{"paper":{"title":"EpiGraph: Building Generalists for Evidence-Intensive Epilepsy Reasoning in the Wild","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"A new epilepsy knowledge graph boosts LLM performance on clinical reasoning tasks by up to 41 percent.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Jathurshan Pradeepkumar, Jimeng Sun, Yasuko Matsubara, Yasushi Sakurai, Yushun Dong, Yuyang Dai, Zheng Chen","submitted_at":"2026-05-10T12:27:32Z","abstract_excerpt":"Epilepsy diagnosis and treatment require evidence-intensive reasoning across heterogeneous clinical knowledge, including biosignal patterns, genetic mechanisms, pharmacogenomics, treatment strategies, and patient outcomes. In this work, we present \\textsc{EpiGraph}, a large-scale epilepsy knowledge graph and benchmark for evaluating knowledge-augmented clinical reasoning. \\textsc{EpiGraph} integrates 48,166 peer-reviewed papers and seven clinical resources into a heterogeneous graph containing 24,324 entities and 32,009 evidence-grounded triplets across five clinical layers. Built upon this gr"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"integrating EpiGraph consistently improves performance across all tasks, with the largest gains observed in pharmacogenomic reasoning (+30--41%).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the automatically extracted triplets and five-layer structure accurately capture clinically reliable knowledge without introducing systematic errors or omissions from the source papers.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EpiGraph creates a heterogeneous epilepsy knowledge graph that boosts LLM performance on clinical reasoning tasks by 30-41% in pharmacogenomics when used with Graph-RAG.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A new epilepsy knowledge graph boosts LLM performance on clinical reasoning tasks by up to 41 percent.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"be3f10b2b69c977f714a91517ed6abd523ca5987135d50a7ebbda265c9bf4bc9"},"source":{"id":"2605.09505","kind":"arxiv","version":2},"verdict":{"id":"86230406-e2b5-4ad3-8a42-63dd5ee671b1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T21:20:48.007939Z","strongest_claim":"integrating EpiGraph consistently improves performance across all tasks, with the largest gains observed in pharmacogenomic reasoning (+30--41%).","one_line_summary":"EpiGraph creates a heterogeneous epilepsy knowledge graph that boosts LLM performance on clinical reasoning tasks by 30-41% in pharmacogenomics when used with Graph-RAG.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the automatically extracted triplets and five-layer structure accurately capture clinically reliable knowledge without introducing systematic errors or omissions from the source papers.","pith_extraction_headline":"A new epilepsy knowledge graph boosts LLM performance on clinical reasoning tasks by up to 41 percent."},"references":{"count":66,"sample":[{"doi":"","year":2025,"title":"Evobrain: Dynamic multi-channel EEG graph modeling for time-evolving brain networks","work_id":"2f700eec-3c27-447e-8c6f-2418e4663365","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2011,"title":"Drug-resistant epilepsy","work_id":"63bdc733-4fd3-4ba2-96a1-15da50de0a7b","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Long-term eeg partitioning for seizure onset detection","work_id":"46c6f1d6-6b32-4f73-bdcd-26f31e32866d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Review of pharmacogenetics of antiseizure medications: focusing on genetic variants of mecha- nistic targets","work_id":"ff1d9200-7592-4e35-b3f7-be57b8a77717","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"A review on knowledge graphs for healthcare: Resources, applications, and promises","work_id":"41e00ad7-c530-4d1c-bf16-d097ae5e7432","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":66,"snapshot_sha256":"75c22219f158aff05da13eaa5abe8250b00b7fec7a9c6035eb38b30a034879fc","internal_anchors":12},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8e08aed194b6709bbcaf4b7ce3740103c10862055a2c0f6cec3ca82d73ae5adb"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}