Researchers Use Deep Learning to Identify New Medications

Researchers at the Gwangju Institute of Science and Technology in Korea have developed a new deep learning model that can predict the binding between a drug and target molecule. The team, which was led by associate professor Hojung Nam and Phd student Ingoo Lee, called the new model “Highlights on Target Sequences” (HoTS). 

The research was published in the Journal of Cheminformatics. 

The Drug Discovery Process

Drugs are tested in the drug discovery process for their ability to bind or interact with target molecules in the body. Deep learning models have proved useful in making this process more effective, but their predictions do not always demonstrate interpretability. That is why the team created HoTS, which makes better predictions of drug-target interactions while also being interpretable. 

It is crucial to determine how well a drug binds to its target molecule, and this usually involves aligning a 3D structure of a drug and its target protein at various configurations. This process is referred to as “docking.”  Following this process, preferred binding sites are then discovered by running docking simulations over and over with multiple drug candidates for a target molecule. Deep learning models are relied on to carry out these simulations. 

HoTS Model

The newly developed model can also predict drug-target interactions (DTIs) without the need for simulations or 3D structures. 

“First, we explicitly teach the model which parts of a protein sequence will interact with the drug using prior knowledge,” Professor Nam explains. “The trained model is then utilized to recognize and predict interactions between drugs and target proteins, giving better prediction performances. Using this, we built a model that can predict the target proteins’ binding regions and their interactions with drugs without a 3D-complex.” 

The model doesn’t have to deal with the complete length of the protein sequence. Instead, it can make predictions based on parts of the protein that are relevant to the DTI interaction. 

“We taught the model where to ‘focus’ to ensure that it can comprehend important sub-regions of proteins in predicting its interaction with candidate drugs,” Professor Nam continues. 

This enables the model to predict DTIs more accurately than existing models. 

These new findings will provide a good starting point for future docking simulations to predict new drug candidates. 

“This model used in our study would make the drug discovery process more transparent as well as low-risk and low-cost. This will allow researchers to discover more drugs for the same amount of budget and time,” Professor Nam concludes.