In the post-genomic era, detecting the function of gene products rises as a new interesting research area. Since a protein shows its specific function by interacting with other proteins in biological process, one way to detect the function of a protein is to look into whether it interacts with another protein whose function is revealed. Since a protein interaction is known as the result of physical interactions between domains of interacting proteins, the domain-domain interaction information provides decisive clues for predicting protein-protein interaction. In our previous research, we proposed that protein-protein interaction prediction system, PreDIN (Prediction-oriented Database of Interaction Network), using domain-domain interaction information. In this paper, we introduced the artificial neural network concept and propose a computational scheme that predicts interactions of protein pairs based on some biological features (e.g., gene expression profile or function category) of proteins as well as domain-domain interactions. We use only three referencing factors of determining protein interactions, but other biological features can be incorporated into our system as interaction determining referencing factors. The easy expandability is the strong points of our system and the augment of referencing factors enforces the predictable power of the system.