What it is
Biomedical research is constrained by the time and expertise needed to analyse ever larger and more complex datasets. The authors present OpenScientist, an open-source agentic AI co-scientist that investigates questions a scientist defines and produces verifiable, data-grounded analyses: it researches the literature, generates hypotheses, executes analyses and reports insights. Domain experts evaluated it across 5 biomedical case studies. It accurately quantified how well Alzheimer's blood biomarkers perform, generated a plasma proteomic survival model that predicted survival in a hidden evaluation cohort, proposed a reproducible mechanism linking tau pathology to altered lysosomal acidification, generated multiple myeloma hypotheses that were validated in an external cohort with an attenuated signal in randomized controls, and agreed with an independent human analysis in a blinded, prespecified proteomic comparison.
Why it matters
Analyses the authors say would otherwise take months of human time and expertise were completed in minutes, and domain experts judged each case study's output. It can also run in a fully air-gapped environment with open-source models, which the authors offer to support research on sensitive clinical datasets, and they frame it as assisting biomedical data analysis in concert with human experts.
Underlined numbers link to their source. Every metric and quoted figure is listed under Sources and data below.
Filed underArtificial Intelligence in Healthcare and Education, Artificial Intelligence in Healthcare, Machine Learning in Healthcare