Drug-drug interactions (DDIs) are a great concern to the selection of new drug candidates. While in vitro screening assays for DDI are a routine procedure in preclinical research, their interpretation and relevance for the in vivo situation still represent a major challenge. The objective of the present study was to develop a novel mechanistic modeling approach to quantitatively predict DDI solely based upon in vitro data. The overall strategy consisted of developing a model of the liver with physiological details on three subcompartments: the sinusoidal space, the space of Disse, and the cellular matrix. The substrate and inhibitor concentrations available to the metabolizing enzyme were modeled with respect to time and were used to relate the in vitro inhibition constant (K(i)) to the in vivo situation. The development of the liver model was supported by experimental studies in a stepwise fashion: (i) characterizing the interactions between the three selected drugs (R-bufuralol (BUF), bunitrolol (BUN), and debrisoquine (DBQ)) in microsomal incubations, (ii) modeling DDI based on binary mixtures model for all the possible pairs of interactions (BUF-BUN, BUF-DBQ, BUN-DBQ) describing a mutual competitive inhibition between the compounds, (iii) incorporating in the binary mixtures model the related constants determined in vitro for the inhibition, metabolism, transport, and partition coefficients of each compound, and (iv) validating the overall liver model for the prediction of the perfusate kinetics of each drug determined in isolated perfused rat liver (IPRL) for the single and paired compounds. Results from microsomal coincubations showed that competitive inhibition was the mechanism of interactions between all three compounds, as expected since those compounds are all substrates of rat CYP2D2. For each drug, the K(i) values estimated were similar to their K(m) values for CYP2D2 indicative of a competition for the same substrate-binding site. Comparison of the performance between the novel liver physiologically based pharmacokinetic (PBPK) model and published empirical models in simulating the perfusate concentration-time profile was based on the area under the curve (AUC) and the shape of the curve of the perfusate time course. The present liver PBPK model was able to quantitatively predict the metabolic interactions determined during the perfusions of mixtures of BUF-DBQ and BUN-DBQ. However, a lower degree of accuracy was obtained for the mixtures of BUF-BUN, potentially due to some interindividual variability in the relative proportion of CYP2D1 and CYP2D2 isoenzymes, both involved in BUF metabolism. Overall, in this metabolic interaction prediction exercise, the PBPK model clearly showed to be the best predictor of perfusate kinetics compared to more empirical models. The present study demonstrated the potential of the mechanistic liver model to enable predictions of metabolic DDI under in vivo condition solely from in vitro information.