video1.0<iframe src="https://www.loom.com/embed/7facea5f333440d1be07d5f6a40dd3df" frameborder="0" width="1268" height="951" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>9511268Loomhttps://www.loom.com9511268https://cdn.loom.com/sessions/thumbnails/7facea5f333440d1be07d5f6a40dd3df-938e52dd6fc8c243.gif258.432This Loom presents CardoGraph, a system for mapping SARS-CoV-2 protein interactions to identify missing edges and generate testable mechanistic hypotheses. The author explains that large “history” maps of 332 contacts are hard to read, so their model reads the graph, proposes an interaction path, and retrieves deposited structural and coordinate evidence down to atomic details such as ORF6 wedged into RAE1 at 2.8 angstrom. In evaluation, it achieved Precision at 24 of 45%, ROC 0.845, and found that adding a structural evidence channel did not improve results, while using conservation across SARS-1 and MERS increased precision at 10 from 30% to 60%. The tool can operate on new organisms and maps, with Claude writing mechanisms only from citations retrieved and vetted against the fetched papers.