Spotlight on the Friedberg & Rued Labs

Dr. Iddo Friedberg
Associate Professor of Veterinary Microbiology & Preventive Medicine
Dr. Britta Rued
Assistant Professor of Veterinary Microbiology & Preventive Medicine

Research Area
As antibiotic resistance continues to threaten both human and animal health, researchers at the Department of Veterinary Microbiology and Preventive Medicine are looking for alternatives to conventional antibiotics. A team led by Drs. Iddo Friedberg and Britta Rued, has developed PLUM: Peptide modeLs for Understanding and engineering antiMicrobial therapeutics, an artificial-intelligence system designed to generate entirely new antimicrobial peptides. Antimicrobial peptides, or AMPs, are short chains of amino acids that occur naturally across all domains of life and can act against bacterial pathogens.

Projects Origins
Drs. Friedberg and Rued first came up with the idea of PLUM last year, during a research and innovation round table organized by Iowa State University’s Vice President for Research office. Following a discussion at that meeting, they have applied for and received a one-year grant from USDA and ISU. “At the time we had a completely different machine learning model in mind”, they say. But as work progressed then narrowed in on a model called a structured conditional variational autoencoder to generate new antimicrobial peptides.

Looking Ahead
Their AI model separates different aspects of the peptides it generates into distinct components of their internal representation. This allowed Drs. Friedberg and Rued to separately shape the peptide characteristics: whether a peptide is antimicrobial or a benign control, and its length. PLUM can generate completely novel peptides ranging from 5 to 35 amino acids and can also create new variants based on existing peptide prototypes.

In large-scale computational tests, PLUM generated the highest fraction of predicted antimicrobial peptides among the models tested. Of 45,000 generated sequences, 82.9% were classified as antimicrobial by the industry standard AMP classifier AMPLify, compared with 65.2% by a leading previous approach. Importantly, the generated peptides were not simply copies of known antimicrobial peptides. Analyses showed substantial differences between PLUM’s generated sequences and the sequences used to train the model, while also showing considerable diversity within the generated population.

A particularly useful function of the model is PLUM’s ability to start with an existing peptide and generate variants while independently controlling their desired functional class and length. In tests using known peptide prototypes, PLUM produced substantially larger pools of predicted antimicrobial variants than competing approaches, including when starting from peptides that were originally classified as non-antimicrobial. This joint project between a “wet” and “dry” lab illustrated how bringing computational biology and veterinary microbiology together can help address a fundamental problem that crosses species boundaries; as antimicrobial resistance is also a major concern in agriculture and food production.

Collaborations
Priyanka Banerjee, a PhD computer science student in Friedberg’s lab, led the development of the AI model, and Khawla Alharbi from Rued’s lab performed tests demonstrating effectiveness of the PLUM designed peptides against Salmonella and Campylobacter, two major foodborne pathogens. Researchers at the USDA, Dr. Torey Looft and Dr. Christopher Anderson, will be performing toxicity assays with poultry cell lines and ex-vivo poultry gut bioreactor experiments to examine the feasibility of the use of these peptides in livestock applications. The goal is to be able to use these as potential therapeutics specifically to drive down Salmonella and Campylobacter colonization in livestock, first targeting poultry. Given that poultry are the major reservoir for these organisms in terms of foodborne illness, this work represents a novel interdisciplinary and cross-institutional approach to a major problem. Eventually, the group aims to extend PLUM to design peptides for targeted use against other bacterial pathogens.

Results
The researchers emphasize that PLUM is a computational discovery and prioritization tool, not a complete translational drug-development platform. Its generated peptides will require experimental testing to determine whether their predicted antimicrobial activity, safety and stability translate into real biological effects. In preliminary tests done in Rued’s lab, PLUM peptides have been shown to have effective antibacterial toxicity.

This work demonstrates how generative AI can expand the search space for antimicrobial molecules. Rather than testing modifications of a relatively small number of known peptides one at a time, researchers can computationally explore thousands of potential sequences and prioritize candidates for laboratory investigation.

September 2026