Background Bedaquiline is classified as group A by WHO, to be used for the first-line treatment of drug-resistant tuberculosis (DR-TB). It is metabolized primarily by the cytochrome P450 3A4 to a 4- to 6-fold less-active N-monodesmethyl metabolite (M2). To individualize the bedaquiline dose regimen, a quantitative relationship between the exposure of bedaquiline and the antimycobacterial activity in the infection site must be revealed. Method A physiologically based pharmacokinetic (PBPK) model for bedaquiline and M2 was established and validated using different clinical datasets. A population pharmacokinetic-pharmacodynamic (PK/PD) model for bedaquiline, including minimal inhibitory concentration (MIC) and the parameters of the host immune status, was developed based on the whole-blood bactericidal activity (WBA) in ex vivo culture. Results The developed PBPK model successfully characterized the typical trends of the available bedaquiline and M2 data. The population PK/PD model was also well validated. The two models were thereafter combined to predict the time course of bacterial killing in the lung - the infection site. The resulting PBPK/PD model allowed a 1000-subject Monte-Carlo simulation to determine the proportion of subjects who would keep WBA below zero, which is needed for bactericidal activity. For bedaquiline doses of 200, 400, and 600 mg/day, their target attainment probability was provided according to MIC. Conclusion Our analysis outlines how PBPK/PD modeling may be used to design and individualize tuberculosis treatment regimens.