Prescription drugs

3D neural networks improve protein structure investigations for drugs

Image credit: Marilyn Gould - Dreamstime

Researchers have used deep learning and 3D technology to better understand how proteins interact in the body – paving way for the development of more effective prescription drugs.

The team at Purdue University, Indiana, hope their new method will advance developments in producing accurate structure models of protein interactions involved in various diseases. They also aim to design better drugs that specifically target protein interactions.

Proteins are often called the working molecules of the human body. A typical body has more than 20,000 different types of proteins, each of which are involved in many functions essential to human life.

“To understand molecular mechanisms of functions of protein complexes, biologists have been using experimental methods such as X-rays and microscopes, but they are time- and resource-intensive efforts,” said Daisuke Kihara, a professor of biological sciences and computer science in Purdue’s College of Science.

He added: “Bioinformatics researchers in our lab and other institutions have been developing computational methods for modelling protein complexes. One big challenge is that a computational method usually generates thousands of models, and choosing the correct one or ranking the models can be difficult.”


DOVE, created by Purdue researchers, captures structural and energetic features of the interface of a protein docking model with a 3D box and judges if the model is more likely to be correct or incorrect using 3D convolutional neural network.

Image credit: Purdue University

To overcome this challenge, Kihara and his team developed a system known as DOVE (DOcking decoy selection with Voxel-based deep neural nEtwork) which applies deep learning model principles to virtual 3D models of protein interactions.

DOVE scans the protein-protein interface of a model and then uses deep learning model principles to distinguish and capture structural features of correct and incorrect models.

This system is a 3D convolutional neural network (CNN) which uses 3D convolutional kernels to make segmentation predictions for a volumetric patch of a scan, or in this case the virtual models.

 “Our work represents a major advancement in the field of bioinformatics,” said graduate student Xiao Wang. “This may be the first time researchers have successfully used deep learning and 3D features to quickly understand the effectiveness of certain protein models.”

“Then, this information can be used in the creation of targeted drugs to block certain protein-protein interactions,” he added.

The study, ‘Protein docking model evaluation by 3D deep convolutional neural networks’, was published in the journal Bioinformatics.

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles