From networks of protein interactions to networks of functional dependencies

Abstract

Background: As protein-protein interactions connect proteins that participate in either the same or different functions, networks of interacting and functionally annotated proteins can be converted into process graphs of inter-dependent function nodes (each node corresponding to interacting proteins with the same functional annotation). However, as proteins have multiple annotations, the process graph is non-redundant, if only proteins participating directly in a given function are included in the related function node.

Results: Reasoning that topological features (e.g., clusters of highly inter-connected proteins) might help approaching structured and non-redundant understanding of molecular function, an algorithm was developed that prioritizes inclusion of proteins into the function nodes that best overlap protein clusters. Specifically, the algorithm identifies function nodes (and their mutual relations), based on the topological analysis of a protein interaction network, which can be related to various biological domains, such as cellular components (e.g., peroxisome and cellular bud) or biological processes (e.g., cell budding) of the model organism S. cerevisiae.

Conclusions: The method we have described allows converting a protein interaction network into a non-redundant process graph of inter-dependent function nodes. The examples we have described show that the resulting graph allows researchers to formulate testable hypotheses about dependencies among functions and the underlying mechanisms.

Figure 1

From the protein-protein interaction network to the function nodes and the process graph. (A, D) When two (or more) interacting proteins (circles) within a PPI network share the same GO annotation, they originate a FN (rounded squares) by virtue of internal PPI (solid lines). In many cases (B, E), it may happen that a protein has more annotations (asterisk) and is therefore assigned to more FN. (C, F) Eventually, two FN, which share crossing PPI (dashed lines) and/or proteins, are linked in a PG.

Figure 2

Controlling annotation redundancy based on protein topology. (A) Schematic overview of the procedure adopted for retaining a given protein only in those FN that satisfactorily overlap (as assessed by the PMS) the topological structures to which the protein belongs. In the example shown in (B), the protein doubly annotated with the blue and green terms is initially included into the two relevant FN (rounded rectangles with dashed lines). Subsequently, however, the protein is retained in the blue node (but excluded from the green node), because (compared with the green node) the blue node overlaps better the 3-protein clique (i.e., the triangle), to which the protein belongs. As a result, in the final PG, no edge is established between the green and red nodes. In (C), the example of the Cta1p catalase is shown, while (D) shows the procedure of enucleating a function from a FN, which then undergoes relabeling. See also Additional file 3 for the distribution of FN and edges at different NTS.

Figure 3

The peroxisome process graph at high topological score. The PG shows the FN that represent peroxisome-specific functions and extra-peroxisomal functions. Specifically, core and neighbor FN were chosen because of their highly connected protein content, as reflected in NTS ≥ 30 (core) or NTS ≥ 60 (neighbors). See also Additional file 6 and Additional file 4, for a detailed analysis of the FN and the edges, respectively.

Figure 4

Examples of experimentally-testable hypotheses. (A) The general strategy and (B-D) different examples of experimentally-testable hypotheses derived from the PG of Figure 3 are shown. See also Additional file 7 for a list of selected DAG.

Figure 5

The budding-related process graphs. The PG shows the FN that represent functions related to the cellular bud (A) and to cell budding (B), two examples of a cellular component and a biological process in the budding yeast, respectively. See also Additional file 5 for a detailed analysis of the FN and the edges.