The term "AI" or artificial intelligence has become ubiquitous, with applications from facial recognition on smartphones to online language translators. But the possibilities go far beyond the convenience it provides for day-to-day tasks.

Scientists at the Stowers Institute for Medical Research are harnessing novel technology to benefit foundational science and enhance our understanding of human health and disease. With enough data and artificial intelligence tools, researchers can build predictive models to better understand outcomes, accelerate research, and further innovate and test scientific theories.

What is AI, machine learning, and deep learning?

“AI is a broad term encompassing the use of code and algorithms to make decisions in a manner similar to human decision-making. An example of AI in action is automated tax software,” explained Melanie Weilert, a bioinformatician in the lab of Julia Zeitlinger, Ph.D. at the Stowers Institute.

When it comes to AI, there are two learning models that people often refer to: machine learning and more specifically, deep learning.

Machine learning is a subset of AI that combines code with provided data to make progressively smarter decisions. Within machine learning, there is a subset known as deep learning, which employs complex mathematical models to make decisions and predictions.

“AI ‘learning’ is essentially a mathematical process. Each machine learning algorithm consists of interconnected mathematical equations,” Weilert said. “Just as electrical impulses flow through neurons in the human brain, data flows through these interconnected equations to make decisions. When new data is introduced, the equations are adjusted accordingly.”

In essence, AI "learns" by refining these mathematical equations based on the data it processes. The more data the algorithm is exposed to, the more precise its decision-making becomes, ultimately enabling it to perform specific tasks effectively.

How is AI used in foundational research?

Scientists have harnessed the power of AI to tackle an array of tasks, ranging from diagnosing diseases to mapping the stars. At the Stowers Institute, researchers in the Zeitlinger Lab employ deep learning to address a fundamental question: What makes us who we are? Answering this question relies on learning more about our DNA and deciphering the instruction for gene regulation in the human genome.

“The human genome is an extensive 3.2-billion-letter book, and understanding it promises significant advances in the field of life science. Deep learning is particularly suited for this task because it can be tailored to interpret specific data and thrives on large datasets,” Weilert said.

In the Zeitlinger Lab, deep learning algorithms are trained to understand the intricate language written in our DNA.

“Remarkably, AI technology can readily decipher many of the words, sentences, and even paragraphs embedded within our genetic code,” Weilert expanded. “Once trained, these algorithms are used to gain a more human-like understanding of the intricate patterns hidden within our DNA.”

“It's crucial to note that AI is not sentient in the way humans are. AI is a tool designed for specific tasks, much like a car is designed for transportation or an oven for heating food. Scientists meticulously design each AI algorithm for a particular purpose, and it operates within those predefined boundaries,” Weilert said.

Weilert explained just like many people are weary of the nearly unavoidable presence of AI in their own personal lives today – from chatbots to targeted advertisements – she as a scientist was also hesitant of its use in scientific research at first, but quickly saw benefits.

A collaboration with Anshul Kundaje, Ph.D., Associate Professor of Genetics and Computer Science at Stanford University, yielded Weilert and her teammates their first deep learning application towards their genomics data. “AI seemed like a really scary subject to delve into then,” she explained. “My role in that project began as a “translator” across the fields of biology and computer science. But overtime, I realized I had an instinct for how data would flow through neural networks. Deep learning became deeply intuitive to me as a tool for understanding and recognizing patterns – and the benefits were clear.”

The benefits of AI moving forward

Without deep learning, deciphering the complexities of the genome would be an insurmountable challenge. By harnessing the mathematical prowess of deep learning algorithms, scientists can explore the unknown of the human genome and unlock valuable insights into human health and disease.

“Working with this data has also made me deeply appreciate the interdisciplinary nature of our work. Without the careful, gold-standard experiments conducted by the biologists in our lab, none of my research would be possible,” Weilert said.

As AI continues to evolve, its role in advancing scientific knowledge and innovation will likely become even more profound.

“I never run out of questions, and the work never stops challenging me to be a better programmer and better scientist. I have excellent role models in the Zeitlinger Lab at Stowers and am grateful to be a part of the work we are doing.”