Building evidence-based knowledge graphs from full-text literature for disease-specific biomedical reasoning

Abstract

Biomedical knowledge resources often either preserve evidence as unstructured text or compress it into flat triples that omit study design, provenance, and quantitative support. Here we present EvidenceNet, a framework and dataset for building disease-specific knowledge graphs from full-text biomedical literature. EvidenceNet uses a large language model (LLM)-assisted pipeline to extract experimentally grounded findings as structured evidence nodes, normalize biomedical entities, score evidence quality, and connect evidence records through typed semantic relations. We release two resources: EvidenceNet-HCC with 7,872 evidence records, 10,328 graph nodes, and 49,756 edges, and EvidenceNet-CRC with 6,622 records, 8,795 nodes, and 39,361 edges. Technical validation shows high component fidelity, including 98.3% field-level extraction accuracy, 100.0% high-confidence entity-link accuracy, 87.5% fusion integrity, and 90.0% semantic relation-type accuracy. In downstream evaluation, EvidenceNet improves internal and external retrieval-augmented question answering and retains structural signal for future link prediction and target prioritization. These results establish EvidenceNet as a disease-specific resource for evidence-aware biomedical reasoning and hypothesis generation.

Figures (5)

• Figure 1: Overview of the EvidenceNet workflow. The pipeline proceeds through four stages---data preprocessing, LLM-driven evidence extraction, normalization and scoring, and integration and graph construction---to convert full-text biomedical literature into evidence nodes linked to normalized entities and cross-paper semantic relations.
• Figure 2: Representative transformation of a literature statement into a graph-native evidence object. A colorectal cancer example is shown from source text to a structured evidence record and then to its graph representation, illustrating how provenance, study context, quantitative outcomes, normalized entities, and typed evidence relations are retained in the released schema.
• Figure 3: Global overview of the released HCC and CRC EvidenceNet resources. Nodes represent filtered evidence records and are coloured by study design. Edges represent cross-paper semantic relations and are coloured by relation type. Both diseases show a densely connected backbone together with smaller peripheral modules. Very small disconnected components are omitted for visual clarity.
• Figure 4: Representative local motifs in the HCC and CRC EvidenceNet resources. The four panels show example subgraphs selected from the full evidence graphs, including contradiction-rich neighbourhoods and more specialized peripheral modules. These motifs illustrate how EvidenceNet supports both global inspection and fine-grained exploration of disease-relevant evidence patterns.
• Figure 5: Quantitative overview of the released HCC and CRC EvidenceNet resources. (a) Annual evidence-record counts by source year. (b) Study-design composition of evidence records. (c) Composite-score distributions across records. (d) Composition of semantic relation types among evidence--evidence edges.