<p>Process mining enables the discovery and analysis of business processes from event logs. Once the log has been collected and preprocessed, a typical initial step is to automatically discover a process model. However, insight generation often requires analyzing and comparing multiple process models derived from systematically filtering the event log into several subsets (e.g., based on attribute values). Despite being common practice, current process mining tools lack effective support for interactively handling such collections of process models, making the process time-consuming and error-prone. In this paper, we argue that collections of process models should be treated as first-class objects and investigate what functionalities software tools should offer to support user-driven analysis workflows over these collections. Based on a qualitative study of visual query tools, domain-specific query languages, and real-world process mining reports, we identify three key fundamental tasks (context creation, results visualization, and pattern search) and derive 14 functional requirements to support them. To validate the feasibility and utility of these requirements, we developed VISCoPro, a prototypical implementation that facilitates the interactive handling of collections of Directly-Follows Graphs through filtering, grouping, abstraction, and pattern-based querying. We demonstrate the approach through a real-world use case and a user study with 18 participants, providing evidence of its usefulness, ease of use, and capacity to support real-world exploratory analysis workflows.</p>