A Knowledge Graph Framework for Systematic Collaboration Intelligence in Biomedical Research Portfolios

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Collaboration discovery in large publicly funded research portfolios depends on fragmented information distributed across project records, deliverables, and personal networks. We present a knowledge graph framework that makes collaboration-relevant structure in funding portfolios systematically observable and queryable. Starting from public funding records, the graph is enriched with disambiguated organisations, linked publications, researcher identifiers, and data assets recovered from project documents via large language models.Applied to 235 biomedical research projects from the Innovative Health Initiative, the resulting portfolio graph reveals both dense organisational connectivity and distinct thematic structure. We show that structural pathways and thematic alignment provide complementary signals for collaboration brokerage. We evaluate lexical, graph-structural, and embedding-based scoring configurations against a structural pseudo-benchmark and show that hybrid approaches outperform individual channels. Document-level extraction recovers hundreds of hidden data assets absent from structured metadata, and a cross-portfolio comparison with the German Research Foundation highlights structural differences between public-private and academic-led funding models. Our work demonstrates that a portfolio graph built from public records can support systematic brokerage of data collaboration and research interest at scale, with implications for research information management and science-of-science indicator design.