Many cellular proteins are oligomeric. In a high-throughput small angel x-ray scattering (SAXS) study, we found that 60% of pure proteins were homo-multimers in solution (Nature Methods 6:606-12. PMCID: PMC3094553). The number obviously gets higher for hetero-multimers. Interactions between different proteins can be inferred based on their phylogenetic distances, the existence of fused orthologues in other organisms, co-evolution, association methods based upon characteristic sequences or motifs, Bayesian network methods to predict likelihood of interactions by integrating data from experimental and computational predictions, and computational docking. These methods have false positives and negatives so can be useful in the design of experiments but cannot provide definitive results without experimental validation by methods such as SAXS, immunoprecipitation, or co-purification. This is not an easy problem, and if you man define rather than predict you will need to do experiments on the proteins in question.