Los Angeles-based Cedars-Sinai has developed three free, publicly available software tools designed to build, modify and monitor agentic artificial intelligence systems used in biomedical research.
The health system announced the tools Nov. 9, saying they aim to improve transparency in AI-driven scientific research by giving investigators greater visibility into how AI agents perform tasks and generate results.
The tools, called ASAREE, Memstara and GENASYS, are available online at no cost and can be used and adapted by researchers and other members of the public.
Jason Moore, PhD, chair of the Department of Computational Biomedicine at Cedars-Sinai, said commercial agentic AI systems often lack transparency, raising concerns about the reliability of their results.
“Agentic AI systems are complicated,” Dr. Moore said. “They must choose and organize AI agents — programs that can autonomously complete complex tasks — to do the work, and then also critique the work, perform quality control and evaluate ethics.”
“Our software aims to combat this lack of transparency,” Dr. Moore said.
The three software packages serve different functions in developing and refining agentic AI systems.
ASAREE, short for Analytical Sandbox for Agentic Research, Engineering, and Experimentation, allows researchers to test hypotheses about AI workflows by evaluating different approaches. For example, investigators can use the tool to determine whether a particular software configuration improves workflow reliability.
Memstara uses existing knowledge about relationships among genes, diseases and biomarkers to generate biological hypotheses that can be tested against real-world biomedical data.
GENASYS, short for Genetic Evolution of Networked Agent Systems, is designed to improve AI systems after they have been created. It automatically modifies agentic AI workflows, evaluates alternative configurations and selects those that perform best for a specific biomedical task.
Although ASAREE and GENASYS were developed for biomedical applications, Dr. Moore said they could also be applied to research in engineering, economics and other scientific fields. Dr. Moore said Memstara is currently more specialized because it relies on a specific knowledge base, though it could be adapted to work with other knowledge bases.
Cedars-Sinai researchers hope making the tools publicly available will also allow outside users to contribute ideas that inform future improvements to the software.
