What this is
BioResearch Trend follows biotech and biomedical research as it appears in PubMed. It shows what was published this month, and how research areas have moved over the 33 months it has collected (123,399 papers so far).
Where the data comes from
Every paper comes from PubMed, the U.S. National Library of Medicine's database. Seven topic searches, built on MeSH headings (drug development, advanced therapeutics, clinical and disease, AI and genomics, metabolism and aging, drug delivery, targeted small molecules), select the papers. The site keeps each paper's title, journal, authors, abstract, keywords and MeSH headings, and links every paper back to PubMed.
How it is updated
- Every day, the last three days of new papers are fetched. The month so far is rebuilt as a knowledge graph of papers and the concepts they share, grouped into research communities, and the pages are rebuilt.
- On the 1st of each month, the month just ended is built as a whole into the monthly graph behind the Community Timeline and the long-run trends.
Papers are filed by the month they were collected, not their publication date. Until March 2026 months were collected in bulk, well after publication; since April 2026 papers are collected daily, before PubMed has finished indexing many of them. Trend pages note where that change shows.
What the model does
gpt-4o-mini extracts each paper's concepts from its abstract, names the research communities, scores papers as drug development or basic research, and writes the key findings and trial readouts. Every sentence of a finding or readout is checked against the abstracts it cites. Evidence levels, trial phases and disease areas come from PubMed's publication types and MeSH, not from the model. See the Disclaimer for what that means for accuracy.
The BioLLM Wiki's keyword entries are written by gpt-5.6-luna from a sample of each keyword's papers, checked the same way, and rewritten on the 1st of each month.
Built with
Graphify and NetworkX (Louvain community detection) for the graphs, NCBI E-utilities for PubMed, and Astro for the site.