Wei-Hsuan “Jenny” Lo-Ciganic’s introduction to artificial intelligence came in 2009. As a doctoral student at the University of Pittsburgh, she pursued a master’s degree in biostatistics, where she learned about tree-based machine learning methods.
“At the time, I was like ‘why am I learning this? Am I going to use this in the future?’” said Lo-Ciganic, Ph.D., M.S., M.S.Pharm., an associate professor of pharmaceutical outcomes and policy in the UF College of Pharmacy. “Now when I look back, I am thankful. It put me on the top of the curve. I’m a pioneer in our field using machine learning approaches.”
Machine learning is a subset of AI. Whereas AI features techniques that enable computers to mimic human behavior, machine learning uses computer systems that learn and adapt without explicit instructions. Machine learning allows scientists to discover hidden patterns and incorporate complex interactions in large data to generate more accurate predictions in clinical settings.
Lo-Ciganic applies machine learning in three areas of research — identifying individuals at high risk of substance abuse disorders, improving medication adherence and predicting treatment failures. She has secured federal funding from the National Institute on Drug Abuse and the National Institute on Aging and successfully developed machine learning algorithms using health care claims data to predict patients who are at high risk for opioid overdose and opioid use disorder. The models have outperformed traditional statistical approaches, which tend to target people who are not truly at risk and miss the majority of individuals who are at risk.
“If we can more effectively predict who poses the greatest risk of an opioid overdose, then we can help clinicians allocate their time and resources to patients who need intervention the most — rather than targeting an entire population,” Lo-Ciganic said.
Incorporating machine learning into clinical decision-making is an idea also being studied by Caitrin McDonough, Ph.D., M.S., an assistant professor of pharmacotherapy and translational research. She is adopting new AI strategies into her data analysis to predict patients at high risk for cardiovascular disease. Machine learning and AI tools are helping her go beyond traditional statistical models in predicting cardiovascular events.
“With traditional models, I’m often using my own expertise and prior literature to determine factors for analysis,” McDonough said. “AI and machine learning have the computational power to build really robust models, and include additional factors, which could make a significant impact on pharmacotherapy.”
McDonough’s long-term goal is to build models that run continuously over data living within the electronic health records and flag high-risk hypertension patients. Providing this clinical decision support to physicians and caregivers has immense potential, especially in the area of adherence. She said machine learning technology could help layer electronic health record and insurance claims data to understand why patients are not filling their prescriptions or taking blood pressure medications.
“I was drawn to study cardiovascular disease by the sheer number of people impacted by this health condition,” McDonough said. “Our blood pressure control rates across this country are not great, but if we can provide patients the right medication and get them to take their medication, it could prevent a lot of other complications. The AI-supported modeling I’m developing around cardiovascular disease is a step forward toward improving patient care.”
While Lo-Ciganic and McDonough have their sights set on using AI to improve health outcomes and patient care, Chenglong Li, Ph.D., a professor of medicinal chemistry, has embraced an AI-based approach for drug discovery. His lab allocates a lot of energy into optimizing a compound to become a drug. It’s easy to find an initial hit compound, but it’s hard to turn the compound into a drug.
AI is assisting Li’s lab in building neural network models to predict the binding of small molecules to their disease targets and in developing a computational small molecule drug design platform to optimize newly developed hit compounds. These compounds could one day lead to new drug therapies to treat pancreatic, liver, prostate or breast cancers.
“We have the potential to generate a more efficient molecule using AI,” Li said. “For example, if we have a lead compound, and we don’t know how to make a better one, the traditional way is to rely on organic chemistry instinct and make a lot of similar compounds. We might need to make 100 compounds to design a better one, but with our newly designed platform supported by AI, we can select the top five compounds to synthesize and make a better compound. That’s enhanced efficiency and improved productivity.”
Li, who serves as the Nicholas Bodor Professor in Drug Discovery in the UF College of Pharmacy, began incorporating AI into his lab in 2018. He acknowledges that the structure-based drug design field is limited right now by the amount of high-quality data available, but he credits AI for improving drug optimization.
“In the traditional way of doing drug design, you are basically shooting things in the dark and hoping to capture something. There’s a lot of serendipity,” Li said. “If we can take computing AI approaches and combine with lab experiments, we can really expand the drug optimization options and find new drugs more effectively and efficiently.”