Microbial interactions are ubiquitous in nature, and are equally as relevant to human wellbeing as the identities of the interacting microbes. However, microbial interactions are difficult to measure and characterize. Furthermore, there is growing evidence that they are not fixed, but dependent on environmental context. We present a novel workflow for inferring microbial interactions that integrates semi-automated image analysis with a colony stamping mechanism, with the overall effect of improving throughput and reproducibility of colony interaction assays. We apply our approach to infer interactions among bacterial species associated with the normal lung microbiome, and how those interactions are altered by the presence of benzo[a]pyrene, a carcinogenic compound found in cigarettes. We found that the presence of this single compound changed the interaction network, demonstrating that microbial interactions are indeed dynamic and responsive to local chemical context.