Two AI assistants generate drug‑retargeting hypotheses; one also evaluates experiment data

This week two papers reported AI systems that produced drug‑retargeting hypotheses and completed related tasks, with one system also capable of assessing biological experiment data. The result is notable because these tools move beyond text generation toward working alongside laboratory workflows, potentially speeding hypothesis generation and prioritization for researchers. One system, Google Co — Scientist, is presented as a 'scientist in the loop' tool that operates under researcher direction rather than replacing human judgment. The authors describe it as a collaborative assistant that generates hypotheses for drug repurposing while keeping scientists central to experimental decision‑making and interpretation.

The other paper describes work from nonprofit FutureHouse, which trained a model able to evaluate biological data from certain classes of experiments in addition to proposing hypotheses. That capability allows the model to consider experimental readouts when forming and assessing retargeting ideas, extending its role from ideation to preliminary data appraisal.

Both systems use agentic designs that call external tools in the background to carry out tasks such as querying databases or running analyses. The papers note this approach aligns with similar engineering choices elsewhere in the field: Microsoft has deployed comparable agentic strategies, and OpenAI has experimented with tuning large language models for related workflows.