECTRA: An Integrated Knowledge Base for Comparing Tissue and Tumor-Specific PPI Networks in Human

Abstract

Protein-protein interaction (PPI) networks available in public repositories usually represent relationships between proteins within the cell. They ignore the specific set of tissues or tumors where the interactions take place. Indeed, proteins can form tissue-selective complexes, while they remain inactive in other tissues. For these reasons, a great attention has been recently paid to tissue-specific PPI networks, in which nodes are proteins of the global PPI network whose corresponding genes are preferentially expressed in specific tissues. In this paper, we present SPECTRA, a knowledge base to build and compare tissue or tumor-specific PPI networks. SPECTRA integrates gene expression and protein interaction data from the most authoritative online repositories. We also provide tools for visualizing and comparing such networks, in order to identify the expression and interaction changes of proteins across tissues, or between the normal and pathological states of the same tissue. SPECTRA is available as a web server at http://alpha.dmi.unict.it/spectra.

Keywords: database; interactions; proteins; tissue; tumor.

Figure 1

Venn diagram showing the number of common interactions across PPI datasets in SPECTRA.

Figure 2

Log–log plot of degree distribution of the final integrated PPI network in SPECTRA.

Figure 3

Venn diagrams depicting: (A) the number of common tissues and (B) the number of common cancer types across expression datasets in SPECTRA. The zero overlap among the cancer datasets is due to the fact that cancers are considered according to their specific names and not to their class (breast cancer, prostate cancer, etc.).

Figure 4

General description of GASOLINE.

Figure 5

SPECTRA search tabbed panel. Red boxes highlight the three sections: (A) “Gene data,” (B) “Expression data,” and (C) “Interaction data.” In this case, the parameters have been set to indicate that we want to retrieve all the interactions that are present at least in Havugimana and HPRD, involving genes that are expressed in adrenal gland” tissue according at least to GDS3113 and ProteinAtlas. In this example, we neither restrict our search to a predefined set of genes nor provide a threshold for interaction weights, dataset coverage, and expression scores.

Figure 6

Result table for the query of Figure 5. For each interaction in the table we report, the tissue, the average expression scores of interacting genes, and the total interaction weight. Expression scores, weights, and dataset coverage are represented with a colored progress bar (from cyan to red). By selecting a row (in the example the interaction between NCBP2 and NCBP1), detailed data about the interaction are shown to the right. For each dataset, the corresponding interaction weight (when available) is reported (for example, 0.71 for IntAct database).

Figure 7

The panel with detailed information of a gene. When an interaction is selected from the result table (Figure 6), two panels with additional data, one for each interacting gene, are shown. This example refers to the detailed panel for gene NCBP2, which appears when the row table of Figure 6 is selected. In the detailed panel, the gene symbol, the description, the corresponding ID in Entrez Gene database (when available), and aliases (including references in other databases) are reported. Finally, two tables with the set of tissues and tumors where the gene is expressed are shown. These are shown in decreasing order with respect to the average expression scores.

Figure 8

The SPECTRA compare panel. In this example, we first loaded 4 different TS-PPI networks from files using the “Add networks” button. Then by clicking on “Run Gasoline” the form for the selection of the adapted GASOLINE input parameters appears.

Figure 9

Result table for the differential local alignment of the four TS-PPI networks of Figure 8 with the Adaptive GASOLINE. The table reports, for each alignment, the size (i.e., the number of aligned nodes), the average expression difference between aligned nodes, and the ISC (Index of Structural Conservation) score. When the user selects a row in the table, a panel with alignment details is shown to the right. Details include the list of aligned subnetworks (defined by the set of nodes and edges) and the mapping between aligned nodes. Nodes of aligned networks are represented by the corresponding ids, followed by their weights, while edges are represented by the ids of interacting proteins, followed by the interaction weights and the corresponding tissues. Alignment mapping is represented as a matrix where rows contain aligned proteins and columns represent nodes of the same subnetwork.

Figure 10

The Network visualization in SPECTRA. (A) A TS-PPI network for a single tissue; (B) A TS-PPI network for multiple tissues. In this case, nodes are represented as pies with slice sizes proportional to the expression of corresponding gene in a tissue. Nodes and edges are colored according to the corresponding tissue and node dimensions are proportional to the total gene expression score.

Figure 11

The two biggest local differential alignments found by the Adaptive GASOLINE for the TS-PPI networks of normal breast cells (grade 0), well-differentiated cells (grade 1), moderately differentiated cells (grade 2), and poorly differentiated cells (grade 3). (A) A complex of chemokine proteins. (B) The Human Leukocyte Antigen (HLA) system. Nodes and edges are colored according to the corresponding network. Edge widths are proportional to the strength of interaction. Node dimensions are proportional to the gene expressions. Solid lines (intra-edges) connect the nodes of the same network, while dashed lines (inter-edges) connect the aligned nodes.