The project starts from a familiar bottleneck in research: there is simply too much literature to review paper by paper. To help researchers scan broad patterns, Boyu Xu uses topic-based exploration that highlights how concepts are linked across many publications. If two topics are mentioned together - for example, a brain region and a disorder appearing in the same statement in a title or abstract - this counts as a direct relation in the literature.
Indirect relations
Xu focuses on indirect relations: cases where two topics do not show up together directly, but both connect to a third “bridge” topic such as a gene or a mental process. These bridge-based links can be useful early signals: they do not prove a relationship, but they can point researchers towards combinations that may be worth a closer look.
Building intuition
AR is a type of extended reality (XR) that overlays digital information onto the physical world. Xu uses AR because brain anatomy and many neuroscientific concepts are inherently three-dimensional. Seeing regions and their connections in 3D space can help researchers build intuition, while still keeping their regular desk workflow - reading papers on a screen and using the surrounding space to compare multiple relationships at once.
Bridge topics
Xu built on an existing 3D AR prototype called DatAR and extended it with features co-designed and evaluated with neuroscientists. The system supports three key tasks: finding indirect relations via bridge topics, seeing how those links develop over time (for example, when an indirect hint later becomes a direct relation in the literature), and drilling down into the specific bridge topic that connects two areas.
Under the hood, the tool relies on a large, structured map of topics and links extracted from neuroscience papers in PubMed (2010 - early 2022). In practice, the system looks at titles and abstracts and records when two topics are mentioned together in the same sentence, for instance, a brain region named in the same sentence as a gene or condition. Those repeated “mentioned-together” links form the backbone of the dataset Xu explores in AR.
Support tool
Xu positions this as a support tool for early-stage experimental design. A way to narrow down where to read more deeply, generate candidate hypotheses, and identify possible variables (such as genes or mental processes) that could be relevant. Even when no single paper spells out the connection yet.
