What was the challenge?
The Institute's researchers work with published studies from NCBI's Gene Expression Omnibus (GEO), many of them ribosome profiling (Ribo-seq) experiments. Each study, and every sample within it, describes its species, tissue, cell lines, genotype and treatment in free text and in its own way. Comparing experiments across studies needed that information in one consistent, structured form.
What did Druid Learning do?
- 1
Took 1,035 studies from NCBI GEO, linked to 660 published research papers, and every sample within them: 20,000+ research samples in total.
- 2
Extracted study-level data from each study: species, organ, tissue and cell type, cell lines, intervention, experimental design and publication date, plus a one-line summary and the study's key insight.
- 3
Extracted sample-level data for every sample: Control or Test classification, genotype, treatment, a one-sentence summary and condition tags.
- 4
Scored every extracted field for confidence, re-ran low-confidence results, and cross-checked each study's sample list against GEO.
What were the results?
- 1,000+ genomics studies and 20,000+ research samples turned into structured data.
- 100k+ structured data points generated, each with a confidence score.
- The structured data now feeds the Institute's deep research and knowledge discovery workflows.
Want the full details?
This is a summary of our work with Netherlands Cancer Institute. Talk to us for the full story, including the approach, the integrations and how it could apply to your content.
