Evidence-Grounded Tree-Based Hypothesis Generation for Scientific Discovery

AI scientists are advancing rapidly, yet most entangle evidence management with reasoning, limiting hypothesis generation to narrow, prior-driven contexts. To enable the sustained reasoning chains required for scientific ideation, we propose the Evidence-Grounded AI Scientist (EGAS), which strictly separates evidence curation from reasoning. A Context Optimizer curates and grounds evidence without interpretation, enabling a Tree-of-Thoughts Hypothesis Generator to explore diverse reasoning paths and produce novel, falsifiable hypotheses. Experimental outcomes feed back into the system, with a Diagnostician that flags anomalous signals before they corrupt downstream interpretation. This growing body of validated evidence improves subsequent ideation cycles. As a proof of concept, we demonstrate this closed-loop architecture as an integrated member of a drug discovery team, from target suggestion through wet-lab validation.