A new class of artificial intelligence (AI) algorithms known as hypothesis-driven AI was recently developed by Mayo Clinic researchers. This is a major break from conventional AI models, which only learn from facts.
The researchers write in a review that will be published in Cancers that this new type of AI provides a novel approach to leverage large datasets in order to better understand the intricate causes of diseases like cancer and develop treatment plans.
The senior author and co-inventor of the work, Hu Li, Ph.D., is a Mayo Clinic Systems biology and AI researcher in the Department of Molecular Pharmacology and Experimental Therapeutics. “This fosters a new era in designing targeted and informed AI algorithms to solve scientific questions, better understand diseases, and guide individualized medicine,” Li says. “It has the potential to uncover insights missed by conventional AI.”
The main applications of conventional AI are in classification and recognition tasks, including face recognition and imaging classification for medical diagnosis. However, it is also being used more and more in generative tasks, such writing literature that appears human. Scholars observe that traditional learning algorithms frequently do not take into account current scientific theories or information. Rather, these mainly depend on substantial, impartial datasets, which might be hard to come by.
Dr. Li claims that this restriction significantly limits the adaptability of AI techniques and their applications in fields such as medicine that necessitate knowledge discovery.
Artificial intelligence (AI) is a useful tool for finding patterns in big, complicated datasets, such as those used in cancer research. Leveraging the embedded information in those datasets has been the main obstacle to employing conventional AI.
“One issue may be a lack of integration between the theory and the body of current knowledge. We call this the “rubbish in rubbish out” dilemma because AI models may deliver results without proper design from academics and physicians, according to Dr. Li. “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.”
Using hypothesis-driven AI, scientists hunt for methods to build learning algorithms that take into account knowledge of a condition, such as incorporating relationships between specific genes in cancer and known dangerous genetic variations. This will improve interpretability by allowing researchers and clinicians to identify the elements that influence model performance. This approach can also address problems with the datasets and encourage us to concentrate on open scientific questions.
“This new class of AI holds great promise not only to test medical hypotheses but also to predict and explain how patients will respond to immunotherapies,” says Daniel Billadeau, Ph.D., a professor in the Department of Immunology at Mayo Clinic. “It opens a new avenue for better understanding the interactions between cancer and the immune system.” Billadeau has a lengthy history of studying cancer immunology and is a co-author and co-inventor of the work.
The research team claims that a wide range of cancer research applications, including as tumor categorization, patient stratification, cancer gene discovery, treatment response prediction, and tumor spatial structure, can be achieved with hypothesis-driven AI.
AI driven by hypotheses has advantages.
aimed at: concentrates on particular research issues or hypotheses.
Utilizes current information leads the search for connections that were previously overlooked.
Greater interpretability Comparing the results to traditional AI, they are simpler to understand.
Decreased requirement for resources: uses less computational power and data.
“Machine-based reasoning”: Assists scientists in testing and validating theories by integrating theories with knowledge of biology and medicine while creating learning algorithms.
The drawback of this tool, according to Dr. Li, is that developing these kinds of algorithms calls for experience and specialized knowledge, which can restrict its general accessibility. When applying various types of information, researchers should be aware of the possibility of bias creeping in. Furthermore, researchers typically work within a narrow scope and may not consider every circumstance, which could lead to the omission of some important and unexpected associations.
“Nonetheless, hypothesis-driven AI facilitates active interactions between human experts and AI, that relieve the worries that AI will eventually eliminate some professional jobs,” according to Dr. Li.
As hypothesis-driven AI is still in its infancy, there are still issues to be resolved, like how to combine biological data and knowledge in the most effective way to reduce bias and enhance interpretation. Dr. Li asserts that hypothesis-driven AI is a positive development despite its difficulties.
“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,” according to Dr. Li.
