Dr. Sriram Chandrasekaran, associate professor of biomedical engineering at the University of Michigan, Ann Arbor, discussed the use of AI models to create new drugs at an Oct. 10 event. 

The talk, titled “Next Generation Drug Discovery Using Mechanistic AI,” was the latest installment of the Bhussry Seminar Series, sponsored by the department of biochemistry and molecular & cellular biology.

Chandrasekaran spoke about two main research topics: drug discovery and precision health for infectious diseases, and systems biology tools for studying metabolic regulation.

Chandrasekaran gave a timeline of the development of antibiotics from the 1920s to the present. 

“Unfortunately, we hit the peak 50 years ago,” Chandrasekaran said. “Even the antibiotics that have been recently introduced, they’re all sort of analogs of previously existing ones.” 

Chandrasekaran said this research is especially important in the era of antibiotic resistance, where treatments are becoming less effective. 

“It’s alarming because on one hand, we have this rapid spread of these pathogens, but on the other hand, we don’t have any new treatments to solve them. And so we need systems, biology and computational tools to stay ahead of these pathogens,” Chandrasekaran said. 

Chandrasekaran advocated for the use of combination therapy, which combines either existing antibiotics or other novel agents together, and noted its many benefits. 

“The idea is that the probability of a pathogen evolving resistance to two or more antibiotics is lower than evolving resistance to any single agent. Another cool aspect is that you can also repurpose drugs that are being used for other diseases,” Chandrasekaran said.

Chandrasekaran said he was able to create a series of hybrid mechanistic AI models that could make drug combinations and predict their efficacy by integrating both engineering and artificial intelligence principles.

One of these models, INDIGO, made five million combinations from a group of 200 drugs designed to combat E. coli, eventually deciding on a combination of spectinomycin and chlorpromazine, two drugs that Chandrasekaran had not initially considered combining.

Chandrasekaran expanded on the applications of INDIGO, illustrating how the AI model, once trained on the E. coli data, could be leveraged to find drug combinations for other pathogens, such as Staphylococcus aureus, a bacteria that can lead to serious skin and bloodstream infections.

Another mechanistic AI model, MAGENTA, could take into account not only the properties of the drug, but also the metabolites or conditions present in the environment. 

“MAGENTA makes more context specific predictions. So now we know that no drug response is universal,” Chandrasekaran said. 

Chandrasekaran said that he hopes his research can contribute toward solving diseases that are currently impacting millions of people around the world, such as cancer and tuberculosis. He envisions a world where personalized medical simulations, in which the properties of the drug, pathogen and patient data are inputted to get the predicted treatment outcome, are the norm. 

When asked about whether the mechanistic AI models could be used to predict the severity of side effects that a drug or drug combination might have, Chandrasekaran acknowledged that certain drug combinations could be effective against the disease but also potentially toxic. 

Chandrasekaran said that lowering the dosage of the drug could reduce the severity of the toxicities. However, he is still actively developing solutions to this problem. 

Yoontae Kim, a Georgetown postdoctoral fellow currently using 3D printing to develop biomimetic models to better understand pathogens, observed that Chandrasekaran’s presentation relates to his own scientific field of study.

“Recently, I have encountered various types of research that combine patient data, mainly images, and AI to improve diagnosis accuracy and shorten prediction time. I was very surprised to learn that AI-related research is also actively underway in the pharmaceutical and bioinformatics fields,” Kim said. 

Kim said Chandrasekaran’s research opens up numerous possibilities for scientific discovery.

“If bioinformatics tools such as INDIGO were actively utilized in the development of patient tailored anticancer drugs, it would be possible to develop much better patient-tailored antibiotics.”