New AI tool can generate novel scientific hypotheses and demonstrates capabilities in drug repurposing and proposing novel treatment targets.
Another tech giant has revealed new progress in the mission to harness the power of AI to advance biomedical research. Following the recent revelation that OpenAI’s new GPT-4b micro model had been developed with longevity biotech Retro Bio, Google just unveiled “co-scientist” – an AI tool designed to accelerate scientific discovery.
Designed as a virtual collaborator, the “multi-agent” system leverages new AI advances to assist biomedical scientists in tackling complex research challenges. By mirroring the reasoning process of the scientific method, Google claims its AI can generate novel hypotheses, refine research strategies and facilitate experimental design. All of which could reasonably be expected to help accelerate new discoveries in aging and age-related diseases.
Developed on Google’s Gemini 2.0 architecture, the co-scientist tool is structured around specialized agents that reflect core elements of the scientific method. The system employs a feedback-driven approach, continuously refining its outputs through a process of self-critique and iteration. It also integrates external tools, such as web searches and specialized AI models, to enhance the accuracy of its hypotheses.
Designed to be collaborative in nature, scientists can interact with the system by feeding it their own ideas or providing feedback on its outputs, making it an adaptable tool for researchers at various stages of inquiry, from initial hypothesis generation to experimental validation. The system’s ability to refine its reasoning over time is supported by its use of an Elo-based auto-evaluation metric, a method for calculating the relative skill levels of players in games like chess.
According to Google, the effectiveness of co-scientist has already been demonstrated through real-world applications in biomedical research, including drug repurposing and proposing novel treatment targets. In drug repurposing, it identified potential treatments for acute myeloid leukemia, with subsequent laboratory experiments confirming that the suggested compounds inhibited tumor viability in relevant cell lines.
In target discovery, co-scientist proposed novel epigenetic targets for liver fibrosis, with experimental validation supporting their therapeutic potential. The AI also showed its potential in antimicrobial resistance by independently identifying mechanisms that had already been uncovered in laboratory experiments, reinforcing its utility as an assistive technology.
While the progress is clearly exciting, Google acknowledges that co-scientist is still in its early stages, highlighting the need for improvements in the tool’s literature review capabilities, factual verification and more expert evaluations to strengthen its reliability.
