A machine learning model screened over 16,000 peptide sequences and identified two novel antimicrobial peptides that killed drug-resistant ESKAPE pathogens, eliminated biofilms, and showed no resistance development after 22 passages.
0 resistance in 22 passagesBacteria failed to develop resistance to the ML-identified peptides even after 22 rounds of exposure — a stark contrast to conventional antibiotics where resistance often emerges within a few passages.
What the researchers found
Two ML-identified peptides (GDST-038 and GDST-045) killed ESKAPE pathogens, achieved >3-log reduction in biofilm bacteria, eliminated S. aureus in 3D human skin models at ≥15 μM, and resisted bacterial resistance development through 22 passages.
Why it matters
Antibiotic resistance is projected to kill millions annually by 2050. Traditional antibiotic discovery takes years and billions of dollars. Using machine learning to rapidly identify effective antimicrobial peptides — ones that bacteria struggle to develop resistance against — could dramatically accelerate the pipeline of new anti-infective treatments.
How the study worked
Machine learning-based screening of 16,384 peptide sequences using CalcAMP model, followed by in vitro antimicrobial testing against MDR bacteria, biofilm killing assays, hemolysis testing, 3D human skin infection model, and 22-passage resistance development assessment.
What this study cannot tell us
Testing has been limited to in vitro and 3D skin model settings — no animal or human trials yet. The 14-amino-acid peptides may face stability and delivery challenges in vivo. Hemolysis was minimal but needs to be confirmed at therapeutic doses in living systems. Long-term resistance potential beyond 22 passages is unknown.
How to read the evidence
This is a preclinical study combining computational screening with in vitro and 3D tissue model validation. Results are promising but entirely pre-animal and pre-clinical in terms of therapeutic development.
When this study was published
Published in 2025, representing the current frontier of AI-driven antimicrobial peptide discovery.
The bigger picture
Antimicrobial peptides have long been seen as promising but difficult to develop. Machine learning is now making it practical to search enormous peptide sequence spaces quickly. The fact that these AI-discovered peptides avoid triggering resistance after 22 passages is particularly significant — resistance to conventional antibiotics often emerges within a handful of exposures.
Questions still open
- Can these peptides maintain their effectiveness in systemic (not just topical) applications?
- Will the retro-inverso variants show improved stability and half-life in vivo?
- Could the CalcAMP platform be expanded to discover peptides targeting viral or fungal infections?
Common questions
Why can't bacteria resist these peptides?
How is machine learning used to find new antibiotics?
Read the original research
Machine Learning-Identified Potent Antimicrobial Peptides Against Multidrug-Resistant Bacteria and Skin Infections.
Antibiotics (Basel, Switzerland), 14(11)
Citation
Babuççu, Gizem; Vavilthota, Nikitha; Bournez, Colin; de Boer, Leonie; Cordfunke, Robert A; Nibbering, Peter H; van Westen, Gerard J P; Drijfhout, Jan W; Zaat, Sebastian A J; Riool, Martijn. (2025). Machine Learning-Identified Potent Antimicrobial Peptides Against Multidrug-Resistant Bacteria and Skin Infections.. Antibiotics (Basel, Switzerland), 14(11). https://doi.org/10.3390/antibiotics14111172