In this paper, we present an interdisciplinary computational framework towards modeling and integrating recurrent biological interaction network in time and space dimensions by applying dynamic Bayesian methods to a set of biological qualitative hypotheses. Our approach uses a previously proposed qualitative knowledge model to translate qualitative hypotheses into a set of constraints which restrain the uncertainty of dynamic Bayesian models. The biological entities at different abstract levels are combined hierarchically into a single network and the complementary molecular interaction networks in space-time dimension can be integrated consistently into a uniform representation. Quantitative in-silico inference is performed by model averaging with Monte Carlo simulation. We apply our method to model the TGFβ-Smad signaling pathway in the breast cancer bone metastasis by integrating independent models of the signaling pathway and the breast cancer bone metastasis network. We show that our method can integrate a set of complementary Bayesian models consistently and produce reasonable quantitative predictions.