Making drug discovery faster and more affordable through artificial intelligence

UM researchers have developed a deep learning model to predict compound protein interactions. GraphBAN is an inductive graph-based approach. The model is all about discovering new drug candidates in the pre-clinical stage. This means speeding up the drug discovery process and making it more affordable.

The work is published in the journal Nature Communications.

“One proven approach in drug discovery is to find the proteins that play a key role in a disease or help harmful microbes survive. If we can target those proteins with the right small molecules, we can disrupt the disease process,” says Hamid Hadipour, data scientist.

Hadipour conceptualized the idea and designed the algorithms along with Dr. Pingzhao Hu. Hu is an adjunct professor at UM Max Rady College of Medicine.

Hadipour explains that GraphBAN predicts if a small molecule can bind to a protein. It can also tell us which parts of it the protein interacts with. This deep-learning model speeds up the prediction process by doing a visual test using AI. It saves time and money, helping researchers focus on the best drug candidates. These can be antibiotics or cancer treatments.

GraphBAN reflects a strong interdisciplinary collaboration between chemistry, biochemistry, microbiology, and computer science. The project was made possible with Dr. Silvia Cardona’s contributions and co-supervision. Cardona is a professor and associate head graduate at the Department of Microbiology. Her lab studies molecular microbiology and microbial genomics, all with a focus on antibiotic discovery.

Cardona tells us that we are going to see more AI predictions in science—predictions that we then have to confirm with experimental research. In a way, AI won’t replace experimental research but rather complement it.