Skip to content. | Skip to navigation

Personal tools
Log in
Home > News & Publications > Press Releases > Close Collaboration Sheds Light on Collective Behaviors

Close Collaboration Sheds Light on Collective Behaviors

Nina Welding • DATE: November 21, 2017

Categories:  Press Release
Close Collaboration Sheds Light on Collective Behaviors

Infected "zombie" ant. Photo courtesy of Hughes Lab, Pennsylvania State University

From the earliest of days, researchers have been recording their observations, analyzing what they see to interpret and apply the facts before them. Today, however, imaging especially in biomedical communities requires more than the human eye or even incredibly accurate “cameras.” In cases such as the joint project between the University of Notre Dame and Pennsylvania State University (PSU), it requires close collaboration between biologists and computer scientists using deep-learning methods for artificial intelligence to speed up and improve the process.

The joint project, titled “From Cells to Societies: Mechanisms by which Microbial Parasites Control Host Phenotypes,” studies the collective social behaviors of fungal cells inside host ants. Called “zombie ants,” the insects’ bodies are basically hijacked by a fungus, which compels them to act in a certain way in order to spread fungal spores.

Entomologist David Hughes at PSU had been studying the phenomenon for years, searching for clues as to how the fungus gains control over an ant’s body without infecting its brain. Hughes and his team have dissected colonies of infected ants, studying each slice to identify ant cells versus fungal cells. However, a single ant image would take months to identify and analyze.

Danny Z. Chen
Enter Danny Z. Chen, professor of computer science and engineering (CSE) at the University of Notre Dame, and Yizhe Zhang, a CSE Ph.D. student at Notre Dame. Hughes and Chen met at the National Academies Keck Futures Initiatives Conference on Collective Behaviors in 2014. Within a year not only were they working together to find ways to better image and analyze fungal cells within ants, but they had also been awarded a 5-year National Institutes of Health Research Grant to continue their efforts. One of their first papers from this project, titled “3D Visualization and a Deep Learning Model Reveal Complex Fungal Parasite Networks in Behaviorally Manipulated Ants,” was published in the Early Edition of the Proceedings of the National Academy of Sciences of the United States of America, November 7, 2017.

Yizhe Zhang
“My research team has been conducting biomedical imaging research for more than 15 years,” says Chen. “We have also been actively developing deep-learning approaches using artificial intelligence specifically for biomedical imaging problems.”

According to Chen, unlike natural scene images which are commonly two-dimensional, biomedical images are often three-dimensional, so the first order of business was to process fungal cells and ant tissues (muscles) in three-dimensional images.

The next challenge the team faced was to create a new machine learning model for identifying the individual cells. Chen says that using a deep-learning method to train an intelligent model to identify a specific phenomenon typically requires a large training data set. The Notre Dame team overcame the problem of having only a small training data set. Using the images and data collected by the PSU group, including many hours of videos recording how the infected ants functioned in nature, the Notre Dame team was able to develop a machine learning model to analyze the structure and relationships between the two types of cells, specifically how the fungal cells were able to form a network and collaboratively control the host ants.

As they continue working on this particular project, Chen and Hughes are also collaborating on other projects, all focusing on collective biological systems. “There is still quite a bit of work to accomplish,” says Chen, “but we are excited to be working on these common and important biological systems in nature and health studies and believe that the deep-learning models we are developing will help overcome many biomedical imaging challenges.”

Filed under: