A new class of artificial intelligence (AI) called hypothesis-driven AI, invented recently by Mayo Clinic researchers, offers an innovative way to discover the complex causes of cancer and improve treatment strategies.
The team notes that conventional AI is used primarily for classification and recognition, such as facial recognition, or providing imaging classification in diagnosis, while also being increasingly used for generative applications like creating text. However, many of the algorithms associated with these conventional tasks often fail to include existing knowledge or hypotheses, relying instead on hard-to-obtain, large unbiased datasets. This limitation has restricted the flexibility of AI models and methods and their use in areas that require knowledge discovery such as medicine.
“We showcase the power of hypothesis-driven AI in making novel cancer discoveries that can be overlooked by conventional AI methods,” the researchers wrote in their review published in the journal Cancers.
According to Hu Li, PhD, a systems biology and AI researcher at Mayo their new hypothesis-driven AI approach “fosters a new era in designing targeted and informed AI algorithms to solve scientific questions, better understand diseases, and guide individualized medicine. It has the potential to uncover insights missed by conventional AI.”
Artificial intelligence has already shown its utility as a tool that can identify broad, often overlooked, patterns when analyzing large, complex datasets for cancer research, but one of the challenges in this area has been finding a way to full leverage the embedded information within these large datasets. Hypothesis-driven AI allows researchers to incorporate currently knowledge of a disease and its drivers and integrating these into the design of a learning algorithm.
“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,” saysaids 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.”
The Mayo team highlight a number of benefits of hypothesis-driven AI, noting it:
- Focuses on specific hypotheses or research questions;
- Leverages existing knowledge to find previously missed connections;
- Is more interpretable, providing results that are easier to understand than with conventional AI;
- Requires less data and computing power; and
- Helps scientists test and validate hypotheses by incorporating hypotheses, and biological and medical knowledge, into the design of the learning algorithm.
While Li and team point out these potential benefits of hypothesis-driven AI, they are also cognizant of its potential limitations, the most notable of which is that creating these kinds of learning algorithms require specialized knowledge, which may limit accessibility, as well the potential for the algorithms to build in bias. Also, researchers may not have a broad enough view, which may prevent them from formulating hypotheses that take into account all possible scenarios, which may result in missing some unforeseen relationships that could be critical to broader understanding of disease.
“Since hypothesis-driven AI is still in its infancy, open questions such as how to better incorporate new knowledge and biological perspectives to ameliorate bias and improve interpretability in the design of AI algorithms still need to be addressed,” the investigators noted.
Regardless, the development of hypothesis-driven AI, its ability to ask questions, and learn from them, while lessening the burden of access to huge datasets, provides a step toward the use of AI in scientific and clinical research.
“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,” Li concluded.