Predicting the presence of major depressive disorder (MDD) using speech is highly non-trivial. The heterogeneous clinical profile of MDD means that any given speech pattern may be associated with a unique combination of depressive symptoms. Conventional discriminative machine learning models may lack the complexity to robustly model this heterogeneity. Bayesian networks, however, are well-suited to such a scenario. They provide further advantages over standard discriminative modeling by offering the possibility to (i) fuse with other data streams; (ii) incorporate expert opinion into the models; (iii) generate explainable model predictions, inform about the uncertainty of predictions, and (iv) handle missing data. In this study, we apply a Bayesian framework to capture the relationships between depression, depression symptoms, and features derived from speech, facial expression and cognitive game data. Presented results also highlight our model is not subject to demographic biases.