ESCARGOT Proves Agentic AI Is the Future of Biomedical Research

Cedars-Sinai scientists unveil a powerful agentic AI model, ESCARGOT, that harnesses multiple intelligent agents and biomedical knowledge graphs to deliver sharper, faster, and more reliable answers than ChatGPT in Alzheimer’s research.

Algorithm flow chart (above) describing ESCARGOT’s approach to strategize, create python executable code, convert to machine readable XML code, deploy the Graph of Thoughts, and return the output.

According to Jason Moore, PhD, chair of the Department of Computational Biomedicine at Cedars-Sinai, agentic AI is the hottest trend in artificial intelligence (AI).

Unlike traditional AI, which is primarily designed to complete a single task, agentic AI is a new generation of AI models that can independently perform multiple tasks simultaneously to achieve specific objectives.

At Cedars-Sinai, Moore and colleagues are immersed in agentic AI models, creating algorithms that make faster and more accurate decisions while combing through large datasets. Moore, a professor of Computational Biomedicine and Medicine, sat down with the Cedars-Sinai Newsroom to explain the potential and recent boom in the use of agentic AI.

What is agentic AI, and how does it differ from existing AI models?

For the past 10 years, we have been developing state-of-the-art AI methods, including deep-learning algorithms and large language models for natural language processing. These methods have been designed to complete one specific task, such as analyzing an echocardiogram image of the heart to find defects.

Agentic AI, however, assembles teams of AI specialists to complete specific tasks-then collectively brings these teams together to solve a complex problem. Using the same echocardiogram example, an agentic AI model can simultaneously analyze echocardiogram images, laboratory tests, vital signs, medication history, and clinical notes to provide a comprehensive picture of a patient in a fraction of the time it would take multiple clinicians to review results.

This technique mirrors the way humans solve complex problems.

What makes agentic AI the next hottest trend in AI?

ChatGPT has shown that we can use powerful algorithms for specific tasks. With agentic AI, algorithms are tailored to specific needs and adapt strategies independently to achieve predefined goals. It can assemble teams of AI agents to handle various tasks.

In my laboratory, for example, we work with big data. So, we need people whose expertise is in cleaning data, preparing it for analysis, building computational models with the data, and providing statistical analysis. We also need people who can interpret the data for us; what does the data tell us about biology, clinical care, etc.? We then need someone to summarize all of these results in written form and prepare graphs and figures to communicate these results.

Agentic AI builds teams of AI agents that handle each of these respective areas, with the goal of providing understanding of the data and explaining the results.

The field is advancing so that individuals may soon use these methods at home. I expect to see many tools coming out in the next year or so that will make our lives easier. One can imagine an AI agent helping prepare your taxes, family budget, or weekly grocery list.

Is there published research happening in agentic AI?

We are seeing an uptick in published research studies involving agentic AI. Our laboratory recently published a study in Bioinformatics about an agentic AI model we created called ESCARGOT (Enhanced Strategy and Cypher-driven Analysis and Reasoning using Graph Of Thoughts).

The ESCARGOT model combines large language models with a dynamic "graph of thoughts" and biomedical knowledge graphs- an approach that improved output reliability and reduced inaccuracies. To do this, we input existing data we have procured about Alzheimer's disease, then asked the agentic AI model to provide several things: genes associated with the disease, drugs and therapies that may offer the best treatments for these genetic variations, etc.

We compared these findings to the responses ChatGPT produced, and, not shockingly, agentic AI provided answers with 80%-90 % accuracy, compared to ChatGPT, which scored about 50%.

We strongly believe in making our models open-access to ensure science progresses. The ESCARGOT model is public, free, and available on GitHub.