Mayo Clinic develops novel AI class to advance cancer research, care –

3 minutes, 40 seconds Read

image By Shania Kennedy

March 12, 2024 – Mayo Clinic researchers have developed a new class of artificial intelligence (AI) – known as hypothesis-driven AI – to advance oncology research, according to a study published recently in Cancers.

The research team underscored that this new type of AI is designed to incorporate specific hypotheses or research questions, rather than simply relying on data alone, as in traditional models. They posited that the approach could help enhance knowledge discovery efforts in medicine.

“This fosters a new era in designing targeted and informed AI algorithms to solve scientific questions, better understand diseases, and guide individualized medicine,” explained senior author and co-inventor Hu Li, PhD, a Mayo Clinic Systems biology and AI researcher in the Department of Molecular Pharmacology and Experimental Therapeutics, in the press release. “It has the potential to uncover insights missed by conventional AI.”

The research team noted that conventional AI systems excel in recognition, classification, prediction and generative tasks, but they typically do not incorporate existing scientific knowledge. Instead, these tools rely on large, difficult-to-obtain datasets, limiting the models’ knowledge discovery capabilities.

For use cases that require researchers to be able to identify patterns in complex datasets and use those insights to drive knowledge discovery, conventional AI often falls short. Hypothesis-driven AI is designed to address this challenge by using a specific research question to help maximize the utility of the information within datasets.

“Lack of integration between existing knowledge and hypothesis can be a problem. AI models may produce results without careful design from researchers and clinicians what we refer to as the ‘rubbish in rubbish out’ problem,” Li noted. “Without being guided by scientific questions, AI may provide less efficient analyses and struggle to yield significant insights that can help form testable hypotheses and move medicine forward.”

For example, a research team could incorporate existing knowledge about a disease like cancer, such as information on known pathogenic genetic variants, into the design of a hypothesis-driven AI algorithm. Doing so would allow the researchers to better understand which factors contribute to the tool’s performance and improve its interpretability.

“This new class of AI opens a new avenue for better understanding the interactions between cancer and the immune system and holds great promise not only to test medical hypotheses but also predict and explain how patients will respond to immunotherapies,” said co-author and co-inventor Daniel Billadeau, PhD, a professor in Mayo Clinic’s Department of Immunology.

The researchers indicated that their approach provides multiple benefits: hypothesis-driven AI can help research teams take a more focused approach to answering their research question; the AI uses existing scientific knowledge to explore previously unknown connections in data; and the results yielded by hypothesis-driven AI are more interpretable than those of traditional tools.

The research team highlighted that hypothesis-driven AI’s capacity for ‘machine-based reasoning’ could help validate hypotheses and surface new insights in oncology applications like cancer gene discovery, drug response prediction, tumor classification, tumor spatial organization and patient stratification.

However, hypothesis-driven AI also presents limitations related to accessibility, bias and research scope. The researchers explained that developing hypothesis-driven algorithms requires specialized expertise that many healthcare organizations may not possess. Further, like all AI, hypothesis-driven approaches have the potential for bias.

Researchers using hypothesis-driven AI also won’t be simulating all possible scenarios during their analyses because of the limited scope of their research question, which the study authors cautioned could result in some data relationships being missed.

“Nonetheless, hypothesis-driven AI facilitates active interactions between human experts and AI, that relieve the worries that AI will eventually eliminate some professional jobs,” stated Li.

“It can significantly advance medical research by leading to deeper understanding and improved treatment strategies, potentially charting a new roadmap to improve treatment regimens for patients,” he continued.

This research is the latest to explore how AI could advance oncology.

Last week, Cedars-Sinai detailed the development of an AI to make pathology reports machine-readable, which could improve cancer patient recruitment in clinical trials.

The researchers emphasized that pathologists’ notes contain a wealth of data that could significantly advance research, but these data are typically not able to be captured using traditional data-mining approaches.

To overcome this challenge, the research team built a set of 10,000 machine-readable pathology reports using The Cancer Genome Atlas (TCGA), optical character recognition (OCR) techniques, and AI. The resulting dataset could help researchers train models to extract relevant pathology data and support clinical trial recruitment.

This post was originally published on this site

Similar Posts