A 27B Gemma‑based model trained on transcriptomics and bio text hypothesized that inhibiting CK2 (via silmitasertib) would enhance MHC‑I antigen presentation—making tumors more visible to the immune system. Yale labs tested the prediction and confirmed it in vitro, and are now probing the mechanism and related hypotheses.
— If small, domain‑trained LLMs can reliably generate testable, validated biomedical insights, AI will reshape scientific workflow, credit, and regulation while potentially speeding new immunotherapy strategies.
Steve Hsu
2025.12.02
70% relevant
This prior idea records LLMs producing testable, validated scientific hypotheses; Hsu's report extends that pattern into theoretical physics (GPT‑5 originating the main idea), showing the phenomenon is cross‑disciplinary and not limited to biomedical lab leads.
Alexander Kruel
2025.10.16
100% relevant
Google’s report that the Yale team validated the model’s CK2→MHC‑I prediction and is expanding testing of AI‑generated hypotheses.
← Back to All Ideas