(Based on NHGRI: https://www.genome.gov/about-genomics/educational-resources/fact-sheets/artificial-intelligence-machine-learning-and-genomics)
There are many definitions of artificial intelligence (AI). It is a technology that can be hard to explain or fully understand. But its value in the field of medicine keeps growing as we discover the many ways it supports our work.
One definition, created by AI itself, is, “AI is a broad field that involves the use of machines and computers. It is a combination of computer science, cognitive science, and math, and it uses a variety of technologies to mimic human behavior.”
AI can be created as software or tools that mimic human intelligence or go beyond it. While most of us have only been aware of AI recently, it was developed in 1955 by Professor John McCarthy at Stanford University.
To create and refine AI, scientists start with large, well-defined datasets that allow them to understand the techniques and processes people use to analyze and interpret complex information. For example, a dataset could be thousands and thousands of de-identified broken leg x-rays. Using this dataset, a computer can be “trained” to understand what it is “seeing” regarding leg fractures. AI is constantly developing and changing as researchers learn more and develop new techniques and tools.
What are machine learning and deep learning?
Machine learning and deep learning are frequently mentioned along with AI. Both are subfields of AI. Machine learning is a process by which computers can be given the capability to “learn” about a given dataset without being directly programmed on what to learn.
Machines can develop these capabilities in what’s called a “supervised” or “unsupervised” manner. Under “supervised learning,” scientists provide machines with specific “training” (i.e., computer programming) and test datasets. The training data has defined categories, such as “people with coronary heart disease (CHD)” and “people without CHD.” The computer can use that programming to uncover hidden qualities of the data and figure out different categories. For example, it may discover CHD risk factors we don’t know about yet. The computer can use this new information to work on the test data and make informed predictions, such as which people are likely to develop CHD.
In “unsupervised learning,” computers that are already highly developed can recognize patterns in large datasets and make predictions about the real world without requiring any additional input from researchers.
When machines can learn like this, they are considered to be learning “deeply.” Deep learning is a relatively modern technique used to implement machine learning, but its beginnings can be traced back to 1943 when scientists created a computer model based on how the human brain works. Deep-learning algorithms can take a dataset and find patterns and key information by imitating how a human brain’s neurons interact with each other. The deep-learning algorithms are artificial neural networks — a computing system that can imitate the human brain’s ability to weigh the importance of some information versus other information and make decisions based on that.
Why use artificial intelligence/machine learning in genomics?
The landmark completion of the human genome sequence happened in 2001. This huge accomplishment led to the creation of an extraordinary amount of genomic data over the past 24 years. Experts predict that genomics research will generate between 2 and 40 exabytes of data within the next decade. The graphic below gives us an understanding of the enormity of the data:
Ongoing DNA sequencing and other scientific advances will continue to increase the number and complexity of genomic datasets. This is why genomics researchers need AI/machine-learning tools that can deal with and interpret the information hidden within this truly huge amount of data.
What are some ways in which artificial intelligence/machine learning are being used in medicine?
Although the use of AI/machine-learning tools in genomics is still at an early stage, researchers have already benefited from developing programs that assist in specific ways.
Some examples of machine-learning techniques include:
- Identifying the primary kind of cancer from a liquid biopsy.
- Predicting how a certain kind of cancer will progress in a patient.
Some examples of deep-learning techniques include:
- Identifying disease-causing genomic variants compared to non-harmful variants.
- Improving the function of gene-editing tools such as CRISPR.
For more on AI and medicine generally, these additional sources may be of interest:
https://about.kaiserpermanente.org/news/ai-in-health-care-7-principles-of-responsible-use
https://www.massgeneralbrigham.org/en/about/newsroom/articles/how-is-ai-used-in-health-care
