Mayo Clinic Researchers Invent Hypothesis-Driven AI for Cancer Research Breakthroughs

A team of Mayo Clinic researchers have recently unveiled a groundbreaking class of artificial intelligence (AI) algorithms known as hypothesis-driven AI. Unlike conventional AI models that rely solely on large datasets, these new algorithms incorporate existing scientific knowledge and hypotheses. In a review paper published in the journal Cancers, Mayo’s researchers illustrated the potential of this innovative computational approach in utilizing massive datasets to unravel the intricate causes of cancer. 

Dr. Hu Li, the senior author of the review and co-inventor of the hypothesis-driven AI, noted that these novel algorithms have the potential to uncover insights missed by conventional AI, fostering a new era for designing targeted and informed algorithms, facilitating a deeper understanding of diseases and guiding individualized medicine. 

Related article: Researchers Predict Drug Interactions with Machine Learning to Enhance Patient Safety

Overcoming Limitations of Conventional AI with Targeted and Informed Algorithms

As AI technology advances, conventional AI models can already perform sufficiently well in recognition and classification tasks including face recognition and imaging classification in clinical diagnosis, and it has been increasingly applied to generative tasks, such as creating human-like text. However, due to the heavy reliance on large, unbiased datasets which can be challenging to obtain, conventional AI models usually lack interpretability and fail to integrate existing scientific knowledge and hypotheses, thereby hindering their flexibility and adoption in areas that demand knowledge discovery, like medicine since they can hardly lead to testable hypotheses. “AI models may produce results without careful design from researchers and clinicians what we refer to as the ‘rubbish in rubbish out’ problem,” commented Dr. Li.

Hypothesis-driven AI, on the other hand, aims to bridge this gap by “machine-based reasoning”, that is, incorporating an understanding of diseases into the design of learning algorithms. In the realm of cancer research, such understanding may refer to, for example, the information about pathogenic genetic variants and interactions between certain genes in cancer. The targeted and informed approach enables researchers and clinicians to enhance the interpretability of AI systems, address dataset limitations, and focus on open scientific questions. 

Transforming Cancer Research With Hypothesis-Driven AI

According to Prof. Daniel D. Billadeau, Mayo’s chair of the Department of Immunology, and another co-inventor of the hypothesis-driven AI, this new class of AI can expand horizons in cancer research by depicting a clearer picture of how cancer cells interact with the immune system. It offers great promise in testing medical hypotheses and predicting the impact of immunotherapies on patients.

Mayo’s researchers describe hypothesis-driven artificial intelligence as a powerful tool for cancer research in numerous areas, including tumor classification, patient stratification, the discovery of cancer genes, prediction of drug response, and spatial organization of tumors. Apart from the above-mentioned benefits, hypothesis-driven AI also requires less data and computing power than conventional models, reducing the need for resources. 

Foreseeable Challenges and Future Directions

Being a new technology at its nascent stage, hypothesis-driven AI presents exciting possibilities for cancer research. Yet challenges remain in its implementation, including the limited accessibility of expertise required for creating novel algorithms, potential biases, and the limitation of not covering all possible scenarios. 

Despite these challenges, Dr. Li showed optimism about the potential of this innovative class of AI to advance cancer research and improve treatment strategies, potentially charting a new roadmap that leads to enhanced treatment regimens for individual patients. Moreover, he expressed confidence that hypothesis-driven AI will facilitate active interactions between human experts and AI, which will in turn alleviate concerns over the possible elimination of professional jobs due to AI in the future.

Author

Richard Chau