We define the reliability score S of an interactions between proteins p_i and p_j as a weighted sum of three subscores of domain (S_d), function (S_f), and sub-cellular localization (S_l). This scoring method is more general than a typical method of selecting co-functional or co-localized protein pairs since some pairs of proteins interact frequently even they are not co-functional or co-localized in a cell.

(1)

The domain score (S_d) is defined by equation 2, and the information of domain-domain pairs can be extracted from InterPro, which are determined by the information of BIND, DIP, and HPRD.

(2)

where ∑I(d_i,d_j) is the number of interactions between domains d_i and d_j, and ∑d_i and ∑d_j are the number of domains d_iand d_j, respectively, participating in the interactions.

For the function score (S_f), we first assign the Gene Ontology (http://www.geneontology.org/GO.doc.html) molecular functions to all human proteins in HPRD, and then compute the matrix M_GO (equation 3), where each entry M_GO represents the ratio of the number of the interactions between GO functions f_iand f_j to the number of proteins with function f_i or f_j, or both.

(3)

Suppose that protein p_a has functions f₁, f₂, and f₃, and protein p_b has function f₁ and f₄, and protein p_c has function f₄ (see Figure 1). If there are interactions (p_a, p_b) and (p_b, p_c), then M_GO (f₁, f₄)=1/3 since there is a single interaction between functional groups f₁ and f₄ and there are three proteins with function f₁ or f₄.

The function score (S_f) was computed using equation 4, in which k is the number of M_GO entrieswith nonzero values.

(4)

The sub-cellular localization (S_l) is computed using equation 5, in whichSI_l(p_i, p_j) is the number of interactions between proteins p_i and p_j in a same compartment l, and N is the number of proteins that participate in protein-protein interactions. Interactions with S_l=0 are cases in which either source or target protein has no sub-cellular localization information.

(5)

Figure 1. Example of interpreting protein interactions as molecular function interactions