Antibiotic-resistant bacteria cause an estimated 2.8 million infections and more than 35,000 deaths annually in the United States.
A series of studies published in scientific journals have reported the development of artificial intelligence platforms designed to identify and improve antimicrobial peptides for use against antibiotic-resistant bacteria. The research, conducted by separate teams at Houston Methodist and the University of Pennsylvania, was published in Nature Communications and Nature Machine Intelligence, respectively.
Houston Methodist Study: The CAMPER Platform
Platform Overview
Researchers at Houston Methodist developed an AI-powered tool called CAMPER (Constraint-driven AMP Engineering with Ranking). The platform integrates machine learning with biologically informed features to evaluate and rank libraries of candidate antimicrobial peptides based on physical and chemical properties and predicted performance.
Study Results
Using CAMPER, researchers identified a peptide candidate named WP-CAMPER1. In laboratory tests, this peptide demonstrated activity against methicillin-resistant Staphylococcus aureus (MRSA) at low concentrations.
"We validated the CAMPER methodology and demonstrated its ability to generate peptides effective against difficult-to-treat and persistent infections."
Publication and Authors
The study was published in Nature Communications. The research was led by Eleftherios Mylonakis, M.D., Ph.D. , chair of the Houston Methodist Charles W. Duncan Jr. Department of Medicine. First authors were Fadi Shehadeh and Biswajit Mishra.
Collaborating Institutions
The study involved researchers from:
- Houston Methodist
- Harvard Medical School
- Rajiv Gandhi Technological University
- The University of Texas Medical Branch at Galveston
- Brown University
- National Technical University of Athens and Archimedes-Athena Research Center
University of Pennsylvania Studies: ApexGO and Prionin Discovery
ApexGO Platform Development
Researchers at the University of Pennsylvania developed an AI-driven method called ApexGO that iteratively improves imperfect antibiotic candidates by predicting which molecular modifications will enhance antimicrobial activity.
Methodology: ApexGO combines a generative component that suggests molecular edits with a predictive model (APEX) that estimates antimicrobial activity. It uses Bayesian optimization to explore possible peptide modifications, balancing exploitation of promising regions with exploration of uncertain ones.
Results:
- 85% of AI-generated molecules halted bacterial growth
- 72% of AI-generated molecules outperformed their parent peptides
- In mice, two ApexGO-designed peptides reduced bacterial counts at levels comparable to polymyxin B, an FDA-approved last-resort antibiotic
"Antibiotic discovery is fundamentally a search problem across an enormous molecular space. ApexGO gives us a way to navigate that space with far more direction."
— César de la Fuente
Publication and Authors: The study was published in Nature Machine Intelligence. Co-senior authors were César de la Fuente (Presidential Associate Professor) and Jacob R. Gardner (Assistant Professor in Computer and Information Science). Co-first authors were Marcelo Torres (Research Assistant Professor) and Yimeng Zeng (doctoral student).
Funding: National Institutes of Health, Defense Threat Reduction Agency, National Science Foundation.
Prionin Discovery Study
Researchers at the University of Pennsylvania used AI to screen 19.3 million peptide fragments from 2,897 prion and prion-like proteins.
Findings:
- 1,179 candidate antimicrobial peptides, termed 'prionins', were identified
- 75 peptides were selected for experimental testing
- 59 inhibited at least one bacterial pathogen
- 42 showed strong activity at low concentrations
Animal Testing: Two prionins (one from a fungus, one from a roundworm) were tested in a mouse skin infection model with Acinetobacter baumannii. They reduced bacterial levels comparably to polymyxin B, with no treatment-related weight loss observed.
The study suggests prion proteins may contain hidden molecular features with antimicrobial potential.
Future Directions
Researchers noted that the best-performing peptides identified in these studies remain early-stage candidates requiring further optimization for safety and stability before potential clinical use. Researchers suggested the ApexGO approach could be applied to other biological functions, such as immune modulation or tumor targeting.