Combining biology and algorithms to solve complex medical challenges
Can powerful machine learning algorithms and artificial intelligence (AI) techniques be used to understand what genes are responsible for disease?
And can that knowledge help doctors prepare better treatment plans for patients or prevent some diseases completely?
Dr. Ting Hu thinks so.
That’s why the assistant professor in the Department of Computer Science and her Machine Intelligence and Biocomputing Laboratory are working to develop new, intelligent learning methods and complex network models for biomedical knowledge discovery.
A graduate of Memorial (PhD’10), she competed a post-doctoral fellowship in computational genetics at Dartmouth College.
“A baby learns to walk and talk based on experiences and observations. It’s the same idea for machine learning.”
The experience now allows her to combine a knowledge of biology and an expertise in algorithms with the goal of solving some of medicine’s most challenging problems.
What is AI?
Dr. Hu says the concept of artificial intelligence has evolved over the years, but the idea is to get machines to develop human-like intelligence and figure out how to solve complex problems for themselves.
“Humans aren’t pre-programmed,” she said. “A baby learns to walk and talk based on experiences and observations. It’s the same idea for machine learning.”
“If you want a machine to recognize whether a picture is of a dog or cat, you show it thousands of pictures of cats and dogs and when the algorithm makes a prediction, tell it if it is right or not,” said Dr. Hu.
“After training on hundreds or thousands of instances, the algorithm can learn and do a good job at identifying different objects.”
This application has many uses, including making your smartphone camera auto focus on a human face or allowing Netflix to predict what show you might want to binge watch next.
Dr. Hu’s specialty is in evolutionary computing, a creative approach to AI. It uses a learning process that is similar to natural evolution, an iterative process.
“You try something, see how it works, make changes and then try again. That’s nature’s method for solving problems.
“If we can combine AI and medicine, perhaps you could have your DNA tested to see how you might respond to a particular treatment before you even take it.”
Humans are the products of evolution and we know this method works well. We’re highly adaptive to our environment and we have a high-level of intelligence.”
One key to the current success of machine learning is having access to vast amounts of data. One field where data is in great supply is in medicine. In recent years, new biotechnologies have advanced the collection of biological and medical data.
“Ten years ago, it cost tens of thousands of dollars to genotype your DNA, but today it can be done for less than $100,” said Dr. Hu. “So now we have this huge amount of data to explore. The problem is, we don’t have enough tools or methods to help us understand or take useful information from it.”
Personalized and effective
Scientists have some understanding of which genes are associated with what disease, but they also recognize biology is more complicated than that.
Disease can be caused by a group of genes or things like environmental factors and lifestyle. There are thousands to a million variables that can play a role and often they are intertwined. That makes the problem more complicated.
If science can use technology to identify those factors, says Dr. Hu, patients who know they are at risk can take preventative action, or doctors can develop personalized and effective treatments – also known as precision medicine.
“When you are diagnosed with a disease, it’s often the same treatment for everyone. But if we can combine AI and medicine, perhaps you could have your DNA tested to see how you might respond to a particular treatment before you even take it. Precision medicine may take a while to get there, but that’s the goal.”