Dr. Olivier Elemento, the director of the Englander Institute for Precision Medicine at Weill Cornell Medicine, has been awarded an NIH Director's Pioneer Award to build an artificial intelligence (AI) tool to accelerate cancer research, starting with lung cancer.
Part of the Common Fund’s High-Risk, High-Reward Research program, the award supports scientists with outstanding records of creativity to develop pioneering approaches to major biomedical challenges. Only seven such awards were issued this year. The five-year, nearly $6 million grant will fund an AI-human discovery engine that uses large language model agents, such as those used in AI tools like Claude or ChatGPT, to generate hypotheses from human cell and tissue data. The system will then test the hypotheses using robotics testing platforms that can run hundreds to thousands of experiments in parallel on human cells or organoids, miniature three-dimensional tumors grown from patients’ own cells.
“It’s a real honor and a privilege to receive this award,” said Dr. Elemento, who is a professor of systems and computational biomedicine at Weill Cornell Medicine. “It will support the creation of an AI system to drive research from the initial data analysis through the validation stage, with human scientists’ oversight.”
Building a Better Tool
Previous efforts to build an “AI scientist” have focused on creating agents that use the scientific literature to build hypotheses. Dr. Elemento said those efforts are valuable but are limited to what has already been published, which is a secondhand and sometimes incomplete record of human biology. Additionally, he said most research focuses on testing potential hypotheses in animal models one at a time, a time-consuming process that can produce results that do not always translate to humans.
He and his colleagues plan to take a different approach, having the AI agents look for insights from human samples and large volumes of deidentified human data to generate hypotheses, run computer modeling experiments with virtual tumors, and then test several hypotheses simultaneously in large-scale robotics experiments using human cells or organoids. The virtual tumors are AI models built from data on how real human cells respond when they are perturbed, allowing the team to test an idea in simulation before running an experiment. Each round of experimental results will be fed back into the AI, which will learn from the outcomes and decide where the next experiments matter most. “Prediction is not discovery,” Dr. Elemento said. “We want the system to propose, test and revise biological mechanisms, the way scientists do.”
Dr. Elemento and his team at the Englander Institute are well positioned to take on this challenge. Under his leadership, the institute has made major investments over the past five years in data systems, data collection, robotics and the development of disease models for many types of cancer. These include a collection of more than 300 patient-derived organoid models spanning more than a dozen cancer types.
“We have been able in the past few years to create a very large dataset of information on the molecular and cellular features of lung cancer,” he said. “We can now put everything together and let AI run the scientific process, with human oversight, and hopefully make some major discoveries along the way.”
Scientists will supervise the AI tools, setting research objectives, rejecting spurious hypotheses, or rejecting experiments that are too costly or unlikely to be successful. They will also ensure that the work adheres to strict safety and ethical standards.
Accelerating Progress
The team chose to focus first on lung cancer due to the high unmet needs of patients, who currently face an overall five-year survival rate of fewer than 3 in 10, according to the American Cancer Society. Dr. Elemento and his team will study why only some early lung lesions, such as the hazy spots known as ground-glass opacities seen on CT scans, progress to invasive cancer, which molecular and cellular changes drive that progression, and where it might be stopped.
“There is a real need to discover new mechanisms that can be targeted pharmacologically, especially in the early stages of lung cancer,” he said.
Eventually, the goal is to expand the platform's use for other types of cancer. Dr. Elemento also hopes the tools he is building will transform science by helping customize AI agents for scientific use. He envisions separate AI agents specialized for hypothesis generation, critique, experimental design or data analysis. He noted that these tools may complement the expertise of human scientists by analyzing large, complex datasets beyond human capabilities and enable the testing of personalized treatments. He also wants to “democratize” access to AI tools for research to help other scientists, who may not have computer training, leverage these tools to advance and accelerate discovery.
“For decades we have upgraded our instruments while keeping the same scientific method,” he said. “This award lets us change the method itself. I want to create a new AI-powered paradigm for research that makes these large language models work for every scientist.”
This work is supported by the NIH Common Fund through grant DP1CA324880, administered by the National Cancer Institute.
