Advances in data collection and processing in biomedical research allow different data types to be measured on the same subjects, each representing different sets of characteristics, but collectively helping to explain underlying complex mechanisms. In some instances, phenotypic data are available. The main goal is to study the overall dependency structure among the data types, and to develop a model for predicting future phenotypes. We present a Bayesian factor analysis method that simultaneously models the overall association between data types using only relevant variables, while also predicting future outcomes using factor loadings. Through prior distributions, we incorporate structural information (e.g., biological networks) in our model that allows us to select functionally meaningful networks involved in the determination of factor loadings. We demonstrate the effectiveness of the proposed approach using simulations and observed data.