Food contamination is always a serious issue, due to its severe effects on humanhealth, starting from the early times. Environmental monitoring and pathogen detection in food processing facilities are key tools to reduce the risk of food contamination. The key challenge is the high cost in sampling and testing the entire facility on a regular basis. As a result, there is a need for sampling strategies that offers reliable pathogen detection with a small set of samples. In this thesis, we develop efficient sampling strategies for pathogen detection. We leverage recent development in digital-twin models that capture the rich interactions, e.g., the cross-contamination patterns, across various components (referred to as agents) of a food processing facility. Such a digital-twin model gives a graph representation of the cross-contamination interactions and corrective measures (e.g., cleaning processes). Our technical approach is to exploit this graph representation to identify information-critical agents for efficient sampling and detection strategies. We are especially interested in sequential sampling strategies that learn from past test outcomes and improve performance as data accumulating over time. We also leverage various centrality measures developed in network science to identify most important/influential agents. We conclude with a discussion of future research directions in terms of alternative solution methods such as deep reinforcement learning and group testing.