Louis Dumontet and Mingon Kang (both Computer Science) published a paper titled, "Plastic hydrolytic enzyme classification using explainable deep learning," in ACS Applied and Environmental Microbiology. The team developed PEPIC, an explainable AI framework that predicts plastic-degrading enzymes across 9 substrate types. By identifying "hot spots" in protein sequences, PEPIC fast-tracks the discovery of biological solutions to the plastic crisis. The study validated the model not only using publicly available dataset (SwissProt) but also new external enzymes.

In the study, when the team mapped PEPIC’s high-score amino acids onto 3D structural models, they found that the AI correctly identified the active catalytic sites and binding regions already known by biologists. Using the tool, the team identified a previously uncurated enzyme as a strong candidate for PET degradation, proving the model can find "hidden" solutions in existing genomic data.