Science labs often face a tradeoff between exploration and coordination. On the one hand, the uncertainty of research necessitates rapid, sometimes parallel, exploration of diverse project ideas; on the other hand, lab members who explore disparate projects can struggle to ensure their work advances the overall research direction of the team. Traditional top-down, hierarchical approaches to management can help with cohesion and coordination, but may unduly bottleneck exploration. Self-organized work may enable rapid parallel exploration, but may be bottlenecked by the lack of structure, context or granular credit for both individual project ideas as well as the overall landscape of exploration. How might members of a science lab self-organize their work in a way that advances the overall research direction of the lab?

To overcome these challenges, we present Lab Discourse Graphs, a collaborative knowledge management system designed to support effective self-organized team science. A Lab Discourse Graph is a shared lab notebook that includes a mechanism for representing individual project ideas as issue nodes in a knowledge graph of domain-specific semantic nodes like questions, hypotheses, results, and experiments, with associated metadata on creators/updaters of those nodes. From this base graph structure, team members can 1) browse a running query of project ideas on a graph-enabled issues board, 2) gain rich context for issues as well as their relationship to the overall research landscape the lab is exploring, via the network of links to related semantic nodes, and 3) receive granular credit attributions for contributions to larger research projects based on operations on individual nodes, such as claiming issues, or linking new evidence to existing issues.

We report on an in-depth case study of how Lab Discourse Graphs has enabled self-organized team science in a cell biology lab. Our findings include quantitative metrics from dozens of issue exchanges between researchers, including frequency of creation, claiming rates, and time-to-completion data, as well as qualitative case studies of several successful knowledge transfers. We describe how undergraduate researchers identified contextually-rich microprojects through the graph-enabled issues board and completed them with minimal direct supervision. For example, one undergraduate researcher independently selected an analysis task from the issues board, connected with another lab member who had bookmarked it, and successfully completed the project before the principal investigator was even aware of the interaction. Our findings further highlight how the system transformed idea documentation from a chore or potential competitive risk into a collaborative asset: researchers reported increased willingness to externalize their surplus ideas when the system explicitly tracked idea provenance and enabled revisiting and building on past ideas into larger research contributions.

Our findings demonstrate how structured knowledge graph representations can enable self-organized team science by maximizing intellectual resource utilization across experience levels and creating visible pathways for contribution and recognition. Our ongoing work focuses on scaling these graph-based collaboration tools to distributed scientific collaborations with varying degrees of shared context. We envision that this approach will support more fluid cross-laboratory collaborations while maintaining critical attribution structures that encourage open knowledge sharing.