Univ Texas: Plastic-Eating Enzyme Breakthrough Enabled by Supercomputing

[Hamartia Antidote]

Plastic-Eating Enzyme Breakthrough Enabled by Supercomputing

More than 90 percent of plastic remains unrecycled, and as a result, billions of tons of plastic reside in landfills (or worse, in natural environments)

www.hpcwire.com

More than 90 percent of plastic remains unrecycled, and as a result, billions of tons of plastic reside in landfills (or worse, in natural environments) around the world. This snowballing environmental disaster has caught the attention of researchers from a wide range of fields, who have proposed solutions ranging from the robotic to the chemical for collecting and breaking down plastic waste. Now, researchers from the University of Texas at Austin have created an enzyme variant that can break down these plastics—a discovery made possible by supercomputing.

The researchers sought to break down polyethylene terephthalate, more popularly known as PET, which pervades consumer packaging and constitutes a double-digit percentage of global waste. At UT Austin’s Cockrell School of Engineering and College of Natural Sciences, the university’s researchers worked to hone an enzyme called PETase that enables PET degradation by bacteria.

To generate novel mutations for this enzyme, Raghav Shroff (formerly a member of a lab in the UT Austin Center for Systems and Synthetic Biology) created a 3DCNN-based deep learning model to predict which mutations would generate faster and faster depolymerization of PET.


The Maverick2 supercomputer. Image courtesy of TACC.

“The Texas Advanced Computing Center (TACC) enabled the deep learning part of this paper,” explained Danny Diaz, a current member of the same lab (which is led by Andrew Ellington, a professor at UT Austin and project lead for the enzyme development). At TACC, the team used the Maverick2 supercomputer. Maverick2, which specializes in deep learning applications, is equipped with 30 main nodes, all equipped with Intel CPUs: 23 are also equipped with quadruple Nvidia 1080 Ti GPUs; 4 with dual Nvidia V100 GPUs; and 3 with dual Nvidia P100 GPUs. The system uses Mellanox networking.

“Shroff could easily train models in parallel and have GPUs that support extremely fast training, quickening the time between iterations of model testing, which allowed publishing of the first 3DCNNs guided engineered proteins,” Diaz said. “In this paper, we leveraged Raghav’s model to generate the mutation-generating predictions pivotal to engineering the plastic-eating protein, further demonstrating the power of deep learning models to guide protein engineering.”

The researchers say that the improved enzyme, called FAST-PETase, can break down PET in hours or days at less than 50°C (the need for high temperatures can be a hurdle for enzyme-based plastic waste solutions). Next, the team is planning to scale up production of FAST-PETase.

“The possibilities are endless across industries to leverage this leading-edge recycling process,” said Hal Alper, a professor of chemical engineering at UT Austin. “Beyond the obvious waste management industry, this also provides corporations from every sector the opportunity to take a lead in recycling their products. Through these more sustainable enzyme approaches, we can begin to envision a true circular plastics economy.”