An estimated one in five Americans lives with chronic pain, and current treatment options leave much to be desired. Feixiong Cheng, PhD, director of Cleveland Clinic’s Genome Center, and IBM are using artificial intelligence (AI) for drug discovery in advanced pain management. The team’s deep learning framework identified multiple gut microbiome-derived metabolites and FDA-approved drugs that can be repurposed to select non-addictive, non-opioid options to treat chronic pain.
The findings, published in Celpers, represent one of the many ways the organizations’ Discovery Accelerator partnership is helping advance research in healthcare and life sciences.
Treating chronic pain with opioids is still a challenge due to the risk of serious side effects and dependency, says co-first author Yunguang Qiu, PhD, a postdoctoral researcher in Dr. Cheng, whose research program focuses on developing therapies for nervous system disorders. Recent evidence has shown that drugging a specific subset of pain receptors in a class of proteins called G protein-coupled receptors (GPCRs) can provide non-addictive, non-opioid pain relief. The question is how we can target these receptors, explains Dr. Qiu out.
Rather than inventing new molecules from scratch, the team wondered if they could apply research methods they had already developed to find pre-existing, FDA-approved drugs for potential pain indication. Part of this process involves mapping gut metabolites to identify drug targets.
To identify these molecules, first author and computational scientist Yuxin Yang, PhD, a former graduate student at Kent State University. Dr. Yang completed his thesis research in the laboratory of Dr. Cheng and continues to work there as a data scientist. Drs. Yang and Qiu led a team to update an earlier AI drug discovery algorithm that the Cheng Lab had developed. IBM employees helped write and edit the manuscript.
Our IBM employees gave us valuable advice and perspective to develop advanced computing techniques. I am pleased to have the opportunity to work with and learn from colleagues in the industrial sector.”
Dr. Yuxin Yang, PhD, first author and computational scientist
To determine whether a molecule will work as a drug, researchers must predict how it will physically interact with and affect proteins in our bodies (in this case, our pain receptors). To do this, the researchers need a 3D understanding of both molecules, based on extensive 2D data on their physical, structural and chemical properties.
“Even using current computational methods, combining the amount of data we need for our predictive analytics is extremely complex and time-consuming,” explains Dr. Cheng out. “AI can quickly make full use of compound and protein data obtained from imaging, evolutionary and chemical experiments to predict which compound has the best chance of appropriately affecting our pain receptors.”
The research team’s tool, called LISA-CPI (Ligand Image- and receptor’s three-dimensional (3D) Structures-Aware framework to predict Compound-Protein Interactions) uses a form of artificial intelligence called deep learning to predict:
The team used LISA-CPI to predict how 369 gut microbial metabolites and 2,308 FDA-approved drugs would interact with 13 pain-associated receptors. The AI framework identified several compounds that could be repurposed to treat pain. Studies are underway to validate these compounds in the laboratory.
“The predictions of this algorithm could reduce the experimental burden that researchers must overcome to even come up with a list of drug candidates for further testing,” says Dr. Yang. “We can use this tool to test even more drugs, metabolites, GPCRs and other receptors to find therapies that treat diseases beyond pain, such as Alzheimer’s disease.”
Dr. Cheng added that this is just one example of how the team is working with IBM to develop basic small molecule models for drug development – including both drug repurposing in this study and an ongoing drug discovery project.
“We believe these foundational models will provide powerful AI technologies to rapidly develop therapies for multiple challenging human health problems,” he says.
Source:
Magazine reference:
Yang, Y., et al. (2024). A deep learning framework combining molecular images and structural protein representations identifies drug candidates for pain. Cell report methods. doi.org/10.1016/j.crmeth.2024.100865.