EvidenceNet: Building Evidence-Based Knowledge Bases from Full-Text Literature for Disease-Specific Biomedical Reasoning

This repository contains the public data release for EvidenceNet, a disease-specific evidence-centric biomedical knowledge resource for hepatocellular carcinoma (HCC) and colorectal cancer (CRC). The release includes four components: 1. Record-level structured evidence datasets in records/, which are the primary reusable datasets underlying the study. Each evidence record contains source metadata, PICO-style study context, extracted biomedical entities, quantitative statistics, evidence-quality scores, supporting source text, and semantic relations to other evidence records. 2. Graph-level serializations in graphs/, providing network representations derived from the record-level datasets. 3. Source article metadata tables in source_metadata/, listing the unique literature sources represented in the released datasets, including article-level provenance fields and the number of evidence records derived from each source. 4. Derived evaluation outputs in evaluation/, including validation, QA, link prediction, and target prioritization results used in the project. Release scope: - HCC: 7,872 evidence records, 10,328 graph nodes, 49,756 graph edges, 470 unique source articles - CRC: 6,622 evidence records, 8,795 graph nodes, 39,361 graph edges, 472 unique source articles Field-level documentation for all released files is provided in SCHEMA.md.